diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5355 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 3756,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.003993610223642172,
+      "grad_norm": 124.79125135092546,
+      "learning_rate": 9.308510638297872e-08,
+      "loss": 1.1351,
+      "step": 5
+    },
+    {
+      "epoch": 0.007987220447284345,
+      "grad_norm": 53.854891738476,
+      "learning_rate": 1.8617021276595745e-07,
+      "loss": 1.1123,
+      "step": 10
+    },
+    {
+      "epoch": 0.011980830670926517,
+      "grad_norm": 35.55984338795738,
+      "learning_rate": 2.7925531914893617e-07,
+      "loss": 1.1318,
+      "step": 15
+    },
+    {
+      "epoch": 0.01597444089456869,
+      "grad_norm": 53.424023941482716,
+      "learning_rate": 3.723404255319149e-07,
+      "loss": 1.1141,
+      "step": 20
+    },
+    {
+      "epoch": 0.019968051118210862,
+      "grad_norm": 61.65516106085094,
+      "learning_rate": 4.654255319148936e-07,
+      "loss": 1.0358,
+      "step": 25
+    },
+    {
+      "epoch": 0.023961661341853034,
+      "grad_norm": 24.644625180093044,
+      "learning_rate": 5.585106382978723e-07,
+      "loss": 0.9898,
+      "step": 30
+    },
+    {
+      "epoch": 0.027955271565495207,
+      "grad_norm": 35.18949474973098,
+      "learning_rate": 6.515957446808511e-07,
+      "loss": 0.7832,
+      "step": 35
+    },
+    {
+      "epoch": 0.03194888178913738,
+      "grad_norm": 21.730902642961105,
+      "learning_rate": 7.446808510638298e-07,
+      "loss": 0.7333,
+      "step": 40
+    },
+    {
+      "epoch": 0.035942492012779555,
+      "grad_norm": 9.326268068588167,
+      "learning_rate": 8.377659574468085e-07,
+      "loss": 0.5276,
+      "step": 45
+    },
+    {
+      "epoch": 0.039936102236421724,
+      "grad_norm": 4.219331242308076,
+      "learning_rate": 9.308510638297872e-07,
+      "loss": 0.443,
+      "step": 50
+    },
+    {
+      "epoch": 0.0439297124600639,
+      "grad_norm": 3.828106728488943,
+      "learning_rate": 1.023936170212766e-06,
+      "loss": 0.3801,
+      "step": 55
+    },
+    {
+      "epoch": 0.04792332268370607,
+      "grad_norm": 26.95012737394801,
+      "learning_rate": 1.1170212765957447e-06,
+      "loss": 0.3434,
+      "step": 60
+    },
+    {
+      "epoch": 0.051916932907348244,
+      "grad_norm": 3.0269858271064702,
+      "learning_rate": 1.2101063829787234e-06,
+      "loss": 0.3256,
+      "step": 65
+    },
+    {
+      "epoch": 0.05591054313099041,
+      "grad_norm": 2.9924716110173173,
+      "learning_rate": 1.3031914893617021e-06,
+      "loss": 0.3081,
+      "step": 70
+    },
+    {
+      "epoch": 0.05990415335463259,
+      "grad_norm": 3.332606341722849,
+      "learning_rate": 1.3962765957446809e-06,
+      "loss": 0.2981,
+      "step": 75
+    },
+    {
+      "epoch": 0.06389776357827476,
+      "grad_norm": 2.9070407704071846,
+      "learning_rate": 1.4893617021276596e-06,
+      "loss": 0.2841,
+      "step": 80
+    },
+    {
+      "epoch": 0.06789137380191693,
+      "grad_norm": 2.8536909013493745,
+      "learning_rate": 1.5824468085106383e-06,
+      "loss": 0.2818,
+      "step": 85
+    },
+    {
+      "epoch": 0.07188498402555911,
+      "grad_norm": 3.0044229558990367,
+      "learning_rate": 1.675531914893617e-06,
+      "loss": 0.2738,
+      "step": 90
+    },
+    {
+      "epoch": 0.07587859424920128,
+      "grad_norm": 2.8334575842346887,
+      "learning_rate": 1.7686170212765958e-06,
+      "loss": 0.2648,
+      "step": 95
+    },
+    {
+      "epoch": 0.07987220447284345,
+      "grad_norm": 2.634618871564216,
+      "learning_rate": 1.8617021276595745e-06,
+      "loss": 0.2629,
+      "step": 100
+    },
+    {
+      "epoch": 0.08386581469648563,
+      "grad_norm": 2.6590626092313014,
+      "learning_rate": 1.9547872340425528e-06,
+      "loss": 0.2472,
+      "step": 105
+    },
+    {
+      "epoch": 0.0878594249201278,
+      "grad_norm": 2.90486484818493,
+      "learning_rate": 2.047872340425532e-06,
+      "loss": 0.2495,
+      "step": 110
+    },
+    {
+      "epoch": 0.09185303514376997,
+      "grad_norm": 2.8099143216926317,
+      "learning_rate": 2.1409574468085107e-06,
+      "loss": 0.2327,
+      "step": 115
+    },
+    {
+      "epoch": 0.09584664536741214,
+      "grad_norm": 2.8136050336335434,
+      "learning_rate": 2.2340425531914894e-06,
+      "loss": 0.2338,
+      "step": 120
+    },
+    {
+      "epoch": 0.09984025559105432,
+      "grad_norm": 2.4697542437553768,
+      "learning_rate": 2.327127659574468e-06,
+      "loss": 0.2339,
+      "step": 125
+    },
+    {
+      "epoch": 0.10383386581469649,
+      "grad_norm": 2.6351363837368136,
+      "learning_rate": 2.420212765957447e-06,
+      "loss": 0.2262,
+      "step": 130
+    },
+    {
+      "epoch": 0.10782747603833866,
+      "grad_norm": 2.613372563691564,
+      "learning_rate": 2.5132978723404256e-06,
+      "loss": 0.2206,
+      "step": 135
+    },
+    {
+      "epoch": 0.11182108626198083,
+      "grad_norm": 2.597752416642004,
+      "learning_rate": 2.6063829787234043e-06,
+      "loss": 0.2083,
+      "step": 140
+    },
+    {
+      "epoch": 0.11581469648562301,
+      "grad_norm": 2.4178777850084767,
+      "learning_rate": 2.699468085106383e-06,
+      "loss": 0.2039,
+      "step": 145
+    },
+    {
+      "epoch": 0.11980830670926518,
+      "grad_norm": 2.249342405853297,
+      "learning_rate": 2.7925531914893617e-06,
+      "loss": 0.2003,
+      "step": 150
+    },
+    {
+      "epoch": 0.12380191693290735,
+      "grad_norm": 2.050068983302969,
+      "learning_rate": 2.8856382978723405e-06,
+      "loss": 0.2024,
+      "step": 155
+    },
+    {
+      "epoch": 0.12779552715654952,
+      "grad_norm": 1.7186257113193864,
+      "learning_rate": 2.978723404255319e-06,
+      "loss": 0.1958,
+      "step": 160
+    },
+    {
+      "epoch": 0.13178913738019168,
+      "grad_norm": 1.6632638689827575,
+      "learning_rate": 3.071808510638298e-06,
+      "loss": 0.1961,
+      "step": 165
+    },
+    {
+      "epoch": 0.13578274760383385,
+      "grad_norm": 1.4976658741753586,
+      "learning_rate": 3.1648936170212766e-06,
+      "loss": 0.1881,
+      "step": 170
+    },
+    {
+      "epoch": 0.13977635782747605,
+      "grad_norm": 1.5262647756049854,
+      "learning_rate": 3.2579787234042553e-06,
+      "loss": 0.1882,
+      "step": 175
+    },
+    {
+      "epoch": 0.14376996805111822,
+      "grad_norm": 1.5576363273434715,
+      "learning_rate": 3.351063829787234e-06,
+      "loss": 0.1861,
+      "step": 180
+    },
+    {
+      "epoch": 0.1477635782747604,
+      "grad_norm": 1.4077910584213011,
+      "learning_rate": 3.444148936170213e-06,
+      "loss": 0.1888,
+      "step": 185
+    },
+    {
+      "epoch": 0.15175718849840256,
+      "grad_norm": 1.44303025282263,
+      "learning_rate": 3.5372340425531915e-06,
+      "loss": 0.1887,
+      "step": 190
+    },
+    {
+      "epoch": 0.15575079872204473,
+      "grad_norm": 1.3605098039485544,
+      "learning_rate": 3.63031914893617e-06,
+      "loss": 0.1809,
+      "step": 195
+    },
+    {
+      "epoch": 0.1597444089456869,
+      "grad_norm": 1.2997453068507787,
+      "learning_rate": 3.723404255319149e-06,
+      "loss": 0.1877,
+      "step": 200
+    },
+    {
+      "epoch": 0.16373801916932906,
+      "grad_norm": 1.2577351321884693,
+      "learning_rate": 3.816489361702128e-06,
+      "loss": 0.18,
+      "step": 205
+    },
+    {
+      "epoch": 0.16773162939297126,
+      "grad_norm": 1.2562249686118996,
+      "learning_rate": 3.9095744680851056e-06,
+      "loss": 0.1846,
+      "step": 210
+    },
+    {
+      "epoch": 0.17172523961661343,
+      "grad_norm": 1.3018013547612421,
+      "learning_rate": 4.002659574468085e-06,
+      "loss": 0.1853,
+      "step": 215
+    },
+    {
+      "epoch": 0.1757188498402556,
+      "grad_norm": 1.2860696034925374,
+      "learning_rate": 4.095744680851064e-06,
+      "loss": 0.183,
+      "step": 220
+    },
+    {
+      "epoch": 0.17971246006389777,
+      "grad_norm": 1.3940860735108558,
+      "learning_rate": 4.188829787234043e-06,
+      "loss": 0.1749,
+      "step": 225
+    },
+    {
+      "epoch": 0.18370607028753994,
+      "grad_norm": 1.266616659720171,
+      "learning_rate": 4.281914893617021e-06,
+      "loss": 0.1812,
+      "step": 230
+    },
+    {
+      "epoch": 0.1876996805111821,
+      "grad_norm": 1.1679912165565878,
+      "learning_rate": 4.375e-06,
+      "loss": 0.1764,
+      "step": 235
+    },
+    {
+      "epoch": 0.19169329073482427,
+      "grad_norm": 1.2063141158373527,
+      "learning_rate": 4.468085106382979e-06,
+      "loss": 0.1725,
+      "step": 240
+    },
+    {
+      "epoch": 0.19568690095846644,
+      "grad_norm": 1.3549203536924732,
+      "learning_rate": 4.561170212765957e-06,
+      "loss": 0.1811,
+      "step": 245
+    },
+    {
+      "epoch": 0.19968051118210864,
+      "grad_norm": 1.3023445750500837,
+      "learning_rate": 4.654255319148936e-06,
+      "loss": 0.1732,
+      "step": 250
+    },
+    {
+      "epoch": 0.2036741214057508,
+      "grad_norm": 1.1897447106164616,
+      "learning_rate": 4.747340425531915e-06,
+      "loss": 0.1768,
+      "step": 255
+    },
+    {
+      "epoch": 0.20766773162939298,
+      "grad_norm": 1.3217343761947535,
+      "learning_rate": 4.840425531914894e-06,
+      "loss": 0.1876,
+      "step": 260
+    },
+    {
+      "epoch": 0.21166134185303515,
+      "grad_norm": 1.1882978703516491,
+      "learning_rate": 4.933510638297872e-06,
+      "loss": 0.1849,
+      "step": 265
+    },
+    {
+      "epoch": 0.21565495207667731,
+      "grad_norm": 1.2988676577170748,
+      "learning_rate": 5.026595744680851e-06,
+      "loss": 0.1742,
+      "step": 270
+    },
+    {
+      "epoch": 0.21964856230031948,
+      "grad_norm": 1.3576611779609293,
+      "learning_rate": 5.11968085106383e-06,
+      "loss": 0.1915,
+      "step": 275
+    },
+    {
+      "epoch": 0.22364217252396165,
+      "grad_norm": 1.324286801402471,
+      "learning_rate": 5.2127659574468086e-06,
+      "loss": 0.1849,
+      "step": 280
+    },
+    {
+      "epoch": 0.22763578274760382,
+      "grad_norm": 1.1601790573503918,
+      "learning_rate": 5.305851063829787e-06,
+      "loss": 0.176,
+      "step": 285
+    },
+    {
+      "epoch": 0.23162939297124602,
+      "grad_norm": 1.2319656878269474,
+      "learning_rate": 5.398936170212766e-06,
+      "loss": 0.179,
+      "step": 290
+    },
+    {
+      "epoch": 0.2356230031948882,
+      "grad_norm": 1.1947505074640163,
+      "learning_rate": 5.492021276595744e-06,
+      "loss": 0.1805,
+      "step": 295
+    },
+    {
+      "epoch": 0.23961661341853036,
+      "grad_norm": 1.1865842816293095,
+      "learning_rate": 5.5851063829787235e-06,
+      "loss": 0.1798,
+      "step": 300
+    },
+    {
+      "epoch": 0.24361022364217252,
+      "grad_norm": 1.2294425800486448,
+      "learning_rate": 5.678191489361702e-06,
+      "loss": 0.176,
+      "step": 305
+    },
+    {
+      "epoch": 0.2476038338658147,
+      "grad_norm": 1.1750231455196776,
+      "learning_rate": 5.771276595744681e-06,
+      "loss": 0.1798,
+      "step": 310
+    },
+    {
+      "epoch": 0.2515974440894569,
+      "grad_norm": 1.2451767206261621,
+      "learning_rate": 5.864361702127659e-06,
+      "loss": 0.1791,
+      "step": 315
+    },
+    {
+      "epoch": 0.25559105431309903,
+      "grad_norm": 1.2675583808344273,
+      "learning_rate": 5.957446808510638e-06,
+      "loss": 0.1798,
+      "step": 320
+    },
+    {
+      "epoch": 0.2595846645367412,
+      "grad_norm": 1.179067660563583,
+      "learning_rate": 6.0505319148936175e-06,
+      "loss": 0.1709,
+      "step": 325
+    },
+    {
+      "epoch": 0.26357827476038337,
+      "grad_norm": 1.2526557533302813,
+      "learning_rate": 6.143617021276596e-06,
+      "loss": 0.1751,
+      "step": 330
+    },
+    {
+      "epoch": 0.26757188498402557,
+      "grad_norm": 1.1681286446845103,
+      "learning_rate": 6.236702127659574e-06,
+      "loss": 0.1656,
+      "step": 335
+    },
+    {
+      "epoch": 0.2715654952076677,
+      "grad_norm": 1.1728046626126167,
+      "learning_rate": 6.329787234042553e-06,
+      "loss": 0.1798,
+      "step": 340
+    },
+    {
+      "epoch": 0.2755591054313099,
+      "grad_norm": 1.1964596076655767,
+      "learning_rate": 6.4228723404255316e-06,
+      "loss": 0.1775,
+      "step": 345
+    },
+    {
+      "epoch": 0.2795527156549521,
+      "grad_norm": 1.2208354176011962,
+      "learning_rate": 6.515957446808511e-06,
+      "loss": 0.1779,
+      "step": 350
+    },
+    {
+      "epoch": 0.28354632587859424,
+      "grad_norm": 1.196375387187927,
+      "learning_rate": 6.609042553191489e-06,
+      "loss": 0.1779,
+      "step": 355
+    },
+    {
+      "epoch": 0.28753993610223644,
+      "grad_norm": 1.299599153387323,
+      "learning_rate": 6.702127659574468e-06,
+      "loss": 0.1793,
+      "step": 360
+    },
+    {
+      "epoch": 0.2915335463258786,
+      "grad_norm": 1.2361399705105702,
+      "learning_rate": 6.7952127659574464e-06,
+      "loss": 0.1772,
+      "step": 365
+    },
+    {
+      "epoch": 0.2955271565495208,
+      "grad_norm": 1.1486851112889924,
+      "learning_rate": 6.888297872340426e-06,
+      "loss": 0.171,
+      "step": 370
+    },
+    {
+      "epoch": 0.2995207667731629,
+      "grad_norm": 1.2309787273113733,
+      "learning_rate": 6.981382978723405e-06,
+      "loss": 0.169,
+      "step": 375
+    },
+    {
+      "epoch": 0.3035143769968051,
+      "grad_norm": 1.1069081217064252,
+      "learning_rate": 6.999975810667964e-06,
+      "loss": 0.174,
+      "step": 380
+    },
+    {
+      "epoch": 0.3075079872204473,
+      "grad_norm": 1.0934902660778836,
+      "learning_rate": 6.999877542079611e-06,
+      "loss": 0.1749,
+      "step": 385
+    },
+    {
+      "epoch": 0.31150159744408945,
+      "grad_norm": 1.1647975706264049,
+      "learning_rate": 6.99970368452242e-06,
+      "loss": 0.1682,
+      "step": 390
+    },
+    {
+      "epoch": 0.31549520766773165,
+      "grad_norm": 1.150865125937996,
+      "learning_rate": 6.9994542417513e-06,
+      "loss": 0.1797,
+      "step": 395
+    },
+    {
+      "epoch": 0.3194888178913738,
+      "grad_norm": 1.0698977727857675,
+      "learning_rate": 6.999129219153615e-06,
+      "loss": 0.1729,
+      "step": 400
+    },
+    {
+      "epoch": 0.323482428115016,
+      "grad_norm": 1.1500528795772351,
+      "learning_rate": 6.998728623749078e-06,
+      "loss": 0.1824,
+      "step": 405
+    },
+    {
+      "epoch": 0.3274760383386581,
+      "grad_norm": 1.1428662750727947,
+      "learning_rate": 6.9982524641895845e-06,
+      "loss": 0.1705,
+      "step": 410
+    },
+    {
+      "epoch": 0.3314696485623003,
+      "grad_norm": 1.1575666664852504,
+      "learning_rate": 6.997700750759044e-06,
+      "loss": 0.1769,
+      "step": 415
+    },
+    {
+      "epoch": 0.3354632587859425,
+      "grad_norm": 1.115412562169602,
+      "learning_rate": 6.997073495373145e-06,
+      "loss": 0.1734,
+      "step": 420
+    },
+    {
+      "epoch": 0.33945686900958466,
+      "grad_norm": 1.151869893863153,
+      "learning_rate": 6.996370711579098e-06,
+      "loss": 0.1789,
+      "step": 425
+    },
+    {
+      "epoch": 0.34345047923322686,
+      "grad_norm": 1.1454117113132618,
+      "learning_rate": 6.99559241455535e-06,
+      "loss": 0.166,
+      "step": 430
+    },
+    {
+      "epoch": 0.347444089456869,
+      "grad_norm": 1.0873453028650606,
+      "learning_rate": 6.994738621111253e-06,
+      "loss": 0.1654,
+      "step": 435
+    },
+    {
+      "epoch": 0.3514376996805112,
+      "grad_norm": 0.99906893867618,
+      "learning_rate": 6.993809349686699e-06,
+      "loss": 0.1751,
+      "step": 440
+    },
+    {
+      "epoch": 0.35543130990415334,
+      "grad_norm": 1.157573643863577,
+      "learning_rate": 6.992804620351724e-06,
+      "loss": 0.1728,
+      "step": 445
+    },
+    {
+      "epoch": 0.35942492012779553,
+      "grad_norm": 1.0531396174518157,
+      "learning_rate": 6.991724454806074e-06,
+      "loss": 0.1765,
+      "step": 450
+    },
+    {
+      "epoch": 0.3634185303514377,
+      "grad_norm": 1.1386062653595457,
+      "learning_rate": 6.990568876378738e-06,
+      "loss": 0.1761,
+      "step": 455
+    },
+    {
+      "epoch": 0.36741214057507987,
+      "grad_norm": 1.2777140078161082,
+      "learning_rate": 6.989337910027439e-06,
+      "loss": 0.1772,
+      "step": 460
+    },
+    {
+      "epoch": 0.37140575079872207,
+      "grad_norm": 1.1495981738430336,
+      "learning_rate": 6.988031582338101e-06,
+      "loss": 0.1685,
+      "step": 465
+    },
+    {
+      "epoch": 0.3753993610223642,
+      "grad_norm": 1.6362808683762557,
+      "learning_rate": 6.986649921524274e-06,
+      "loss": 0.1717,
+      "step": 470
+    },
+    {
+      "epoch": 0.3793929712460064,
+      "grad_norm": 0.9691070079618013,
+      "learning_rate": 6.9851929574265194e-06,
+      "loss": 0.1682,
+      "step": 475
+    },
+    {
+      "epoch": 0.38338658146964855,
+      "grad_norm": 1.024317390117457,
+      "learning_rate": 6.98366072151177e-06,
+      "loss": 0.167,
+      "step": 480
+    },
+    {
+      "epoch": 0.38738019169329074,
+      "grad_norm": 1.2702847001741562,
+      "learning_rate": 6.982053246872652e-06,
+      "loss": 0.1703,
+      "step": 485
+    },
+    {
+      "epoch": 0.3913738019169329,
+      "grad_norm": 1.0391707960731285,
+      "learning_rate": 6.9803705682267635e-06,
+      "loss": 0.1637,
+      "step": 490
+    },
+    {
+      "epoch": 0.3953674121405751,
+      "grad_norm": 1.0070748510170011,
+      "learning_rate": 6.978612721915935e-06,
+      "loss": 0.1677,
+      "step": 495
+    },
+    {
+      "epoch": 0.3993610223642173,
+      "grad_norm": 1.0402855278956755,
+      "learning_rate": 6.976779745905432e-06,
+      "loss": 0.1654,
+      "step": 500
+    },
+    {
+      "epoch": 0.3993610223642173,
+      "eval_loss": 0.15384411811828613,
+      "eval_runtime": 389.0289,
+      "eval_samples_per_second": 45.773,
+      "eval_steps_per_second": 5.722,
+      "step": 500
+    },
+    {
+      "epoch": 0.4033546325878594,
+      "grad_norm": 1.1163339009849595,
+      "learning_rate": 6.974871679783144e-06,
+      "loss": 0.17,
+      "step": 505
+    },
+    {
+      "epoch": 0.4073482428115016,
+      "grad_norm": 0.9859275646648643,
+      "learning_rate": 6.972888564758729e-06,
+      "loss": 0.1657,
+      "step": 510
+    },
+    {
+      "epoch": 0.41134185303514376,
+      "grad_norm": 0.9394276032165236,
+      "learning_rate": 6.970830443662719e-06,
+      "loss": 0.1685,
+      "step": 515
+    },
+    {
+      "epoch": 0.41533546325878595,
+      "grad_norm": 1.073675948253867,
+      "learning_rate": 6.968697360945598e-06,
+      "loss": 0.1759,
+      "step": 520
+    },
+    {
+      "epoch": 0.4193290734824281,
+      "grad_norm": 1.036048195729432,
+      "learning_rate": 6.966489362676843e-06,
+      "loss": 0.1631,
+      "step": 525
+    },
+    {
+      "epoch": 0.4233226837060703,
+      "grad_norm": 1.034694676356374,
+      "learning_rate": 6.964206496543924e-06,
+      "loss": 0.1693,
+      "step": 530
+    },
+    {
+      "epoch": 0.4273162939297125,
+      "grad_norm": 1.0185215828025804,
+      "learning_rate": 6.96184881185128e-06,
+      "loss": 0.1653,
+      "step": 535
+    },
+    {
+      "epoch": 0.43130990415335463,
+      "grad_norm": 0.9931981204110877,
+      "learning_rate": 6.959416359519253e-06,
+      "loss": 0.168,
+      "step": 540
+    },
+    {
+      "epoch": 0.4353035143769968,
+      "grad_norm": 1.0616121384749488,
+      "learning_rate": 6.956909192082982e-06,
+      "loss": 0.1709,
+      "step": 545
+    },
+    {
+      "epoch": 0.43929712460063897,
+      "grad_norm": 1.0045397760966588,
+      "learning_rate": 6.954327363691278e-06,
+      "loss": 0.1721,
+      "step": 550
+    },
+    {
+      "epoch": 0.44329073482428116,
+      "grad_norm": 1.0341210608686486,
+      "learning_rate": 6.951670930105448e-06,
+      "loss": 0.1637,
+      "step": 555
+    },
+    {
+      "epoch": 0.4472843450479233,
+      "grad_norm": 0.943453900900678,
+      "learning_rate": 6.9489399486980925e-06,
+      "loss": 0.1672,
+      "step": 560
+    },
+    {
+      "epoch": 0.4512779552715655,
+      "grad_norm": 1.0198128200679288,
+      "learning_rate": 6.946134478451864e-06,
+      "loss": 0.1561,
+      "step": 565
+    },
+    {
+      "epoch": 0.45527156549520764,
+      "grad_norm": 0.981934871464691,
+      "learning_rate": 6.943254579958201e-06,
+      "loss": 0.1696,
+      "step": 570
+    },
+    {
+      "epoch": 0.45926517571884984,
+      "grad_norm": 0.9799305322175899,
+      "learning_rate": 6.940300315416007e-06,
+      "loss": 0.1716,
+      "step": 575
+    },
+    {
+      "epoch": 0.46325878594249204,
+      "grad_norm": 1.0222871844508339,
+      "learning_rate": 6.93727174863032e-06,
+      "loss": 0.1732,
+      "step": 580
+    },
+    {
+      "epoch": 0.4672523961661342,
+      "grad_norm": 1.0017118270280685,
+      "learning_rate": 6.934168945010925e-06,
+      "loss": 0.1697,
+      "step": 585
+    },
+    {
+      "epoch": 0.4712460063897764,
+      "grad_norm": 1.0155218802083956,
+      "learning_rate": 6.930991971570945e-06,
+      "loss": 0.1612,
+      "step": 590
+    },
+    {
+      "epoch": 0.4752396166134185,
+      "grad_norm": 0.968986765149799,
+      "learning_rate": 6.9277408969253935e-06,
+      "loss": 0.1624,
+      "step": 595
+    },
+    {
+      "epoch": 0.4792332268370607,
+      "grad_norm": 0.9049030479920512,
+      "learning_rate": 6.924415791289696e-06,
+      "loss": 0.1666,
+      "step": 600
+    },
+    {
+      "epoch": 0.48322683706070285,
+      "grad_norm": 1.0516179412211666,
+      "learning_rate": 6.921016726478164e-06,
+      "loss": 0.1711,
+      "step": 605
+    },
+    {
+      "epoch": 0.48722044728434505,
+      "grad_norm": 0.9668990395103417,
+      "learning_rate": 6.917543775902452e-06,
+      "loss": 0.1726,
+      "step": 610
+    },
+    {
+      "epoch": 0.49121405750798725,
+      "grad_norm": 1.106970562985427,
+      "learning_rate": 6.913997014569974e-06,
+      "loss": 0.1659,
+      "step": 615
+    },
+    {
+      "epoch": 0.4952076677316294,
+      "grad_norm": 0.995421768407361,
+      "learning_rate": 6.910376519082274e-06,
+      "loss": 0.1588,
+      "step": 620
+    },
+    {
+      "epoch": 0.4992012779552716,
+      "grad_norm": 1.0063189979179088,
+      "learning_rate": 6.906682367633382e-06,
+      "loss": 0.1733,
+      "step": 625
+    },
+    {
+      "epoch": 0.5031948881789138,
+      "grad_norm": 1.0076370540912887,
+      "learning_rate": 6.902914640008114e-06,
+      "loss": 0.1583,
+      "step": 630
+    },
+    {
+      "epoch": 0.5071884984025559,
+      "grad_norm": 0.9196883887602973,
+      "learning_rate": 6.899073417580362e-06,
+      "loss": 0.167,
+      "step": 635
+    },
+    {
+      "epoch": 0.5111821086261981,
+      "grad_norm": 0.9516430809786864,
+      "learning_rate": 6.895158783311325e-06,
+      "loss": 0.1727,
+      "step": 640
+    },
+    {
+      "epoch": 0.5151757188498403,
+      "grad_norm": 1.0089849026997129,
+      "learning_rate": 6.891170821747724e-06,
+      "loss": 0.1656,
+      "step": 645
+    },
+    {
+      "epoch": 0.5191693290734825,
+      "grad_norm": 0.9583023331989294,
+      "learning_rate": 6.887109619019972e-06,
+      "loss": 0.1665,
+      "step": 650
+    },
+    {
+      "epoch": 0.5231629392971247,
+      "grad_norm": 1.00161133810888,
+      "learning_rate": 6.882975262840319e-06,
+      "loss": 0.1644,
+      "step": 655
+    },
+    {
+      "epoch": 0.5271565495207667,
+      "grad_norm": 1.051049075871936,
+      "learning_rate": 6.87876784250095e-06,
+      "loss": 0.1668,
+      "step": 660
+    },
+    {
+      "epoch": 0.5311501597444089,
+      "grad_norm": 0.8907899001628331,
+      "learning_rate": 6.874487448872064e-06,
+      "loss": 0.1642,
+      "step": 665
+    },
+    {
+      "epoch": 0.5351437699680511,
+      "grad_norm": 0.9259445853753279,
+      "learning_rate": 6.870134174399907e-06,
+      "loss": 0.1662,
+      "step": 670
+    },
+    {
+      "epoch": 0.5391373801916933,
+      "grad_norm": 0.8930200609325035,
+      "learning_rate": 6.8657081131047766e-06,
+      "loss": 0.1646,
+      "step": 675
+    },
+    {
+      "epoch": 0.5431309904153354,
+      "grad_norm": 0.9561210976739172,
+      "learning_rate": 6.861209360578991e-06,
+      "loss": 0.1627,
+      "step": 680
+    },
+    {
+      "epoch": 0.5471246006389776,
+      "grad_norm": 0.9117689744304602,
+      "learning_rate": 6.856638013984827e-06,
+      "loss": 0.1623,
+      "step": 685
+    },
+    {
+      "epoch": 0.5511182108626198,
+      "grad_norm": 0.9687731076699732,
+      "learning_rate": 6.8519941720524155e-06,
+      "loss": 0.1567,
+      "step": 690
+    },
+    {
+      "epoch": 0.555111821086262,
+      "grad_norm": 0.9600827029154959,
+      "learning_rate": 6.847277935077615e-06,
+      "loss": 0.1608,
+      "step": 695
+    },
+    {
+      "epoch": 0.5591054313099042,
+      "grad_norm": 0.9096771947825363,
+      "learning_rate": 6.842489404919846e-06,
+      "loss": 0.1523,
+      "step": 700
+    },
+    {
+      "epoch": 0.5630990415335463,
+      "grad_norm": 0.9443224796506533,
+      "learning_rate": 6.837628684999885e-06,
+      "loss": 0.1566,
+      "step": 705
+    },
+    {
+      "epoch": 0.5670926517571885,
+      "grad_norm": 0.979707108806894,
+      "learning_rate": 6.832695880297634e-06,
+      "loss": 0.166,
+      "step": 710
+    },
+    {
+      "epoch": 0.5710862619808307,
+      "grad_norm": 1.0043374037070385,
+      "learning_rate": 6.827691097349857e-06,
+      "loss": 0.1627,
+      "step": 715
+    },
+    {
+      "epoch": 0.5750798722044729,
+      "grad_norm": 0.9464996778618299,
+      "learning_rate": 6.8226144442478715e-06,
+      "loss": 0.164,
+      "step": 720
+    },
+    {
+      "epoch": 0.579073482428115,
+      "grad_norm": 0.9416490321614909,
+      "learning_rate": 6.817466030635222e-06,
+      "loss": 0.1618,
+      "step": 725
+    },
+    {
+      "epoch": 0.5830670926517572,
+      "grad_norm": 0.8925338429207857,
+      "learning_rate": 6.812245967705307e-06,
+      "loss": 0.1619,
+      "step": 730
+    },
+    {
+      "epoch": 0.5870607028753994,
+      "grad_norm": 0.9194904859850123,
+      "learning_rate": 6.8069543681989755e-06,
+      "loss": 0.1553,
+      "step": 735
+    },
+    {
+      "epoch": 0.5910543130990416,
+      "grad_norm": 0.895691163374912,
+      "learning_rate": 6.8015913464021e-06,
+      "loss": 0.1606,
+      "step": 740
+    },
+    {
+      "epoch": 0.5950479233226837,
+      "grad_norm": 0.8442230894893846,
+      "learning_rate": 6.796157018143101e-06,
+      "loss": 0.1541,
+      "step": 745
+    },
+    {
+      "epoch": 0.5990415335463258,
+      "grad_norm": 0.911157960300542,
+      "learning_rate": 6.790651500790446e-06,
+      "loss": 0.158,
+      "step": 750
+    },
+    {
+      "epoch": 0.603035143769968,
+      "grad_norm": 0.9270097775375741,
+      "learning_rate": 6.785074913250119e-06,
+      "loss": 0.1578,
+      "step": 755
+    },
+    {
+      "epoch": 0.6070287539936102,
+      "grad_norm": 0.985999202515655,
+      "learning_rate": 6.779427375963048e-06,
+      "loss": 0.1605,
+      "step": 760
+    },
+    {
+      "epoch": 0.6110223642172524,
+      "grad_norm": 0.9816513779676883,
+      "learning_rate": 6.773709010902506e-06,
+      "loss": 0.1579,
+      "step": 765
+    },
+    {
+      "epoch": 0.6150159744408946,
+      "grad_norm": 0.9851452504449081,
+      "learning_rate": 6.767919941571479e-06,
+      "loss": 0.1643,
+      "step": 770
+    },
+    {
+      "epoch": 0.6190095846645367,
+      "grad_norm": 0.932019250350353,
+      "learning_rate": 6.762060292999991e-06,
+      "loss": 0.1641,
+      "step": 775
+    },
+    {
+      "epoch": 0.6230031948881789,
+      "grad_norm": 1.0112706975782446,
+      "learning_rate": 6.756130191742413e-06,
+      "loss": 0.1669,
+      "step": 780
+    },
+    {
+      "epoch": 0.6269968051118211,
+      "grad_norm": 1.0652365454569854,
+      "learning_rate": 6.75012976587472e-06,
+      "loss": 0.1553,
+      "step": 785
+    },
+    {
+      "epoch": 0.6309904153354633,
+      "grad_norm": 0.9293147453915607,
+      "learning_rate": 6.7440591449917345e-06,
+      "loss": 0.1629,
+      "step": 790
+    },
+    {
+      "epoch": 0.6349840255591054,
+      "grad_norm": 0.9124885849493852,
+      "learning_rate": 6.737918460204323e-06,
+      "loss": 0.1615,
+      "step": 795
+    },
+    {
+      "epoch": 0.6389776357827476,
+      "grad_norm": 0.9576719497775147,
+      "learning_rate": 6.731707844136561e-06,
+      "loss": 0.166,
+      "step": 800
+    },
+    {
+      "epoch": 0.6429712460063898,
+      "grad_norm": 0.8872595723908449,
+      "learning_rate": 6.725427430922875e-06,
+      "loss": 0.1542,
+      "step": 805
+    },
+    {
+      "epoch": 0.646964856230032,
+      "grad_norm": 0.9935128432068615,
+      "learning_rate": 6.719077356205143e-06,
+      "loss": 0.1574,
+      "step": 810
+    },
+    {
+      "epoch": 0.6509584664536742,
+      "grad_norm": 0.931891327221199,
+      "learning_rate": 6.712657757129762e-06,
+      "loss": 0.1646,
+      "step": 815
+    },
+    {
+      "epoch": 0.6549520766773163,
+      "grad_norm": 0.8727423850294462,
+      "learning_rate": 6.7061687723446925e-06,
+      "loss": 0.1588,
+      "step": 820
+    },
+    {
+      "epoch": 0.6589456869009584,
+      "grad_norm": 0.9689019750184885,
+      "learning_rate": 6.699610541996453e-06,
+      "loss": 0.1544,
+      "step": 825
+    },
+    {
+      "epoch": 0.6629392971246006,
+      "grad_norm": 0.9135093758857515,
+      "learning_rate": 6.69298320772711e-06,
+      "loss": 0.1633,
+      "step": 830
+    },
+    {
+      "epoch": 0.6669329073482428,
+      "grad_norm": 0.8841844046649162,
+      "learning_rate": 6.6862869126712006e-06,
+      "loss": 0.1517,
+      "step": 835
+    },
+    {
+      "epoch": 0.670926517571885,
+      "grad_norm": 0.9197410885367548,
+      "learning_rate": 6.6795218014526535e-06,
+      "loss": 0.1582,
+      "step": 840
+    },
+    {
+      "epoch": 0.6749201277955271,
+      "grad_norm": 0.9071224596757496,
+      "learning_rate": 6.672688020181662e-06,
+      "loss": 0.1568,
+      "step": 845
+    },
+    {
+      "epoch": 0.6789137380191693,
+      "grad_norm": 0.8384791088112575,
+      "learning_rate": 6.665785716451527e-06,
+      "loss": 0.161,
+      "step": 850
+    },
+    {
+      "epoch": 0.6829073482428115,
+      "grad_norm": 0.9697839295867846,
+      "learning_rate": 6.658815039335469e-06,
+      "loss": 0.1663,
+      "step": 855
+    },
+    {
+      "epoch": 0.6869009584664537,
+      "grad_norm": 0.93519835830344,
+      "learning_rate": 6.651776139383413e-06,
+      "loss": 0.1635,
+      "step": 860
+    },
+    {
+      "epoch": 0.6908945686900958,
+      "grad_norm": 0.8980357502380838,
+      "learning_rate": 6.644669168618731e-06,
+      "loss": 0.1614,
+      "step": 865
+    },
+    {
+      "epoch": 0.694888178913738,
+      "grad_norm": 0.9335437765568465,
+      "learning_rate": 6.637494280534963e-06,
+      "loss": 0.1629,
+      "step": 870
+    },
+    {
+      "epoch": 0.6988817891373802,
+      "grad_norm": 0.8803593207070962,
+      "learning_rate": 6.630251630092499e-06,
+      "loss": 0.1477,
+      "step": 875
+    },
+    {
+      "epoch": 0.7028753993610224,
+      "grad_norm": 0.904997999578463,
+      "learning_rate": 6.622941373715235e-06,
+      "loss": 0.1575,
+      "step": 880
+    },
+    {
+      "epoch": 0.7068690095846646,
+      "grad_norm": 0.9235638802056787,
+      "learning_rate": 6.615563669287194e-06,
+      "loss": 0.1584,
+      "step": 885
+    },
+    {
+      "epoch": 0.7108626198083067,
+      "grad_norm": 0.9076843251345146,
+      "learning_rate": 6.6081186761491116e-06,
+      "loss": 0.1557,
+      "step": 890
+    },
+    {
+      "epoch": 0.7148562300319489,
+      "grad_norm": 0.9740399647119267,
+      "learning_rate": 6.600606555095002e-06,
+      "loss": 0.1604,
+      "step": 895
+    },
+    {
+      "epoch": 0.7188498402555911,
+      "grad_norm": 0.928718969148393,
+      "learning_rate": 6.593027468368679e-06,
+      "loss": 0.1571,
+      "step": 900
+    },
+    {
+      "epoch": 0.7228434504792333,
+      "grad_norm": 0.9098156003642778,
+      "learning_rate": 6.585381579660256e-06,
+      "loss": 0.1572,
+      "step": 905
+    },
+    {
+      "epoch": 0.7268370607028753,
+      "grad_norm": 0.9164438133952856,
+      "learning_rate": 6.577669054102609e-06,
+      "loss": 0.1549,
+      "step": 910
+    },
+    {
+      "epoch": 0.7308306709265175,
+      "grad_norm": 0.8693093471558418,
+      "learning_rate": 6.56989005826781e-06,
+      "loss": 0.1644,
+      "step": 915
+    },
+    {
+      "epoch": 0.7348242811501597,
+      "grad_norm": 0.8225661786689502,
+      "learning_rate": 6.5620447601635305e-06,
+      "loss": 0.1598,
+      "step": 920
+    },
+    {
+      "epoch": 0.7388178913738019,
+      "grad_norm": 0.9061170444615521,
+      "learning_rate": 6.55413332922941e-06,
+      "loss": 0.1507,
+      "step": 925
+    },
+    {
+      "epoch": 0.7428115015974441,
+      "grad_norm": 1.2155491539418426,
+      "learning_rate": 6.546155936333401e-06,
+      "loss": 0.1531,
+      "step": 930
+    },
+    {
+      "epoch": 0.7468051118210862,
+      "grad_norm": 0.8958306454880214,
+      "learning_rate": 6.538112753768076e-06,
+      "loss": 0.1543,
+      "step": 935
+    },
+    {
+      "epoch": 0.7507987220447284,
+      "grad_norm": 0.9538107994557221,
+      "learning_rate": 6.530003955246904e-06,
+      "loss": 0.1552,
+      "step": 940
+    },
+    {
+      "epoch": 0.7547923322683706,
+      "grad_norm": 0.9543004936787222,
+      "learning_rate": 6.521829715900506e-06,
+      "loss": 0.1579,
+      "step": 945
+    },
+    {
+      "epoch": 0.7587859424920128,
+      "grad_norm": 0.9325838144594478,
+      "learning_rate": 6.5135902122728635e-06,
+      "loss": 0.1556,
+      "step": 950
+    },
+    {
+      "epoch": 0.762779552715655,
+      "grad_norm": 0.9186355171261438,
+      "learning_rate": 6.505285622317513e-06,
+      "loss": 0.1589,
+      "step": 955
+    },
+    {
+      "epoch": 0.7667731629392971,
+      "grad_norm": 0.9099495360242699,
+      "learning_rate": 6.496916125393698e-06,
+      "loss": 0.1587,
+      "step": 960
+    },
+    {
+      "epoch": 0.7707667731629393,
+      "grad_norm": 0.8392658119656629,
+      "learning_rate": 6.4884819022625e-06,
+      "loss": 0.1619,
+      "step": 965
+    },
+    {
+      "epoch": 0.7747603833865815,
+      "grad_norm": 0.8457903416638893,
+      "learning_rate": 6.479983135082927e-06,
+      "loss": 0.1516,
+      "step": 970
+    },
+    {
+      "epoch": 0.7787539936102237,
+      "grad_norm": 0.9121023222975818,
+      "learning_rate": 6.4714200074079885e-06,
+      "loss": 0.157,
+      "step": 975
+    },
+    {
+      "epoch": 0.7827476038338658,
+      "grad_norm": 0.8553962057764459,
+      "learning_rate": 6.462792704180721e-06,
+      "loss": 0.1517,
+      "step": 980
+    },
+    {
+      "epoch": 0.786741214057508,
+      "grad_norm": 0.9373339203123983,
+      "learning_rate": 6.454101411730205e-06,
+      "loss": 0.1492,
+      "step": 985
+    },
+    {
+      "epoch": 0.7907348242811502,
+      "grad_norm": 0.8309793749606978,
+      "learning_rate": 6.445346317767532e-06,
+      "loss": 0.1475,
+      "step": 990
+    },
+    {
+      "epoch": 0.7947284345047924,
+      "grad_norm": 0.8968174110255692,
+      "learning_rate": 6.436527611381752e-06,
+      "loss": 0.1554,
+      "step": 995
+    },
+    {
+      "epoch": 0.7987220447284346,
+      "grad_norm": 0.8719592876952176,
+      "learning_rate": 6.427645483035793e-06,
+      "loss": 0.1531,
+      "step": 1000
+    },
+    {
+      "epoch": 0.7987220447284346,
+      "eval_loss": 0.13947652280330658,
+      "eval_runtime": 388.4879,
+      "eval_samples_per_second": 45.837,
+      "eval_steps_per_second": 5.73,
+      "step": 1000
+    },
+    {
+      "epoch": 0.8027156549520766,
+      "grad_norm": 0.8945078745681652,
+      "learning_rate": 6.418700124562346e-06,
+      "loss": 0.1511,
+      "step": 1005
+    },
+    {
+      "epoch": 0.8067092651757188,
+      "grad_norm": 0.9037985750009097,
+      "learning_rate": 6.409691729159718e-06,
+      "loss": 0.158,
+      "step": 1010
+    },
+    {
+      "epoch": 0.810702875399361,
+      "grad_norm": 0.9841558619803956,
+      "learning_rate": 6.400620491387666e-06,
+      "loss": 0.1572,
+      "step": 1015
+    },
+    {
+      "epoch": 0.8146964856230032,
+      "grad_norm": 0.8856887705179193,
+      "learning_rate": 6.39148660716319e-06,
+      "loss": 0.1564,
+      "step": 1020
+    },
+    {
+      "epoch": 0.8186900958466453,
+      "grad_norm": 0.8458094119326396,
+      "learning_rate": 6.382290273756303e-06,
+      "loss": 0.1564,
+      "step": 1025
+    },
+    {
+      "epoch": 0.8226837060702875,
+      "grad_norm": 0.8951775862164378,
+      "learning_rate": 6.373031689785771e-06,
+      "loss": 0.1567,
+      "step": 1030
+    },
+    {
+      "epoch": 0.8266773162939297,
+      "grad_norm": 0.8814016163914961,
+      "learning_rate": 6.363711055214824e-06,
+      "loss": 0.1559,
+      "step": 1035
+    },
+    {
+      "epoch": 0.8306709265175719,
+      "grad_norm": 0.8434088161690633,
+      "learning_rate": 6.354328571346836e-06,
+      "loss": 0.154,
+      "step": 1040
+    },
+    {
+      "epoch": 0.8346645367412141,
+      "grad_norm": 0.8465270748003091,
+      "learning_rate": 6.344884440820976e-06,
+      "loss": 0.1621,
+      "step": 1045
+    },
+    {
+      "epoch": 0.8386581469648562,
+      "grad_norm": 0.8575168379781873,
+      "learning_rate": 6.335378867607834e-06,
+      "loss": 0.1494,
+      "step": 1050
+    },
+    {
+      "epoch": 0.8426517571884984,
+      "grad_norm": 0.9030974173812433,
+      "learning_rate": 6.325812057005017e-06,
+      "loss": 0.1509,
+      "step": 1055
+    },
+    {
+      "epoch": 0.8466453674121406,
+      "grad_norm": 0.9655594054454741,
+      "learning_rate": 6.316184215632708e-06,
+      "loss": 0.1574,
+      "step": 1060
+    },
+    {
+      "epoch": 0.8506389776357828,
+      "grad_norm": 0.8882567467448087,
+      "learning_rate": 6.306495551429215e-06,
+      "loss": 0.1523,
+      "step": 1065
+    },
+    {
+      "epoch": 0.854632587859425,
+      "grad_norm": 0.8714942442812957,
+      "learning_rate": 6.296746273646466e-06,
+      "loss": 0.1494,
+      "step": 1070
+    },
+    {
+      "epoch": 0.8586261980830671,
+      "grad_norm": 0.9184869505642743,
+      "learning_rate": 6.286936592845503e-06,
+      "loss": 0.1553,
+      "step": 1075
+    },
+    {
+      "epoch": 0.8626198083067093,
+      "grad_norm": 0.9132914637442158,
+      "learning_rate": 6.2770667208919265e-06,
+      "loss": 0.1507,
+      "step": 1080
+    },
+    {
+      "epoch": 0.8666134185303515,
+      "grad_norm": 0.8713397978039784,
+      "learning_rate": 6.26713687095132e-06,
+      "loss": 0.1495,
+      "step": 1085
+    },
+    {
+      "epoch": 0.8706070287539937,
+      "grad_norm": 0.8854147468102307,
+      "learning_rate": 6.2571472574846515e-06,
+      "loss": 0.1486,
+      "step": 1090
+    },
+    {
+      "epoch": 0.8746006389776357,
+      "grad_norm": 0.9369163481137155,
+      "learning_rate": 6.247098096243633e-06,
+      "loss": 0.1506,
+      "step": 1095
+    },
+    {
+      "epoch": 0.8785942492012779,
+      "grad_norm": 0.9543529445696081,
+      "learning_rate": 6.23698960426607e-06,
+      "loss": 0.1547,
+      "step": 1100
+    },
+    {
+      "epoch": 0.8825878594249201,
+      "grad_norm": 0.9373088824897908,
+      "learning_rate": 6.2268219998711676e-06,
+      "loss": 0.1488,
+      "step": 1105
+    },
+    {
+      "epoch": 0.8865814696485623,
+      "grad_norm": 0.9422023974896926,
+      "learning_rate": 6.216595502654819e-06,
+      "loss": 0.1573,
+      "step": 1110
+    },
+    {
+      "epoch": 0.8905750798722045,
+      "grad_norm": 0.9696343156334155,
+      "learning_rate": 6.206310333484859e-06,
+      "loss": 0.1625,
+      "step": 1115
+    },
+    {
+      "epoch": 0.8945686900958466,
+      "grad_norm": 0.8142960364081798,
+      "learning_rate": 6.1959667144962984e-06,
+      "loss": 0.1525,
+      "step": 1120
+    },
+    {
+      "epoch": 0.8985623003194888,
+      "grad_norm": 0.8889141559955737,
+      "learning_rate": 6.185564869086523e-06,
+      "loss": 0.1493,
+      "step": 1125
+    },
+    {
+      "epoch": 0.902555910543131,
+      "grad_norm": 0.9104469356832359,
+      "learning_rate": 6.175105021910469e-06,
+      "loss": 0.144,
+      "step": 1130
+    },
+    {
+      "epoch": 0.9065495207667732,
+      "grad_norm": 0.8784613561636412,
+      "learning_rate": 6.164587398875773e-06,
+      "loss": 0.1553,
+      "step": 1135
+    },
+    {
+      "epoch": 0.9105431309904153,
+      "grad_norm": 0.9170546731493038,
+      "learning_rate": 6.1540122271378905e-06,
+      "loss": 0.1501,
+      "step": 1140
+    },
+    {
+      "epoch": 0.9145367412140575,
+      "grad_norm": 0.8596576185799807,
+      "learning_rate": 6.143379735095194e-06,
+      "loss": 0.1514,
+      "step": 1145
+    },
+    {
+      "epoch": 0.9185303514376997,
+      "grad_norm": 0.8780099284965823,
+      "learning_rate": 6.1326901523840325e-06,
+      "loss": 0.1485,
+      "step": 1150
+    },
+    {
+      "epoch": 0.9225239616613419,
+      "grad_norm": 0.8740379156570441,
+      "learning_rate": 6.12194370987378e-06,
+      "loss": 0.1528,
+      "step": 1155
+    },
+    {
+      "epoch": 0.9265175718849841,
+      "grad_norm": 0.9196888261832091,
+      "learning_rate": 6.111140639661846e-06,
+      "loss": 0.149,
+      "step": 1160
+    },
+    {
+      "epoch": 0.9305111821086262,
+      "grad_norm": 0.898041148881679,
+      "learning_rate": 6.10028117506866e-06,
+      "loss": 0.1588,
+      "step": 1165
+    },
+    {
+      "epoch": 0.9345047923322684,
+      "grad_norm": 0.8638459142453613,
+      "learning_rate": 6.089365550632639e-06,
+      "loss": 0.1513,
+      "step": 1170
+    },
+    {
+      "epoch": 0.9384984025559105,
+      "grad_norm": 0.9015055071062944,
+      "learning_rate": 6.078394002105115e-06,
+      "loss": 0.1561,
+      "step": 1175
+    },
+    {
+      "epoch": 0.9424920127795527,
+      "grad_norm": 0.8542741624271732,
+      "learning_rate": 6.067366766445247e-06,
+      "loss": 0.1519,
+      "step": 1180
+    },
+    {
+      "epoch": 0.9464856230031949,
+      "grad_norm": 0.9418131743815946,
+      "learning_rate": 6.056284081814902e-06,
+      "loss": 0.1445,
+      "step": 1185
+    },
+    {
+      "epoch": 0.950479233226837,
+      "grad_norm": 0.8274533049204239,
+      "learning_rate": 6.045146187573513e-06,
+      "loss": 0.1508,
+      "step": 1190
+    },
+    {
+      "epoch": 0.9544728434504792,
+      "grad_norm": 0.8788233365616475,
+      "learning_rate": 6.0339533242729105e-06,
+      "loss": 0.1457,
+      "step": 1195
+    },
+    {
+      "epoch": 0.9584664536741214,
+      "grad_norm": 0.8978374665514917,
+      "learning_rate": 6.022705733652118e-06,
+      "loss": 0.1607,
+      "step": 1200
+    },
+    {
+      "epoch": 0.9624600638977636,
+      "grad_norm": 0.9265956339818142,
+      "learning_rate": 6.0114036586321476e-06,
+      "loss": 0.1606,
+      "step": 1205
+    },
+    {
+      "epoch": 0.9664536741214057,
+      "grad_norm": 0.8913889253563607,
+      "learning_rate": 6.000047343310736e-06,
+      "loss": 0.1446,
+      "step": 1210
+    },
+    {
+      "epoch": 0.9704472843450479,
+      "grad_norm": 0.9162076451056782,
+      "learning_rate": 5.988637032957083e-06,
+      "loss": 0.1472,
+      "step": 1215
+    },
+    {
+      "epoch": 0.9744408945686901,
+      "grad_norm": 0.9037423765093193,
+      "learning_rate": 5.977172974006552e-06,
+      "loss": 0.1495,
+      "step": 1220
+    },
+    {
+      "epoch": 0.9784345047923323,
+      "grad_norm": 0.8965101197123303,
+      "learning_rate": 5.965655414055348e-06,
+      "loss": 0.1514,
+      "step": 1225
+    },
+    {
+      "epoch": 0.9824281150159745,
+      "grad_norm": 0.8363882753686213,
+      "learning_rate": 5.954084601855169e-06,
+      "loss": 0.1517,
+      "step": 1230
+    },
+    {
+      "epoch": 0.9864217252396166,
+      "grad_norm": 0.8508334278041144,
+      "learning_rate": 5.942460787307833e-06,
+      "loss": 0.1545,
+      "step": 1235
+    },
+    {
+      "epoch": 0.9904153354632588,
+      "grad_norm": 0.8440202005698025,
+      "learning_rate": 5.930784221459883e-06,
+      "loss": 0.154,
+      "step": 1240
+    },
+    {
+      "epoch": 0.994408945686901,
+      "grad_norm": 0.8641285063450203,
+      "learning_rate": 5.919055156497162e-06,
+      "loss": 0.1575,
+      "step": 1245
+    },
+    {
+      "epoch": 0.9984025559105432,
+      "grad_norm": 0.8395654019211741,
+      "learning_rate": 5.907273845739368e-06,
+      "loss": 0.1433,
+      "step": 1250
+    },
+    {
+      "epoch": 1.0023961661341854,
+      "grad_norm": 0.7574896989713115,
+      "learning_rate": 5.895440543634588e-06,
+      "loss": 0.123,
+      "step": 1255
+    },
+    {
+      "epoch": 1.0063897763578276,
+      "grad_norm": 0.7001918457109648,
+      "learning_rate": 5.88355550575379e-06,
+      "loss": 0.0967,
+      "step": 1260
+    },
+    {
+      "epoch": 1.0103833865814698,
+      "grad_norm": 0.8541947338770247,
+      "learning_rate": 5.871618988785316e-06,
+      "loss": 0.0961,
+      "step": 1265
+    },
+    {
+      "epoch": 1.0143769968051117,
+      "grad_norm": 0.7524342512654641,
+      "learning_rate": 5.8596312505293315e-06,
+      "loss": 0.0947,
+      "step": 1270
+    },
+    {
+      "epoch": 1.018370607028754,
+      "grad_norm": 0.7718835887830281,
+      "learning_rate": 5.847592549892258e-06,
+      "loss": 0.0945,
+      "step": 1275
+    },
+    {
+      "epoch": 1.0223642172523961,
+      "grad_norm": 0.8408835100355743,
+      "learning_rate": 5.835503146881185e-06,
+      "loss": 0.1015,
+      "step": 1280
+    },
+    {
+      "epoch": 1.0263578274760383,
+      "grad_norm": 0.8763612828817176,
+      "learning_rate": 5.823363302598248e-06,
+      "loss": 0.0968,
+      "step": 1285
+    },
+    {
+      "epoch": 1.0303514376996805,
+      "grad_norm": 0.8178713771043465,
+      "learning_rate": 5.811173279234996e-06,
+      "loss": 0.0962,
+      "step": 1290
+    },
+    {
+      "epoch": 1.0343450479233227,
+      "grad_norm": 0.9101366597977056,
+      "learning_rate": 5.798933340066726e-06,
+      "loss": 0.0983,
+      "step": 1295
+    },
+    {
+      "epoch": 1.038338658146965,
+      "grad_norm": 0.753210589288919,
+      "learning_rate": 5.786643749446795e-06,
+      "loss": 0.0923,
+      "step": 1300
+    },
+    {
+      "epoch": 1.042332268370607,
+      "grad_norm": 0.8263779547352994,
+      "learning_rate": 5.774304772800912e-06,
+      "loss": 0.0978,
+      "step": 1305
+    },
+    {
+      "epoch": 1.0463258785942493,
+      "grad_norm": 0.835011482405387,
+      "learning_rate": 5.76191667662141e-06,
+      "loss": 0.0917,
+      "step": 1310
+    },
+    {
+      "epoch": 1.0503194888178913,
+      "grad_norm": 0.8228387988406632,
+      "learning_rate": 5.74947972846148e-06,
+      "loss": 0.0977,
+      "step": 1315
+    },
+    {
+      "epoch": 1.0543130990415335,
+      "grad_norm": 0.7510886619859262,
+      "learning_rate": 5.736994196929405e-06,
+      "loss": 0.0934,
+      "step": 1320
+    },
+    {
+      "epoch": 1.0583067092651757,
+      "grad_norm": 0.8004708846108053,
+      "learning_rate": 5.724460351682749e-06,
+      "loss": 0.0946,
+      "step": 1325
+    },
+    {
+      "epoch": 1.0623003194888179,
+      "grad_norm": 0.8382732441836329,
+      "learning_rate": 5.711878463422534e-06,
+      "loss": 0.0925,
+      "step": 1330
+    },
+    {
+      "epoch": 1.06629392971246,
+      "grad_norm": 0.8616180876824737,
+      "learning_rate": 5.699248803887402e-06,
+      "loss": 0.0933,
+      "step": 1335
+    },
+    {
+      "epoch": 1.0702875399361023,
+      "grad_norm": 0.8300175767051388,
+      "learning_rate": 5.686571645847735e-06,
+      "loss": 0.0964,
+      "step": 1340
+    },
+    {
+      "epoch": 1.0742811501597445,
+      "grad_norm": 0.8522423266080669,
+      "learning_rate": 5.673847263099774e-06,
+      "loss": 0.0917,
+      "step": 1345
+    },
+    {
+      "epoch": 1.0782747603833867,
+      "grad_norm": 0.7589979341067651,
+      "learning_rate": 5.6610759304596946e-06,
+      "loss": 0.0914,
+      "step": 1350
+    },
+    {
+      "epoch": 1.0822683706070289,
+      "grad_norm": 0.8241214078415051,
+      "learning_rate": 5.64825792375768e-06,
+      "loss": 0.0972,
+      "step": 1355
+    },
+    {
+      "epoch": 1.0862619808306708,
+      "grad_norm": 0.7525508184236472,
+      "learning_rate": 5.6353935198319636e-06,
+      "loss": 0.091,
+      "step": 1360
+    },
+    {
+      "epoch": 1.090255591054313,
+      "grad_norm": 0.8304249863595534,
+      "learning_rate": 5.622482996522844e-06,
+      "loss": 0.094,
+      "step": 1365
+    },
+    {
+      "epoch": 1.0942492012779552,
+      "grad_norm": 0.814379429479353,
+      "learning_rate": 5.6095266326666915e-06,
+      "loss": 0.0988,
+      "step": 1370
+    },
+    {
+      "epoch": 1.0982428115015974,
+      "grad_norm": 0.7740762735894441,
+      "learning_rate": 5.5965247080899184e-06,
+      "loss": 0.0936,
+      "step": 1375
+    },
+    {
+      "epoch": 1.1022364217252396,
+      "grad_norm": 0.7765444476341753,
+      "learning_rate": 5.583477503602944e-06,
+      "loss": 0.0949,
+      "step": 1380
+    },
+    {
+      "epoch": 1.1062300319488818,
+      "grad_norm": 0.7960825199906291,
+      "learning_rate": 5.570385300994121e-06,
+      "loss": 0.0973,
+      "step": 1385
+    },
+    {
+      "epoch": 1.110223642172524,
+      "grad_norm": 0.8211272733995488,
+      "learning_rate": 5.557248383023656e-06,
+      "loss": 0.0989,
+      "step": 1390
+    },
+    {
+      "epoch": 1.1142172523961662,
+      "grad_norm": 0.8230061747958791,
+      "learning_rate": 5.5440670334175e-06,
+      "loss": 0.0959,
+      "step": 1395
+    },
+    {
+      "epoch": 1.1182108626198084,
+      "grad_norm": 0.7930641006824678,
+      "learning_rate": 5.5308415368612194e-06,
+      "loss": 0.0941,
+      "step": 1400
+    },
+    {
+      "epoch": 1.1222044728434506,
+      "grad_norm": 0.7986731856931012,
+      "learning_rate": 5.5175721789938525e-06,
+      "loss": 0.0964,
+      "step": 1405
+    },
+    {
+      "epoch": 1.1261980830670926,
+      "grad_norm": 0.8180516432624779,
+      "learning_rate": 5.504259246401733e-06,
+      "loss": 0.0992,
+      "step": 1410
+    },
+    {
+      "epoch": 1.1301916932907348,
+      "grad_norm": 0.8080517814845294,
+      "learning_rate": 5.490903026612306e-06,
+      "loss": 0.0954,
+      "step": 1415
+    },
+    {
+      "epoch": 1.134185303514377,
+      "grad_norm": 0.7894397030245984,
+      "learning_rate": 5.477503808087915e-06,
+      "loss": 0.0939,
+      "step": 1420
+    },
+    {
+      "epoch": 1.1381789137380192,
+      "grad_norm": 0.8266502827314096,
+      "learning_rate": 5.464061880219575e-06,
+      "loss": 0.0972,
+      "step": 1425
+    },
+    {
+      "epoch": 1.1421725239616614,
+      "grad_norm": 0.7695060771844507,
+      "learning_rate": 5.450577533320721e-06,
+      "loss": 0.0889,
+      "step": 1430
+    },
+    {
+      "epoch": 1.1461661341853036,
+      "grad_norm": 0.7975054651942407,
+      "learning_rate": 5.437051058620934e-06,
+      "loss": 0.0912,
+      "step": 1435
+    },
+    {
+      "epoch": 1.1501597444089458,
+      "grad_norm": 0.835290771956647,
+      "learning_rate": 5.423482748259657e-06,
+      "loss": 0.1003,
+      "step": 1440
+    },
+    {
+      "epoch": 1.154153354632588,
+      "grad_norm": 0.796837303035922,
+      "learning_rate": 5.4098728952798835e-06,
+      "loss": 0.0924,
+      "step": 1445
+    },
+    {
+      "epoch": 1.15814696485623,
+      "grad_norm": 0.7979782248094183,
+      "learning_rate": 5.396221793621827e-06,
+      "loss": 0.097,
+      "step": 1450
+    },
+    {
+      "epoch": 1.1621405750798721,
+      "grad_norm": 0.7733365430240714,
+      "learning_rate": 5.382529738116575e-06,
+      "loss": 0.0943,
+      "step": 1455
+    },
+    {
+      "epoch": 1.1661341853035143,
+      "grad_norm": 0.792955069966389,
+      "learning_rate": 5.3687970244797175e-06,
+      "loss": 0.0907,
+      "step": 1460
+    },
+    {
+      "epoch": 1.1701277955271565,
+      "grad_norm": 0.7955575959074169,
+      "learning_rate": 5.355023949304962e-06,
+      "loss": 0.0935,
+      "step": 1465
+    },
+    {
+      "epoch": 1.1741214057507987,
+      "grad_norm": 0.773556306974029,
+      "learning_rate": 5.3412108100577344e-06,
+      "loss": 0.0961,
+      "step": 1470
+    },
+    {
+      "epoch": 1.178115015974441,
+      "grad_norm": 0.7701308254197153,
+      "learning_rate": 5.327357905068743e-06,
+      "loss": 0.091,
+      "step": 1475
+    },
+    {
+      "epoch": 1.182108626198083,
+      "grad_norm": 0.8555538690082194,
+      "learning_rate": 5.3134655335275425e-06,
+      "loss": 0.097,
+      "step": 1480
+    },
+    {
+      "epoch": 1.1861022364217253,
+      "grad_norm": 0.8028090926659807,
+      "learning_rate": 5.29953399547607e-06,
+      "loss": 0.0991,
+      "step": 1485
+    },
+    {
+      "epoch": 1.1900958466453675,
+      "grad_norm": 0.8257848244515429,
+      "learning_rate": 5.285563591802168e-06,
+      "loss": 0.0951,
+      "step": 1490
+    },
+    {
+      "epoch": 1.1940894568690097,
+      "grad_norm": 0.7918972043683726,
+      "learning_rate": 5.271554624233079e-06,
+      "loss": 0.094,
+      "step": 1495
+    },
+    {
+      "epoch": 1.1980830670926517,
+      "grad_norm": 0.8361622190630472,
+      "learning_rate": 5.257507395328937e-06,
+      "loss": 0.0938,
+      "step": 1500
+    },
+    {
+      "epoch": 1.1980830670926517,
+      "eval_loss": 0.14057603478431702,
+      "eval_runtime": 388.2062,
+      "eval_samples_per_second": 45.87,
+      "eval_steps_per_second": 5.734,
+      "step": 1500
+    },
+    {
+      "epoch": 1.2020766773162939,
+      "grad_norm": 0.835859137908053,
+      "learning_rate": 5.243422208476228e-06,
+      "loss": 0.0963,
+      "step": 1505
+    },
+    {
+      "epoch": 1.206070287539936,
+      "grad_norm": 0.8373993609652842,
+      "learning_rate": 5.229299367881236e-06,
+      "loss": 0.0931,
+      "step": 1510
+    },
+    {
+      "epoch": 1.2100638977635783,
+      "grad_norm": 0.8468046787504842,
+      "learning_rate": 5.215139178563481e-06,
+      "loss": 0.0983,
+      "step": 1515
+    },
+    {
+      "epoch": 1.2140575079872205,
+      "grad_norm": 0.835364718161382,
+      "learning_rate": 5.200941946349121e-06,
+      "loss": 0.0964,
+      "step": 1520
+    },
+    {
+      "epoch": 1.2180511182108626,
+      "grad_norm": 0.7759190731109368,
+      "learning_rate": 5.186707977864355e-06,
+      "loss": 0.0929,
+      "step": 1525
+    },
+    {
+      "epoch": 1.2220447284345048,
+      "grad_norm": 0.8158943737288525,
+      "learning_rate": 5.172437580528796e-06,
+      "loss": 0.094,
+      "step": 1530
+    },
+    {
+      "epoch": 1.226038338658147,
+      "grad_norm": 0.8729951582908195,
+      "learning_rate": 5.158131062548833e-06,
+      "loss": 0.096,
+      "step": 1535
+    },
+    {
+      "epoch": 1.230031948881789,
+      "grad_norm": 0.8333545374654573,
+      "learning_rate": 5.143788732910973e-06,
+      "loss": 0.0952,
+      "step": 1540
+    },
+    {
+      "epoch": 1.2340255591054312,
+      "grad_norm": 0.8612855433349985,
+      "learning_rate": 5.129410901375172e-06,
+      "loss": 0.0974,
+      "step": 1545
+    },
+    {
+      "epoch": 1.2380191693290734,
+      "grad_norm": 0.8542415949428595,
+      "learning_rate": 5.11499787846814e-06,
+      "loss": 0.0947,
+      "step": 1550
+    },
+    {
+      "epoch": 1.2420127795527156,
+      "grad_norm": 0.8391599714715479,
+      "learning_rate": 5.100549975476637e-06,
+      "loss": 0.0944,
+      "step": 1555
+    },
+    {
+      "epoch": 1.2460063897763578,
+      "grad_norm": 0.8898214771674834,
+      "learning_rate": 5.0860675044407495e-06,
+      "loss": 0.0965,
+      "step": 1560
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 0.8779811650939291,
+      "learning_rate": 5.071550778147149e-06,
+      "loss": 0.098,
+      "step": 1565
+    },
+    {
+      "epoch": 1.2539936102236422,
+      "grad_norm": 0.8272086627597742,
+      "learning_rate": 5.05700011012234e-06,
+      "loss": 0.0935,
+      "step": 1570
+    },
+    {
+      "epoch": 1.2579872204472844,
+      "grad_norm": 0.810515606440639,
+      "learning_rate": 5.042415814625887e-06,
+      "loss": 0.0969,
+      "step": 1575
+    },
+    {
+      "epoch": 1.2619808306709266,
+      "grad_norm": 0.8943157588820453,
+      "learning_rate": 5.027798206643629e-06,
+      "loss": 0.0985,
+      "step": 1580
+    },
+    {
+      "epoch": 1.2659744408945688,
+      "grad_norm": 0.825143864442046,
+      "learning_rate": 5.013147601880871e-06,
+      "loss": 0.0944,
+      "step": 1585
+    },
+    {
+      "epoch": 1.269968051118211,
+      "grad_norm": 0.8256699054355341,
+      "learning_rate": 4.998464316755571e-06,
+      "loss": 0.0932,
+      "step": 1590
+    },
+    {
+      "epoch": 1.273961661341853,
+      "grad_norm": 0.8116571534197091,
+      "learning_rate": 4.983748668391507e-06,
+      "loss": 0.0936,
+      "step": 1595
+    },
+    {
+      "epoch": 1.2779552715654952,
+      "grad_norm": 0.8161005230017339,
+      "learning_rate": 4.969000974611423e-06,
+      "loss": 0.0968,
+      "step": 1600
+    },
+    {
+      "epoch": 1.2819488817891374,
+      "grad_norm": 0.8346647605386535,
+      "learning_rate": 4.954221553930166e-06,
+      "loss": 0.0942,
+      "step": 1605
+    },
+    {
+      "epoch": 1.2859424920127795,
+      "grad_norm": 0.7827131547372781,
+      "learning_rate": 4.939410725547812e-06,
+      "loss": 0.0941,
+      "step": 1610
+    },
+    {
+      "epoch": 1.2899361022364217,
+      "grad_norm": 0.8568068704330788,
+      "learning_rate": 4.9245688093427655e-06,
+      "loss": 0.0926,
+      "step": 1615
+    },
+    {
+      "epoch": 1.293929712460064,
+      "grad_norm": 0.7805159905736859,
+      "learning_rate": 4.909696125864851e-06,
+      "loss": 0.0912,
+      "step": 1620
+    },
+    {
+      "epoch": 1.2979233226837061,
+      "grad_norm": 0.7738061176599114,
+      "learning_rate": 4.894792996328399e-06,
+      "loss": 0.0922,
+      "step": 1625
+    },
+    {
+      "epoch": 1.3019169329073481,
+      "grad_norm": 0.8184213649823732,
+      "learning_rate": 4.879859742605294e-06,
+      "loss": 0.0942,
+      "step": 1630
+    },
+    {
+      "epoch": 1.3059105431309903,
+      "grad_norm": 0.8290453852769142,
+      "learning_rate": 4.864896687218038e-06,
+      "loss": 0.096,
+      "step": 1635
+    },
+    {
+      "epoch": 1.3099041533546325,
+      "grad_norm": 0.8629463577885302,
+      "learning_rate": 4.849904153332774e-06,
+      "loss": 0.0951,
+      "step": 1640
+    },
+    {
+      "epoch": 1.3138977635782747,
+      "grad_norm": 0.8123701678440974,
+      "learning_rate": 4.834882464752308e-06,
+      "loss": 0.0948,
+      "step": 1645
+    },
+    {
+      "epoch": 1.317891373801917,
+      "grad_norm": 0.7528537020864617,
+      "learning_rate": 4.8198319459091215e-06,
+      "loss": 0.0939,
+      "step": 1650
+    },
+    {
+      "epoch": 1.321884984025559,
+      "grad_norm": 0.7800246305597872,
+      "learning_rate": 4.8047529218583586e-06,
+      "loss": 0.0942,
+      "step": 1655
+    },
+    {
+      "epoch": 1.3258785942492013,
+      "grad_norm": 0.8313222255456414,
+      "learning_rate": 4.789645718270808e-06,
+      "loss": 0.0948,
+      "step": 1660
+    },
+    {
+      "epoch": 1.3298722044728435,
+      "grad_norm": 0.8179733784658576,
+      "learning_rate": 4.774510661425869e-06,
+      "loss": 0.0947,
+      "step": 1665
+    },
+    {
+      "epoch": 1.3338658146964857,
+      "grad_norm": 0.7776969162457743,
+      "learning_rate": 4.759348078204504e-06,
+      "loss": 0.09,
+      "step": 1670
+    },
+    {
+      "epoch": 1.3378594249201279,
+      "grad_norm": 0.841754427277134,
+      "learning_rate": 4.744158296082179e-06,
+      "loss": 0.0945,
+      "step": 1675
+    },
+    {
+      "epoch": 1.34185303514377,
+      "grad_norm": 0.8370881837026815,
+      "learning_rate": 4.7289416431217925e-06,
+      "loss": 0.0904,
+      "step": 1680
+    },
+    {
+      "epoch": 1.3458466453674123,
+      "grad_norm": 0.8562758183663967,
+      "learning_rate": 4.713698447966586e-06,
+      "loss": 0.0977,
+      "step": 1685
+    },
+    {
+      "epoch": 1.3498402555910542,
+      "grad_norm": 0.7855201641830065,
+      "learning_rate": 4.6984290398330524e-06,
+      "loss": 0.0943,
+      "step": 1690
+    },
+    {
+      "epoch": 1.3538338658146964,
+      "grad_norm": 0.8763494885051274,
+      "learning_rate": 4.6831337485038196e-06,
+      "loss": 0.0979,
+      "step": 1695
+    },
+    {
+      "epoch": 1.3578274760383386,
+      "grad_norm": 0.8144070746777402,
+      "learning_rate": 4.6678129043205315e-06,
+      "loss": 0.0913,
+      "step": 1700
+    },
+    {
+      "epoch": 1.3618210862619808,
+      "grad_norm": 0.7944782321390458,
+      "learning_rate": 4.652466838176711e-06,
+      "loss": 0.0938,
+      "step": 1705
+    },
+    {
+      "epoch": 1.365814696485623,
+      "grad_norm": 0.8864834970350833,
+      "learning_rate": 4.6370958815106175e-06,
+      "loss": 0.0944,
+      "step": 1710
+    },
+    {
+      "epoch": 1.3698083067092652,
+      "grad_norm": 0.8054616729912718,
+      "learning_rate": 4.621700366298084e-06,
+      "loss": 0.0939,
+      "step": 1715
+    },
+    {
+      "epoch": 1.3738019169329074,
+      "grad_norm": 0.8454180008589398,
+      "learning_rate": 4.60628062504535e-06,
+      "loss": 0.093,
+      "step": 1720
+    },
+    {
+      "epoch": 1.3777955271565494,
+      "grad_norm": 0.7990549534085654,
+      "learning_rate": 4.590836990781877e-06,
+      "loss": 0.0886,
+      "step": 1725
+    },
+    {
+      "epoch": 1.3817891373801916,
+      "grad_norm": 0.8070045082712269,
+      "learning_rate": 4.575369797053165e-06,
+      "loss": 0.0882,
+      "step": 1730
+    },
+    {
+      "epoch": 1.3857827476038338,
+      "grad_norm": 0.8612739828428202,
+      "learning_rate": 4.559879377913534e-06,
+      "loss": 0.0964,
+      "step": 1735
+    },
+    {
+      "epoch": 1.389776357827476,
+      "grad_norm": 0.8180434067759751,
+      "learning_rate": 4.544366067918922e-06,
+      "loss": 0.0947,
+      "step": 1740
+    },
+    {
+      "epoch": 1.3937699680511182,
+      "grad_norm": 0.7929033707791658,
+      "learning_rate": 4.5288302021196525e-06,
+      "loss": 0.0934,
+      "step": 1745
+    },
+    {
+      "epoch": 1.3977635782747604,
+      "grad_norm": 0.8213908701484177,
+      "learning_rate": 4.513272116053203e-06,
+      "loss": 0.092,
+      "step": 1750
+    },
+    {
+      "epoch": 1.4017571884984026,
+      "grad_norm": 0.8758208928982602,
+      "learning_rate": 4.497692145736952e-06,
+      "loss": 0.0984,
+      "step": 1755
+    },
+    {
+      "epoch": 1.4057507987220448,
+      "grad_norm": 0.8498430639803995,
+      "learning_rate": 4.482090627660928e-06,
+      "loss": 0.0948,
+      "step": 1760
+    },
+    {
+      "epoch": 1.409744408945687,
+      "grad_norm": 0.8357042132874987,
+      "learning_rate": 4.466467898780538e-06,
+      "loss": 0.0937,
+      "step": 1765
+    },
+    {
+      "epoch": 1.4137380191693292,
+      "grad_norm": 0.8049671549685841,
+      "learning_rate": 4.450824296509293e-06,
+      "loss": 0.0962,
+      "step": 1770
+    },
+    {
+      "epoch": 1.4177316293929714,
+      "grad_norm": 0.762029470780092,
+      "learning_rate": 4.435160158711517e-06,
+      "loss": 0.093,
+      "step": 1775
+    },
+    {
+      "epoch": 1.4217252396166133,
+      "grad_norm": 0.7777991642289898,
+      "learning_rate": 4.419475823695054e-06,
+      "loss": 0.0881,
+      "step": 1780
+    },
+    {
+      "epoch": 1.4257188498402555,
+      "grad_norm": 0.8172923019508063,
+      "learning_rate": 4.40377163020396e-06,
+      "loss": 0.0962,
+      "step": 1785
+    },
+    {
+      "epoch": 1.4297124600638977,
+      "grad_norm": 0.8195226636883436,
+      "learning_rate": 4.388047917411185e-06,
+      "loss": 0.0938,
+      "step": 1790
+    },
+    {
+      "epoch": 1.43370607028754,
+      "grad_norm": 0.8567775821515811,
+      "learning_rate": 4.37230502491125e-06,
+      "loss": 0.094,
+      "step": 1795
+    },
+    {
+      "epoch": 1.4376996805111821,
+      "grad_norm": 0.8975281588105369,
+      "learning_rate": 4.35654329271291e-06,
+      "loss": 0.0931,
+      "step": 1800
+    },
+    {
+      "epoch": 1.4416932907348243,
+      "grad_norm": 0.7980268710605869,
+      "learning_rate": 4.340763061231815e-06,
+      "loss": 0.094,
+      "step": 1805
+    },
+    {
+      "epoch": 1.4456869009584665,
+      "grad_norm": 0.8423324417742822,
+      "learning_rate": 4.3249646712831525e-06,
+      "loss": 0.0934,
+      "step": 1810
+    },
+    {
+      "epoch": 1.4496805111821085,
+      "grad_norm": 0.760524105440743,
+      "learning_rate": 4.309148464074293e-06,
+      "loss": 0.0929,
+      "step": 1815
+    },
+    {
+      "epoch": 1.4536741214057507,
+      "grad_norm": 0.8316187906255309,
+      "learning_rate": 4.293314781197414e-06,
+      "loss": 0.0933,
+      "step": 1820
+    },
+    {
+      "epoch": 1.457667731629393,
+      "grad_norm": 0.8371905859373513,
+      "learning_rate": 4.277463964622125e-06,
+      "loss": 0.0887,
+      "step": 1825
+    },
+    {
+      "epoch": 1.461661341853035,
+      "grad_norm": 0.8364714730137837,
+      "learning_rate": 4.261596356688086e-06,
+      "loss": 0.0962,
+      "step": 1830
+    },
+    {
+      "epoch": 1.4656549520766773,
+      "grad_norm": 0.8141222333202962,
+      "learning_rate": 4.245712300097606e-06,
+      "loss": 0.0925,
+      "step": 1835
+    },
+    {
+      "epoch": 1.4696485623003195,
+      "grad_norm": 0.8478554496354959,
+      "learning_rate": 4.229812137908251e-06,
+      "loss": 0.0904,
+      "step": 1840
+    },
+    {
+      "epoch": 1.4736421725239617,
+      "grad_norm": 0.7540954927526371,
+      "learning_rate": 4.213896213525421e-06,
+      "loss": 0.0892,
+      "step": 1845
+    },
+    {
+      "epoch": 1.4776357827476039,
+      "grad_norm": 0.8171840914315421,
+      "learning_rate": 4.197964870694949e-06,
+      "loss": 0.0959,
+      "step": 1850
+    },
+    {
+      "epoch": 1.481629392971246,
+      "grad_norm": 0.8254400596397985,
+      "learning_rate": 4.182018453495667e-06,
+      "loss": 0.0938,
+      "step": 1855
+    },
+    {
+      "epoch": 1.4856230031948883,
+      "grad_norm": 0.8125736388997561,
+      "learning_rate": 4.166057306331977e-06,
+      "loss": 0.0909,
+      "step": 1860
+    },
+    {
+      "epoch": 1.4896166134185305,
+      "grad_norm": 0.7640120620441644,
+      "learning_rate": 4.150081773926414e-06,
+      "loss": 0.0901,
+      "step": 1865
+    },
+    {
+      "epoch": 1.4936102236421724,
+      "grad_norm": 0.799653826309561,
+      "learning_rate": 4.134092201312196e-06,
+      "loss": 0.0922,
+      "step": 1870
+    },
+    {
+      "epoch": 1.4976038338658146,
+      "grad_norm": 0.8128276034935282,
+      "learning_rate": 4.118088933825783e-06,
+      "loss": 0.093,
+      "step": 1875
+    },
+    {
+      "epoch": 1.5015974440894568,
+      "grad_norm": 0.8834498526452107,
+      "learning_rate": 4.102072317099404e-06,
+      "loss": 0.0953,
+      "step": 1880
+    },
+    {
+      "epoch": 1.505591054313099,
+      "grad_norm": 0.8866139042397759,
+      "learning_rate": 4.086042697053604e-06,
+      "loss": 0.0959,
+      "step": 1885
+    },
+    {
+      "epoch": 1.5095846645367412,
+      "grad_norm": 0.9024588302568654,
+      "learning_rate": 4.0700004198897665e-06,
+      "loss": 0.0929,
+      "step": 1890
+    },
+    {
+      "epoch": 1.5135782747603834,
+      "grad_norm": 0.7631982111355976,
+      "learning_rate": 4.053945832082642e-06,
+      "loss": 0.0907,
+      "step": 1895
+    },
+    {
+      "epoch": 1.5175718849840254,
+      "grad_norm": 0.752187865320078,
+      "learning_rate": 4.037879280372855e-06,
+      "loss": 0.0922,
+      "step": 1900
+    },
+    {
+      "epoch": 1.5215654952076676,
+      "grad_norm": 0.8037283021019334,
+      "learning_rate": 4.0218011117594265e-06,
+      "loss": 0.0887,
+      "step": 1905
+    },
+    {
+      "epoch": 1.5255591054313098,
+      "grad_norm": 0.7822698417920745,
+      "learning_rate": 4.005711673492274e-06,
+      "loss": 0.0874,
+      "step": 1910
+    },
+    {
+      "epoch": 1.529552715654952,
+      "grad_norm": 0.8037410692104164,
+      "learning_rate": 3.989611313064714e-06,
+      "loss": 0.0909,
+      "step": 1915
+    },
+    {
+      "epoch": 1.5335463258785942,
+      "grad_norm": 0.8877519379788956,
+      "learning_rate": 3.97350037820595e-06,
+      "loss": 0.0945,
+      "step": 1920
+    },
+    {
+      "epoch": 1.5375399361022364,
+      "grad_norm": 0.7148728619247341,
+      "learning_rate": 3.9573792168735735e-06,
+      "loss": 0.091,
+      "step": 1925
+    },
+    {
+      "epoch": 1.5415335463258786,
+      "grad_norm": 0.8136672941763351,
+      "learning_rate": 3.9412481772460404e-06,
+      "loss": 0.0917,
+      "step": 1930
+    },
+    {
+      "epoch": 1.5455271565495208,
+      "grad_norm": 0.7546989024401581,
+      "learning_rate": 3.925107607715156e-06,
+      "loss": 0.0918,
+      "step": 1935
+    },
+    {
+      "epoch": 1.549520766773163,
+      "grad_norm": 0.7557424472213525,
+      "learning_rate": 3.908957856878548e-06,
+      "loss": 0.0909,
+      "step": 1940
+    },
+    {
+      "epoch": 1.5535143769968052,
+      "grad_norm": 0.8400496651270898,
+      "learning_rate": 3.892799273532136e-06,
+      "loss": 0.0951,
+      "step": 1945
+    },
+    {
+      "epoch": 1.5575079872204474,
+      "grad_norm": 0.8463913720969456,
+      "learning_rate": 3.876632206662605e-06,
+      "loss": 0.0933,
+      "step": 1950
+    },
+    {
+      "epoch": 1.5615015974440896,
+      "grad_norm": 0.8750388641162147,
+      "learning_rate": 3.860457005439858e-06,
+      "loss": 0.0936,
+      "step": 1955
+    },
+    {
+      "epoch": 1.5654952076677318,
+      "grad_norm": 0.8208298877616164,
+      "learning_rate": 3.844274019209488e-06,
+      "loss": 0.0942,
+      "step": 1960
+    },
+    {
+      "epoch": 1.569488817891374,
+      "grad_norm": 0.8289575426887811,
+      "learning_rate": 3.828083597485219e-06,
+      "loss": 0.0942,
+      "step": 1965
+    },
+    {
+      "epoch": 1.573482428115016,
+      "grad_norm": 0.7742099605330947,
+      "learning_rate": 3.811886089941368e-06,
+      "loss": 0.0917,
+      "step": 1970
+    },
+    {
+      "epoch": 1.5774760383386581,
+      "grad_norm": 0.851766269863927,
+      "learning_rate": 3.7956818464052853e-06,
+      "loss": 0.0922,
+      "step": 1975
+    },
+    {
+      "epoch": 1.5814696485623003,
+      "grad_norm": 0.8425829506887897,
+      "learning_rate": 3.779471216849804e-06,
+      "loss": 0.0957,
+      "step": 1980
+    },
+    {
+      "epoch": 1.5854632587859425,
+      "grad_norm": 0.8510499144275525,
+      "learning_rate": 3.7632545513856812e-06,
+      "loss": 0.0938,
+      "step": 1985
+    },
+    {
+      "epoch": 1.5894568690095847,
+      "grad_norm": 0.7793249157088529,
+      "learning_rate": 3.747032200254035e-06,
+      "loss": 0.087,
+      "step": 1990
+    },
+    {
+      "epoch": 1.5934504792332267,
+      "grad_norm": 0.7890445255133955,
+      "learning_rate": 3.7308045138187764e-06,
+      "loss": 0.0914,
+      "step": 1995
+    },
+    {
+      "epoch": 1.5974440894568689,
+      "grad_norm": 0.8507435937119314,
+      "learning_rate": 3.714571842559053e-06,
+      "loss": 0.0948,
+      "step": 2000
+    },
+    {
+      "epoch": 1.5974440894568689,
+      "eval_loss": 0.135970339179039,
+      "eval_runtime": 388.3901,
+      "eval_samples_per_second": 45.848,
+      "eval_steps_per_second": 5.731,
+      "step": 2000
+    },
+    {
+      "epoch": 1.601437699680511,
+      "grad_norm": 0.777341324033066,
+      "learning_rate": 3.6983345370616666e-06,
+      "loss": 0.0901,
+      "step": 2005
+    },
+    {
+      "epoch": 1.6054313099041533,
+      "grad_norm": 0.8135499478146813,
+      "learning_rate": 3.6820929480135086e-06,
+      "loss": 0.093,
+      "step": 2010
+    },
+    {
+      "epoch": 1.6094249201277955,
+      "grad_norm": 0.8719757100439282,
+      "learning_rate": 3.6658474261939888e-06,
+      "loss": 0.0961,
+      "step": 2015
+    },
+    {
+      "epoch": 1.6134185303514377,
+      "grad_norm": 0.8303170119187012,
+      "learning_rate": 3.6495983224674493e-06,
+      "loss": 0.0964,
+      "step": 2020
+    },
+    {
+      "epoch": 1.6174121405750799,
+      "grad_norm": 0.784594767811727,
+      "learning_rate": 3.633345987775599e-06,
+      "loss": 0.0954,
+      "step": 2025
+    },
+    {
+      "epoch": 1.621405750798722,
+      "grad_norm": 0.7991817936852438,
+      "learning_rate": 3.617090773129923e-06,
+      "loss": 0.0899,
+      "step": 2030
+    },
+    {
+      "epoch": 1.6253993610223643,
+      "grad_norm": 0.8182423808938526,
+      "learning_rate": 3.6008330296041096e-06,
+      "loss": 0.0883,
+      "step": 2035
+    },
+    {
+      "epoch": 1.6293929712460065,
+      "grad_norm": 0.7858919236328515,
+      "learning_rate": 3.584573108326463e-06,
+      "loss": 0.0897,
+      "step": 2040
+    },
+    {
+      "epoch": 1.6333865814696487,
+      "grad_norm": 0.8156791778123983,
+      "learning_rate": 3.5683113604723233e-06,
+      "loss": 0.0897,
+      "step": 2045
+    },
+    {
+      "epoch": 1.6373801916932909,
+      "grad_norm": 0.8881831052271689,
+      "learning_rate": 3.552048137256478e-06,
+      "loss": 0.0935,
+      "step": 2050
+    },
+    {
+      "epoch": 1.641373801916933,
+      "grad_norm": 0.8016008616346979,
+      "learning_rate": 3.53578378992558e-06,
+      "loss": 0.0917,
+      "step": 2055
+    },
+    {
+      "epoch": 1.645367412140575,
+      "grad_norm": 0.8557576853411646,
+      "learning_rate": 3.5195186697505616e-06,
+      "loss": 0.0891,
+      "step": 2060
+    },
+    {
+      "epoch": 1.6493610223642172,
+      "grad_norm": 0.8028621053717053,
+      "learning_rate": 3.503253128019046e-06,
+      "loss": 0.0895,
+      "step": 2065
+    },
+    {
+      "epoch": 1.6533546325878594,
+      "grad_norm": 0.8130307630678764,
+      "learning_rate": 3.48698751602776e-06,
+      "loss": 0.0868,
+      "step": 2070
+    },
+    {
+      "epoch": 1.6573482428115016,
+      "grad_norm": 0.7589933957894843,
+      "learning_rate": 3.470722185074948e-06,
+      "loss": 0.0836,
+      "step": 2075
+    },
+    {
+      "epoch": 1.6613418530351438,
+      "grad_norm": 0.833019057283452,
+      "learning_rate": 3.454457486452786e-06,
+      "loss": 0.0853,
+      "step": 2080
+    },
+    {
+      "epoch": 1.6653354632587858,
+      "grad_norm": 0.8485088182697132,
+      "learning_rate": 3.4381937714397917e-06,
+      "loss": 0.0922,
+      "step": 2085
+    },
+    {
+      "epoch": 1.669329073482428,
+      "grad_norm": 0.8184010129649519,
+      "learning_rate": 3.42193139129324e-06,
+      "loss": 0.0901,
+      "step": 2090
+    },
+    {
+      "epoch": 1.6733226837060702,
+      "grad_norm": 0.8164910293916446,
+      "learning_rate": 3.4056706972415752e-06,
+      "loss": 0.0902,
+      "step": 2095
+    },
+    {
+      "epoch": 1.6773162939297124,
+      "grad_norm": 0.7799450306868917,
+      "learning_rate": 3.3894120404768293e-06,
+      "loss": 0.0879,
+      "step": 2100
+    },
+    {
+      "epoch": 1.6813099041533546,
+      "grad_norm": 0.8213948220491234,
+      "learning_rate": 3.373155772147028e-06,
+      "loss": 0.0866,
+      "step": 2105
+    },
+    {
+      "epoch": 1.6853035143769968,
+      "grad_norm": 0.8140459087876813,
+      "learning_rate": 3.3569022433486152e-06,
+      "loss": 0.0905,
+      "step": 2110
+    },
+    {
+      "epoch": 1.689297124600639,
+      "grad_norm": 0.8090512899368958,
+      "learning_rate": 3.3406518051188697e-06,
+      "loss": 0.093,
+      "step": 2115
+    },
+    {
+      "epoch": 1.6932907348242812,
+      "grad_norm": 0.786423302788487,
+      "learning_rate": 3.3244048084283183e-06,
+      "loss": 0.0886,
+      "step": 2120
+    },
+    {
+      "epoch": 1.6972843450479234,
+      "grad_norm": 0.8010633066357388,
+      "learning_rate": 3.3081616041731606e-06,
+      "loss": 0.0871,
+      "step": 2125
+    },
+    {
+      "epoch": 1.7012779552715656,
+      "grad_norm": 0.8562989387605436,
+      "learning_rate": 3.2919225431676874e-06,
+      "loss": 0.0909,
+      "step": 2130
+    },
+    {
+      "epoch": 1.7052715654952078,
+      "grad_norm": 0.8137206318229443,
+      "learning_rate": 3.2756879761367076e-06,
+      "loss": 0.0877,
+      "step": 2135
+    },
+    {
+      "epoch": 1.70926517571885,
+      "grad_norm": 0.8327531430876184,
+      "learning_rate": 3.259458253707968e-06,
+      "loss": 0.0893,
+      "step": 2140
+    },
+    {
+      "epoch": 1.7132587859424921,
+      "grad_norm": 0.819012210220874,
+      "learning_rate": 3.2432337264045843e-06,
+      "loss": 0.0914,
+      "step": 2145
+    },
+    {
+      "epoch": 1.7172523961661343,
+      "grad_norm": 0.8495772189306744,
+      "learning_rate": 3.2270147446374722e-06,
+      "loss": 0.092,
+      "step": 2150
+    },
+    {
+      "epoch": 1.7212460063897763,
+      "grad_norm": 0.8018376512404187,
+      "learning_rate": 3.210801658697774e-06,
+      "loss": 0.0881,
+      "step": 2155
+    },
+    {
+      "epoch": 1.7252396166134185,
+      "grad_norm": 0.7827860574369773,
+      "learning_rate": 3.1945948187492992e-06,
+      "loss": 0.0893,
+      "step": 2160
+    },
+    {
+      "epoch": 1.7292332268370607,
+      "grad_norm": 0.7731065648493003,
+      "learning_rate": 3.1783945748209557e-06,
+      "loss": 0.0854,
+      "step": 2165
+    },
+    {
+      "epoch": 1.733226837060703,
+      "grad_norm": 0.8628787752311644,
+      "learning_rate": 3.1622012767991967e-06,
+      "loss": 0.0946,
+      "step": 2170
+    },
+    {
+      "epoch": 1.7372204472843449,
+      "grad_norm": 0.7757583751727869,
+      "learning_rate": 3.1460152744204584e-06,
+      "loss": 0.0898,
+      "step": 2175
+    },
+    {
+      "epoch": 1.741214057507987,
+      "grad_norm": 0.773307933535615,
+      "learning_rate": 3.129836917263607e-06,
+      "loss": 0.0833,
+      "step": 2180
+    },
+    {
+      "epoch": 1.7452076677316293,
+      "grad_norm": 0.8402186654533444,
+      "learning_rate": 3.113666554742394e-06,
+      "loss": 0.0885,
+      "step": 2185
+    },
+    {
+      "epoch": 1.7492012779552715,
+      "grad_norm": 0.8286606926069269,
+      "learning_rate": 3.0975045360979025e-06,
+      "loss": 0.0902,
+      "step": 2190
+    },
+    {
+      "epoch": 1.7531948881789137,
+      "grad_norm": 0.8317794334903522,
+      "learning_rate": 3.0813512103910113e-06,
+      "loss": 0.0882,
+      "step": 2195
+    },
+    {
+      "epoch": 1.7571884984025559,
+      "grad_norm": 0.7846450982359997,
+      "learning_rate": 3.065206926494848e-06,
+      "loss": 0.0888,
+      "step": 2200
+    },
+    {
+      "epoch": 1.761182108626198,
+      "grad_norm": 0.8623658744163902,
+      "learning_rate": 3.049072033087264e-06,
+      "loss": 0.0899,
+      "step": 2205
+    },
+    {
+      "epoch": 1.7651757188498403,
+      "grad_norm": 0.7914568138762265,
+      "learning_rate": 3.032946878643295e-06,
+      "loss": 0.0906,
+      "step": 2210
+    },
+    {
+      "epoch": 1.7691693290734825,
+      "grad_norm": 0.8007387802489633,
+      "learning_rate": 3.0168318114276377e-06,
+      "loss": 0.0916,
+      "step": 2215
+    },
+    {
+      "epoch": 1.7731629392971247,
+      "grad_norm": 0.776240184386906,
+      "learning_rate": 3.0007271794871337e-06,
+      "loss": 0.0888,
+      "step": 2220
+    },
+    {
+      "epoch": 1.7771565495207668,
+      "grad_norm": 0.7537760505089393,
+      "learning_rate": 2.9846333306432412e-06,
+      "loss": 0.0853,
+      "step": 2225
+    },
+    {
+      "epoch": 1.781150159744409,
+      "grad_norm": 0.8216268561818008,
+      "learning_rate": 2.968550612484536e-06,
+      "loss": 0.0905,
+      "step": 2230
+    },
+    {
+      "epoch": 1.7851437699680512,
+      "grad_norm": 0.8623008566979109,
+      "learning_rate": 2.9524793723591913e-06,
+      "loss": 0.0884,
+      "step": 2235
+    },
+    {
+      "epoch": 1.7891373801916934,
+      "grad_norm": 0.8570248441650711,
+      "learning_rate": 2.936419957367489e-06,
+      "loss": 0.0955,
+      "step": 2240
+    },
+    {
+      "epoch": 1.7931309904153354,
+      "grad_norm": 0.8090425637420421,
+      "learning_rate": 2.92037271435431e-06,
+      "loss": 0.0929,
+      "step": 2245
+    },
+    {
+      "epoch": 1.7971246006389776,
+      "grad_norm": 0.8662160177063208,
+      "learning_rate": 2.904337989901653e-06,
+      "loss": 0.0872,
+      "step": 2250
+    },
+    {
+      "epoch": 1.8011182108626198,
+      "grad_norm": 0.789286926224427,
+      "learning_rate": 2.888316130321146e-06,
+      "loss": 0.0889,
+      "step": 2255
+    },
+    {
+      "epoch": 1.805111821086262,
+      "grad_norm": 0.8199288743264955,
+      "learning_rate": 2.872307481646565e-06,
+      "loss": 0.0918,
+      "step": 2260
+    },
+    {
+      "epoch": 1.8091054313099042,
+      "grad_norm": 0.8451894181207451,
+      "learning_rate": 2.856312389626365e-06,
+      "loss": 0.0942,
+      "step": 2265
+    },
+    {
+      "epoch": 1.8130990415335462,
+      "grad_norm": 0.831433941954139,
+      "learning_rate": 2.840331199716205e-06,
+      "loss": 0.0873,
+      "step": 2270
+    },
+    {
+      "epoch": 1.8170926517571884,
+      "grad_norm": 0.7904106905610935,
+      "learning_rate": 2.8243642570714958e-06,
+      "loss": 0.0865,
+      "step": 2275
+    },
+    {
+      "epoch": 1.8210862619808306,
+      "grad_norm": 0.8007696286288565,
+      "learning_rate": 2.808411906539941e-06,
+      "loss": 0.0908,
+      "step": 2280
+    },
+    {
+      "epoch": 1.8250798722044728,
+      "grad_norm": 0.7398549977807148,
+      "learning_rate": 2.7924744926540866e-06,
+      "loss": 0.0828,
+      "step": 2285
+    },
+    {
+      "epoch": 1.829073482428115,
+      "grad_norm": 0.8051951554405458,
+      "learning_rate": 2.7765523596238862e-06,
+      "loss": 0.0904,
+      "step": 2290
+    },
+    {
+      "epoch": 1.8330670926517572,
+      "grad_norm": 0.8101567895702511,
+      "learning_rate": 2.7606458513292615e-06,
+      "loss": 0.0877,
+      "step": 2295
+    },
+    {
+      "epoch": 1.8370607028753994,
+      "grad_norm": 0.810684787255368,
+      "learning_rate": 2.744755311312679e-06,
+      "loss": 0.0878,
+      "step": 2300
+    },
+    {
+      "epoch": 1.8410543130990416,
+      "grad_norm": 0.836337456473946,
+      "learning_rate": 2.7288810827717262e-06,
+      "loss": 0.0882,
+      "step": 2305
+    },
+    {
+      "epoch": 1.8450479233226837,
+      "grad_norm": 0.7870319338949235,
+      "learning_rate": 2.713023508551704e-06,
+      "loss": 0.0855,
+      "step": 2310
+    },
+    {
+      "epoch": 1.849041533546326,
+      "grad_norm": 0.8711250048723298,
+      "learning_rate": 2.6971829311382195e-06,
+      "loss": 0.0903,
+      "step": 2315
+    },
+    {
+      "epoch": 1.8530351437699681,
+      "grad_norm": 0.7898917076439477,
+      "learning_rate": 2.6813596926497892e-06,
+      "loss": 0.09,
+      "step": 2320
+    },
+    {
+      "epoch": 1.8570287539936103,
+      "grad_norm": 0.8447633678009813,
+      "learning_rate": 2.66555413483045e-06,
+      "loss": 0.092,
+      "step": 2325
+    },
+    {
+      "epoch": 1.8610223642172525,
+      "grad_norm": 0.8503814184125837,
+      "learning_rate": 2.649766599042378e-06,
+      "loss": 0.0855,
+      "step": 2330
+    },
+    {
+      "epoch": 1.8650159744408947,
+      "grad_norm": 0.7976736872294655,
+      "learning_rate": 2.6339974262585194e-06,
+      "loss": 0.089,
+      "step": 2335
+    },
+    {
+      "epoch": 1.8690095846645367,
+      "grad_norm": 0.7883072515843511,
+      "learning_rate": 2.61824695705522e-06,
+      "loss": 0.0879,
+      "step": 2340
+    },
+    {
+      "epoch": 1.873003194888179,
+      "grad_norm": 0.7546676570625951,
+      "learning_rate": 2.602515531604877e-06,
+      "loss": 0.086,
+      "step": 2345
+    },
+    {
+      "epoch": 1.876996805111821,
+      "grad_norm": 0.8281853119034305,
+      "learning_rate": 2.586803489668584e-06,
+      "loss": 0.0869,
+      "step": 2350
+    },
+    {
+      "epoch": 1.8809904153354633,
+      "grad_norm": 0.8108187343259362,
+      "learning_rate": 2.5711111705888005e-06,
+      "loss": 0.0903,
+      "step": 2355
+    },
+    {
+      "epoch": 1.8849840255591053,
+      "grad_norm": 0.7924520966861539,
+      "learning_rate": 2.555438913282018e-06,
+      "loss": 0.0924,
+      "step": 2360
+    },
+    {
+      "epoch": 1.8889776357827475,
+      "grad_norm": 0.7259687130542957,
+      "learning_rate": 2.5397870562314398e-06,
+      "loss": 0.0852,
+      "step": 2365
+    },
+    {
+      "epoch": 1.8929712460063897,
+      "grad_norm": 0.8020364993306531,
+      "learning_rate": 2.5241559374796786e-06,
+      "loss": 0.085,
+      "step": 2370
+    },
+    {
+      "epoch": 1.8969648562300319,
+      "grad_norm": 0.8025355343637186,
+      "learning_rate": 2.5085458946214414e-06,
+      "loss": 0.082,
+      "step": 2375
+    },
+    {
+      "epoch": 1.900958466453674,
+      "grad_norm": 0.8669003546070674,
+      "learning_rate": 2.4929572647962543e-06,
+      "loss": 0.0889,
+      "step": 2380
+    },
+    {
+      "epoch": 1.9049520766773163,
+      "grad_norm": 0.8325372820796868,
+      "learning_rate": 2.477390384681166e-06,
+      "loss": 0.0863,
+      "step": 2385
+    },
+    {
+      "epoch": 1.9089456869009584,
+      "grad_norm": 0.8158237511694976,
+      "learning_rate": 2.4618455904834918e-06,
+      "loss": 0.0821,
+      "step": 2390
+    },
+    {
+      "epoch": 1.9129392971246006,
+      "grad_norm": 0.7873429759802417,
+      "learning_rate": 2.446323217933536e-06,
+      "loss": 0.0831,
+      "step": 2395
+    },
+    {
+      "epoch": 1.9169329073482428,
+      "grad_norm": 0.8243924857344042,
+      "learning_rate": 2.4308236022773507e-06,
+      "loss": 0.0882,
+      "step": 2400
+    },
+    {
+      "epoch": 1.920926517571885,
+      "grad_norm": 0.8133568397024189,
+      "learning_rate": 2.415347078269501e-06,
+      "loss": 0.0906,
+      "step": 2405
+    },
+    {
+      "epoch": 1.9249201277955272,
+      "grad_norm": 0.8850847431900986,
+      "learning_rate": 2.3998939801658167e-06,
+      "loss": 0.0866,
+      "step": 2410
+    },
+    {
+      "epoch": 1.9289137380191694,
+      "grad_norm": 0.8265464069503446,
+      "learning_rate": 2.3844646417161936e-06,
+      "loss": 0.0893,
+      "step": 2415
+    },
+    {
+      "epoch": 1.9329073482428116,
+      "grad_norm": 0.7667422944606539,
+      "learning_rate": 2.3690593961573685e-06,
+      "loss": 0.0817,
+      "step": 2420
+    },
+    {
+      "epoch": 1.9369009584664538,
+      "grad_norm": 0.7912825165955961,
+      "learning_rate": 2.353678576205736e-06,
+      "loss": 0.0892,
+      "step": 2425
+    },
+    {
+      "epoch": 1.9408945686900958,
+      "grad_norm": 0.8354304573744004,
+      "learning_rate": 2.3383225140501514e-06,
+      "loss": 0.0887,
+      "step": 2430
+    },
+    {
+      "epoch": 1.944888178913738,
+      "grad_norm": 0.8061340310847699,
+      "learning_rate": 2.3229915413447616e-06,
+      "loss": 0.0852,
+      "step": 2435
+    },
+    {
+      "epoch": 1.9488817891373802,
+      "grad_norm": 0.8450228338142728,
+      "learning_rate": 2.3076859892018444e-06,
+      "loss": 0.0856,
+      "step": 2440
+    },
+    {
+      "epoch": 1.9528753993610224,
+      "grad_norm": 0.8337145461940115,
+      "learning_rate": 2.2924061881846505e-06,
+      "loss": 0.0887,
+      "step": 2445
+    },
+    {
+      "epoch": 1.9568690095846646,
+      "grad_norm": 0.803788343960733,
+      "learning_rate": 2.277152468300272e-06,
+      "loss": 0.0847,
+      "step": 2450
+    },
+    {
+      "epoch": 1.9608626198083066,
+      "grad_norm": 0.8037804153473037,
+      "learning_rate": 2.2619251589925065e-06,
+      "loss": 0.0862,
+      "step": 2455
+    },
+    {
+      "epoch": 1.9648562300319488,
+      "grad_norm": 0.7921163159021665,
+      "learning_rate": 2.2467245891347513e-06,
+      "loss": 0.0867,
+      "step": 2460
+    },
+    {
+      "epoch": 1.968849840255591,
+      "grad_norm": 0.8492215680100746,
+      "learning_rate": 2.2315510870228916e-06,
+      "loss": 0.0923,
+      "step": 2465
+    },
+    {
+      "epoch": 1.9728434504792332,
+      "grad_norm": 0.8033174401260257,
+      "learning_rate": 2.2164049803682157e-06,
+      "loss": 0.086,
+      "step": 2470
+    },
+    {
+      "epoch": 1.9768370607028753,
+      "grad_norm": 0.8216004654984708,
+      "learning_rate": 2.2012865962903367e-06,
+      "loss": 0.0858,
+      "step": 2475
+    },
+    {
+      "epoch": 1.9808306709265175,
+      "grad_norm": 0.7388468150996138,
+      "learning_rate": 2.1861962613101236e-06,
+      "loss": 0.0858,
+      "step": 2480
+    },
+    {
+      "epoch": 1.9848242811501597,
+      "grad_norm": 0.7699033555586355,
+      "learning_rate": 2.1711343013426562e-06,
+      "loss": 0.0843,
+      "step": 2485
+    },
+    {
+      "epoch": 1.988817891373802,
+      "grad_norm": 0.8060166372469154,
+      "learning_rate": 2.156101041690178e-06,
+      "loss": 0.0861,
+      "step": 2490
+    },
+    {
+      "epoch": 1.9928115015974441,
+      "grad_norm": 0.7929032945553157,
+      "learning_rate": 2.1410968070350774e-06,
+      "loss": 0.0831,
+      "step": 2495
+    },
+    {
+      "epoch": 1.9968051118210863,
+      "grad_norm": 0.7490472268146842,
+      "learning_rate": 2.1261219214328716e-06,
+      "loss": 0.0844,
+      "step": 2500
+    },
+    {
+      "epoch": 1.9968051118210863,
+      "eval_loss": 0.13146671652793884,
+      "eval_runtime": 388.8053,
+      "eval_samples_per_second": 45.799,
+      "eval_steps_per_second": 5.725,
+      "step": 2500
+    },
+    {
+      "epoch": 2.0007987220447285,
+      "grad_norm": 0.5590440556683411,
+      "learning_rate": 2.111176708305206e-06,
+      "loss": 0.0823,
+      "step": 2505
+    },
+    {
+      "epoch": 2.0047923322683707,
+      "grad_norm": 0.5779473730088169,
+      "learning_rate": 2.096261490432876e-06,
+      "loss": 0.0463,
+      "step": 2510
+    },
+    {
+      "epoch": 2.008785942492013,
+      "grad_norm": 0.5866488405628578,
+      "learning_rate": 2.0813765899488486e-06,
+      "loss": 0.0451,
+      "step": 2515
+    },
+    {
+      "epoch": 2.012779552715655,
+      "grad_norm": 0.6677470451681199,
+      "learning_rate": 2.066522328331308e-06,
+      "loss": 0.0446,
+      "step": 2520
+    },
+    {
+      "epoch": 2.0167731629392973,
+      "grad_norm": 0.6989167945663965,
+      "learning_rate": 2.051699026396713e-06,
+      "loss": 0.0434,
+      "step": 2525
+    },
+    {
+      "epoch": 2.0207667731629395,
+      "grad_norm": 0.7661180883841344,
+      "learning_rate": 2.036907004292868e-06,
+      "loss": 0.0432,
+      "step": 2530
+    },
+    {
+      "epoch": 2.0247603833865813,
+      "grad_norm": 0.8024449864690092,
+      "learning_rate": 2.0221465814920067e-06,
+      "loss": 0.0424,
+      "step": 2535
+    },
+    {
+      "epoch": 2.0287539936102235,
+      "grad_norm": 0.7501816123790741,
+      "learning_rate": 2.007418076783894e-06,
+      "loss": 0.0439,
+      "step": 2540
+    },
+    {
+      "epoch": 2.0327476038338657,
+      "grad_norm": 0.6877745159716515,
+      "learning_rate": 1.99272180826894e-06,
+      "loss": 0.04,
+      "step": 2545
+    },
+    {
+      "epoch": 2.036741214057508,
+      "grad_norm": 0.6608397420428562,
+      "learning_rate": 1.9780580933513303e-06,
+      "loss": 0.0422,
+      "step": 2550
+    },
+    {
+      "epoch": 2.04073482428115,
+      "grad_norm": 0.6603272146294105,
+      "learning_rate": 1.9634272487321734e-06,
+      "loss": 0.043,
+      "step": 2555
+    },
+    {
+      "epoch": 2.0447284345047922,
+      "grad_norm": 0.7257523609250439,
+      "learning_rate": 1.9488295904026556e-06,
+      "loss": 0.0439,
+      "step": 2560
+    },
+    {
+      "epoch": 2.0487220447284344,
+      "grad_norm": 0.6619441686023483,
+      "learning_rate": 1.9342654336372184e-06,
+      "loss": 0.041,
+      "step": 2565
+    },
+    {
+      "epoch": 2.0527156549520766,
+      "grad_norm": 0.775856056665606,
+      "learning_rate": 1.9197350929867496e-06,
+      "loss": 0.0422,
+      "step": 2570
+    },
+    {
+      "epoch": 2.056709265175719,
+      "grad_norm": 0.71946656984334,
+      "learning_rate": 1.905238882271794e-06,
+      "loss": 0.0403,
+      "step": 2575
+    },
+    {
+      "epoch": 2.060702875399361,
+      "grad_norm": 0.8286391951732616,
+      "learning_rate": 1.8907771145757672e-06,
+      "loss": 0.0453,
+      "step": 2580
+    },
+    {
+      "epoch": 2.0646964856230032,
+      "grad_norm": 0.6701516033428961,
+      "learning_rate": 1.8763501022381988e-06,
+      "loss": 0.0432,
+      "step": 2585
+    },
+    {
+      "epoch": 2.0686900958466454,
+      "grad_norm": 0.6951719440989161,
+      "learning_rate": 1.861958156847988e-06,
+      "loss": 0.0401,
+      "step": 2590
+    },
+    {
+      "epoch": 2.0726837060702876,
+      "grad_norm": 0.7071648587283893,
+      "learning_rate": 1.8476015892366678e-06,
+      "loss": 0.0422,
+      "step": 2595
+    },
+    {
+      "epoch": 2.07667731629393,
+      "grad_norm": 0.7088715440583857,
+      "learning_rate": 1.8332807094717024e-06,
+      "loss": 0.0427,
+      "step": 2600
+    },
+    {
+      "epoch": 2.080670926517572,
+      "grad_norm": 0.6796726953293041,
+      "learning_rate": 1.8189958268497788e-06,
+      "loss": 0.0398,
+      "step": 2605
+    },
+    {
+      "epoch": 2.084664536741214,
+      "grad_norm": 0.7275520062937834,
+      "learning_rate": 1.8047472498901338e-06,
+      "loss": 0.0437,
+      "step": 2610
+    },
+    {
+      "epoch": 2.0886581469648564,
+      "grad_norm": 0.6983089859377336,
+      "learning_rate": 1.7905352863278885e-06,
+      "loss": 0.0423,
+      "step": 2615
+    },
+    {
+      "epoch": 2.0926517571884986,
+      "grad_norm": 0.749372173179819,
+      "learning_rate": 1.776360243107401e-06,
+      "loss": 0.0426,
+      "step": 2620
+    },
+    {
+      "epoch": 2.0966453674121404,
+      "grad_norm": 0.6497883844703632,
+      "learning_rate": 1.7622224263756437e-06,
+      "loss": 0.0414,
+      "step": 2625
+    },
+    {
+      "epoch": 2.1006389776357826,
+      "grad_norm": 0.6784997898192929,
+      "learning_rate": 1.7481221414755777e-06,
+      "loss": 0.0424,
+      "step": 2630
+    },
+    {
+      "epoch": 2.1046325878594248,
+      "grad_norm": 0.7187277256802926,
+      "learning_rate": 1.7340596929395748e-06,
+      "loss": 0.0433,
+      "step": 2635
+    },
+    {
+      "epoch": 2.108626198083067,
+      "grad_norm": 0.7412210732229237,
+      "learning_rate": 1.7200353844828254e-06,
+      "loss": 0.0419,
+      "step": 2640
+    },
+    {
+      "epoch": 2.112619808306709,
+      "grad_norm": 0.6639717542375545,
+      "learning_rate": 1.706049518996793e-06,
+      "loss": 0.0403,
+      "step": 2645
+    },
+    {
+      "epoch": 2.1166134185303513,
+      "grad_norm": 0.8077179697601087,
+      "learning_rate": 1.6921023985426556e-06,
+      "loss": 0.0449,
+      "step": 2650
+    },
+    {
+      "epoch": 2.1206070287539935,
+      "grad_norm": 0.7020946900796683,
+      "learning_rate": 1.6781943243447969e-06,
+      "loss": 0.043,
+      "step": 2655
+    },
+    {
+      "epoch": 2.1246006389776357,
+      "grad_norm": 0.7157899672728614,
+      "learning_rate": 1.664325596784297e-06,
+      "loss": 0.0416,
+      "step": 2660
+    },
+    {
+      "epoch": 2.128594249201278,
+      "grad_norm": 0.6802103945208559,
+      "learning_rate": 1.650496515392439e-06,
+      "loss": 0.0399,
+      "step": 2665
+    },
+    {
+      "epoch": 2.13258785942492,
+      "grad_norm": 0.6278491083549733,
+      "learning_rate": 1.6367073788442447e-06,
+      "loss": 0.037,
+      "step": 2670
+    },
+    {
+      "epoch": 2.1365814696485623,
+      "grad_norm": 0.7345746287621161,
+      "learning_rate": 1.6229584849520226e-06,
+      "loss": 0.0407,
+      "step": 2675
+    },
+    {
+      "epoch": 2.1405750798722045,
+      "grad_norm": 0.7382168583442479,
+      "learning_rate": 1.6092501306589388e-06,
+      "loss": 0.0421,
+      "step": 2680
+    },
+    {
+      "epoch": 2.1445686900958467,
+      "grad_norm": 1.0670796847742963,
+      "learning_rate": 1.5955826120325985e-06,
+      "loss": 0.0418,
+      "step": 2685
+    },
+    {
+      "epoch": 2.148562300319489,
+      "grad_norm": 0.748357154888744,
+      "learning_rate": 1.5819562242586558e-06,
+      "loss": 0.042,
+      "step": 2690
+    },
+    {
+      "epoch": 2.152555910543131,
+      "grad_norm": 0.6924925720927515,
+      "learning_rate": 1.5683712616344354e-06,
+      "loss": 0.0413,
+      "step": 2695
+    },
+    {
+      "epoch": 2.1565495207667733,
+      "grad_norm": 0.7321556671315872,
+      "learning_rate": 1.5548280175625781e-06,
+      "loss": 0.0427,
+      "step": 2700
+    },
+    {
+      "epoch": 2.1605431309904155,
+      "grad_norm": 0.7345504259937425,
+      "learning_rate": 1.5413267845447073e-06,
+      "loss": 0.0407,
+      "step": 2705
+    },
+    {
+      "epoch": 2.1645367412140577,
+      "grad_norm": 0.7275376645175038,
+      "learning_rate": 1.527867854175105e-06,
+      "loss": 0.0404,
+      "step": 2710
+    },
+    {
+      "epoch": 2.1685303514377,
+      "grad_norm": 0.7841865628062479,
+      "learning_rate": 1.5144515171344174e-06,
+      "loss": 0.0422,
+      "step": 2715
+    },
+    {
+      "epoch": 2.1725239616613417,
+      "grad_norm": 0.7399131108932046,
+      "learning_rate": 1.5010780631833782e-06,
+      "loss": 0.0396,
+      "step": 2720
+    },
+    {
+      "epoch": 2.176517571884984,
+      "grad_norm": 0.7364100768910417,
+      "learning_rate": 1.4877477811565473e-06,
+      "loss": 0.042,
+      "step": 2725
+    },
+    {
+      "epoch": 2.180511182108626,
+      "grad_norm": 0.7009848783954564,
+      "learning_rate": 1.474460958956078e-06,
+      "loss": 0.0427,
+      "step": 2730
+    },
+    {
+      "epoch": 2.1845047923322682,
+      "grad_norm": 0.6692563631050057,
+      "learning_rate": 1.461217883545492e-06,
+      "loss": 0.0386,
+      "step": 2735
+    },
+    {
+      "epoch": 2.1884984025559104,
+      "grad_norm": 0.7198493387599846,
+      "learning_rate": 1.4480188409434866e-06,
+      "loss": 0.0405,
+      "step": 2740
+    },
+    {
+      "epoch": 2.1924920127795526,
+      "grad_norm": 0.7159870701360656,
+      "learning_rate": 1.4348641162177543e-06,
+      "loss": 0.0421,
+      "step": 2745
+    },
+    {
+      "epoch": 2.196485623003195,
+      "grad_norm": 0.6323328853392578,
+      "learning_rate": 1.4217539934788313e-06,
+      "loss": 0.0379,
+      "step": 2750
+    },
+    {
+      "epoch": 2.200479233226837,
+      "grad_norm": 0.716152180482688,
+      "learning_rate": 1.4086887558739545e-06,
+      "loss": 0.0399,
+      "step": 2755
+    },
+    {
+      "epoch": 2.2044728434504792,
+      "grad_norm": 0.7209751580895504,
+      "learning_rate": 1.395668685580951e-06,
+      "loss": 0.0399,
+      "step": 2760
+    },
+    {
+      "epoch": 2.2084664536741214,
+      "grad_norm": 0.6976023824417826,
+      "learning_rate": 1.3826940638021407e-06,
+      "loss": 0.0405,
+      "step": 2765
+    },
+    {
+      "epoch": 2.2124600638977636,
+      "grad_norm": 0.676003189556891,
+      "learning_rate": 1.369765170758265e-06,
+      "loss": 0.0405,
+      "step": 2770
+    },
+    {
+      "epoch": 2.216453674121406,
+      "grad_norm": 0.6769501596871886,
+      "learning_rate": 1.3568822856824374e-06,
+      "loss": 0.04,
+      "step": 2775
+    },
+    {
+      "epoch": 2.220447284345048,
+      "grad_norm": 0.757978868906908,
+      "learning_rate": 1.3440456868141044e-06,
+      "loss": 0.0427,
+      "step": 2780
+    },
+    {
+      "epoch": 2.22444089456869,
+      "grad_norm": 0.7268819944292451,
+      "learning_rate": 1.3312556513930447e-06,
+      "loss": 0.0421,
+      "step": 2785
+    },
+    {
+      "epoch": 2.2284345047923324,
+      "grad_norm": 0.7714098456385522,
+      "learning_rate": 1.3185124556533755e-06,
+      "loss": 0.0417,
+      "step": 2790
+    },
+    {
+      "epoch": 2.2324281150159746,
+      "grad_norm": 0.7430135294254091,
+      "learning_rate": 1.3058163748175932e-06,
+      "loss": 0.0389,
+      "step": 2795
+    },
+    {
+      "epoch": 2.236421725239617,
+      "grad_norm": 0.7214319576784142,
+      "learning_rate": 1.2931676830906231e-06,
+      "loss": 0.0386,
+      "step": 2800
+    },
+    {
+      "epoch": 2.2404153354632586,
+      "grad_norm": 0.7127065991787688,
+      "learning_rate": 1.280566653653895e-06,
+      "loss": 0.0378,
+      "step": 2805
+    },
+    {
+      "epoch": 2.244408945686901,
+      "grad_norm": 0.6839690662193432,
+      "learning_rate": 1.2680135586594546e-06,
+      "loss": 0.0406,
+      "step": 2810
+    },
+    {
+      "epoch": 2.248402555910543,
+      "grad_norm": 0.7523843789217508,
+      "learning_rate": 1.2555086692240736e-06,
+      "loss": 0.0391,
+      "step": 2815
+    },
+    {
+      "epoch": 2.252396166134185,
+      "grad_norm": 0.6885045467953661,
+      "learning_rate": 1.2430522554234037e-06,
+      "loss": 0.0416,
+      "step": 2820
+    },
+    {
+      "epoch": 2.2563897763578273,
+      "grad_norm": 0.702918819197609,
+      "learning_rate": 1.2306445862861316e-06,
+      "loss": 0.0403,
+      "step": 2825
+    },
+    {
+      "epoch": 2.2603833865814695,
+      "grad_norm": 0.7187953136824686,
+      "learning_rate": 1.2182859297881844e-06,
+      "loss": 0.0387,
+      "step": 2830
+    },
+    {
+      "epoch": 2.2643769968051117,
+      "grad_norm": 0.7261350898101718,
+      "learning_rate": 1.2059765528469286e-06,
+      "loss": 0.0386,
+      "step": 2835
+    },
+    {
+      "epoch": 2.268370607028754,
+      "grad_norm": 0.8144574975767674,
+      "learning_rate": 1.193716721315411e-06,
+      "loss": 0.0407,
+      "step": 2840
+    },
+    {
+      "epoch": 2.272364217252396,
+      "grad_norm": 0.719536955678268,
+      "learning_rate": 1.1815066999766174e-06,
+      "loss": 0.0405,
+      "step": 2845
+    },
+    {
+      "epoch": 2.2763578274760383,
+      "grad_norm": 0.7645179459629033,
+      "learning_rate": 1.1693467525377513e-06,
+      "loss": 0.0405,
+      "step": 2850
+    },
+    {
+      "epoch": 2.2803514376996805,
+      "grad_norm": 0.7168125333161555,
+      "learning_rate": 1.1572371416245424e-06,
+      "loss": 0.041,
+      "step": 2855
+    },
+    {
+      "epoch": 2.2843450479233227,
+      "grad_norm": 0.6809214841238408,
+      "learning_rate": 1.14517812877557e-06,
+      "loss": 0.0372,
+      "step": 2860
+    },
+    {
+      "epoch": 2.288338658146965,
+      "grad_norm": 0.7113403532991136,
+      "learning_rate": 1.1331699744366168e-06,
+      "loss": 0.0419,
+      "step": 2865
+    },
+    {
+      "epoch": 2.292332268370607,
+      "grad_norm": 0.7363573736512515,
+      "learning_rate": 1.1212129379550432e-06,
+      "loss": 0.0401,
+      "step": 2870
+    },
+    {
+      "epoch": 2.2963258785942493,
+      "grad_norm": 0.7386118577478687,
+      "learning_rate": 1.109307277574187e-06,
+      "loss": 0.0413,
+      "step": 2875
+    },
+    {
+      "epoch": 2.3003194888178915,
+      "grad_norm": 0.727683344048625,
+      "learning_rate": 1.0974532504277863e-06,
+      "loss": 0.042,
+      "step": 2880
+    },
+    {
+      "epoch": 2.3043130990415337,
+      "grad_norm": 0.7386228312374142,
+      "learning_rate": 1.0856511125344233e-06,
+      "loss": 0.0407,
+      "step": 2885
+    },
+    {
+      "epoch": 2.308306709265176,
+      "grad_norm": 0.7140159548965673,
+      "learning_rate": 1.0739011187919976e-06,
+      "loss": 0.0408,
+      "step": 2890
+    },
+    {
+      "epoch": 2.312300319488818,
+      "grad_norm": 0.7497153823378125,
+      "learning_rate": 1.0622035229722183e-06,
+      "loss": 0.0418,
+      "step": 2895
+    },
+    {
+      "epoch": 2.31629392971246,
+      "grad_norm": 0.7001040487899245,
+      "learning_rate": 1.0505585777151287e-06,
+      "loss": 0.042,
+      "step": 2900
+    },
+    {
+      "epoch": 2.3202875399361025,
+      "grad_norm": 0.6894355300080932,
+      "learning_rate": 1.038966534523643e-06,
+      "loss": 0.0392,
+      "step": 2905
+    },
+    {
+      "epoch": 2.3242811501597442,
+      "grad_norm": 0.6990435544277329,
+      "learning_rate": 1.0274276437581183e-06,
+      "loss": 0.0381,
+      "step": 2910
+    },
+    {
+      "epoch": 2.3282747603833864,
+      "grad_norm": 0.6909814467001467,
+      "learning_rate": 1.0159421546309467e-06,
+      "loss": 0.0406,
+      "step": 2915
+    },
+    {
+      "epoch": 2.3322683706070286,
+      "grad_norm": 0.7109244643358452,
+      "learning_rate": 1.004510315201172e-06,
+      "loss": 0.041,
+      "step": 2920
+    },
+    {
+      "epoch": 2.336261980830671,
+      "grad_norm": 0.6878258243345245,
+      "learning_rate": 9.931323723691355e-07,
+      "loss": 0.0408,
+      "step": 2925
+    },
+    {
+      "epoch": 2.340255591054313,
+      "grad_norm": 0.6756279917857075,
+      "learning_rate": 9.81808571871139e-07,
+      "loss": 0.0402,
+      "step": 2930
+    },
+    {
+      "epoch": 2.344249201277955,
+      "grad_norm": 0.749467435614857,
+      "learning_rate": 9.70539158274141e-07,
+      "loss": 0.041,
+      "step": 2935
+    },
+    {
+      "epoch": 2.3482428115015974,
+      "grad_norm": 0.752336563070383,
+      "learning_rate": 9.593243749704702e-07,
+      "loss": 0.0421,
+      "step": 2940
+    },
+    {
+      "epoch": 2.3522364217252396,
+      "grad_norm": 0.7364168174436757,
+      "learning_rate": 9.481644641725768e-07,
+      "loss": 0.0413,
+      "step": 2945
+    },
+    {
+      "epoch": 2.356230031948882,
+      "grad_norm": 0.7454044222548492,
+      "learning_rate": 9.370596669077937e-07,
+      "loss": 0.0401,
+      "step": 2950
+    },
+    {
+      "epoch": 2.360223642172524,
+      "grad_norm": 0.6962574991902063,
+      "learning_rate": 9.260102230131306e-07,
+      "loss": 0.039,
+      "step": 2955
+    },
+    {
+      "epoch": 2.364217252396166,
+      "grad_norm": 0.7110768770465447,
+      "learning_rate": 9.150163711301031e-07,
+      "loss": 0.0374,
+      "step": 2960
+    },
+    {
+      "epoch": 2.3682108626198084,
+      "grad_norm": 0.7757468444639659,
+      "learning_rate": 9.040783486995669e-07,
+      "loss": 0.0386,
+      "step": 2965
+    },
+    {
+      "epoch": 2.3722044728434506,
+      "grad_norm": 0.6850534935139829,
+      "learning_rate": 8.931963919566015e-07,
+      "loss": 0.0391,
+      "step": 2970
+    },
+    {
+      "epoch": 2.376198083067093,
+      "grad_norm": 0.7275846393994331,
+      "learning_rate": 8.823707359253941e-07,
+      "loss": 0.0404,
+      "step": 2975
+    },
+    {
+      "epoch": 2.380191693290735,
+      "grad_norm": 0.725401060190788,
+      "learning_rate": 8.716016144141784e-07,
+      "loss": 0.0392,
+      "step": 2980
+    },
+    {
+      "epoch": 2.384185303514377,
+      "grad_norm": 0.719311654275402,
+      "learning_rate": 8.608892600101741e-07,
+      "loss": 0.0408,
+      "step": 2985
+    },
+    {
+      "epoch": 2.3881789137380194,
+      "grad_norm": 0.6867331873092225,
+      "learning_rate": 8.502339040745675e-07,
+      "loss": 0.0385,
+      "step": 2990
+    },
+    {
+      "epoch": 2.392172523961661,
+      "grad_norm": 0.7395697453495743,
+      "learning_rate": 8.396357767375185e-07,
+      "loss": 0.041,
+      "step": 2995
+    },
+    {
+      "epoch": 2.3961661341853033,
+      "grad_norm": 0.7118596021755634,
+      "learning_rate": 8.290951068931795e-07,
+      "loss": 0.0418,
+      "step": 3000
+    },
+    {
+      "epoch": 2.3961661341853033,
+      "eval_loss": 0.16109129786491394,
+      "eval_runtime": 388.1385,
+      "eval_samples_per_second": 45.878,
+      "eval_steps_per_second": 5.735,
+      "step": 3000
+    },
+    {
+      "epoch": 2.4001597444089455,
+      "grad_norm": 0.6579880169340647,
+      "learning_rate": 8.186121221947648e-07,
+      "loss": 0.0412,
+      "step": 3005
+    },
+    {
+      "epoch": 2.4041533546325877,
+      "grad_norm": 0.6563989234205505,
+      "learning_rate": 8.081870490496246e-07,
+      "loss": 0.0392,
+      "step": 3010
+    },
+    {
+      "epoch": 2.40814696485623,
+      "grad_norm": 0.6782118306828682,
+      "learning_rate": 7.978201126143619e-07,
+      "loss": 0.0405,
+      "step": 3015
+    },
+    {
+      "epoch": 2.412140575079872,
+      "grad_norm": 0.7287775052219024,
+      "learning_rate": 7.875115367899602e-07,
+      "loss": 0.0404,
+      "step": 3020
+    },
+    {
+      "epoch": 2.4161341853035143,
+      "grad_norm": 0.6939389857285485,
+      "learning_rate": 7.772615442169569e-07,
+      "loss": 0.0385,
+      "step": 3025
+    },
+    {
+      "epoch": 2.4201277955271565,
+      "grad_norm": 0.6771579465197793,
+      "learning_rate": 7.670703562706338e-07,
+      "loss": 0.0391,
+      "step": 3030
+    },
+    {
+      "epoch": 2.4241214057507987,
+      "grad_norm": 0.7805400199400504,
+      "learning_rate": 7.569381930562301e-07,
+      "loss": 0.0404,
+      "step": 3035
+    },
+    {
+      "epoch": 2.428115015974441,
+      "grad_norm": 0.7430898082732748,
+      "learning_rate": 7.468652734041938e-07,
+      "loss": 0.0384,
+      "step": 3040
+    },
+    {
+      "epoch": 2.432108626198083,
+      "grad_norm": 0.7030712092404906,
+      "learning_rate": 7.368518148654524e-07,
+      "loss": 0.0399,
+      "step": 3045
+    },
+    {
+      "epoch": 2.4361022364217253,
+      "grad_norm": 0.6593873003197812,
+      "learning_rate": 7.268980337067193e-07,
+      "loss": 0.0387,
+      "step": 3050
+    },
+    {
+      "epoch": 2.4400958466453675,
+      "grad_norm": 0.7180247154760864,
+      "learning_rate": 7.17004144905816e-07,
+      "loss": 0.04,
+      "step": 3055
+    },
+    {
+      "epoch": 2.4440894568690097,
+      "grad_norm": 0.6917266349878818,
+      "learning_rate": 7.07170362147034e-07,
+      "loss": 0.0388,
+      "step": 3060
+    },
+    {
+      "epoch": 2.448083067092652,
+      "grad_norm": 0.7191478188087486,
+      "learning_rate": 6.973968978165176e-07,
+      "loss": 0.0388,
+      "step": 3065
+    },
+    {
+      "epoch": 2.452076677316294,
+      "grad_norm": 0.7285189143803087,
+      "learning_rate": 6.876839629976778e-07,
+      "loss": 0.0386,
+      "step": 3070
+    },
+    {
+      "epoch": 2.4560702875399363,
+      "grad_norm": 0.6602961842830507,
+      "learning_rate": 6.78031767466635e-07,
+      "loss": 0.039,
+      "step": 3075
+    },
+    {
+      "epoch": 2.460063897763578,
+      "grad_norm": 0.7397023360105911,
+      "learning_rate": 6.684405196876843e-07,
+      "loss": 0.0397,
+      "step": 3080
+    },
+    {
+      "epoch": 2.4640575079872207,
+      "grad_norm": 0.6184869436420118,
+      "learning_rate": 6.589104268087958e-07,
+      "loss": 0.0365,
+      "step": 3085
+    },
+    {
+      "epoch": 2.4680511182108624,
+      "grad_norm": 0.7393488490012594,
+      "learning_rate": 6.494416946571408e-07,
+      "loss": 0.0389,
+      "step": 3090
+    },
+    {
+      "epoch": 2.4720447284345046,
+      "grad_norm": 0.7764250937534626,
+      "learning_rate": 6.400345277346447e-07,
+      "loss": 0.038,
+      "step": 3095
+    },
+    {
+      "epoch": 2.476038338658147,
+      "grad_norm": 0.6463875407701625,
+      "learning_rate": 6.306891292135743e-07,
+      "loss": 0.0413,
+      "step": 3100
+    },
+    {
+      "epoch": 2.480031948881789,
+      "grad_norm": 0.7320022575736014,
+      "learning_rate": 6.21405700932144e-07,
+      "loss": 0.0375,
+      "step": 3105
+    },
+    {
+      "epoch": 2.484025559105431,
+      "grad_norm": 0.6818592089753043,
+      "learning_rate": 6.121844433901612e-07,
+      "loss": 0.0386,
+      "step": 3110
+    },
+    {
+      "epoch": 2.4880191693290734,
+      "grad_norm": 0.6923239557377695,
+      "learning_rate": 6.030255557446922e-07,
+      "loss": 0.038,
+      "step": 3115
+    },
+    {
+      "epoch": 2.4920127795527156,
+      "grad_norm": 0.7225836003409724,
+      "learning_rate": 5.939292358057669e-07,
+      "loss": 0.0401,
+      "step": 3120
+    },
+    {
+      "epoch": 2.496006389776358,
+      "grad_norm": 0.7224726921229143,
+      "learning_rate": 5.848956800320994e-07,
+      "loss": 0.0396,
+      "step": 3125
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 0.7730383752980494,
+      "learning_rate": 5.75925083526847e-07,
+      "loss": 0.0403,
+      "step": 3130
+    },
+    {
+      "epoch": 2.503993610223642,
+      "grad_norm": 0.7211592353668832,
+      "learning_rate": 5.670176400334023e-07,
+      "loss": 0.0381,
+      "step": 3135
+    },
+    {
+      "epoch": 2.5079872204472844,
+      "grad_norm": 0.7472591071933815,
+      "learning_rate": 5.581735419312e-07,
+      "loss": 0.0401,
+      "step": 3140
+    },
+    {
+      "epoch": 2.5119808306709266,
+      "grad_norm": 0.6705306160257385,
+      "learning_rate": 5.493929802315701e-07,
+      "loss": 0.0404,
+      "step": 3145
+    },
+    {
+      "epoch": 2.515974440894569,
+      "grad_norm": 0.744340088134437,
+      "learning_rate": 5.406761445736025e-07,
+      "loss": 0.0372,
+      "step": 3150
+    },
+    {
+      "epoch": 2.519968051118211,
+      "grad_norm": 0.7438732345302391,
+      "learning_rate": 5.320232232200639e-07,
+      "loss": 0.0411,
+      "step": 3155
+    },
+    {
+      "epoch": 2.523961661341853,
+      "grad_norm": 0.7988995555141504,
+      "learning_rate": 5.234344030533204e-07,
+      "loss": 0.038,
+      "step": 3160
+    },
+    {
+      "epoch": 2.527955271565495,
+      "grad_norm": 0.6711977319305636,
+      "learning_rate": 5.149098695713102e-07,
+      "loss": 0.0352,
+      "step": 3165
+    },
+    {
+      "epoch": 2.5319488817891376,
+      "grad_norm": 0.6847061252244899,
+      "learning_rate": 5.064498068835288e-07,
+      "loss": 0.0388,
+      "step": 3170
+    },
+    {
+      "epoch": 2.5359424920127793,
+      "grad_norm": 0.6841997385045966,
+      "learning_rate": 4.980543977070587e-07,
+      "loss": 0.0385,
+      "step": 3175
+    },
+    {
+      "epoch": 2.539936102236422,
+      "grad_norm": 0.7373519760893469,
+      "learning_rate": 4.897238233626237e-07,
+      "loss": 0.0397,
+      "step": 3180
+    },
+    {
+      "epoch": 2.5439297124600637,
+      "grad_norm": 0.7612391951526681,
+      "learning_rate": 4.814582637706676e-07,
+      "loss": 0.0406,
+      "step": 3185
+    },
+    {
+      "epoch": 2.547923322683706,
+      "grad_norm": 0.6691221826001316,
+      "learning_rate": 4.732578974474729e-07,
+      "loss": 0.0373,
+      "step": 3190
+    },
+    {
+      "epoch": 2.551916932907348,
+      "grad_norm": 0.7504326919914636,
+      "learning_rate": 4.651229015013024e-07,
+      "loss": 0.0395,
+      "step": 3195
+    },
+    {
+      "epoch": 2.5559105431309903,
+      "grad_norm": 0.728197142532798,
+      "learning_rate": 4.5705345162857707e-07,
+      "loss": 0.0404,
+      "step": 3200
+    },
+    {
+      "epoch": 2.5599041533546325,
+      "grad_norm": 0.7638750317365375,
+      "learning_rate": 4.4904972211007877e-07,
+      "loss": 0.0365,
+      "step": 3205
+    },
+    {
+      "epoch": 2.5638977635782747,
+      "grad_norm": 0.6786482009733499,
+      "learning_rate": 4.4111188580718737e-07,
+      "loss": 0.0381,
+      "step": 3210
+    },
+    {
+      "epoch": 2.567891373801917,
+      "grad_norm": 0.6750362855329872,
+      "learning_rate": 4.332401141581471e-07,
+      "loss": 0.0384,
+      "step": 3215
+    },
+    {
+      "epoch": 2.571884984025559,
+      "grad_norm": 0.6434774711250648,
+      "learning_rate": 4.2543457717436274e-07,
+      "loss": 0.0389,
+      "step": 3220
+    },
+    {
+      "epoch": 2.5758785942492013,
+      "grad_norm": 0.678657018918754,
+      "learning_rate": 4.1769544343673265e-07,
+      "loss": 0.0387,
+      "step": 3225
+    },
+    {
+      "epoch": 2.5798722044728435,
+      "grad_norm": 0.7234126311376552,
+      "learning_rate": 4.10022880092e-07,
+      "loss": 0.0374,
+      "step": 3230
+    },
+    {
+      "epoch": 2.5838658146964857,
+      "grad_norm": 0.7172155271619155,
+      "learning_rate": 4.0241705284914966e-07,
+      "loss": 0.0411,
+      "step": 3235
+    },
+    {
+      "epoch": 2.587859424920128,
+      "grad_norm": 0.698249391590756,
+      "learning_rate": 3.9487812597582513e-07,
+      "loss": 0.0408,
+      "step": 3240
+    },
+    {
+      "epoch": 2.59185303514377,
+      "grad_norm": 0.6637186291346941,
+      "learning_rate": 3.874062622947825e-07,
+      "loss": 0.0363,
+      "step": 3245
+    },
+    {
+      "epoch": 2.5958466453674123,
+      "grad_norm": 0.6576549119431471,
+      "learning_rate": 3.800016231803752e-07,
+      "loss": 0.0391,
+      "step": 3250
+    },
+    {
+      "epoch": 2.5998402555910545,
+      "grad_norm": 0.7376314683338078,
+      "learning_rate": 3.7266436855506484e-07,
+      "loss": 0.0373,
+      "step": 3255
+    },
+    {
+      "epoch": 2.6038338658146962,
+      "grad_norm": 0.715005088433797,
+      "learning_rate": 3.6539465688597107e-07,
+      "loss": 0.0374,
+      "step": 3260
+    },
+    {
+      "epoch": 2.607827476038339,
+      "grad_norm": 0.7126399642722202,
+      "learning_rate": 3.5819264518144565e-07,
+      "loss": 0.0399,
+      "step": 3265
+    },
+    {
+      "epoch": 2.6118210862619806,
+      "grad_norm": 0.7304318779489742,
+      "learning_rate": 3.5105848898768615e-07,
+      "loss": 0.0373,
+      "step": 3270
+    },
+    {
+      "epoch": 2.6158146964856233,
+      "grad_norm": 0.7507980457632606,
+      "learning_rate": 3.4399234238537254e-07,
+      "loss": 0.0393,
+      "step": 3275
+    },
+    {
+      "epoch": 2.619808306709265,
+      "grad_norm": 0.6923767027594208,
+      "learning_rate": 3.369943579863379e-07,
+      "loss": 0.0371,
+      "step": 3280
+    },
+    {
+      "epoch": 2.623801916932907,
+      "grad_norm": 0.6964206029100168,
+      "learning_rate": 3.300646869302794e-07,
+      "loss": 0.0362,
+      "step": 3285
+    },
+    {
+      "epoch": 2.6277955271565494,
+      "grad_norm": 0.7352109790702203,
+      "learning_rate": 3.2320347888148644e-07,
+      "loss": 0.0377,
+      "step": 3290
+    },
+    {
+      "epoch": 2.6317891373801916,
+      "grad_norm": 0.6628167533609087,
+      "learning_rate": 3.1641088202561373e-07,
+      "loss": 0.0379,
+      "step": 3295
+    },
+    {
+      "epoch": 2.635782747603834,
+      "grad_norm": 0.6564933786076047,
+      "learning_rate": 3.0968704306647714e-07,
+      "loss": 0.0369,
+      "step": 3300
+    },
+    {
+      "epoch": 2.639776357827476,
+      "grad_norm": 0.6308056550542149,
+      "learning_rate": 3.0303210722288666e-07,
+      "loss": 0.0346,
+      "step": 3305
+    },
+    {
+      "epoch": 2.643769968051118,
+      "grad_norm": 0.6573474228547711,
+      "learning_rate": 2.964462182255117e-07,
+      "loss": 0.0368,
+      "step": 3310
+    },
+    {
+      "epoch": 2.6477635782747604,
+      "grad_norm": 0.6896389035881784,
+      "learning_rate": 2.899295183137723e-07,
+      "loss": 0.0395,
+      "step": 3315
+    },
+    {
+      "epoch": 2.6517571884984026,
+      "grad_norm": 0.7174278292512358,
+      "learning_rate": 2.834821482327745e-07,
+      "loss": 0.0366,
+      "step": 3320
+    },
+    {
+      "epoch": 2.655750798722045,
+      "grad_norm": 0.7073005459759208,
+      "learning_rate": 2.7710424723026073e-07,
+      "loss": 0.0404,
+      "step": 3325
+    },
+    {
+      "epoch": 2.659744408945687,
+      "grad_norm": 0.7369122238419571,
+      "learning_rate": 2.7079595305361237e-07,
+      "loss": 0.0361,
+      "step": 3330
+    },
+    {
+      "epoch": 2.663738019169329,
+      "grad_norm": 0.703027115337837,
+      "learning_rate": 2.645574019468673e-07,
+      "loss": 0.0357,
+      "step": 3335
+    },
+    {
+      "epoch": 2.6677316293929714,
+      "grad_norm": 0.769356516837401,
+      "learning_rate": 2.5838872864778276e-07,
+      "loss": 0.0398,
+      "step": 3340
+    },
+    {
+      "epoch": 2.6717252396166136,
+      "grad_norm": 0.7144690912826553,
+      "learning_rate": 2.522900663849195e-07,
+      "loss": 0.035,
+      "step": 3345
+    },
+    {
+      "epoch": 2.6757188498402558,
+      "grad_norm": 0.7538629864589558,
+      "learning_rate": 2.4626154687476913e-07,
+      "loss": 0.0392,
+      "step": 3350
+    },
+    {
+      "epoch": 2.6797124600638975,
+      "grad_norm": 0.6680095163501195,
+      "learning_rate": 2.403033003189097e-07,
+      "loss": 0.0384,
+      "step": 3355
+    },
+    {
+      "epoch": 2.68370607028754,
+      "grad_norm": 0.7091669937348918,
+      "learning_rate": 2.3441545540118957e-07,
+      "loss": 0.04,
+      "step": 3360
+    },
+    {
+      "epoch": 2.687699680511182,
+      "grad_norm": 0.7387779458841153,
+      "learning_rate": 2.285981392849507e-07,
+      "loss": 0.0404,
+      "step": 3365
+    },
+    {
+      "epoch": 2.6916932907348246,
+      "grad_norm": 0.716479182444788,
+      "learning_rate": 2.2285147761028234e-07,
+      "loss": 0.037,
+      "step": 3370
+    },
+    {
+      "epoch": 2.6956869009584663,
+      "grad_norm": 0.736182154333335,
+      "learning_rate": 2.1717559449130745e-07,
+      "loss": 0.0374,
+      "step": 3375
+    },
+    {
+      "epoch": 2.6996805111821085,
+      "grad_norm": 0.7195196251858598,
+      "learning_rate": 2.115706125135016e-07,
+      "loss": 0.0403,
+      "step": 3380
+    },
+    {
+      "epoch": 2.7036741214057507,
+      "grad_norm": 0.795586487340967,
+      "learning_rate": 2.0603665273104477e-07,
+      "loss": 0.0383,
+      "step": 3385
+    },
+    {
+      "epoch": 2.707667731629393,
+      "grad_norm": 0.6816457489515708,
+      "learning_rate": 2.0057383466420887e-07,
+      "loss": 0.0362,
+      "step": 3390
+    },
+    {
+      "epoch": 2.711661341853035,
+      "grad_norm": 0.7211817535473453,
+      "learning_rate": 1.9518227629677343e-07,
+      "loss": 0.0393,
+      "step": 3395
+    },
+    {
+      "epoch": 2.7156549520766773,
+      "grad_norm": 0.6621050649016567,
+      "learning_rate": 1.898620940734823e-07,
+      "loss": 0.0387,
+      "step": 3400
+    },
+    {
+      "epoch": 2.7196485623003195,
+      "grad_norm": 0.7338961263391105,
+      "learning_rate": 1.8461340289752259e-07,
+      "loss": 0.0386,
+      "step": 3405
+    },
+    {
+      "epoch": 2.7236421725239617,
+      "grad_norm": 0.7256126254117897,
+      "learning_rate": 1.7943631612804777e-07,
+      "loss": 0.0379,
+      "step": 3410
+    },
+    {
+      "epoch": 2.727635782747604,
+      "grad_norm": 0.7117270166197374,
+      "learning_rate": 1.7433094557772664e-07,
+      "loss": 0.0384,
+      "step": 3415
+    },
+    {
+      "epoch": 2.731629392971246,
+      "grad_norm": 0.6841486868214987,
+      "learning_rate": 1.692974015103306e-07,
+      "loss": 0.0363,
+      "step": 3420
+    },
+    {
+      "epoch": 2.7356230031948883,
+      "grad_norm": 0.7446852668120804,
+      "learning_rate": 1.643357926383505e-07,
+      "loss": 0.0383,
+      "step": 3425
+    },
+    {
+      "epoch": 2.7396166134185305,
+      "grad_norm": 0.6829132700695434,
+      "learning_rate": 1.5944622612064964e-07,
+      "loss": 0.0375,
+      "step": 3430
+    },
+    {
+      "epoch": 2.7436102236421727,
+      "grad_norm": 0.6858172526376208,
+      "learning_rate": 1.5462880756014852e-07,
+      "loss": 0.0397,
+      "step": 3435
+    },
+    {
+      "epoch": 2.747603833865815,
+      "grad_norm": 0.67374596113679,
+      "learning_rate": 1.498836410015449e-07,
+      "loss": 0.0382,
+      "step": 3440
+    },
+    {
+      "epoch": 2.751597444089457,
+      "grad_norm": 0.6968820872384376,
+      "learning_rate": 1.4521082892906716e-07,
+      "loss": 0.0375,
+      "step": 3445
+    },
+    {
+      "epoch": 2.755591054313099,
+      "grad_norm": 0.7350519767986172,
+      "learning_rate": 1.4061047226425888e-07,
+      "loss": 0.0401,
+      "step": 3450
+    },
+    {
+      "epoch": 2.7595846645367414,
+      "grad_norm": 0.7301495139163381,
+      "learning_rate": 1.3608267036380177e-07,
+      "loss": 0.0381,
+      "step": 3455
+    },
+    {
+      "epoch": 2.763578274760383,
+      "grad_norm": 0.706129469994641,
+      "learning_rate": 1.3162752101736706e-07,
+      "loss": 0.0403,
+      "step": 3460
+    },
+    {
+      "epoch": 2.7675718849840254,
+      "grad_norm": 0.72296699028865,
+      "learning_rate": 1.2724512044550578e-07,
+      "loss": 0.0373,
+      "step": 3465
+    },
+    {
+      "epoch": 2.7715654952076676,
+      "grad_norm": 0.7326657651133558,
+      "learning_rate": 1.2293556329757026e-07,
+      "loss": 0.039,
+      "step": 3470
+    },
+    {
+      "epoch": 2.77555910543131,
+      "grad_norm": 0.7083978204265703,
+      "learning_rate": 1.1869894264966757e-07,
+      "loss": 0.0365,
+      "step": 3475
+    },
+    {
+      "epoch": 2.779552715654952,
+      "grad_norm": 0.6541405696956989,
+      "learning_rate": 1.1453535000265358e-07,
+      "loss": 0.0363,
+      "step": 3480
+    },
+    {
+      "epoch": 2.783546325878594,
+      "grad_norm": 0.6793468016189803,
+      "learning_rate": 1.1044487528015168e-07,
+      "loss": 0.0389,
+      "step": 3485
+    },
+    {
+      "epoch": 2.7875399361022364,
+      "grad_norm": 0.6830412517472533,
+      "learning_rate": 1.0642760682661645e-07,
+      "loss": 0.0365,
+      "step": 3490
+    },
+    {
+      "epoch": 2.7915335463258786,
+      "grad_norm": 0.7653650618048291,
+      "learning_rate": 1.0248363140541955e-07,
+      "loss": 0.0377,
+      "step": 3495
+    },
+    {
+      "epoch": 2.7955271565495208,
+      "grad_norm": 0.6495511843544473,
+      "learning_rate": 9.861303419697947e-08,
+      "loss": 0.0379,
+      "step": 3500
+    },
+    {
+      "epoch": 2.7955271565495208,
+      "eval_loss": 0.16211672127246857,
+      "eval_runtime": 388.1552,
+      "eval_samples_per_second": 45.876,
+      "eval_steps_per_second": 5.735,
+      "step": 3500
+    },
+    {
+      "epoch": 2.799520766773163,
+      "grad_norm": 0.7226650174013275,
+      "learning_rate": 9.481589879692242e-08,
+      "loss": 0.0378,
+      "step": 3505
+    },
+    {
+      "epoch": 2.803514376996805,
+      "grad_norm": 0.7151599173267984,
+      "learning_rate": 9.109230721427425e-08,
+      "loss": 0.0403,
+      "step": 3510
+    },
+    {
+      "epoch": 2.8075079872204474,
+      "grad_norm": 0.7251578235357717,
+      "learning_rate": 8.744233986969163e-08,
+      "loss": 0.0375,
+      "step": 3515
+    },
+    {
+      "epoch": 2.8115015974440896,
+      "grad_norm": 0.7887886233607367,
+      "learning_rate": 8.386607559372245e-08,
+      "loss": 0.0396,
+      "step": 3520
+    },
+    {
+      "epoch": 2.8154952076677318,
+      "grad_norm": 0.662483687379777,
+      "learning_rate": 8.036359162510726e-08,
+      "loss": 0.0381,
+      "step": 3525
+    },
+    {
+      "epoch": 2.819488817891374,
+      "grad_norm": 0.7331222659123482,
+      "learning_rate": 7.69349636091065e-08,
+      "loss": 0.0384,
+      "step": 3530
+    },
+    {
+      "epoch": 2.8234824281150157,
+      "grad_norm": 0.7383759987861803,
+      "learning_rate": 7.358026559587044e-08,
+      "loss": 0.0371,
+      "step": 3535
+    },
+    {
+      "epoch": 2.8274760383386583,
+      "grad_norm": 0.7270481431745002,
+      "learning_rate": 7.029957003883775e-08,
+      "loss": 0.0393,
+      "step": 3540
+    },
+    {
+      "epoch": 2.8314696485623,
+      "grad_norm": 0.6650945156242567,
+      "learning_rate": 6.709294779317115e-08,
+      "loss": 0.037,
+      "step": 3545
+    },
+    {
+      "epoch": 2.8354632587859427,
+      "grad_norm": 0.7401242798225801,
+      "learning_rate": 6.396046811422879e-08,
+      "loss": 0.0366,
+      "step": 3550
+    },
+    {
+      "epoch": 2.8394568690095845,
+      "grad_norm": 0.7233839042760212,
+      "learning_rate": 6.0902198656065e-08,
+      "loss": 0.0391,
+      "step": 3555
+    },
+    {
+      "epoch": 2.8434504792332267,
+      "grad_norm": 0.6897214426187732,
+      "learning_rate": 5.7918205469972835e-08,
+      "loss": 0.0382,
+      "step": 3560
+    },
+    {
+      "epoch": 2.847444089456869,
+      "grad_norm": 0.7453035813326291,
+      "learning_rate": 5.500855300305485e-08,
+      "loss": 0.0388,
+      "step": 3565
+    },
+    {
+      "epoch": 2.851437699680511,
+      "grad_norm": 0.7866890040979969,
+      "learning_rate": 5.217330409683235e-08,
+      "loss": 0.0394,
+      "step": 3570
+    },
+    {
+      "epoch": 2.8554313099041533,
+      "grad_norm": 0.7099023785681982,
+      "learning_rate": 4.941251998588813e-08,
+      "loss": 0.0391,
+      "step": 3575
+    },
+    {
+      "epoch": 2.8594249201277955,
+      "grad_norm": 0.6644517725486436,
+      "learning_rate": 4.672626029654453e-08,
+      "loss": 0.0382,
+      "step": 3580
+    },
+    {
+      "epoch": 2.8634185303514377,
+      "grad_norm": 0.7086094015075386,
+      "learning_rate": 4.41145830455737e-08,
+      "loss": 0.0393,
+      "step": 3585
+    },
+    {
+      "epoch": 2.86741214057508,
+      "grad_norm": 0.6421742170261078,
+      "learning_rate": 4.1577544638946026e-08,
+      "loss": 0.0325,
+      "step": 3590
+    },
+    {
+      "epoch": 2.871405750798722,
+      "grad_norm": 0.681264218176713,
+      "learning_rate": 3.9115199870613124e-08,
+      "loss": 0.0391,
+      "step": 3595
+    },
+    {
+      "epoch": 2.8753993610223643,
+      "grad_norm": 0.6922943935506756,
+      "learning_rate": 3.672760192132107e-08,
+      "loss": 0.0379,
+      "step": 3600
+    },
+    {
+      "epoch": 2.8793929712460065,
+      "grad_norm": 0.6533088211666771,
+      "learning_rate": 3.44148023574653e-08,
+      "loss": 0.0351,
+      "step": 3605
+    },
+    {
+      "epoch": 2.8833865814696487,
+      "grad_norm": 0.719598133752619,
+      "learning_rate": 3.21768511299742e-08,
+      "loss": 0.0365,
+      "step": 3610
+    },
+    {
+      "epoch": 2.887380191693291,
+      "grad_norm": 0.6712232835091643,
+      "learning_rate": 3.001379657323161e-08,
+      "loss": 0.0375,
+      "step": 3615
+    },
+    {
+      "epoch": 2.891373801916933,
+      "grad_norm": 0.7793799932409174,
+      "learning_rate": 2.7925685404034226e-08,
+      "loss": 0.0397,
+      "step": 3620
+    },
+    {
+      "epoch": 2.8953674121405752,
+      "grad_norm": 0.7349037778253329,
+      "learning_rate": 2.5912562720578625e-08,
+      "loss": 0.0378,
+      "step": 3625
+    },
+    {
+      "epoch": 2.899361022364217,
+      "grad_norm": 0.7347196983541275,
+      "learning_rate": 2.397447200149211e-08,
+      "loss": 0.0372,
+      "step": 3630
+    },
+    {
+      "epoch": 2.9033546325878596,
+      "grad_norm": 0.7399058877098672,
+      "learning_rate": 2.2111455104889653e-08,
+      "loss": 0.0376,
+      "step": 3635
+    },
+    {
+      "epoch": 2.9073482428115014,
+      "grad_norm": 0.6460231812865741,
+      "learning_rate": 2.0323552267473176e-08,
+      "loss": 0.0379,
+      "step": 3640
+    },
+    {
+      "epoch": 2.911341853035144,
+      "grad_norm": 0.7479929434238791,
+      "learning_rate": 1.8610802103659174e-08,
+      "loss": 0.0373,
+      "step": 3645
+    },
+    {
+      "epoch": 2.915335463258786,
+      "grad_norm": 0.7817922135971772,
+      "learning_rate": 1.697324160474678e-08,
+      "loss": 0.0385,
+      "step": 3650
+    },
+    {
+      "epoch": 2.919329073482428,
+      "grad_norm": 0.7824066395040583,
+      "learning_rate": 1.5410906138120406e-08,
+      "loss": 0.0371,
+      "step": 3655
+    },
+    {
+      "epoch": 2.92332268370607,
+      "grad_norm": 0.7186999884806289,
+      "learning_rate": 1.3923829446481128e-08,
+      "loss": 0.0373,
+      "step": 3660
+    },
+    {
+      "epoch": 2.9273162939297124,
+      "grad_norm": 0.6803186595275351,
+      "learning_rate": 1.2512043647123937e-08,
+      "loss": 0.0379,
+      "step": 3665
+    },
+    {
+      "epoch": 2.9313099041533546,
+      "grad_norm": 0.7573331915873533,
+      "learning_rate": 1.117557923123752e-08,
+      "loss": 0.038,
+      "step": 3670
+    },
+    {
+      "epoch": 2.9353035143769968,
+      "grad_norm": 0.6550603996023604,
+      "learning_rate": 9.914465063252608e-09,
+      "loss": 0.0367,
+      "step": 3675
+    },
+    {
+      "epoch": 2.939297124600639,
+      "grad_norm": 0.6971503004155564,
+      "learning_rate": 8.728728380212102e-09,
+      "loss": 0.0382,
+      "step": 3680
+    },
+    {
+      "epoch": 2.943290734824281,
+      "grad_norm": 0.8017190342088107,
+      "learning_rate": 7.618394791188198e-09,
+      "loss": 0.0393,
+      "step": 3685
+    },
+    {
+      "epoch": 2.9472843450479234,
+      "grad_norm": 0.77895577170809,
+      "learning_rate": 6.5834882767251686e-09,
+      "loss": 0.039,
+      "step": 3690
+    },
+    {
+      "epoch": 2.9512779552715656,
+      "grad_norm": 0.746442946729805,
+      "learning_rate": 5.624031188324497e-09,
+      "loss": 0.0387,
+      "step": 3695
+    },
+    {
+      "epoch": 2.9552715654952078,
+      "grad_norm": 0.7104058466917486,
+      "learning_rate": 4.740044247960318e-09,
+      "loss": 0.0382,
+      "step": 3700
+    },
+    {
+      "epoch": 2.95926517571885,
+      "grad_norm": 0.7440341760747408,
+      "learning_rate": 3.931546547632558e-09,
+      "loss": 0.0401,
+      "step": 3705
+    },
+    {
+      "epoch": 2.963258785942492,
+      "grad_norm": 0.6642274503256519,
+      "learning_rate": 3.1985555489534854e-09,
+      "loss": 0.0381,
+      "step": 3710
+    },
+    {
+      "epoch": 2.9672523961661343,
+      "grad_norm": 0.7495614794397316,
+      "learning_rate": 2.541087082773119e-09,
+      "loss": 0.0409,
+      "step": 3715
+    },
+    {
+      "epoch": 2.9712460063897765,
+      "grad_norm": 0.6823205768784191,
+      "learning_rate": 1.959155348834951e-09,
+      "loss": 0.0372,
+      "step": 3720
+    },
+    {
+      "epoch": 2.9752396166134183,
+      "grad_norm": 0.7640147746969532,
+      "learning_rate": 1.4527729154697466e-09,
+      "loss": 0.0373,
+      "step": 3725
+    },
+    {
+      "epoch": 2.979233226837061,
+      "grad_norm": 0.7256837714457507,
+      "learning_rate": 1.0219507193247045e-09,
+      "loss": 0.0378,
+      "step": 3730
+    },
+    {
+      "epoch": 2.9832268370607027,
+      "grad_norm": 0.6990359157973326,
+      "learning_rate": 6.666980651272025e-10,
+      "loss": 0.0389,
+      "step": 3735
+    },
+    {
+      "epoch": 2.987220447284345,
+      "grad_norm": 0.6921384600587103,
+      "learning_rate": 3.870226254831244e-10,
+      "loss": 0.0386,
+      "step": 3740
+    },
+    {
+      "epoch": 2.991214057507987,
+      "grad_norm": 0.7426276592936151,
+      "learning_rate": 1.829304407121035e-10,
+      "loss": 0.0395,
+      "step": 3745
+    },
+    {
+      "epoch": 2.9952076677316293,
+      "grad_norm": 0.6789921703001532,
+      "learning_rate": 5.44259187161833e-11,
+      "loss": 0.0404,
+      "step": 3750
+    },
+    {
+      "epoch": 2.9992012779552715,
+      "grad_norm": 0.7434455972283719,
+      "learning_rate": 1.5118348850040597e-12,
+      "loss": 0.0372,
+      "step": 3755
+    },
+    {
+      "epoch": 3.0,
+      "step": 3756,
+      "total_flos": 1168211181961216.0,
+      "train_loss": 0.11109248143090163,
+      "train_runtime": 45172.9315,
+      "train_samples_per_second": 10.643,
+      "train_steps_per_second": 0.083
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3756,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1168211181961216.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}