diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5035 @@
+{
+  "best_metric": 5.638747215270996,
+  "best_model_checkpoint": "./results/models/mistral-prot/checkpoint-351531",
+  "epoch": 9.0,
+  "eval_steps": 500,
+  "global_step": 351531,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.012801146982769657,
+      "grad_norm": 0.2890625,
+      "learning_rate": 0.001999487954120689,
+      "loss": 5.8774,
+      "step": 500
+    },
+    {
+      "epoch": 0.025602293965539313,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.0019989759082413784,
+      "loss": 5.8257,
+      "step": 1000
+    },
+    {
+      "epoch": 0.03840344094830897,
+      "grad_norm": 2.765625,
+      "learning_rate": 0.001998463862362068,
+      "loss": 5.8373,
+      "step": 1500
+    },
+    {
+      "epoch": 0.05120458793107863,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.001997951816482757,
+      "loss": 5.8402,
+      "step": 2000
+    },
+    {
+      "epoch": 0.06400573491384828,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.0019974397706034462,
+      "loss": 5.8272,
+      "step": 2500
+    },
+    {
+      "epoch": 0.07680688189661794,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.001996927724724135,
+      "loss": 5.8143,
+      "step": 3000
+    },
+    {
+      "epoch": 0.0896080288793876,
+      "grad_norm": 5.6875,
+      "learning_rate": 0.0019964156788448246,
+      "loss": 5.8118,
+      "step": 3500
+    },
+    {
+      "epoch": 0.10240917586215725,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0019959036329655136,
+      "loss": 5.8184,
+      "step": 4000
+    },
+    {
+      "epoch": 0.11521032284492691,
+      "grad_norm": 3.609375,
+      "learning_rate": 0.001995391587086203,
+      "loss": 5.8083,
+      "step": 4500
+    },
+    {
+      "epoch": 0.12801146982769657,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.001994879541206892,
+      "loss": 5.8025,
+      "step": 5000
+    },
+    {
+      "epoch": 0.14081261681046622,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0019943674953275814,
+      "loss": 5.8007,
+      "step": 5500
+    },
+    {
+      "epoch": 0.15361376379323588,
+      "grad_norm": 3.0625,
+      "learning_rate": 0.0019938554494482704,
+      "loss": 5.7932,
+      "step": 6000
+    },
+    {
+      "epoch": 0.16641491077600554,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.00199334340356896,
+      "loss": 5.7984,
+      "step": 6500
+    },
+    {
+      "epoch": 0.1792160577587752,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.001992831357689649,
+      "loss": 5.7895,
+      "step": 7000
+    },
+    {
+      "epoch": 0.19201720474154485,
+      "grad_norm": 0.5,
+      "learning_rate": 0.001992319311810338,
+      "loss": 5.7869,
+      "step": 7500
+    },
+    {
+      "epoch": 0.2048183517243145,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0019918072659310276,
+      "loss": 5.7823,
+      "step": 8000
+    },
+    {
+      "epoch": 0.21761949870708416,
+      "grad_norm": 3.234375,
+      "learning_rate": 0.0019912952200517166,
+      "loss": 5.7778,
+      "step": 8500
+    },
+    {
+      "epoch": 0.23042064568985382,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.001990783174172406,
+      "loss": 5.7752,
+      "step": 9000
+    },
+    {
+      "epoch": 0.24322179267262348,
+      "grad_norm": 0.875,
+      "learning_rate": 0.0019902711282930954,
+      "loss": 5.7739,
+      "step": 9500
+    },
+    {
+      "epoch": 0.25602293965539313,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0019897590824137844,
+      "loss": 5.7784,
+      "step": 10000
+    },
+    {
+      "epoch": 0.26882408663816276,
+      "grad_norm": 1.5,
+      "learning_rate": 0.0019892470365344733,
+      "loss": 5.7735,
+      "step": 10500
+    },
+    {
+      "epoch": 0.28162523362093245,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0019887349906551628,
+      "loss": 5.7704,
+      "step": 11000
+    },
+    {
+      "epoch": 0.2944263806037021,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.001988222944775852,
+      "loss": 5.7677,
+      "step": 11500
+    },
+    {
+      "epoch": 0.30722752758647176,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.001987710898896541,
+      "loss": 5.7653,
+      "step": 12000
+    },
+    {
+      "epoch": 0.3200286745692414,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.0019871988530172306,
+      "loss": 5.7659,
+      "step": 12500
+    },
+    {
+      "epoch": 0.3328298215520111,
+      "grad_norm": 3.375,
+      "learning_rate": 0.0019866868071379195,
+      "loss": 5.7669,
+      "step": 13000
+    },
+    {
+      "epoch": 0.3456309685347807,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.001986174761258609,
+      "loss": 5.7625,
+      "step": 13500
+    },
+    {
+      "epoch": 0.3584321155175504,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.001985662715379298,
+      "loss": 5.762,
+      "step": 14000
+    },
+    {
+      "epoch": 0.37123326250032,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0019851506694999873,
+      "loss": 5.7618,
+      "step": 14500
+    },
+    {
+      "epoch": 0.3840344094830897,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.0019846386236206763,
+      "loss": 5.7574,
+      "step": 15000
+    },
+    {
+      "epoch": 0.39683555646585933,
+      "grad_norm": 0.68359375,
+      "learning_rate": 0.0019841265777413657,
+      "loss": 5.756,
+      "step": 15500
+    },
+    {
+      "epoch": 0.409636703448629,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.0019836145318620547,
+      "loss": 5.7518,
+      "step": 16000
+    },
+    {
+      "epoch": 0.42243785043139864,
+      "grad_norm": 2.265625,
+      "learning_rate": 0.001983102485982744,
+      "loss": 5.7524,
+      "step": 16500
+    },
+    {
+      "epoch": 0.4352389974141683,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.0019825904401034335,
+      "loss": 5.7561,
+      "step": 17000
+    },
+    {
+      "epoch": 0.44804014439693796,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0019820783942241225,
+      "loss": 5.749,
+      "step": 17500
+    },
+    {
+      "epoch": 0.46084129137970764,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0019815663483448115,
+      "loss": 5.7498,
+      "step": 18000
+    },
+    {
+      "epoch": 0.47364243836247727,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.001981054302465501,
+      "loss": 5.7474,
+      "step": 18500
+    },
+    {
+      "epoch": 0.48644358534524695,
+      "grad_norm": 2.34375,
+      "learning_rate": 0.0019805422565861903,
+      "loss": 5.7462,
+      "step": 19000
+    },
+    {
+      "epoch": 0.4992447323280166,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0019800302107068793,
+      "loss": 5.7469,
+      "step": 19500
+    },
+    {
+      "epoch": 0.5120458793107863,
+      "grad_norm": 3.984375,
+      "learning_rate": 0.0019795181648275687,
+      "loss": 5.7461,
+      "step": 20000
+    },
+    {
+      "epoch": 0.524847026293556,
+      "grad_norm": 0.7578125,
+      "learning_rate": 0.0019790061189482577,
+      "loss": 5.7471,
+      "step": 20500
+    },
+    {
+      "epoch": 0.5376481732763255,
+      "grad_norm": 4.90625,
+      "learning_rate": 0.001978494073068947,
+      "loss": 5.7454,
+      "step": 21000
+    },
+    {
+      "epoch": 0.5504493202590952,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.0019779820271896365,
+      "loss": 5.744,
+      "step": 21500
+    },
+    {
+      "epoch": 0.5632504672418649,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0019774699813103255,
+      "loss": 5.7431,
+      "step": 22000
+    },
+    {
+      "epoch": 0.5760516142246346,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.001976957935431015,
+      "loss": 5.7406,
+      "step": 22500
+    },
+    {
+      "epoch": 0.5888527612074042,
+      "grad_norm": 0.9375,
+      "learning_rate": 0.001976445889551704,
+      "loss": 5.7395,
+      "step": 23000
+    },
+    {
+      "epoch": 0.6016539081901738,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.001975933843672393,
+      "loss": 5.7383,
+      "step": 23500
+    },
+    {
+      "epoch": 0.6144550551729435,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.0019754217977930822,
+      "loss": 5.7394,
+      "step": 24000
+    },
+    {
+      "epoch": 0.6272562021557132,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.0019749097519137717,
+      "loss": 5.7354,
+      "step": 24500
+    },
+    {
+      "epoch": 0.6400573491384828,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.0019743977060344606,
+      "loss": 5.7367,
+      "step": 25000
+    },
+    {
+      "epoch": 0.6528584961212525,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.00197388566015515,
+      "loss": 5.7345,
+      "step": 25500
+    },
+    {
+      "epoch": 0.6656596431040221,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.001973373614275839,
+      "loss": 5.7333,
+      "step": 26000
+    },
+    {
+      "epoch": 0.6784607900867918,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0019728615683965284,
+      "loss": 5.7334,
+      "step": 26500
+    },
+    {
+      "epoch": 0.6912619370695614,
+      "grad_norm": 2.125,
+      "learning_rate": 0.001972349522517218,
+      "loss": 5.7334,
+      "step": 27000
+    },
+    {
+      "epoch": 0.7040630840523311,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.001971837476637907,
+      "loss": 5.7328,
+      "step": 27500
+    },
+    {
+      "epoch": 0.7168642310351008,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.001971325430758596,
+      "loss": 5.7329,
+      "step": 28000
+    },
+    {
+      "epoch": 0.7296653780178703,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.001970813384879285,
+      "loss": 5.7324,
+      "step": 28500
+    },
+    {
+      "epoch": 0.74246652500064,
+      "grad_norm": 3.375,
+      "learning_rate": 0.0019703013389999746,
+      "loss": 5.7314,
+      "step": 29000
+    },
+    {
+      "epoch": 0.7552676719834097,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.0019697892931206636,
+      "loss": 5.7307,
+      "step": 29500
+    },
+    {
+      "epoch": 0.7680688189661794,
+      "grad_norm": 1.671875,
+      "learning_rate": 0.001969277247241353,
+      "loss": 5.7291,
+      "step": 30000
+    },
+    {
+      "epoch": 0.780869965948949,
+      "grad_norm": 1.6171875,
+      "learning_rate": 0.001968765201362042,
+      "loss": 5.7313,
+      "step": 30500
+    },
+    {
+      "epoch": 0.7936711129317187,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.0019682531554827314,
+      "loss": 5.7286,
+      "step": 31000
+    },
+    {
+      "epoch": 0.8064722599144883,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0019677411096034204,
+      "loss": 5.732,
+      "step": 31500
+    },
+    {
+      "epoch": 0.819273406897258,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0019672290637241098,
+      "loss": 5.7302,
+      "step": 32000
+    },
+    {
+      "epoch": 0.8320745538800276,
+      "grad_norm": 0.67578125,
+      "learning_rate": 0.0019667170178447988,
+      "loss": 5.7277,
+      "step": 32500
+    },
+    {
+      "epoch": 0.8448757008627973,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.001966204971965488,
+      "loss": 5.728,
+      "step": 33000
+    },
+    {
+      "epoch": 0.857676847845567,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.001965692926086177,
+      "loss": 5.7276,
+      "step": 33500
+    },
+    {
+      "epoch": 0.8704779948283367,
+      "grad_norm": 2.875,
+      "learning_rate": 0.0019651808802068666,
+      "loss": 5.7268,
+      "step": 34000
+    },
+    {
+      "epoch": 0.8832791418111062,
+      "grad_norm": 1.703125,
+      "learning_rate": 0.001964668834327556,
+      "loss": 5.7294,
+      "step": 34500
+    },
+    {
+      "epoch": 0.8960802887938759,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.001964156788448245,
+      "loss": 5.7281,
+      "step": 35000
+    },
+    {
+      "epoch": 0.9088814357766456,
+      "grad_norm": 1.75,
+      "learning_rate": 0.0019636447425689344,
+      "loss": 5.7281,
+      "step": 35500
+    },
+    {
+      "epoch": 0.9216825827594153,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.0019631326966896233,
+      "loss": 5.7254,
+      "step": 36000
+    },
+    {
+      "epoch": 0.9344837297421849,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.0019626206508103127,
+      "loss": 5.7248,
+      "step": 36500
+    },
+    {
+      "epoch": 0.9472848767249545,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.001962108604931002,
+      "loss": 5.7229,
+      "step": 37000
+    },
+    {
+      "epoch": 0.9600860237077242,
+      "grad_norm": 1.890625,
+      "learning_rate": 0.001961596559051691,
+      "loss": 5.7217,
+      "step": 37500
+    },
+    {
+      "epoch": 0.9728871706904939,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.00196108451317238,
+      "loss": 5.7211,
+      "step": 38000
+    },
+    {
+      "epoch": 0.9856883176732635,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0019605724672930695,
+      "loss": 5.7208,
+      "step": 38500
+    },
+    {
+      "epoch": 0.9984894646560332,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.001960060421413759,
+      "loss": 5.7199,
+      "step": 39000
+    },
+    {
+      "epoch": 1.0,
+      "eval_loss": 5.719020366668701,
+      "eval_runtime": 1.3381,
+      "eval_samples_per_second": 747.35,
+      "eval_steps_per_second": 2.989,
+      "step": 39059
+    },
+    {
+      "epoch": 1.0112906116388027,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.001959548375534448,
+      "loss": 5.7187,
+      "step": 39500
+    },
+    {
+      "epoch": 1.0240917586215725,
+      "grad_norm": 2.1875,
+      "learning_rate": 0.0019590363296551373,
+      "loss": 5.7184,
+      "step": 40000
+    },
+    {
+      "epoch": 1.036892905604342,
+      "grad_norm": 2.5,
+      "learning_rate": 0.0019585242837758263,
+      "loss": 5.7179,
+      "step": 40500
+    },
+    {
+      "epoch": 1.049694052587112,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.0019580122378965157,
+      "loss": 5.7175,
+      "step": 41000
+    },
+    {
+      "epoch": 1.0624951995698815,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.0019575001920172047,
+      "loss": 5.7169,
+      "step": 41500
+    },
+    {
+      "epoch": 1.075296346552651,
+      "grad_norm": 0.3984375,
+      "learning_rate": 0.001956988146137894,
+      "loss": 5.7159,
+      "step": 42000
+    },
+    {
+      "epoch": 1.0880974935354208,
+      "grad_norm": 2.265625,
+      "learning_rate": 0.001956476100258583,
+      "loss": 5.7154,
+      "step": 42500
+    },
+    {
+      "epoch": 1.1008986405181904,
+      "grad_norm": 30.875,
+      "learning_rate": 0.0019559640543792725,
+      "loss": 5.7143,
+      "step": 43000
+    },
+    {
+      "epoch": 1.11369978750096,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.0019554520084999615,
+      "loss": 5.7134,
+      "step": 43500
+    },
+    {
+      "epoch": 1.1265009344837298,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.001954939962620651,
+      "loss": 5.713,
+      "step": 44000
+    },
+    {
+      "epoch": 1.1393020814664994,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.0019544279167413403,
+      "loss": 5.7128,
+      "step": 44500
+    },
+    {
+      "epoch": 1.1521032284492692,
+      "grad_norm": 2.96875,
+      "learning_rate": 0.0019539158708620293,
+      "loss": 5.714,
+      "step": 45000
+    },
+    {
+      "epoch": 1.1649043754320387,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.0019534038249827182,
+      "loss": 5.7138,
+      "step": 45500
+    },
+    {
+      "epoch": 1.1777055224148083,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.0019528917791034077,
+      "loss": 5.7114,
+      "step": 46000
+    },
+    {
+      "epoch": 1.190506669397578,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0019523797332240969,
+      "loss": 5.7114,
+      "step": 46500
+    },
+    {
+      "epoch": 1.2033078163803477,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.001951867687344786,
+      "loss": 5.7114,
+      "step": 47000
+    },
+    {
+      "epoch": 1.2161089633631172,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.0019513556414654755,
+      "loss": 5.712,
+      "step": 47500
+    },
+    {
+      "epoch": 1.228910110345887,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.0019508435955861644,
+      "loss": 5.7118,
+      "step": 48000
+    },
+    {
+      "epoch": 1.2417112573286566,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0019503315497068536,
+      "loss": 5.7105,
+      "step": 48500
+    },
+    {
+      "epoch": 1.2545124043114262,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.001949819503827543,
+      "loss": 5.709,
+      "step": 49000
+    },
+    {
+      "epoch": 1.267313551294196,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0019493074579482322,
+      "loss": 5.7086,
+      "step": 49500
+    },
+    {
+      "epoch": 1.2801146982769656,
+      "grad_norm": 1.953125,
+      "learning_rate": 0.0019487954120689214,
+      "loss": 5.7086,
+      "step": 50000
+    },
+    {
+      "epoch": 1.2929158452597354,
+      "grad_norm": 1.5703125,
+      "learning_rate": 0.0019482833661896106,
+      "loss": 5.7083,
+      "step": 50500
+    },
+    {
+      "epoch": 1.305716992242505,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0019477713203102998,
+      "loss": 5.7073,
+      "step": 51000
+    },
+    {
+      "epoch": 1.3185181392252745,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0019472592744309892,
+      "loss": 5.7056,
+      "step": 51500
+    },
+    {
+      "epoch": 1.3313192862080443,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.0019467472285516782,
+      "loss": 5.7052,
+      "step": 52000
+    },
+    {
+      "epoch": 1.3441204331908139,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0019462351826723674,
+      "loss": 5.7075,
+      "step": 52500
+    },
+    {
+      "epoch": 1.3569215801735837,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.0019457231367930568,
+      "loss": 5.7071,
+      "step": 53000
+    },
+    {
+      "epoch": 1.3697227271563532,
+      "grad_norm": 1.703125,
+      "learning_rate": 0.001945211090913746,
+      "loss": 5.7071,
+      "step": 53500
+    },
+    {
+      "epoch": 1.3825238741391228,
+      "grad_norm": 2.46875,
+      "learning_rate": 0.001944699045034435,
+      "loss": 5.7083,
+      "step": 54000
+    },
+    {
+      "epoch": 1.3953250211218924,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.0019441869991551244,
+      "loss": 5.7072,
+      "step": 54500
+    },
+    {
+      "epoch": 1.4081261681046622,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.0019436749532758136,
+      "loss": 5.7068,
+      "step": 55000
+    },
+    {
+      "epoch": 1.4209273150874318,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.0019431629073965028,
+      "loss": 5.7085,
+      "step": 55500
+    },
+    {
+      "epoch": 1.4337284620702015,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.001942650861517192,
+      "loss": 5.707,
+      "step": 56000
+    },
+    {
+      "epoch": 1.4465296090529711,
+      "grad_norm": 0.83984375,
+      "learning_rate": 0.0019421388156378812,
+      "loss": 5.7061,
+      "step": 56500
+    },
+    {
+      "epoch": 1.4593307560357407,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.0019416267697585704,
+      "loss": 5.7068,
+      "step": 57000
+    },
+    {
+      "epoch": 1.4721319030185105,
+      "grad_norm": 3.984375,
+      "learning_rate": 0.0019411147238792598,
+      "loss": 5.7078,
+      "step": 57500
+    },
+    {
+      "epoch": 1.48493305000128,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.0019406026779999488,
+      "loss": 5.7061,
+      "step": 58000
+    },
+    {
+      "epoch": 1.4977341969840499,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.001940090632120638,
+      "loss": 5.7045,
+      "step": 58500
+    },
+    {
+      "epoch": 1.5105353439668194,
+      "grad_norm": 2.34375,
+      "learning_rate": 0.0019395785862413274,
+      "loss": 5.7041,
+      "step": 59000
+    },
+    {
+      "epoch": 1.523336490949589,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.0019390665403620166,
+      "loss": 5.7041,
+      "step": 59500
+    },
+    {
+      "epoch": 1.5361376379323586,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.0019385544944827055,
+      "loss": 5.7034,
+      "step": 60000
+    },
+    {
+      "epoch": 1.5489387849151284,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.001938042448603395,
+      "loss": 5.7031,
+      "step": 60500
+    },
+    {
+      "epoch": 1.5617399318978982,
+      "grad_norm": 3.6875,
+      "learning_rate": 0.0019375304027240841,
+      "loss": 5.7022,
+      "step": 61000
+    },
+    {
+      "epoch": 1.5745410788806677,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.0019370183568447733,
+      "loss": 5.7014,
+      "step": 61500
+    },
+    {
+      "epoch": 1.5873422258634373,
+      "grad_norm": 1.28125,
+      "learning_rate": 0.0019365063109654625,
+      "loss": 5.702,
+      "step": 62000
+    },
+    {
+      "epoch": 1.600143372846207,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.0019359942650861517,
+      "loss": 5.7015,
+      "step": 62500
+    },
+    {
+      "epoch": 1.6129445198289767,
+      "grad_norm": 28.125,
+      "learning_rate": 0.001935482219206841,
+      "loss": 5.7002,
+      "step": 63000
+    },
+    {
+      "epoch": 1.6257456668117465,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.0019349701733275303,
+      "loss": 5.7002,
+      "step": 63500
+    },
+    {
+      "epoch": 1.638546813794516,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0019344581274482193,
+      "loss": 5.7003,
+      "step": 64000
+    },
+    {
+      "epoch": 1.6513479607772856,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.0019339460815689087,
+      "loss": 5.7007,
+      "step": 64500
+    },
+    {
+      "epoch": 1.6641491077600552,
+      "grad_norm": 0.68359375,
+      "learning_rate": 0.001933434035689598,
+      "loss": 5.6992,
+      "step": 65000
+    },
+    {
+      "epoch": 1.676950254742825,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.0019329219898102869,
+      "loss": 5.6976,
+      "step": 65500
+    },
+    {
+      "epoch": 1.6897514017255946,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.0019324099439309763,
+      "loss": 5.6983,
+      "step": 66000
+    },
+    {
+      "epoch": 1.7025525487083644,
+      "grad_norm": 3.1875,
+      "learning_rate": 0.0019318978980516655,
+      "loss": 5.6977,
+      "step": 66500
+    },
+    {
+      "epoch": 1.715353695691134,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0019313858521723547,
+      "loss": 5.6967,
+      "step": 67000
+    },
+    {
+      "epoch": 1.7281548426739035,
+      "grad_norm": 5.5625,
+      "learning_rate": 0.0019308738062930439,
+      "loss": 5.6962,
+      "step": 67500
+    },
+    {
+      "epoch": 1.740955989656673,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.001930361760413733,
+      "loss": 5.6962,
+      "step": 68000
+    },
+    {
+      "epoch": 1.7537571366394429,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.0019298497145344223,
+      "loss": 5.6962,
+      "step": 68500
+    },
+    {
+      "epoch": 1.7665582836222127,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.0019293376686551117,
+      "loss": 5.6965,
+      "step": 69000
+    },
+    {
+      "epoch": 1.7793594306049823,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.0019288256227758007,
+      "loss": 5.6967,
+      "step": 69500
+    },
+    {
+      "epoch": 1.7921605775877518,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.0019283135768964899,
+      "loss": 5.6962,
+      "step": 70000
+    },
+    {
+      "epoch": 1.8049617245705214,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0019278015310171793,
+      "loss": 5.6972,
+      "step": 70500
+    },
+    {
+      "epoch": 1.8177628715532912,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.0019272894851378685,
+      "loss": 5.6971,
+      "step": 71000
+    },
+    {
+      "epoch": 1.830564018536061,
+      "grad_norm": 1.2265625,
+      "learning_rate": 0.0019267774392585574,
+      "loss": 5.6958,
+      "step": 71500
+    },
+    {
+      "epoch": 1.8433651655188306,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.0019262653933792469,
+      "loss": 5.6957,
+      "step": 72000
+    },
+    {
+      "epoch": 1.8561663125016001,
+      "grad_norm": 6.03125,
+      "learning_rate": 0.001925753347499936,
+      "loss": 5.6965,
+      "step": 72500
+    },
+    {
+      "epoch": 1.8689674594843697,
+      "grad_norm": 34.5,
+      "learning_rate": 0.0019252413016206252,
+      "loss": 5.6964,
+      "step": 73000
+    },
+    {
+      "epoch": 1.8817686064671395,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.0019247292557413144,
+      "loss": 5.6968,
+      "step": 73500
+    },
+    {
+      "epoch": 1.894569753449909,
+      "grad_norm": 3.0,
+      "learning_rate": 0.0019242172098620036,
+      "loss": 5.6961,
+      "step": 74000
+    },
+    {
+      "epoch": 1.9073709004326789,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0019237051639826928,
+      "loss": 5.6974,
+      "step": 74500
+    },
+    {
+      "epoch": 1.9201720474154484,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0019231931181033822,
+      "loss": 5.697,
+      "step": 75000
+    },
+    {
+      "epoch": 1.932973194398218,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0019226810722240712,
+      "loss": 5.695,
+      "step": 75500
+    },
+    {
+      "epoch": 1.9457743413809876,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0019221690263447604,
+      "loss": 5.6965,
+      "step": 76000
+    },
+    {
+      "epoch": 1.9585754883637574,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0019216569804654498,
+      "loss": 5.6989,
+      "step": 76500
+    },
+    {
+      "epoch": 1.9713766353465272,
+      "grad_norm": 0.65234375,
+      "learning_rate": 0.001921144934586139,
+      "loss": 5.6968,
+      "step": 77000
+    },
+    {
+      "epoch": 1.9841777823292968,
+      "grad_norm": 1.8359375,
+      "learning_rate": 0.001920632888706828,
+      "loss": 5.6953,
+      "step": 77500
+    },
+    {
+      "epoch": 1.9969789293120663,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.0019201208428275174,
+      "loss": 5.6954,
+      "step": 78000
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 5.694720268249512,
+      "eval_runtime": 1.2928,
+      "eval_samples_per_second": 773.49,
+      "eval_steps_per_second": 3.094,
+      "step": 78118
+    },
+    {
+      "epoch": 2.009780076294836,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.0019196087969482066,
+      "loss": 5.6937,
+      "step": 78500
+    },
+    {
+      "epoch": 2.0225812232776055,
+      "grad_norm": 1.8515625,
+      "learning_rate": 0.001919096751068896,
+      "loss": 5.6939,
+      "step": 79000
+    },
+    {
+      "epoch": 2.0353823702603755,
+      "grad_norm": 3.171875,
+      "learning_rate": 0.001918584705189585,
+      "loss": 5.6957,
+      "step": 79500
+    },
+    {
+      "epoch": 2.048183517243145,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0019180726593102742,
+      "loss": 5.6932,
+      "step": 80000
+    },
+    {
+      "epoch": 2.0609846642259146,
+      "grad_norm": 1.703125,
+      "learning_rate": 0.0019175606134309636,
+      "loss": 5.6931,
+      "step": 80500
+    },
+    {
+      "epoch": 2.073785811208684,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.0019170485675516528,
+      "loss": 5.6922,
+      "step": 81000
+    },
+    {
+      "epoch": 2.086586958191454,
+      "grad_norm": 29.625,
+      "learning_rate": 0.0019165365216723418,
+      "loss": 5.6921,
+      "step": 81500
+    },
+    {
+      "epoch": 2.099388105174224,
+      "grad_norm": 1.625,
+      "learning_rate": 0.0019160244757930312,
+      "loss": 5.691,
+      "step": 82000
+    },
+    {
+      "epoch": 2.1121892521569934,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.0019155124299137204,
+      "loss": 5.6897,
+      "step": 82500
+    },
+    {
+      "epoch": 2.124990399139763,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.0019150003840344096,
+      "loss": 5.6913,
+      "step": 83000
+    },
+    {
+      "epoch": 2.1377915461225325,
+      "grad_norm": 1.0,
+      "learning_rate": 0.0019144883381550988,
+      "loss": 5.693,
+      "step": 83500
+    },
+    {
+      "epoch": 2.150592693105302,
+      "grad_norm": 1.9609375,
+      "learning_rate": 0.001913976292275788,
+      "loss": 5.691,
+      "step": 84000
+    },
+    {
+      "epoch": 2.163393840088072,
+      "grad_norm": 0.70703125,
+      "learning_rate": 0.0019134642463964771,
+      "loss": 5.6912,
+      "step": 84500
+    },
+    {
+      "epoch": 2.1761949870708417,
+      "grad_norm": 37.0,
+      "learning_rate": 0.0019129522005171666,
+      "loss": 5.6893,
+      "step": 85000
+    },
+    {
+      "epoch": 2.1889961340536113,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.0019124401546378555,
+      "loss": 5.6903,
+      "step": 85500
+    },
+    {
+      "epoch": 2.201797281036381,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.0019119281087585447,
+      "loss": 5.6916,
+      "step": 86000
+    },
+    {
+      "epoch": 2.2145984280191504,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.0019114160628792341,
+      "loss": 5.6917,
+      "step": 86500
+    },
+    {
+      "epoch": 2.22739957500192,
+      "grad_norm": 2.71875,
+      "learning_rate": 0.0019109040169999231,
+      "loss": 5.6898,
+      "step": 87000
+    },
+    {
+      "epoch": 2.24020072198469,
+      "grad_norm": 0.68359375,
+      "learning_rate": 0.0019103919711206123,
+      "loss": 5.69,
+      "step": 87500
+    },
+    {
+      "epoch": 2.2530018689674596,
+      "grad_norm": 3.46875,
+      "learning_rate": 0.0019098799252413017,
+      "loss": 5.6909,
+      "step": 88000
+    },
+    {
+      "epoch": 2.265803015950229,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.001909367879361991,
+      "loss": 5.6894,
+      "step": 88500
+    },
+    {
+      "epoch": 2.2786041629329987,
+      "grad_norm": 1.4453125,
+      "learning_rate": 0.00190885583348268,
+      "loss": 5.6896,
+      "step": 89000
+    },
+    {
+      "epoch": 2.2914053099157683,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0019083437876033693,
+      "loss": 5.6879,
+      "step": 89500
+    },
+    {
+      "epoch": 2.3042064568985383,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.0019078317417240585,
+      "loss": 5.6889,
+      "step": 90000
+    },
+    {
+      "epoch": 2.317007603881308,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.0019073196958447477,
+      "loss": 5.6891,
+      "step": 90500
+    },
+    {
+      "epoch": 2.3298087508640775,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.0019068076499654369,
+      "loss": 5.6879,
+      "step": 91000
+    },
+    {
+      "epoch": 2.342609897846847,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.001906295604086126,
+      "loss": 5.6875,
+      "step": 91500
+    },
+    {
+      "epoch": 2.3554110448296166,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.0019057835582068153,
+      "loss": 5.6873,
+      "step": 92000
+    },
+    {
+      "epoch": 2.368212191812386,
+      "grad_norm": 1.6640625,
+      "learning_rate": 0.0019052715123275047,
+      "loss": 5.6876,
+      "step": 92500
+    },
+    {
+      "epoch": 2.381013338795156,
+      "grad_norm": 2.015625,
+      "learning_rate": 0.0019047594664481937,
+      "loss": 5.6874,
+      "step": 93000
+    },
+    {
+      "epoch": 2.3938144857779258,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.001904247420568883,
+      "loss": 5.687,
+      "step": 93500
+    },
+    {
+      "epoch": 2.4066156327606953,
+      "grad_norm": 6.125,
+      "learning_rate": 0.0019037353746895723,
+      "loss": 5.6859,
+      "step": 94000
+    },
+    {
+      "epoch": 2.419416779743465,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.0019032233288102615,
+      "loss": 5.6844,
+      "step": 94500
+    },
+    {
+      "epoch": 2.4322179267262345,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.0019027112829309507,
+      "loss": 5.6844,
+      "step": 95000
+    },
+    {
+      "epoch": 2.4450190737090045,
+      "grad_norm": 1.140625,
+      "learning_rate": 0.0019021992370516399,
+      "loss": 5.6854,
+      "step": 95500
+    },
+    {
+      "epoch": 2.457820220691774,
+      "grad_norm": 3.640625,
+      "learning_rate": 0.001901687191172329,
+      "loss": 5.6862,
+      "step": 96000
+    },
+    {
+      "epoch": 2.4706213676745437,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0019011751452930185,
+      "loss": 5.6857,
+      "step": 96500
+    },
+    {
+      "epoch": 2.4834225146573132,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.0019006630994137074,
+      "loss": 5.683,
+      "step": 97000
+    },
+    {
+      "epoch": 2.496223661640083,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.0019001510535343966,
+      "loss": 5.683,
+      "step": 97500
+    },
+    {
+      "epoch": 2.5090248086228524,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.001899639007655086,
+      "loss": 5.6825,
+      "step": 98000
+    },
+    {
+      "epoch": 2.5218259556056224,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.0018991269617757752,
+      "loss": 5.6821,
+      "step": 98500
+    },
+    {
+      "epoch": 2.534627102588392,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0018986149158964642,
+      "loss": 5.6817,
+      "step": 99000
+    },
+    {
+      "epoch": 2.5474282495711615,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.0018981028700171536,
+      "loss": 5.6814,
+      "step": 99500
+    },
+    {
+      "epoch": 2.560229396553931,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.0018975908241378428,
+      "loss": 5.6819,
+      "step": 100000
+    },
+    {
+      "epoch": 2.573030543536701,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.001897078778258532,
+      "loss": 5.6803,
+      "step": 100500
+    },
+    {
+      "epoch": 2.5858316905194707,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.0018965667323792212,
+      "loss": 5.6809,
+      "step": 101000
+    },
+    {
+      "epoch": 2.5986328375022403,
+      "grad_norm": 1.6484375,
+      "learning_rate": 0.0018960546864999104,
+      "loss": 5.6822,
+      "step": 101500
+    },
+    {
+      "epoch": 2.61143398448501,
+      "grad_norm": 1.6015625,
+      "learning_rate": 0.0018955426406205996,
+      "loss": 5.6804,
+      "step": 102000
+    },
+    {
+      "epoch": 2.6242351314677794,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.001895030594741289,
+      "loss": 5.6815,
+      "step": 102500
+    },
+    {
+      "epoch": 2.637036278450549,
+      "grad_norm": 1.859375,
+      "learning_rate": 0.001894518548861978,
+      "loss": 5.6808,
+      "step": 103000
+    },
+    {
+      "epoch": 2.6498374254333186,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.0018940065029826672,
+      "loss": 5.6811,
+      "step": 103500
+    },
+    {
+      "epoch": 2.6626385724160886,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.0018934944571033566,
+      "loss": 5.6795,
+      "step": 104000
+    },
+    {
+      "epoch": 2.675439719398858,
+      "grad_norm": 3.640625,
+      "learning_rate": 0.0018929824112240458,
+      "loss": 5.6799,
+      "step": 104500
+    },
+    {
+      "epoch": 2.6882408663816277,
+      "grad_norm": 2.015625,
+      "learning_rate": 0.0018924703653447348,
+      "loss": 5.6811,
+      "step": 105000
+    },
+    {
+      "epoch": 2.7010420133643973,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.0018919583194654242,
+      "loss": 5.6798,
+      "step": 105500
+    },
+    {
+      "epoch": 2.7138431603471673,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.0018914462735861134,
+      "loss": 5.6799,
+      "step": 106000
+    },
+    {
+      "epoch": 2.726644307329937,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0018909342277068028,
+      "loss": 5.6795,
+      "step": 106500
+    },
+    {
+      "epoch": 2.7394454543127065,
+      "grad_norm": 1.984375,
+      "learning_rate": 0.0018904221818274918,
+      "loss": 5.6804,
+      "step": 107000
+    },
+    {
+      "epoch": 2.752246601295476,
+      "grad_norm": 1.640625,
+      "learning_rate": 0.001889910135948181,
+      "loss": 5.6813,
+      "step": 107500
+    },
+    {
+      "epoch": 2.7650477482782456,
+      "grad_norm": 1.4296875,
+      "learning_rate": 0.0018893980900688704,
+      "loss": 5.6802,
+      "step": 108000
+    },
+    {
+      "epoch": 2.777848895261015,
+      "grad_norm": 6.59375,
+      "learning_rate": 0.0018888860441895596,
+      "loss": 5.6814,
+      "step": 108500
+    },
+    {
+      "epoch": 2.7906500422437848,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0018883739983102485,
+      "loss": 5.6799,
+      "step": 109000
+    },
+    {
+      "epoch": 2.803451189226555,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.001887861952430938,
+      "loss": 5.6792,
+      "step": 109500
+    },
+    {
+      "epoch": 2.8162523362093244,
+      "grad_norm": 0.875,
+      "learning_rate": 0.0018873499065516271,
+      "loss": 5.6795,
+      "step": 110000
+    },
+    {
+      "epoch": 2.829053483192094,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.0018868378606723161,
+      "loss": 5.6809,
+      "step": 110500
+    },
+    {
+      "epoch": 2.8418546301748635,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.0018863258147930055,
+      "loss": 5.6804,
+      "step": 111000
+    },
+    {
+      "epoch": 2.8546557771576335,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0018858137689136947,
+      "loss": 5.6817,
+      "step": 111500
+    },
+    {
+      "epoch": 2.867456924140403,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.001885301723034384,
+      "loss": 5.6811,
+      "step": 112000
+    },
+    {
+      "epoch": 2.8802580711231727,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.0018847896771550731,
+      "loss": 5.6811,
+      "step": 112500
+    },
+    {
+      "epoch": 2.8930592181059422,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.0018842776312757623,
+      "loss": 5.6822,
+      "step": 113000
+    },
+    {
+      "epoch": 2.905860365088712,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0018837655853964515,
+      "loss": 5.6842,
+      "step": 113500
+    },
+    {
+      "epoch": 2.9186615120714814,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.001883253539517141,
+      "loss": 5.6826,
+      "step": 114000
+    },
+    {
+      "epoch": 2.9314626590542514,
+      "grad_norm": 16.75,
+      "learning_rate": 0.0018827414936378299,
+      "loss": 5.681,
+      "step": 114500
+    },
+    {
+      "epoch": 2.944263806037021,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.001882229447758519,
+      "loss": 5.6806,
+      "step": 115000
+    },
+    {
+      "epoch": 2.9570649530197906,
+      "grad_norm": 5.96875,
+      "learning_rate": 0.0018817174018792085,
+      "loss": 5.6805,
+      "step": 115500
+    },
+    {
+      "epoch": 2.96986610000256,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.0018812053559998977,
+      "loss": 5.6798,
+      "step": 116000
+    },
+    {
+      "epoch": 2.98266724698533,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.0018806933101205867,
+      "loss": 5.6785,
+      "step": 116500
+    },
+    {
+      "epoch": 2.9954683939680997,
+      "grad_norm": 0.69140625,
+      "learning_rate": 0.001880181264241276,
+      "loss": 5.6787,
+      "step": 117000
+    },
+    {
+      "epoch": 3.0,
+      "eval_loss": 5.67839241027832,
+      "eval_runtime": 1.3366,
+      "eval_samples_per_second": 748.15,
+      "eval_steps_per_second": 2.993,
+      "step": 117177
+    },
+    {
+      "epoch": 3.0082695409508693,
+      "grad_norm": 9.75,
+      "learning_rate": 0.0018796692183619653,
+      "loss": 5.6776,
+      "step": 117500
+    },
+    {
+      "epoch": 3.021070687933639,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.0018791571724826545,
+      "loss": 5.6778,
+      "step": 118000
+    },
+    {
+      "epoch": 3.0338718349164084,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0018786451266033437,
+      "loss": 5.6775,
+      "step": 118500
+    },
+    {
+      "epoch": 3.046672981899178,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0018781330807240329,
+      "loss": 5.6785,
+      "step": 119000
+    },
+    {
+      "epoch": 3.059474128881948,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.001877621034844722,
+      "loss": 5.6803,
+      "step": 119500
+    },
+    {
+      "epoch": 3.0722752758647176,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.0018771089889654115,
+      "loss": 5.68,
+      "step": 120000
+    },
+    {
+      "epoch": 3.085076422847487,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0018765969430861004,
+      "loss": 5.68,
+      "step": 120500
+    },
+    {
+      "epoch": 3.0978775698302567,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0018760848972067899,
+      "loss": 5.6809,
+      "step": 121000
+    },
+    {
+      "epoch": 3.1106787168130263,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.001875572851327479,
+      "loss": 5.6795,
+      "step": 121500
+    },
+    {
+      "epoch": 3.123479863795796,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.0018750608054481682,
+      "loss": 5.6798,
+      "step": 122000
+    },
+    {
+      "epoch": 3.136281010778566,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0018745487595688574,
+      "loss": 5.6797,
+      "step": 122500
+    },
+    {
+      "epoch": 3.1490821577613355,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.0018740367136895466,
+      "loss": 5.6787,
+      "step": 123000
+    },
+    {
+      "epoch": 3.161883304744105,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0018735246678102358,
+      "loss": 5.6778,
+      "step": 123500
+    },
+    {
+      "epoch": 3.1746844517268746,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.0018730126219309252,
+      "loss": 5.6766,
+      "step": 124000
+    },
+    {
+      "epoch": 3.187485598709644,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.0018725005760516142,
+      "loss": 5.677,
+      "step": 124500
+    },
+    {
+      "epoch": 3.2002867456924142,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.0018719885301723034,
+      "loss": 5.6768,
+      "step": 125000
+    },
+    {
+      "epoch": 3.213087892675184,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.0018714764842929928,
+      "loss": 5.6777,
+      "step": 125500
+    },
+    {
+      "epoch": 3.2258890396579534,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.001870964438413682,
+      "loss": 5.6744,
+      "step": 126000
+    },
+    {
+      "epoch": 3.238690186640723,
+      "grad_norm": 9.1875,
+      "learning_rate": 0.001870452392534371,
+      "loss": 5.6761,
+      "step": 126500
+    },
+    {
+      "epoch": 3.2514913336234925,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.0018699403466550604,
+      "loss": 5.6755,
+      "step": 127000
+    },
+    {
+      "epoch": 3.2642924806062625,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.0018694283007757496,
+      "loss": 5.6765,
+      "step": 127500
+    },
+    {
+      "epoch": 3.277093627589032,
+      "grad_norm": 4.8125,
+      "learning_rate": 0.0018689162548964388,
+      "loss": 5.6759,
+      "step": 128000
+    },
+    {
+      "epoch": 3.2898947745718017,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.001868404209017128,
+      "loss": 5.6742,
+      "step": 128500
+    },
+    {
+      "epoch": 3.3026959215545713,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.0018678921631378172,
+      "loss": 5.6755,
+      "step": 129000
+    },
+    {
+      "epoch": 3.315497068537341,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0018673801172585064,
+      "loss": 5.6754,
+      "step": 129500
+    },
+    {
+      "epoch": 3.3282982155201104,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0018668680713791958,
+      "loss": 5.6747,
+      "step": 130000
+    },
+    {
+      "epoch": 3.3410993625028804,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0018663560254998848,
+      "loss": 5.6731,
+      "step": 130500
+    },
+    {
+      "epoch": 3.35390050948565,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.001865843979620574,
+      "loss": 5.6741,
+      "step": 131000
+    },
+    {
+      "epoch": 3.3667016564684196,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0018653319337412634,
+      "loss": 5.6728,
+      "step": 131500
+    },
+    {
+      "epoch": 3.379502803451189,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.0018648198878619523,
+      "loss": 5.6729,
+      "step": 132000
+    },
+    {
+      "epoch": 3.3923039504339587,
+      "grad_norm": 0.875,
+      "learning_rate": 0.0018643078419826415,
+      "loss": 5.6729,
+      "step": 132500
+    },
+    {
+      "epoch": 3.4051050974167287,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.001863795796103331,
+      "loss": 5.6722,
+      "step": 133000
+    },
+    {
+      "epoch": 3.4179062443994983,
+      "grad_norm": 0.9765625,
+      "learning_rate": 0.0018632837502240201,
+      "loss": 5.6719,
+      "step": 133500
+    },
+    {
+      "epoch": 3.430707391382268,
+      "grad_norm": 1.0,
+      "learning_rate": 0.0018627717043447091,
+      "loss": 5.6708,
+      "step": 134000
+    },
+    {
+      "epoch": 3.4435085383650375,
+      "grad_norm": 7.90625,
+      "learning_rate": 0.0018622596584653985,
+      "loss": 5.6711,
+      "step": 134500
+    },
+    {
+      "epoch": 3.456309685347807,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.0018617476125860877,
+      "loss": 5.6712,
+      "step": 135000
+    },
+    {
+      "epoch": 3.4691108323305766,
+      "grad_norm": 2.625,
+      "learning_rate": 0.0018612355667067771,
+      "loss": 5.6725,
+      "step": 135500
+    },
+    {
+      "epoch": 3.4819119793133466,
+      "grad_norm": 2.125,
+      "learning_rate": 0.0018607235208274661,
+      "loss": 5.6723,
+      "step": 136000
+    },
+    {
+      "epoch": 3.494713126296116,
+      "grad_norm": 0.8359375,
+      "learning_rate": 0.0018602114749481553,
+      "loss": 5.6712,
+      "step": 136500
+    },
+    {
+      "epoch": 3.5075142732788858,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.0018596994290688447,
+      "loss": 5.6703,
+      "step": 137000
+    },
+    {
+      "epoch": 3.5203154202616553,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.001859187383189534,
+      "loss": 5.671,
+      "step": 137500
+    },
+    {
+      "epoch": 3.5331165672444254,
+      "grad_norm": 34.75,
+      "learning_rate": 0.001858675337310223,
+      "loss": 5.669,
+      "step": 138000
+    },
+    {
+      "epoch": 3.545917714227195,
+      "grad_norm": 1.3046875,
+      "learning_rate": 0.0018581632914309123,
+      "loss": 5.6699,
+      "step": 138500
+    },
+    {
+      "epoch": 3.5587188612099645,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.0018576512455516015,
+      "loss": 5.6708,
+      "step": 139000
+    },
+    {
+      "epoch": 3.571520008192734,
+      "grad_norm": 1.8984375,
+      "learning_rate": 0.0018571391996722907,
+      "loss": 5.6699,
+      "step": 139500
+    },
+    {
+      "epoch": 3.5843211551755036,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.0018566271537929799,
+      "loss": 5.6693,
+      "step": 140000
+    },
+    {
+      "epoch": 3.597122302158273,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.001856115107913669,
+      "loss": 5.6688,
+      "step": 140500
+    },
+    {
+      "epoch": 3.609923449141043,
+      "grad_norm": 1.5859375,
+      "learning_rate": 0.0018556030620343583,
+      "loss": 5.6677,
+      "step": 141000
+    },
+    {
+      "epoch": 3.622724596123813,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.0018550910161550477,
+      "loss": 5.6676,
+      "step": 141500
+    },
+    {
+      "epoch": 3.6355257431065824,
+      "grad_norm": 0.478515625,
+      "learning_rate": 0.0018545789702757367,
+      "loss": 5.6683,
+      "step": 142000
+    },
+    {
+      "epoch": 3.648326890089352,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.0018540669243964259,
+      "loss": 5.6689,
+      "step": 142500
+    },
+    {
+      "epoch": 3.6611280370721215,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.0018535548785171153,
+      "loss": 5.6685,
+      "step": 143000
+    },
+    {
+      "epoch": 3.6739291840548916,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0018530428326378045,
+      "loss": 5.669,
+      "step": 143500
+    },
+    {
+      "epoch": 3.686730331037661,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.0018525307867584934,
+      "loss": 5.668,
+      "step": 144000
+    },
+    {
+      "epoch": 3.6995314780204307,
+      "grad_norm": 4.34375,
+      "learning_rate": 0.0018520187408791829,
+      "loss": 5.6681,
+      "step": 144500
+    },
+    {
+      "epoch": 3.7123326250032003,
+      "grad_norm": 65.0,
+      "learning_rate": 0.001851506694999872,
+      "loss": 5.6675,
+      "step": 145000
+    },
+    {
+      "epoch": 3.72513377198597,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0018509946491205612,
+      "loss": 5.6677,
+      "step": 145500
+    },
+    {
+      "epoch": 3.7379349189687394,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.0018504826032412504,
+      "loss": 5.668,
+      "step": 146000
+    },
+    {
+      "epoch": 3.750736065951509,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.0018499705573619396,
+      "loss": 5.668,
+      "step": 146500
+    },
+    {
+      "epoch": 3.763537212934279,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.0018494585114826288,
+      "loss": 5.6679,
+      "step": 147000
+    },
+    {
+      "epoch": 3.7763383599170486,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.0018489464656033182,
+      "loss": 5.6673,
+      "step": 147500
+    },
+    {
+      "epoch": 3.789139506899818,
+      "grad_norm": 2.28125,
+      "learning_rate": 0.0018484344197240072,
+      "loss": 5.6681,
+      "step": 148000
+    },
+    {
+      "epoch": 3.8019406538825877,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0018479223738446964,
+      "loss": 5.6675,
+      "step": 148500
+    },
+    {
+      "epoch": 3.8147418008653577,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.0018474103279653858,
+      "loss": 5.667,
+      "step": 149000
+    },
+    {
+      "epoch": 3.8275429478481273,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.001846898282086075,
+      "loss": 5.6675,
+      "step": 149500
+    },
+    {
+      "epoch": 3.840344094830897,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0018463862362067642,
+      "loss": 5.6677,
+      "step": 150000
+    },
+    {
+      "epoch": 3.8531452418136665,
+      "grad_norm": 2.46875,
+      "learning_rate": 0.0018458741903274534,
+      "loss": 5.6678,
+      "step": 150500
+    },
+    {
+      "epoch": 3.865946388796436,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.0018453621444481426,
+      "loss": 5.6668,
+      "step": 151000
+    },
+    {
+      "epoch": 3.8787475357792056,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.001844850098568832,
+      "loss": 5.6688,
+      "step": 151500
+    },
+    {
+      "epoch": 3.8915486827619756,
+      "grad_norm": 1.484375,
+      "learning_rate": 0.001844338052689521,
+      "loss": 5.6689,
+      "step": 152000
+    },
+    {
+      "epoch": 3.904349829744745,
+      "grad_norm": 2.828125,
+      "learning_rate": 0.0018438260068102102,
+      "loss": 5.6685,
+      "step": 152500
+    },
+    {
+      "epoch": 3.9171509767275148,
+      "grad_norm": 4.84375,
+      "learning_rate": 0.0018433139609308996,
+      "loss": 5.6674,
+      "step": 153000
+    },
+    {
+      "epoch": 3.9299521237102844,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.0018428019150515886,
+      "loss": 5.6678,
+      "step": 153500
+    },
+    {
+      "epoch": 3.9427532706930544,
+      "grad_norm": 0.8125,
+      "learning_rate": 0.0018422898691722778,
+      "loss": 5.6673,
+      "step": 154000
+    },
+    {
+      "epoch": 3.955554417675824,
+      "grad_norm": 0.67578125,
+      "learning_rate": 0.0018417778232929672,
+      "loss": 5.6667,
+      "step": 154500
+    },
+    {
+      "epoch": 3.9683555646585935,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0018412657774136564,
+      "loss": 5.6673,
+      "step": 155000
+    },
+    {
+      "epoch": 3.981156711641363,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0018407537315343453,
+      "loss": 5.6677,
+      "step": 155500
+    },
+    {
+      "epoch": 3.9939578586241327,
+      "grad_norm": 3.03125,
+      "learning_rate": 0.0018402416856550348,
+      "loss": 5.6668,
+      "step": 156000
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 5.668220520019531,
+      "eval_runtime": 1.2989,
+      "eval_samples_per_second": 769.908,
+      "eval_steps_per_second": 3.08,
+      "step": 156236
+    },
+    {
+      "epoch": 4.006759005606902,
+      "grad_norm": 17.75,
+      "learning_rate": 0.001839729639775724,
+      "loss": 5.667,
+      "step": 156500
+    },
+    {
+      "epoch": 4.019560152589672,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.0018392175938964131,
+      "loss": 5.6679,
+      "step": 157000
+    },
+    {
+      "epoch": 4.032361299572441,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0018387055480171023,
+      "loss": 5.6671,
+      "step": 157500
+    },
+    {
+      "epoch": 4.045162446555211,
+      "grad_norm": 1.375,
+      "learning_rate": 0.0018381935021377915,
+      "loss": 5.6664,
+      "step": 158000
+    },
+    {
+      "epoch": 4.057963593537981,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.0018376814562584807,
+      "loss": 5.6662,
+      "step": 158500
+    },
+    {
+      "epoch": 4.070764740520751,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.0018371694103791701,
+      "loss": 5.6671,
+      "step": 159000
+    },
+    {
+      "epoch": 4.083565887503521,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.0018366573644998591,
+      "loss": 5.6659,
+      "step": 159500
+    },
+    {
+      "epoch": 4.09636703448629,
+      "grad_norm": 2.484375,
+      "learning_rate": 0.0018361453186205483,
+      "loss": 5.6665,
+      "step": 160000
+    },
+    {
+      "epoch": 4.10916818146906,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0018356332727412377,
+      "loss": 5.6667,
+      "step": 160500
+    },
+    {
+      "epoch": 4.121969328451829,
+      "grad_norm": 100.5,
+      "learning_rate": 0.001835121226861927,
+      "loss": 5.6653,
+      "step": 161000
+    },
+    {
+      "epoch": 4.134770475434599,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.001834609180982616,
+      "loss": 5.665,
+      "step": 161500
+    },
+    {
+      "epoch": 4.147571622417368,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.0018340971351033053,
+      "loss": 5.6653,
+      "step": 162000
+    },
+    {
+      "epoch": 4.160372769400138,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.0018335850892239945,
+      "loss": 5.6656,
+      "step": 162500
+    },
+    {
+      "epoch": 4.173173916382908,
+      "grad_norm": 1.59375,
+      "learning_rate": 0.001833073043344684,
+      "loss": 5.6667,
+      "step": 163000
+    },
+    {
+      "epoch": 4.185975063365677,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.0018325609974653729,
+      "loss": 5.6655,
+      "step": 163500
+    },
+    {
+      "epoch": 4.198776210348448,
+      "grad_norm": 18.0,
+      "learning_rate": 0.001832048951586062,
+      "loss": 5.6638,
+      "step": 164000
+    },
+    {
+      "epoch": 4.211577357331217,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.0018315369057067515,
+      "loss": 5.6652,
+      "step": 164500
+    },
+    {
+      "epoch": 4.224378504313987,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0018310248598274407,
+      "loss": 5.6653,
+      "step": 165000
+    },
+    {
+      "epoch": 4.237179651296756,
+      "grad_norm": 2.0625,
+      "learning_rate": 0.0018305128139481297,
+      "loss": 5.6651,
+      "step": 165500
+    },
+    {
+      "epoch": 4.249980798279526,
+      "grad_norm": 1.5390625,
+      "learning_rate": 0.001830000768068819,
+      "loss": 5.6659,
+      "step": 166000
+    },
+    {
+      "epoch": 4.2627819452622955,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.0018294887221895083,
+      "loss": 5.666,
+      "step": 166500
+    },
+    {
+      "epoch": 4.275583092245065,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.0018289766763101975,
+      "loss": 5.6654,
+      "step": 167000
+    },
+    {
+      "epoch": 4.288384239227835,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0018284646304308867,
+      "loss": 5.6649,
+      "step": 167500
+    },
+    {
+      "epoch": 4.301185386210604,
+      "grad_norm": 2.1875,
+      "learning_rate": 0.0018279525845515759,
+      "loss": 5.6648,
+      "step": 168000
+    },
+    {
+      "epoch": 4.313986533193374,
+      "grad_norm": 1.5546875,
+      "learning_rate": 0.001827440538672265,
+      "loss": 5.6647,
+      "step": 168500
+    },
+    {
+      "epoch": 4.326787680176144,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.0018269284927929545,
+      "loss": 5.6647,
+      "step": 169000
+    },
+    {
+      "epoch": 4.339588827158914,
+      "grad_norm": 1.890625,
+      "learning_rate": 0.0018264164469136434,
+      "loss": 5.6648,
+      "step": 169500
+    },
+    {
+      "epoch": 4.352389974141683,
+      "grad_norm": 1.640625,
+      "learning_rate": 0.0018259044010343326,
+      "loss": 5.6653,
+      "step": 170000
+    },
+    {
+      "epoch": 4.365191121124453,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.001825392355155022,
+      "loss": 5.6641,
+      "step": 170500
+    },
+    {
+      "epoch": 4.3779922681072225,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0018248803092757112,
+      "loss": 5.6647,
+      "step": 171000
+    },
+    {
+      "epoch": 4.390793415089992,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.0018243682633964002,
+      "loss": 5.6635,
+      "step": 171500
+    },
+    {
+      "epoch": 4.403594562072762,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.0018238562175170896,
+      "loss": 5.6634,
+      "step": 172000
+    },
+    {
+      "epoch": 4.416395709055531,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.0018233441716377788,
+      "loss": 5.6635,
+      "step": 172500
+    },
+    {
+      "epoch": 4.429196856038301,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.0018228321257584678,
+      "loss": 5.6632,
+      "step": 173000
+    },
+    {
+      "epoch": 4.44199800302107,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.0018223200798791572,
+      "loss": 5.6634,
+      "step": 173500
+    },
+    {
+      "epoch": 4.45479915000384,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.0018218080339998464,
+      "loss": 5.6636,
+      "step": 174000
+    },
+    {
+      "epoch": 4.46760029698661,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0018212959881205356,
+      "loss": 5.6626,
+      "step": 174500
+    },
+    {
+      "epoch": 4.48040144396938,
+      "grad_norm": 6.09375,
+      "learning_rate": 0.0018207839422412248,
+      "loss": 5.6632,
+      "step": 175000
+    },
+    {
+      "epoch": 4.49320259095215,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.001820271896361914,
+      "loss": 5.6624,
+      "step": 175500
+    },
+    {
+      "epoch": 4.506003737934919,
+      "grad_norm": 3.78125,
+      "learning_rate": 0.0018197598504826032,
+      "loss": 5.6623,
+      "step": 176000
+    },
+    {
+      "epoch": 4.518804884917689,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0018192478046032926,
+      "loss": 5.6628,
+      "step": 176500
+    },
+    {
+      "epoch": 4.531606031900458,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.0018187357587239816,
+      "loss": 5.6627,
+      "step": 177000
+    },
+    {
+      "epoch": 4.544407178883228,
+      "grad_norm": 1.6328125,
+      "learning_rate": 0.001818223712844671,
+      "loss": 5.6615,
+      "step": 177500
+    },
+    {
+      "epoch": 4.5572083258659974,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.0018177116669653602,
+      "loss": 5.662,
+      "step": 178000
+    },
+    {
+      "epoch": 4.570009472848767,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0018171996210860494,
+      "loss": 5.6625,
+      "step": 178500
+    },
+    {
+      "epoch": 4.582810619831537,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.0018166875752067386,
+      "loss": 5.6618,
+      "step": 179000
+    },
+    {
+      "epoch": 4.595611766814306,
+      "grad_norm": 0.81640625,
+      "learning_rate": 0.0018161755293274278,
+      "loss": 5.6621,
+      "step": 179500
+    },
+    {
+      "epoch": 4.608412913797077,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.001815663483448117,
+      "loss": 5.6621,
+      "step": 180000
+    },
+    {
+      "epoch": 4.621214060779846,
+      "grad_norm": 3.34375,
+      "learning_rate": 0.0018151514375688064,
+      "loss": 5.6606,
+      "step": 180500
+    },
+    {
+      "epoch": 4.634015207762616,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.0018146393916894953,
+      "loss": 5.6616,
+      "step": 181000
+    },
+    {
+      "epoch": 4.646816354745385,
+      "grad_norm": 13.5625,
+      "learning_rate": 0.0018141273458101845,
+      "loss": 5.6626,
+      "step": 181500
+    },
+    {
+      "epoch": 4.659617501728155,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.001813615299930874,
+      "loss": 5.661,
+      "step": 182000
+    },
+    {
+      "epoch": 4.6724186487109245,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.0018131032540515631,
+      "loss": 5.6614,
+      "step": 182500
+    },
+    {
+      "epoch": 4.685219795693694,
+      "grad_norm": 3.03125,
+      "learning_rate": 0.0018125912081722521,
+      "loss": 5.6612,
+      "step": 183000
+    },
+    {
+      "epoch": 4.698020942676464,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0018120791622929415,
+      "loss": 5.66,
+      "step": 183500
+    },
+    {
+      "epoch": 4.710822089659233,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.0018115671164136307,
+      "loss": 5.6592,
+      "step": 184000
+    },
+    {
+      "epoch": 4.723623236642003,
+      "grad_norm": 4.21875,
+      "learning_rate": 0.00181105507053432,
+      "loss": 5.659,
+      "step": 184500
+    },
+    {
+      "epoch": 4.736424383624772,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.0018105430246550091,
+      "loss": 5.6591,
+      "step": 185000
+    },
+    {
+      "epoch": 4.749225530607543,
+      "grad_norm": 5.0625,
+      "learning_rate": 0.0018100309787756983,
+      "loss": 5.6591,
+      "step": 185500
+    },
+    {
+      "epoch": 4.762026677590312,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.0018095189328963875,
+      "loss": 5.659,
+      "step": 186000
+    },
+    {
+      "epoch": 4.774827824573082,
+      "grad_norm": 0.640625,
+      "learning_rate": 0.001809006887017077,
+      "loss": 5.6603,
+      "step": 186500
+    },
+    {
+      "epoch": 4.7876289715558515,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.001808494841137766,
+      "loss": 5.6594,
+      "step": 187000
+    },
+    {
+      "epoch": 4.800430118538621,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.001807982795258455,
+      "loss": 5.6604,
+      "step": 187500
+    },
+    {
+      "epoch": 4.813231265521391,
+      "grad_norm": 3.40625,
+      "learning_rate": 0.0018074707493791445,
+      "loss": 5.6598,
+      "step": 188000
+    },
+    {
+      "epoch": 4.82603241250416,
+      "grad_norm": 2.109375,
+      "learning_rate": 0.0018069587034998337,
+      "loss": 5.6597,
+      "step": 188500
+    },
+    {
+      "epoch": 4.83883355948693,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0018064466576205227,
+      "loss": 5.6594,
+      "step": 189000
+    },
+    {
+      "epoch": 4.851634706469699,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.001805934611741212,
+      "loss": 5.6584,
+      "step": 189500
+    },
+    {
+      "epoch": 4.864435853452469,
+      "grad_norm": 0.59765625,
+      "learning_rate": 0.0018054225658619013,
+      "loss": 5.6582,
+      "step": 190000
+    },
+    {
+      "epoch": 4.877237000435239,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.0018049105199825905,
+      "loss": 5.6581,
+      "step": 190500
+    },
+    {
+      "epoch": 4.890038147418009,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0018043984741032797,
+      "loss": 5.658,
+      "step": 191000
+    },
+    {
+      "epoch": 4.902839294400779,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0018038864282239689,
+      "loss": 5.6576,
+      "step": 191500
+    },
+    {
+      "epoch": 4.915640441383548,
+      "grad_norm": 2.453125,
+      "learning_rate": 0.0018033743823446583,
+      "loss": 5.6576,
+      "step": 192000
+    },
+    {
+      "epoch": 4.928441588366318,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.0018028623364653475,
+      "loss": 5.6574,
+      "step": 192500
+    },
+    {
+      "epoch": 4.941242735349087,
+      "grad_norm": 2.4375,
+      "learning_rate": 0.0018023502905860364,
+      "loss": 5.6571,
+      "step": 193000
+    },
+    {
+      "epoch": 4.954043882331857,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.0018018382447067259,
+      "loss": 5.6567,
+      "step": 193500
+    },
+    {
+      "epoch": 4.9668450293146265,
+      "grad_norm": 0.79296875,
+      "learning_rate": 0.001801326198827415,
+      "loss": 5.6561,
+      "step": 194000
+    },
+    {
+      "epoch": 4.979646176297396,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.0018008141529481042,
+      "loss": 5.6571,
+      "step": 194500
+    },
+    {
+      "epoch": 4.992447323280166,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.0018003021070687934,
+      "loss": 5.6571,
+      "step": 195000
+    },
+    {
+      "epoch": 5.0,
+      "eval_loss": 5.6548848152160645,
+      "eval_runtime": 1.2976,
+      "eval_samples_per_second": 770.633,
+      "eval_steps_per_second": 3.083,
+      "step": 195295
+    },
+    {
+      "epoch": 5.005248470262935,
+      "grad_norm": 2.171875,
+      "learning_rate": 0.0017997900611894826,
+      "loss": 5.6575,
+      "step": 195500
+    },
+    {
+      "epoch": 5.018049617245706,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0017992780153101718,
+      "loss": 5.6569,
+      "step": 196000
+    },
+    {
+      "epoch": 5.030850764228475,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.001798765969430861,
+      "loss": 5.6566,
+      "step": 196500
+    },
+    {
+      "epoch": 5.043651911211245,
+      "grad_norm": 14.4375,
+      "learning_rate": 0.0017982539235515502,
+      "loss": 5.6565,
+      "step": 197000
+    },
+    {
+      "epoch": 5.056453058194014,
+      "grad_norm": 2.984375,
+      "learning_rate": 0.0017977418776722394,
+      "loss": 5.6566,
+      "step": 197500
+    },
+    {
+      "epoch": 5.069254205176784,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.0017972298317929288,
+      "loss": 5.6572,
+      "step": 198000
+    },
+    {
+      "epoch": 5.0820553521595535,
+      "grad_norm": 1.5078125,
+      "learning_rate": 0.0017967177859136178,
+      "loss": 5.6556,
+      "step": 198500
+    },
+    {
+      "epoch": 5.094856499142323,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.001796205740034307,
+      "loss": 5.6562,
+      "step": 199000
+    },
+    {
+      "epoch": 5.107657646125093,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0017956936941549964,
+      "loss": 5.656,
+      "step": 199500
+    },
+    {
+      "epoch": 5.120458793107862,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.0017951816482756856,
+      "loss": 5.6562,
+      "step": 200000
+    },
+    {
+      "epoch": 5.133259940090632,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0017946696023963746,
+      "loss": 5.656,
+      "step": 200500
+    },
+    {
+      "epoch": 5.146061087073401,
+      "grad_norm": 4.75,
+      "learning_rate": 0.001794157556517064,
+      "loss": 5.6567,
+      "step": 201000
+    },
+    {
+      "epoch": 5.158862234056172,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.0017936455106377532,
+      "loss": 5.6557,
+      "step": 201500
+    },
+    {
+      "epoch": 5.171663381038941,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0017931334647584424,
+      "loss": 5.6558,
+      "step": 202000
+    },
+    {
+      "epoch": 5.184464528021711,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0017926214188791316,
+      "loss": 5.6565,
+      "step": 202500
+    },
+    {
+      "epoch": 5.197265675004481,
+      "grad_norm": 2.375,
+      "learning_rate": 0.0017921093729998208,
+      "loss": 5.6562,
+      "step": 203000
+    },
+    {
+      "epoch": 5.21006682198725,
+      "grad_norm": 1.25,
+      "learning_rate": 0.00179159732712051,
+      "loss": 5.6569,
+      "step": 203500
+    },
+    {
+      "epoch": 5.22286796897002,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0017910852812411994,
+      "loss": 5.6573,
+      "step": 204000
+    },
+    {
+      "epoch": 5.235669115952789,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.0017905732353618883,
+      "loss": 5.6566,
+      "step": 204500
+    },
+    {
+      "epoch": 5.248470262935559,
+      "grad_norm": 1.796875,
+      "learning_rate": 0.0017900611894825775,
+      "loss": 5.6567,
+      "step": 205000
+    },
+    {
+      "epoch": 5.261271409918328,
+      "grad_norm": 3.203125,
+      "learning_rate": 0.001789549143603267,
+      "loss": 5.6555,
+      "step": 205500
+    },
+    {
+      "epoch": 5.274072556901098,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.0017890370977239561,
+      "loss": 5.656,
+      "step": 206000
+    },
+    {
+      "epoch": 5.2868737038838685,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.0017885250518446453,
+      "loss": 5.655,
+      "step": 206500
+    },
+    {
+      "epoch": 5.299674850866638,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.0017880130059653345,
+      "loss": 5.657,
+      "step": 207000
+    },
+    {
+      "epoch": 5.312475997849408,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.0017875009600860237,
+      "loss": 5.6563,
+      "step": 207500
+    },
+    {
+      "epoch": 5.325277144832177,
+      "grad_norm": 2.8125,
+      "learning_rate": 0.0017869889142067131,
+      "loss": 5.6554,
+      "step": 208000
+    },
+    {
+      "epoch": 5.338078291814947,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.0017864768683274021,
+      "loss": 5.6564,
+      "step": 208500
+    },
+    {
+      "epoch": 5.350879438797716,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.0017859648224480913,
+      "loss": 5.6555,
+      "step": 209000
+    },
+    {
+      "epoch": 5.363680585780486,
+      "grad_norm": 1.53125,
+      "learning_rate": 0.0017854527765687807,
+      "loss": 5.6548,
+      "step": 209500
+    },
+    {
+      "epoch": 5.3764817327632555,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.00178494073068947,
+      "loss": 5.6547,
+      "step": 210000
+    },
+    {
+      "epoch": 5.389282879746025,
+      "grad_norm": 5.03125,
+      "learning_rate": 0.001784428684810159,
+      "loss": 5.6543,
+      "step": 210500
+    },
+    {
+      "epoch": 5.402084026728795,
+      "grad_norm": 1.5,
+      "learning_rate": 0.0017839166389308483,
+      "loss": 5.6549,
+      "step": 211000
+    },
+    {
+      "epoch": 5.414885173711564,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.0017834045930515375,
+      "loss": 5.6545,
+      "step": 211500
+    },
+    {
+      "epoch": 5.427686320694335,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0017828925471722267,
+      "loss": 5.6552,
+      "step": 212000
+    },
+    {
+      "epoch": 5.440487467677104,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.0017823805012929159,
+      "loss": 5.6537,
+      "step": 212500
+    },
+    {
+      "epoch": 5.453288614659874,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.001781868455413605,
+      "loss": 5.6547,
+      "step": 213000
+    },
+    {
+      "epoch": 5.466089761642643,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.0017813564095342943,
+      "loss": 5.6555,
+      "step": 213500
+    },
+    {
+      "epoch": 5.478890908625413,
+      "grad_norm": 0.796875,
+      "learning_rate": 0.0017808443636549837,
+      "loss": 5.656,
+      "step": 214000
+    },
+    {
+      "epoch": 5.4916920556081825,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.0017803323177756727,
+      "loss": 5.6533,
+      "step": 214500
+    },
+    {
+      "epoch": 5.504493202590952,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.0017798202718963619,
+      "loss": 5.6543,
+      "step": 215000
+    },
+    {
+      "epoch": 5.517294349573722,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.0017793082260170513,
+      "loss": 5.6539,
+      "step": 215500
+    },
+    {
+      "epoch": 5.530095496556491,
+      "grad_norm": 2.9375,
+      "learning_rate": 0.0017787961801377405,
+      "loss": 5.6534,
+      "step": 216000
+    },
+    {
+      "epoch": 5.542896643539261,
+      "grad_norm": 2.3125,
+      "learning_rate": 0.0017782841342584294,
+      "loss": 5.654,
+      "step": 216500
+    },
+    {
+      "epoch": 5.55569779052203,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.0017777720883791189,
+      "loss": 5.6548,
+      "step": 217000
+    },
+    {
+      "epoch": 5.568498937504801,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.001777260042499808,
+      "loss": 5.6524,
+      "step": 217500
+    },
+    {
+      "epoch": 5.58130008448757,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.001776747996620497,
+      "loss": 5.6526,
+      "step": 218000
+    },
+    {
+      "epoch": 5.59410123147034,
+      "grad_norm": 1.5,
+      "learning_rate": 0.0017762359507411864,
+      "loss": 5.6529,
+      "step": 218500
+    },
+    {
+      "epoch": 5.60690237845311,
+      "grad_norm": 0.7265625,
+      "learning_rate": 0.0017757239048618756,
+      "loss": 5.6527,
+      "step": 219000
+    },
+    {
+      "epoch": 5.619703525435879,
+      "grad_norm": 3.125,
+      "learning_rate": 0.0017752118589825648,
+      "loss": 5.6531,
+      "step": 219500
+    },
+    {
+      "epoch": 5.632504672418649,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.001774699813103254,
+      "loss": 5.6524,
+      "step": 220000
+    },
+    {
+      "epoch": 5.645305819401418,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0017741877672239432,
+      "loss": 5.6532,
+      "step": 220500
+    },
+    {
+      "epoch": 5.658106966384188,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.0017736757213446326,
+      "loss": 5.6526,
+      "step": 221000
+    },
+    {
+      "epoch": 5.670908113366957,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.0017731636754653218,
+      "loss": 5.6519,
+      "step": 221500
+    },
+    {
+      "epoch": 5.683709260349727,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0017726516295860108,
+      "loss": 5.6525,
+      "step": 222000
+    },
+    {
+      "epoch": 5.696510407332497,
+      "grad_norm": 1.3125,
+      "learning_rate": 0.0017721395837067002,
+      "loss": 5.6529,
+      "step": 222500
+    },
+    {
+      "epoch": 5.709311554315267,
+      "grad_norm": 0.5390625,
+      "learning_rate": 0.0017716275378273894,
+      "loss": 5.6526,
+      "step": 223000
+    },
+    {
+      "epoch": 5.722112701298037,
+      "grad_norm": 1.375,
+      "learning_rate": 0.0017711154919480786,
+      "loss": 5.6515,
+      "step": 223500
+    },
+    {
+      "epoch": 5.734913848280806,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.0017706034460687678,
+      "loss": 5.6514,
+      "step": 224000
+    },
+    {
+      "epoch": 5.747714995263576,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.001770091400189457,
+      "loss": 5.6523,
+      "step": 224500
+    },
+    {
+      "epoch": 5.760516142246345,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0017695793543101462,
+      "loss": 5.652,
+      "step": 225000
+    },
+    {
+      "epoch": 5.773317289229115,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.0017690673084308356,
+      "loss": 5.6518,
+      "step": 225500
+    },
+    {
+      "epoch": 5.7861184362118845,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.0017685552625515246,
+      "loss": 5.6524,
+      "step": 226000
+    },
+    {
+      "epoch": 5.798919583194654,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.0017680432166722138,
+      "loss": 5.652,
+      "step": 226500
+    },
+    {
+      "epoch": 5.811720730177424,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.0017675311707929032,
+      "loss": 5.6516,
+      "step": 227000
+    },
+    {
+      "epoch": 5.824521877160193,
+      "grad_norm": 0.421875,
+      "learning_rate": 0.0017670191249135924,
+      "loss": 5.6511,
+      "step": 227500
+    },
+    {
+      "epoch": 5.837323024142963,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.0017665070790342813,
+      "loss": 5.6505,
+      "step": 228000
+    },
+    {
+      "epoch": 5.850124171125733,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.0017659950331549708,
+      "loss": 5.6527,
+      "step": 228500
+    },
+    {
+      "epoch": 5.862925318108503,
+      "grad_norm": 5.4375,
+      "learning_rate": 0.00176548298727566,
+      "loss": 5.6513,
+      "step": 229000
+    },
+    {
+      "epoch": 5.875726465091272,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.0017649709413963491,
+      "loss": 5.6503,
+      "step": 229500
+    },
+    {
+      "epoch": 5.888527612074042,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.0017644588955170383,
+      "loss": 5.6508,
+      "step": 230000
+    },
+    {
+      "epoch": 5.9013287590568115,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.0017639468496377275,
+      "loss": 5.6517,
+      "step": 230500
+    },
+    {
+      "epoch": 5.914129906039581,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.0017634348037584167,
+      "loss": 5.6514,
+      "step": 231000
+    },
+    {
+      "epoch": 5.926931053022351,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0017629227578791061,
+      "loss": 5.6513,
+      "step": 231500
+    },
+    {
+      "epoch": 5.93973220000512,
+      "grad_norm": 3.78125,
+      "learning_rate": 0.0017624107119997951,
+      "loss": 5.6507,
+      "step": 232000
+    },
+    {
+      "epoch": 5.95253334698789,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0017618986661204843,
+      "loss": 5.6494,
+      "step": 232500
+    },
+    {
+      "epoch": 5.965334493970659,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.0017613866202411737,
+      "loss": 5.6515,
+      "step": 233000
+    },
+    {
+      "epoch": 5.978135640953429,
+      "grad_norm": 2.515625,
+      "learning_rate": 0.001760874574361863,
+      "loss": 5.6506,
+      "step": 233500
+    },
+    {
+      "epoch": 5.990936787936199,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.0017603625284825521,
+      "loss": 5.6504,
+      "step": 234000
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 5.650012016296387,
+      "eval_runtime": 1.2888,
+      "eval_samples_per_second": 775.89,
+      "eval_steps_per_second": 3.104,
+      "step": 234354
+    },
+    {
+      "epoch": 6.003737934918969,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.0017598504826032413,
+      "loss": 5.6498,
+      "step": 234500
+    },
+    {
+      "epoch": 6.016539081901739,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.0017593384367239305,
+      "loss": 5.6508,
+      "step": 235000
+    },
+    {
+      "epoch": 6.029340228884508,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.00175882639084462,
+      "loss": 5.6499,
+      "step": 235500
+    },
+    {
+      "epoch": 6.042141375867278,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.001758314344965309,
+      "loss": 5.6495,
+      "step": 236000
+    },
+    {
+      "epoch": 6.054942522850047,
+      "grad_norm": 2.265625,
+      "learning_rate": 0.001757802299085998,
+      "loss": 5.6499,
+      "step": 236500
+    },
+    {
+      "epoch": 6.067743669832817,
+      "grad_norm": 0.71875,
+      "learning_rate": 0.0017572902532066875,
+      "loss": 5.6505,
+      "step": 237000
+    },
+    {
+      "epoch": 6.0805448168155865,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.0017567782073273767,
+      "loss": 5.6499,
+      "step": 237500
+    },
+    {
+      "epoch": 6.093345963798356,
+      "grad_norm": 0.8125,
+      "learning_rate": 0.0017562661614480657,
+      "loss": 5.6518,
+      "step": 238000
+    },
+    {
+      "epoch": 6.106147110781126,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.001755754115568755,
+      "loss": 5.6496,
+      "step": 238500
+    },
+    {
+      "epoch": 6.118948257763896,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.0017552420696894443,
+      "loss": 5.6497,
+      "step": 239000
+    },
+    {
+      "epoch": 6.131749404746666,
+      "grad_norm": 1.3828125,
+      "learning_rate": 0.0017547300238101333,
+      "loss": 5.6508,
+      "step": 239500
+    },
+    {
+      "epoch": 6.144550551729435,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.0017542179779308227,
+      "loss": 5.6498,
+      "step": 240000
+    },
+    {
+      "epoch": 6.157351698712205,
+      "grad_norm": 3.296875,
+      "learning_rate": 0.0017537059320515119,
+      "loss": 5.6499,
+      "step": 240500
+    },
+    {
+      "epoch": 6.170152845694974,
+      "grad_norm": 1.0390625,
+      "learning_rate": 0.001753193886172201,
+      "loss": 5.6491,
+      "step": 241000
+    },
+    {
+      "epoch": 6.182953992677744,
+      "grad_norm": 1.34375,
+      "learning_rate": 0.0017526818402928902,
+      "loss": 5.6488,
+      "step": 241500
+    },
+    {
+      "epoch": 6.1957551396605135,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.0017521697944135794,
+      "loss": 5.6494,
+      "step": 242000
+    },
+    {
+      "epoch": 6.208556286643283,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0017516577485342686,
+      "loss": 5.6489,
+      "step": 242500
+    },
+    {
+      "epoch": 6.221357433626053,
+      "grad_norm": 2.140625,
+      "learning_rate": 0.001751145702654958,
+      "loss": 5.6495,
+      "step": 243000
+    },
+    {
+      "epoch": 6.234158580608822,
+      "grad_norm": 2.109375,
+      "learning_rate": 0.001750633656775647,
+      "loss": 5.6492,
+      "step": 243500
+    },
+    {
+      "epoch": 6.246959727591592,
+      "grad_norm": 0.80859375,
+      "learning_rate": 0.0017501216108963362,
+      "loss": 5.6487,
+      "step": 244000
+    },
+    {
+      "epoch": 6.259760874574362,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.0017496095650170256,
+      "loss": 5.6481,
+      "step": 244500
+    },
+    {
+      "epoch": 6.272562021557132,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.0017490975191377148,
+      "loss": 5.6491,
+      "step": 245000
+    },
+    {
+      "epoch": 6.285363168539901,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.0017485854732584038,
+      "loss": 5.6486,
+      "step": 245500
+    },
+    {
+      "epoch": 6.298164315522671,
+      "grad_norm": 0.55078125,
+      "learning_rate": 0.0017480734273790932,
+      "loss": 5.6478,
+      "step": 246000
+    },
+    {
+      "epoch": 6.3109654625054405,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.0017475613814997824,
+      "loss": 5.6478,
+      "step": 246500
+    },
+    {
+      "epoch": 6.32376660948821,
+      "grad_norm": 1.84375,
+      "learning_rate": 0.0017470493356204716,
+      "loss": 5.6494,
+      "step": 247000
+    },
+    {
+      "epoch": 6.33656775647098,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.0017465372897411608,
+      "loss": 5.6489,
+      "step": 247500
+    },
+    {
+      "epoch": 6.349368903453749,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00174602524386185,
+      "loss": 5.6482,
+      "step": 248000
+    },
+    {
+      "epoch": 6.362170050436519,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.0017455131979825394,
+      "loss": 5.6487,
+      "step": 248500
+    },
+    {
+      "epoch": 6.374971197419288,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.0017450011521032286,
+      "loss": 5.6485,
+      "step": 249000
+    },
+    {
+      "epoch": 6.387772344402059,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.0017444891062239176,
+      "loss": 5.6488,
+      "step": 249500
+    },
+    {
+      "epoch": 6.4005734913848285,
+      "grad_norm": 0.90234375,
+      "learning_rate": 0.001743977060344607,
+      "loss": 5.6484,
+      "step": 250000
+    },
+    {
+      "epoch": 6.413374638367598,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.0017434650144652962,
+      "loss": 5.6473,
+      "step": 250500
+    },
+    {
+      "epoch": 6.426175785350368,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.0017429529685859854,
+      "loss": 5.649,
+      "step": 251000
+    },
+    {
+      "epoch": 6.438976932333137,
+      "grad_norm": 1.5078125,
+      "learning_rate": 0.0017424409227066746,
+      "loss": 5.648,
+      "step": 251500
+    },
+    {
+      "epoch": 6.451778079315907,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.0017419288768273638,
+      "loss": 5.6473,
+      "step": 252000
+    },
+    {
+      "epoch": 6.464579226298676,
+      "grad_norm": 1.2734375,
+      "learning_rate": 0.001741416830948053,
+      "loss": 5.6482,
+      "step": 252500
+    },
+    {
+      "epoch": 6.477380373281446,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.0017409047850687424,
+      "loss": 5.6474,
+      "step": 253000
+    },
+    {
+      "epoch": 6.4901815202642155,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.0017403927391894313,
+      "loss": 5.6468,
+      "step": 253500
+    },
+    {
+      "epoch": 6.502982667246985,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.0017398806933101205,
+      "loss": 5.6471,
+      "step": 254000
+    },
+    {
+      "epoch": 6.515783814229755,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00173936864743081,
+      "loss": 5.6476,
+      "step": 254500
+    },
+    {
+      "epoch": 6.528584961212525,
+      "grad_norm": 8.0625,
+      "learning_rate": 0.0017388566015514991,
+      "loss": 5.6478,
+      "step": 255000
+    },
+    {
+      "epoch": 6.541386108195295,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0017383445556721881,
+      "loss": 5.648,
+      "step": 255500
+    },
+    {
+      "epoch": 6.554187255178064,
+      "grad_norm": 0.8671875,
+      "learning_rate": 0.0017378325097928775,
+      "loss": 5.648,
+      "step": 256000
+    },
+    {
+      "epoch": 6.566988402160834,
+      "grad_norm": 0.69921875,
+      "learning_rate": 0.0017373204639135667,
+      "loss": 5.6467,
+      "step": 256500
+    },
+    {
+      "epoch": 6.579789549143603,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.001736808418034256,
+      "loss": 5.6476,
+      "step": 257000
+    },
+    {
+      "epoch": 6.592590696126373,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.0017362963721549451,
+      "loss": 5.6456,
+      "step": 257500
+    },
+    {
+      "epoch": 6.6053918431091425,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.0017357843262756343,
+      "loss": 5.6473,
+      "step": 258000
+    },
+    {
+      "epoch": 6.618192990091912,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.0017352722803963235,
+      "loss": 5.6466,
+      "step": 258500
+    },
+    {
+      "epoch": 6.630994137074682,
+      "grad_norm": 1.9921875,
+      "learning_rate": 0.001734760234517013,
+      "loss": 5.6469,
+      "step": 259000
+    },
+    {
+      "epoch": 6.643795284057451,
+      "grad_norm": 0.87890625,
+      "learning_rate": 0.001734248188637702,
+      "loss": 5.6466,
+      "step": 259500
+    },
+    {
+      "epoch": 6.656596431040221,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.001733736142758391,
+      "loss": 5.6469,
+      "step": 260000
+    },
+    {
+      "epoch": 6.669397578022991,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.0017332240968790805,
+      "loss": 5.6476,
+      "step": 260500
+    },
+    {
+      "epoch": 6.682198725005761,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.0017327120509997695,
+      "loss": 5.6453,
+      "step": 261000
+    },
+    {
+      "epoch": 6.69499987198853,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.0017322000051204587,
+      "loss": 5.6458,
+      "step": 261500
+    },
+    {
+      "epoch": 6.7078010189713,
+      "grad_norm": 2.53125,
+      "learning_rate": 0.001731687959241148,
+      "loss": 5.6461,
+      "step": 262000
+    },
+    {
+      "epoch": 6.72060216595407,
+      "grad_norm": 0.396484375,
+      "learning_rate": 0.0017311759133618373,
+      "loss": 5.6458,
+      "step": 262500
+    },
+    {
+      "epoch": 6.733403312936839,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.0017306638674825265,
+      "loss": 5.6464,
+      "step": 263000
+    },
+    {
+      "epoch": 6.746204459919609,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.0017301518216032157,
+      "loss": 5.647,
+      "step": 263500
+    },
+    {
+      "epoch": 6.759005606902378,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0017296397757239049,
+      "loss": 5.6468,
+      "step": 264000
+    },
+    {
+      "epoch": 6.771806753885148,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.0017291277298445943,
+      "loss": 5.6472,
+      "step": 264500
+    },
+    {
+      "epoch": 6.784607900867917,
+      "grad_norm": 3.59375,
+      "learning_rate": 0.0017286156839652833,
+      "loss": 5.6471,
+      "step": 265000
+    },
+    {
+      "epoch": 6.797409047850687,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.0017281036380859724,
+      "loss": 5.6455,
+      "step": 265500
+    },
+    {
+      "epoch": 6.8102101948334575,
+      "grad_norm": 7.21875,
+      "learning_rate": 0.0017275915922066619,
+      "loss": 5.6456,
+      "step": 266000
+    },
+    {
+      "epoch": 6.823011341816227,
+      "grad_norm": 5.6875,
+      "learning_rate": 0.001727079546327351,
+      "loss": 5.6455,
+      "step": 266500
+    },
+    {
+      "epoch": 6.835812488798997,
+      "grad_norm": 2.578125,
+      "learning_rate": 0.00172656750044804,
+      "loss": 5.6462,
+      "step": 267000
+    },
+    {
+      "epoch": 6.848613635781766,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.0017260554545687294,
+      "loss": 5.6463,
+      "step": 267500
+    },
+    {
+      "epoch": 6.861414782764536,
+      "grad_norm": 4.1875,
+      "learning_rate": 0.0017255434086894186,
+      "loss": 5.6465,
+      "step": 268000
+    },
+    {
+      "epoch": 6.874215929747305,
+      "grad_norm": 1.21875,
+      "learning_rate": 0.0017250313628101078,
+      "loss": 5.6455,
+      "step": 268500
+    },
+    {
+      "epoch": 6.887017076730075,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.001724519316930797,
+      "loss": 5.646,
+      "step": 269000
+    },
+    {
+      "epoch": 6.8998182237128445,
+      "grad_norm": 0.578125,
+      "learning_rate": 0.0017240072710514862,
+      "loss": 5.6466,
+      "step": 269500
+    },
+    {
+      "epoch": 6.912619370695614,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.0017234952251721754,
+      "loss": 5.6451,
+      "step": 270000
+    },
+    {
+      "epoch": 6.925420517678384,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.0017229831792928648,
+      "loss": 5.6459,
+      "step": 270500
+    },
+    {
+      "epoch": 6.938221664661153,
+      "grad_norm": 3.8125,
+      "learning_rate": 0.0017224711334135538,
+      "loss": 5.6458,
+      "step": 271000
+    },
+    {
+      "epoch": 6.951022811643924,
+      "grad_norm": 0.48828125,
+      "learning_rate": 0.001721959087534243,
+      "loss": 5.646,
+      "step": 271500
+    },
+    {
+      "epoch": 6.963823958626693,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.0017214470416549324,
+      "loss": 5.6456,
+      "step": 272000
+    },
+    {
+      "epoch": 6.976625105609463,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.0017209349957756216,
+      "loss": 5.6446,
+      "step": 272500
+    },
+    {
+      "epoch": 6.989426252592232,
+      "grad_norm": 1.6953125,
+      "learning_rate": 0.0017204229498963106,
+      "loss": 5.646,
+      "step": 273000
+    },
+    {
+      "epoch": 7.0,
+      "eval_loss": 5.643965244293213,
+      "eval_runtime": 1.3794,
+      "eval_samples_per_second": 724.964,
+      "eval_steps_per_second": 2.9,
+      "step": 273413
+    },
+    {
+      "epoch": 7.002227399575002,
+      "grad_norm": 0.392578125,
+      "learning_rate": 0.001719910904017,
+      "loss": 5.6451,
+      "step": 273500
+    },
+    {
+      "epoch": 7.0150285465577715,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.0017193988581376892,
+      "loss": 5.6448,
+      "step": 274000
+    },
+    {
+      "epoch": 7.027829693540541,
+      "grad_norm": 0.671875,
+      "learning_rate": 0.0017188868122583784,
+      "loss": 5.646,
+      "step": 274500
+    },
+    {
+      "epoch": 7.040630840523311,
+      "grad_norm": 10.625,
+      "learning_rate": 0.0017183747663790676,
+      "loss": 5.6441,
+      "step": 275000
+    },
+    {
+      "epoch": 7.05343198750608,
+      "grad_norm": 0.6484375,
+      "learning_rate": 0.0017178627204997568,
+      "loss": 5.6451,
+      "step": 275500
+    },
+    {
+      "epoch": 7.06623313448885,
+      "grad_norm": 0.84375,
+      "learning_rate": 0.001717350674620446,
+      "loss": 5.645,
+      "step": 276000
+    },
+    {
+      "epoch": 7.07903428147162,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.0017168386287411354,
+      "loss": 5.6449,
+      "step": 276500
+    },
+    {
+      "epoch": 7.09183542845439,
+      "grad_norm": 0.74609375,
+      "learning_rate": 0.0017163265828618243,
+      "loss": 5.6449,
+      "step": 277000
+    },
+    {
+      "epoch": 7.104636575437159,
+      "grad_norm": 1.1015625,
+      "learning_rate": 0.0017158145369825138,
+      "loss": 5.6445,
+      "step": 277500
+    },
+    {
+      "epoch": 7.117437722419929,
+      "grad_norm": 5.0,
+      "learning_rate": 0.001715302491103203,
+      "loss": 5.6439,
+      "step": 278000
+    },
+    {
+      "epoch": 7.130238869402699,
+      "grad_norm": 3.640625,
+      "learning_rate": 0.0017147904452238921,
+      "loss": 5.6445,
+      "step": 278500
+    },
+    {
+      "epoch": 7.143040016385468,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.0017142783993445813,
+      "loss": 5.6462,
+      "step": 279000
+    },
+    {
+      "epoch": 7.155841163368238,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.0017137663534652705,
+      "loss": 5.6442,
+      "step": 279500
+    },
+    {
+      "epoch": 7.168642310351007,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.0017132543075859597,
+      "loss": 5.6443,
+      "step": 280000
+    },
+    {
+      "epoch": 7.181443457333777,
+      "grad_norm": 0.75,
+      "learning_rate": 0.0017127422617066491,
+      "loss": 5.6449,
+      "step": 280500
+    },
+    {
+      "epoch": 7.194244604316546,
+      "grad_norm": 4.25,
+      "learning_rate": 0.0017122302158273381,
+      "loss": 5.6439,
+      "step": 281000
+    },
+    {
+      "epoch": 7.207045751299316,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.0017117181699480273,
+      "loss": 5.6456,
+      "step": 281500
+    },
+    {
+      "epoch": 7.2198468982820865,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.0017112061240687167,
+      "loss": 5.6457,
+      "step": 282000
+    },
+    {
+      "epoch": 7.232648045264856,
+      "grad_norm": 1.015625,
+      "learning_rate": 0.0017106940781894057,
+      "loss": 5.6448,
+      "step": 282500
+    },
+    {
+      "epoch": 7.245449192247626,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.001710182032310095,
+      "loss": 5.6431,
+      "step": 283000
+    },
+    {
+      "epoch": 7.258250339230395,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.0017096699864307843,
+      "loss": 5.6432,
+      "step": 283500
+    },
+    {
+      "epoch": 7.271051486213165,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.0017091579405514735,
+      "loss": 5.6444,
+      "step": 284000
+    },
+    {
+      "epoch": 7.283852633195934,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.0017086458946721625,
+      "loss": 5.6444,
+      "step": 284500
+    },
+    {
+      "epoch": 7.296653780178704,
+      "grad_norm": 0.462890625,
+      "learning_rate": 0.001708133848792852,
+      "loss": 5.645,
+      "step": 285000
+    },
+    {
+      "epoch": 7.3094549271614735,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.001707621802913541,
+      "loss": 5.645,
+      "step": 285500
+    },
+    {
+      "epoch": 7.322256074144243,
+      "grad_norm": 1.453125,
+      "learning_rate": 0.0017071097570342303,
+      "loss": 5.6457,
+      "step": 286000
+    },
+    {
+      "epoch": 7.335057221127013,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.0017065977111549195,
+      "loss": 5.6445,
+      "step": 286500
+    },
+    {
+      "epoch": 7.347858368109783,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.0017060856652756087,
+      "loss": 5.6456,
+      "step": 287000
+    },
+    {
+      "epoch": 7.360659515092553,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.0017055736193962979,
+      "loss": 5.6455,
+      "step": 287500
+    },
+    {
+      "epoch": 7.373460662075322,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.0017050615735169873,
+      "loss": 5.6451,
+      "step": 288000
+    },
+    {
+      "epoch": 7.386261809058092,
+      "grad_norm": 0.734375,
+      "learning_rate": 0.0017045495276376763,
+      "loss": 5.644,
+      "step": 288500
+    },
+    {
+      "epoch": 7.399062956040861,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.0017040374817583654,
+      "loss": 5.6448,
+      "step": 289000
+    },
+    {
+      "epoch": 7.411864103023631,
+      "grad_norm": 1.4375,
+      "learning_rate": 0.0017035254358790549,
+      "loss": 5.6455,
+      "step": 289500
+    },
+    {
+      "epoch": 7.4246652500064005,
+      "grad_norm": 0.9296875,
+      "learning_rate": 0.001703013389999744,
+      "loss": 5.6435,
+      "step": 290000
+    },
+    {
+      "epoch": 7.43746639698917,
+      "grad_norm": 1.1875,
+      "learning_rate": 0.001702501344120433,
+      "loss": 5.6444,
+      "step": 290500
+    },
+    {
+      "epoch": 7.45026754397194,
+      "grad_norm": 0.7734375,
+      "learning_rate": 0.0017019892982411224,
+      "loss": 5.6444,
+      "step": 291000
+    },
+    {
+      "epoch": 7.463068690954709,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.0017014772523618116,
+      "loss": 5.6432,
+      "step": 291500
+    },
+    {
+      "epoch": 7.475869837937479,
+      "grad_norm": 1.703125,
+      "learning_rate": 0.001700965206482501,
+      "loss": 5.644,
+      "step": 292000
+    },
+    {
+      "epoch": 7.488670984920249,
+      "grad_norm": 4.5,
+      "learning_rate": 0.00170045316060319,
+      "loss": 5.6447,
+      "step": 292500
+    },
+    {
+      "epoch": 7.501472131903019,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.0016999411147238792,
+      "loss": 5.644,
+      "step": 293000
+    },
+    {
+      "epoch": 7.514273278885788,
+      "grad_norm": 2.109375,
+      "learning_rate": 0.0016994290688445686,
+      "loss": 5.6436,
+      "step": 293500
+    },
+    {
+      "epoch": 7.527074425868558,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.0016989170229652578,
+      "loss": 5.6443,
+      "step": 294000
+    },
+    {
+      "epoch": 7.539875572851328,
+      "grad_norm": 1.3203125,
+      "learning_rate": 0.0016984049770859468,
+      "loss": 5.6431,
+      "step": 294500
+    },
+    {
+      "epoch": 7.552676719834097,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.0016978929312066362,
+      "loss": 5.6443,
+      "step": 295000
+    },
+    {
+      "epoch": 7.565477866816867,
+      "grad_norm": 1.125,
+      "learning_rate": 0.0016973808853273254,
+      "loss": 5.6428,
+      "step": 295500
+    },
+    {
+      "epoch": 7.578279013799636,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0016968688394480146,
+      "loss": 5.6434,
+      "step": 296000
+    },
+    {
+      "epoch": 7.591080160782406,
+      "grad_norm": 0.703125,
+      "learning_rate": 0.0016963567935687038,
+      "loss": 5.643,
+      "step": 296500
+    },
+    {
+      "epoch": 7.6038813077651755,
+      "grad_norm": 0.6328125,
+      "learning_rate": 0.001695844747689393,
+      "loss": 5.6441,
+      "step": 297000
+    },
+    {
+      "epoch": 7.616682454747945,
+      "grad_norm": 0.78515625,
+      "learning_rate": 0.0016953327018100822,
+      "loss": 5.6428,
+      "step": 297500
+    },
+    {
+      "epoch": 7.6294836017307155,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.0016948206559307716,
+      "loss": 5.6421,
+      "step": 298000
+    },
+    {
+      "epoch": 7.642284748713485,
+      "grad_norm": 0.49609375,
+      "learning_rate": 0.0016943086100514606,
+      "loss": 5.6424,
+      "step": 298500
+    },
+    {
+      "epoch": 7.655085895696255,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.0016937965641721498,
+      "loss": 5.6422,
+      "step": 299000
+    },
+    {
+      "epoch": 7.667887042679024,
+      "grad_norm": 0.64453125,
+      "learning_rate": 0.0016932845182928392,
+      "loss": 5.6428,
+      "step": 299500
+    },
+    {
+      "epoch": 7.680688189661794,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.0016927724724135284,
+      "loss": 5.6436,
+      "step": 300000
+    },
+    {
+      "epoch": 7.693489336644563,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.0016922604265342174,
+      "loss": 5.6432,
+      "step": 300500
+    },
+    {
+      "epoch": 7.706290483627333,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0016917483806549068,
+      "loss": 5.6432,
+      "step": 301000
+    },
+    {
+      "epoch": 7.7190916306101025,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.001691236334775596,
+      "loss": 5.6434,
+      "step": 301500
+    },
+    {
+      "epoch": 7.731892777592872,
+      "grad_norm": 1.2890625,
+      "learning_rate": 0.0016907242888962852,
+      "loss": 5.6424,
+      "step": 302000
+    },
+    {
+      "epoch": 7.744693924575642,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0016902122430169743,
+      "loss": 5.6427,
+      "step": 302500
+    },
+    {
+      "epoch": 7.757495071558411,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.0016897001971376635,
+      "loss": 5.6427,
+      "step": 303000
+    },
+    {
+      "epoch": 7.770296218541182,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.0016891881512583527,
+      "loss": 5.6424,
+      "step": 303500
+    },
+    {
+      "epoch": 7.783097365523951,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.001688676105379042,
+      "loss": 5.6429,
+      "step": 304000
+    },
+    {
+      "epoch": 7.795898512506721,
+      "grad_norm": 1.8203125,
+      "learning_rate": 0.0016881640594997311,
+      "loss": 5.6426,
+      "step": 304500
+    },
+    {
+      "epoch": 7.80869965948949,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.0016876520136204205,
+      "loss": 5.6418,
+      "step": 305000
+    },
+    {
+      "epoch": 7.82150080647226,
+      "grad_norm": 0.37890625,
+      "learning_rate": 0.0016871399677411097,
+      "loss": 5.6419,
+      "step": 305500
+    },
+    {
+      "epoch": 7.8343019534550296,
+      "grad_norm": 0.4296875,
+      "learning_rate": 0.0016866279218617987,
+      "loss": 5.6429,
+      "step": 306000
+    },
+    {
+      "epoch": 7.847103100437799,
+      "grad_norm": 0.3828125,
+      "learning_rate": 0.0016861158759824881,
+      "loss": 5.6431,
+      "step": 306500
+    },
+    {
+      "epoch": 7.859904247420569,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.0016856038301031773,
+      "loss": 5.642,
+      "step": 307000
+    },
+    {
+      "epoch": 7.872705394403338,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.0016850917842238665,
+      "loss": 5.6423,
+      "step": 307500
+    },
+    {
+      "epoch": 7.885506541386108,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.0016845797383445557,
+      "loss": 5.6436,
+      "step": 308000
+    },
+    {
+      "epoch": 7.898307688368877,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.001684067692465245,
+      "loss": 5.642,
+      "step": 308500
+    },
+    {
+      "epoch": 7.911108835351648,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.001683555646585934,
+      "loss": 5.6431,
+      "step": 309000
+    },
+    {
+      "epoch": 7.9239099823344175,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.0016830436007066235,
+      "loss": 5.6436,
+      "step": 309500
+    },
+    {
+      "epoch": 7.936711129317187,
+      "grad_norm": 0.92578125,
+      "learning_rate": 0.0016825315548273125,
+      "loss": 5.6429,
+      "step": 310000
+    },
+    {
+      "epoch": 7.949512276299957,
+      "grad_norm": 0.72265625,
+      "learning_rate": 0.0016820195089480017,
+      "loss": 5.6429,
+      "step": 310500
+    },
+    {
+      "epoch": 7.962313423282726,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.001681507463068691,
+      "loss": 5.6424,
+      "step": 311000
+    },
+    {
+      "epoch": 7.975114570265496,
+      "grad_norm": 0.84765625,
+      "learning_rate": 0.0016809954171893803,
+      "loss": 5.6424,
+      "step": 311500
+    },
+    {
+      "epoch": 7.987915717248265,
+      "grad_norm": 2.640625,
+      "learning_rate": 0.0016804833713100693,
+      "loss": 5.642,
+      "step": 312000
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 5.642501354217529,
+      "eval_runtime": 1.2812,
+      "eval_samples_per_second": 780.497,
+      "eval_steps_per_second": 3.122,
+      "step": 312472
+    },
+    {
+      "epoch": 8.000716864231036,
+      "grad_norm": 0.6328125,
+      "learning_rate": 0.0016799713254307587,
+      "loss": 5.6415,
+      "step": 312500
+    },
+    {
+      "epoch": 8.013518011213804,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0016794592795514479,
+      "loss": 5.6414,
+      "step": 313000
+    },
+    {
+      "epoch": 8.026319158196575,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.001678947233672137,
+      "loss": 5.6422,
+      "step": 313500
+    },
+    {
+      "epoch": 8.039120305179344,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.0016784351877928263,
+      "loss": 5.6416,
+      "step": 314000
+    },
+    {
+      "epoch": 8.051921452162114,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.0016779231419135154,
+      "loss": 5.6421,
+      "step": 314500
+    },
+    {
+      "epoch": 8.064722599144883,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.0016774110960342046,
+      "loss": 5.642,
+      "step": 315000
+    },
+    {
+      "epoch": 8.077523746127653,
+      "grad_norm": 1.4140625,
+      "learning_rate": 0.001676899050154894,
+      "loss": 5.6418,
+      "step": 315500
+    },
+    {
+      "epoch": 8.090324893110422,
+      "grad_norm": 0.9453125,
+      "learning_rate": 0.001676387004275583,
+      "loss": 5.6408,
+      "step": 316000
+    },
+    {
+      "epoch": 8.103126040093192,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0016758749583962722,
+      "loss": 5.6421,
+      "step": 316500
+    },
+    {
+      "epoch": 8.115927187075963,
+      "grad_norm": 0.99609375,
+      "learning_rate": 0.0016753629125169616,
+      "loss": 5.6421,
+      "step": 317000
+    },
+    {
+      "epoch": 8.128728334058732,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.0016748508666376508,
+      "loss": 5.6412,
+      "step": 317500
+    },
+    {
+      "epoch": 8.141529481041502,
+      "grad_norm": 1.2265625,
+      "learning_rate": 0.0016743388207583398,
+      "loss": 5.6415,
+      "step": 318000
+    },
+    {
+      "epoch": 8.15433062802427,
+      "grad_norm": 2.375,
+      "learning_rate": 0.0016738267748790292,
+      "loss": 5.6419,
+      "step": 318500
+    },
+    {
+      "epoch": 8.167131775007041,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.0016733147289997184,
+      "loss": 5.6406,
+      "step": 319000
+    },
+    {
+      "epoch": 8.17993292198981,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.0016728026831204078,
+      "loss": 5.6412,
+      "step": 319500
+    },
+    {
+      "epoch": 8.19273406897258,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.0016722906372410968,
+      "loss": 5.6412,
+      "step": 320000
+    },
+    {
+      "epoch": 8.205535215955349,
+      "grad_norm": 0.66796875,
+      "learning_rate": 0.001671778591361786,
+      "loss": 5.6424,
+      "step": 320500
+    },
+    {
+      "epoch": 8.21833636293812,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.0016712665454824754,
+      "loss": 5.6402,
+      "step": 321000
+    },
+    {
+      "epoch": 8.231137509920888,
+      "grad_norm": 1.0703125,
+      "learning_rate": 0.0016707544996031646,
+      "loss": 5.6406,
+      "step": 321500
+    },
+    {
+      "epoch": 8.243938656903659,
+      "grad_norm": 2.53125,
+      "learning_rate": 0.0016702424537238536,
+      "loss": 5.6405,
+      "step": 322000
+    },
+    {
+      "epoch": 8.256739803886429,
+      "grad_norm": 0.83203125,
+      "learning_rate": 0.001669730407844543,
+      "loss": 5.6413,
+      "step": 322500
+    },
+    {
+      "epoch": 8.269540950869198,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0016692183619652322,
+      "loss": 5.64,
+      "step": 323000
+    },
+    {
+      "epoch": 8.282342097851968,
+      "grad_norm": 0.66015625,
+      "learning_rate": 0.0016687063160859214,
+      "loss": 5.6413,
+      "step": 323500
+    },
+    {
+      "epoch": 8.295143244834737,
+      "grad_norm": 0.984375,
+      "learning_rate": 0.0016681942702066106,
+      "loss": 5.641,
+      "step": 324000
+    },
+    {
+      "epoch": 8.307944391817507,
+      "grad_norm": 1.5390625,
+      "learning_rate": 0.0016676822243272998,
+      "loss": 5.6406,
+      "step": 324500
+    },
+    {
+      "epoch": 8.320745538800276,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.001667170178447989,
+      "loss": 5.6407,
+      "step": 325000
+    },
+    {
+      "epoch": 8.333546685783046,
+      "grad_norm": 1.5625,
+      "learning_rate": 0.0016666581325686784,
+      "loss": 5.6405,
+      "step": 325500
+    },
+    {
+      "epoch": 8.346347832765815,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.0016661460866893673,
+      "loss": 5.6405,
+      "step": 326000
+    },
+    {
+      "epoch": 8.359148979748586,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.0016656340408100565,
+      "loss": 5.642,
+      "step": 326500
+    },
+    {
+      "epoch": 8.371950126731354,
+      "grad_norm": 2.96875,
+      "learning_rate": 0.001665121994930746,
+      "loss": 5.6405,
+      "step": 327000
+    },
+    {
+      "epoch": 8.384751273714125,
+      "grad_norm": 0.6328125,
+      "learning_rate": 0.001664609949051435,
+      "loss": 5.6414,
+      "step": 327500
+    },
+    {
+      "epoch": 8.397552420696895,
+      "grad_norm": 1.515625,
+      "learning_rate": 0.0016640979031721241,
+      "loss": 5.6414,
+      "step": 328000
+    },
+    {
+      "epoch": 8.410353567679664,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.0016635858572928135,
+      "loss": 5.6399,
+      "step": 328500
+    },
+    {
+      "epoch": 8.423154714662434,
+      "grad_norm": 1.53125,
+      "learning_rate": 0.0016630738114135027,
+      "loss": 5.6418,
+      "step": 329000
+    },
+    {
+      "epoch": 8.435955861645203,
+      "grad_norm": 1.640625,
+      "learning_rate": 0.0016625617655341917,
+      "loss": 5.6404,
+      "step": 329500
+    },
+    {
+      "epoch": 8.448757008627974,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0016620497196548811,
+      "loss": 5.6408,
+      "step": 330000
+    },
+    {
+      "epoch": 8.461558155610742,
+      "grad_norm": 0.625,
+      "learning_rate": 0.0016615376737755703,
+      "loss": 5.6409,
+      "step": 330500
+    },
+    {
+      "epoch": 8.474359302593513,
+      "grad_norm": 0.8984375,
+      "learning_rate": 0.0016610256278962595,
+      "loss": 5.6404,
+      "step": 331000
+    },
+    {
+      "epoch": 8.487160449576281,
+      "grad_norm": 0.921875,
+      "learning_rate": 0.0016605135820169487,
+      "loss": 5.6401,
+      "step": 331500
+    },
+    {
+      "epoch": 8.499961596559052,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.001660001536137638,
+      "loss": 5.642,
+      "step": 332000
+    },
+    {
+      "epoch": 8.51276274354182,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.001659489490258327,
+      "loss": 5.6404,
+      "step": 332500
+    },
+    {
+      "epoch": 8.525563890524591,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.0016589774443790165,
+      "loss": 5.6402,
+      "step": 333000
+    },
+    {
+      "epoch": 8.538365037507361,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.0016584653984997055,
+      "loss": 5.6396,
+      "step": 333500
+    },
+    {
+      "epoch": 8.55116618449013,
+      "grad_norm": 1.6640625,
+      "learning_rate": 0.001657953352620395,
+      "loss": 5.6403,
+      "step": 334000
+    },
+    {
+      "epoch": 8.5639673314729,
+      "grad_norm": 2.8125,
+      "learning_rate": 0.001657441306741084,
+      "loss": 5.6409,
+      "step": 334500
+    },
+    {
+      "epoch": 8.57676847845567,
+      "grad_norm": 1.265625,
+      "learning_rate": 0.0016569292608617733,
+      "loss": 5.6405,
+      "step": 335000
+    },
+    {
+      "epoch": 8.58956962543844,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.0016564172149824625,
+      "loss": 5.6403,
+      "step": 335500
+    },
+    {
+      "epoch": 8.602370772421208,
+      "grad_norm": 1.953125,
+      "learning_rate": 0.0016559051691031517,
+      "loss": 5.6397,
+      "step": 336000
+    },
+    {
+      "epoch": 8.615171919403979,
+      "grad_norm": 0.98828125,
+      "learning_rate": 0.0016553931232238409,
+      "loss": 5.6407,
+      "step": 336500
+    },
+    {
+      "epoch": 8.627973066386748,
+      "grad_norm": 3.015625,
+      "learning_rate": 0.0016548810773445303,
+      "loss": 5.6401,
+      "step": 337000
+    },
+    {
+      "epoch": 8.640774213369518,
+      "grad_norm": 0.455078125,
+      "learning_rate": 0.0016543690314652193,
+      "loss": 5.6412,
+      "step": 337500
+    },
+    {
+      "epoch": 8.653575360352288,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.0016538569855859084,
+      "loss": 5.6402,
+      "step": 338000
+    },
+    {
+      "epoch": 8.666376507335057,
+      "grad_norm": 23.5,
+      "learning_rate": 0.0016533449397065979,
+      "loss": 5.6404,
+      "step": 338500
+    },
+    {
+      "epoch": 8.679177654317828,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.001652832893827287,
+      "loss": 5.6408,
+      "step": 339000
+    },
+    {
+      "epoch": 8.691978801300596,
+      "grad_norm": 1.171875,
+      "learning_rate": 0.001652320847947976,
+      "loss": 5.6404,
+      "step": 339500
+    },
+    {
+      "epoch": 8.704779948283367,
+      "grad_norm": 0.56640625,
+      "learning_rate": 0.0016518088020686654,
+      "loss": 5.6401,
+      "step": 340000
+    },
+    {
+      "epoch": 8.717581095266135,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.0016512967561893546,
+      "loss": 5.6404,
+      "step": 340500
+    },
+    {
+      "epoch": 8.730382242248906,
+      "grad_norm": 2.0,
+      "learning_rate": 0.0016507847103100438,
+      "loss": 5.6409,
+      "step": 341000
+    },
+    {
+      "epoch": 8.743183389231675,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.001650272664430733,
+      "loss": 5.6401,
+      "step": 341500
+    },
+    {
+      "epoch": 8.755984536214445,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.0016497606185514222,
+      "loss": 5.6403,
+      "step": 342000
+    },
+    {
+      "epoch": 8.768785683197214,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0016492485726721114,
+      "loss": 5.6393,
+      "step": 342500
+    },
+    {
+      "epoch": 8.781586830179984,
+      "grad_norm": 1.984375,
+      "learning_rate": 0.0016487365267928008,
+      "loss": 5.6401,
+      "step": 343000
+    },
+    {
+      "epoch": 8.794387977162753,
+      "grad_norm": 2.640625,
+      "learning_rate": 0.0016482244809134898,
+      "loss": 5.64,
+      "step": 343500
+    },
+    {
+      "epoch": 8.807189124145523,
+      "grad_norm": 0.80078125,
+      "learning_rate": 0.001647712435034179,
+      "loss": 5.6399,
+      "step": 344000
+    },
+    {
+      "epoch": 8.819990271128294,
+      "grad_norm": 0.96484375,
+      "learning_rate": 0.0016472003891548684,
+      "loss": 5.641,
+      "step": 344500
+    },
+    {
+      "epoch": 8.832791418111062,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.0016466883432755576,
+      "loss": 5.6391,
+      "step": 345000
+    },
+    {
+      "epoch": 8.845592565093833,
+      "grad_norm": 1.84375,
+      "learning_rate": 0.0016461762973962466,
+      "loss": 5.6407,
+      "step": 345500
+    },
+    {
+      "epoch": 8.858393712076602,
+      "grad_norm": 5.90625,
+      "learning_rate": 0.001645664251516936,
+      "loss": 5.6409,
+      "step": 346000
+    },
+    {
+      "epoch": 8.871194859059372,
+      "grad_norm": 0.71484375,
+      "learning_rate": 0.0016451522056376252,
+      "loss": 5.6398,
+      "step": 346500
+    },
+    {
+      "epoch": 8.88399600604214,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.0016446401597583142,
+      "loss": 5.6401,
+      "step": 347000
+    },
+    {
+      "epoch": 8.896797153024911,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.0016441281138790036,
+      "loss": 5.6395,
+      "step": 347500
+    },
+    {
+      "epoch": 8.90959830000768,
+      "grad_norm": 0.87109375,
+      "learning_rate": 0.0016436160679996928,
+      "loss": 5.6391,
+      "step": 348000
+    },
+    {
+      "epoch": 8.92239944699045,
+      "grad_norm": 1.8046875,
+      "learning_rate": 0.0016431040221203822,
+      "loss": 5.6393,
+      "step": 348500
+    },
+    {
+      "epoch": 8.93520059397322,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.0016425919762410712,
+      "loss": 5.6395,
+      "step": 349000
+    },
+    {
+      "epoch": 8.94800174095599,
+      "grad_norm": 2.90625,
+      "learning_rate": 0.0016420799303617604,
+      "loss": 5.6394,
+      "step": 349500
+    },
+    {
+      "epoch": 8.96080288793876,
+      "grad_norm": 0.88671875,
+      "learning_rate": 0.0016415678844824498,
+      "loss": 5.6396,
+      "step": 350000
+    },
+    {
+      "epoch": 8.973604034921529,
+      "grad_norm": 2.15625,
+      "learning_rate": 0.001641055838603139,
+      "loss": 5.6387,
+      "step": 350500
+    },
+    {
+      "epoch": 8.9864051819043,
+      "grad_norm": 1.578125,
+      "learning_rate": 0.001640543792723828,
+      "loss": 5.6394,
+      "step": 351000
+    },
+    {
+      "epoch": 8.999206328887068,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.0016400317468445173,
+      "loss": 5.6387,
+      "step": 351500
+    },
+    {
+      "epoch": 9.0,
+      "eval_loss": 5.638747215270996,
+      "eval_runtime": 1.2906,
+      "eval_samples_per_second": 774.854,
+      "eval_steps_per_second": 3.099,
+      "step": 351531
+    }
+  ],
+  "logging_steps": 500,
+  "max_steps": 1952950,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 50,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "EarlyStoppingCallback": {
+      "args": {
+        "early_stopping_patience": 3,
+        "early_stopping_threshold": 0.0
+      },
+      "attributes": {
+        "early_stopping_patience_counter": 0
+      }
+    },
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2189688215083418e+20,
+  "train_batch_size": 256,
+  "trial_name": null,
+  "trial_params": null
+}