diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,9033 @@
+{
+  "best_metric": 0.2536342144012451,
+  "best_model_checkpoint": "learning_source_20260316/compounds/bert-output/compounds-medium/checkpoint-57000",
+  "epoch": 5.783313694163914,
+  "eval_steps": 100,
+  "global_step": 60000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.009638856156939856,
+      "grad_norm": 7.281230449676514,
+      "learning_rate": 3e-06,
+      "loss": 3.8704,
+      "step": 100
+    },
+    {
+      "epoch": 0.009638856156939856,
+      "eval_loss": 2.631474494934082,
+      "eval_runtime": 40.2777,
+      "eval_samples_per_second": 248.276,
+      "eval_steps_per_second": 31.035,
+      "step": 100
+    },
+    {
+      "epoch": 0.019277712313879713,
+      "grad_norm": 11.868888854980469,
+      "learning_rate": 6e-06,
+      "loss": 2.555,
+      "step": 200
+    },
+    {
+      "epoch": 0.019277712313879713,
+      "eval_loss": 2.3589468002319336,
+      "eval_runtime": 40.128,
+      "eval_samples_per_second": 249.202,
+      "eval_steps_per_second": 31.15,
+      "step": 200
+    },
+    {
+      "epoch": 0.028916568470819567,
+      "grad_norm": 6.551002025604248,
+      "learning_rate": 5.989966555183947e-06,
+      "loss": 2.3092,
+      "step": 300
+    },
+    {
+      "epoch": 0.028916568470819567,
+      "eval_loss": 2.1789472103118896,
+      "eval_runtime": 40.1341,
+      "eval_samples_per_second": 249.164,
+      "eval_steps_per_second": 31.146,
+      "step": 300
+    },
+    {
+      "epoch": 0.038555424627759426,
+      "grad_norm": 8.500665664672852,
+      "learning_rate": 5.979933110367893e-06,
+      "loss": 2.1633,
+      "step": 400
+    },
+    {
+      "epoch": 0.038555424627759426,
+      "eval_loss": 2.072150945663452,
+      "eval_runtime": 40.1372,
+      "eval_samples_per_second": 249.146,
+      "eval_steps_per_second": 31.143,
+      "step": 400
+    },
+    {
+      "epoch": 0.04819428078469928,
+      "grad_norm": 13.641960144042969,
+      "learning_rate": 5.96989966555184e-06,
+      "loss": 2.0894,
+      "step": 500
+    },
+    {
+      "epoch": 0.04819428078469928,
+      "eval_loss": 2.0400867462158203,
+      "eval_runtime": 40.2974,
+      "eval_samples_per_second": 248.155,
+      "eval_steps_per_second": 31.019,
+      "step": 500
+    },
+    {
+      "epoch": 0.057833136941639135,
+      "grad_norm": 5.950716495513916,
+      "learning_rate": 5.959866220735786e-06,
+      "loss": 2.0528,
+      "step": 600
+    },
+    {
+      "epoch": 0.057833136941639135,
+      "eval_loss": 2.0093040466308594,
+      "eval_runtime": 40.1283,
+      "eval_samples_per_second": 249.2,
+      "eval_steps_per_second": 31.15,
+      "step": 600
+    },
+    {
+      "epoch": 0.06747199309857899,
+      "grad_norm": 9.331167221069336,
+      "learning_rate": 5.949832775919732e-06,
+      "loss": 2.0191,
+      "step": 700
+    },
+    {
+      "epoch": 0.06747199309857899,
+      "eval_loss": 1.9733953475952148,
+      "eval_runtime": 40.1378,
+      "eval_samples_per_second": 249.142,
+      "eval_steps_per_second": 31.143,
+      "step": 700
+    },
+    {
+      "epoch": 0.07711084925551885,
+      "grad_norm": 3.786700963973999,
+      "learning_rate": 5.939799331103679e-06,
+      "loss": 1.972,
+      "step": 800
+    },
+    {
+      "epoch": 0.07711084925551885,
+      "eval_loss": 1.9259250164031982,
+      "eval_runtime": 40.1305,
+      "eval_samples_per_second": 249.187,
+      "eval_steps_per_second": 31.148,
+      "step": 800
+    },
+    {
+      "epoch": 0.0867497054124587,
+      "grad_norm": 3.534658432006836,
+      "learning_rate": 5.929765886287626e-06,
+      "loss": 1.9195,
+      "step": 900
+    },
+    {
+      "epoch": 0.0867497054124587,
+      "eval_loss": 1.8756057024002075,
+      "eval_runtime": 40.1635,
+      "eval_samples_per_second": 248.982,
+      "eval_steps_per_second": 31.123,
+      "step": 900
+    },
+    {
+      "epoch": 0.09638856156939855,
+      "grad_norm": 6.269195556640625,
+      "learning_rate": 5.919732441471572e-06,
+      "loss": 1.8699,
+      "step": 1000
+    },
+    {
+      "epoch": 0.09638856156939855,
+      "eval_loss": 1.847890853881836,
+      "eval_runtime": 40.2907,
+      "eval_samples_per_second": 248.196,
+      "eval_steps_per_second": 31.025,
+      "step": 1000
+    },
+    {
+      "epoch": 0.10602741772633842,
+      "grad_norm": 3.2279481887817383,
+      "learning_rate": 5.9096989966555185e-06,
+      "loss": 1.828,
+      "step": 1100
+    },
+    {
+      "epoch": 0.10602741772633842,
+      "eval_loss": 1.8036913871765137,
+      "eval_runtime": 40.1142,
+      "eval_samples_per_second": 249.288,
+      "eval_steps_per_second": 31.161,
+      "step": 1100
+    },
+    {
+      "epoch": 0.11566627388327827,
+      "grad_norm": 3.0232768058776855,
+      "learning_rate": 5.899665551839465e-06,
+      "loss": 1.7788,
+      "step": 1200
+    },
+    {
+      "epoch": 0.11566627388327827,
+      "eval_loss": 1.7329744100570679,
+      "eval_runtime": 40.1179,
+      "eval_samples_per_second": 249.265,
+      "eval_steps_per_second": 31.158,
+      "step": 1200
+    },
+    {
+      "epoch": 0.12530513004021812,
+      "grad_norm": 3.1454615592956543,
+      "learning_rate": 5.889632107023412e-06,
+      "loss": 1.7228,
+      "step": 1300
+    },
+    {
+      "epoch": 0.12530513004021812,
+      "eval_loss": 1.6728882789611816,
+      "eval_runtime": 40.1057,
+      "eval_samples_per_second": 249.341,
+      "eval_steps_per_second": 31.168,
+      "step": 1300
+    },
+    {
+      "epoch": 0.13494398619715797,
+      "grad_norm": 2.789080858230591,
+      "learning_rate": 5.879598662207358e-06,
+      "loss": 1.6748,
+      "step": 1400
+    },
+    {
+      "epoch": 0.13494398619715797,
+      "eval_loss": 1.6212955713272095,
+      "eval_runtime": 40.2793,
+      "eval_samples_per_second": 248.267,
+      "eval_steps_per_second": 31.033,
+      "step": 1400
+    },
+    {
+      "epoch": 0.14458284235409785,
+      "grad_norm": 3.62225604057312,
+      "learning_rate": 5.869565217391305e-06,
+      "loss": 1.6172,
+      "step": 1500
+    },
+    {
+      "epoch": 0.14458284235409785,
+      "eval_loss": 1.5460147857666016,
+      "eval_runtime": 40.2924,
+      "eval_samples_per_second": 248.186,
+      "eval_steps_per_second": 31.023,
+      "step": 1500
+    },
+    {
+      "epoch": 0.1542216985110377,
+      "grad_norm": 2.5482733249664307,
+      "learning_rate": 5.8595317725752514e-06,
+      "loss": 1.5611,
+      "step": 1600
+    },
+    {
+      "epoch": 0.1542216985110377,
+      "eval_loss": 1.485274076461792,
+      "eval_runtime": 40.1252,
+      "eval_samples_per_second": 249.22,
+      "eval_steps_per_second": 31.153,
+      "step": 1600
+    },
+    {
+      "epoch": 0.16386055466797755,
+      "grad_norm": 1.7880430221557617,
+      "learning_rate": 5.849498327759197e-06,
+      "loss": 1.5098,
+      "step": 1700
+    },
+    {
+      "epoch": 0.16386055466797755,
+      "eval_loss": 1.4245718717575073,
+      "eval_runtime": 40.128,
+      "eval_samples_per_second": 249.202,
+      "eval_steps_per_second": 31.15,
+      "step": 1700
+    },
+    {
+      "epoch": 0.1734994108249174,
+      "grad_norm": 2.3302555084228516,
+      "learning_rate": 5.839464882943144e-06,
+      "loss": 1.4607,
+      "step": 1800
+    },
+    {
+      "epoch": 0.1734994108249174,
+      "eval_loss": 1.3619263172149658,
+      "eval_runtime": 40.1162,
+      "eval_samples_per_second": 249.276,
+      "eval_steps_per_second": 31.159,
+      "step": 1800
+    },
+    {
+      "epoch": 0.18313826698185726,
+      "grad_norm": 3.174801826477051,
+      "learning_rate": 5.829431438127091e-06,
+      "loss": 1.4193,
+      "step": 1900
+    },
+    {
+      "epoch": 0.18313826698185726,
+      "eval_loss": 1.2982574701309204,
+      "eval_runtime": 40.283,
+      "eval_samples_per_second": 248.244,
+      "eval_steps_per_second": 31.03,
+      "step": 1900
+    },
+    {
+      "epoch": 0.1927771231387971,
+      "grad_norm": 2.9484267234802246,
+      "learning_rate": 5.819397993311037e-06,
+      "loss": 1.3769,
+      "step": 2000
+    },
+    {
+      "epoch": 0.1927771231387971,
+      "eval_loss": 1.2819702625274658,
+      "eval_runtime": 40.3215,
+      "eval_samples_per_second": 248.007,
+      "eval_steps_per_second": 31.001,
+      "step": 2000
+    },
+    {
+      "epoch": 0.202415979295737,
+      "grad_norm": 1.8851428031921387,
+      "learning_rate": 5.8093645484949836e-06,
+      "loss": 1.34,
+      "step": 2100
+    },
+    {
+      "epoch": 0.202415979295737,
+      "eval_loss": 1.2403301000595093,
+      "eval_runtime": 40.2922,
+      "eval_samples_per_second": 248.187,
+      "eval_steps_per_second": 31.023,
+      "step": 2100
+    },
+    {
+      "epoch": 0.21205483545267684,
+      "grad_norm": 2.5137407779693604,
+      "learning_rate": 5.79933110367893e-06,
+      "loss": 1.3021,
+      "step": 2200
+    },
+    {
+      "epoch": 0.21205483545267684,
+      "eval_loss": 1.2023345232009888,
+      "eval_runtime": 40.1117,
+      "eval_samples_per_second": 249.304,
+      "eval_steps_per_second": 31.163,
+      "step": 2200
+    },
+    {
+      "epoch": 0.2216936916096167,
+      "grad_norm": 2.809570789337158,
+      "learning_rate": 5.789297658862876e-06,
+      "loss": 1.2718,
+      "step": 2300
+    },
+    {
+      "epoch": 0.2216936916096167,
+      "eval_loss": 1.1547870635986328,
+      "eval_runtime": 40.114,
+      "eval_samples_per_second": 249.29,
+      "eval_steps_per_second": 31.161,
+      "step": 2300
+    },
+    {
+      "epoch": 0.23133254776655654,
+      "grad_norm": 1.812084674835205,
+      "learning_rate": 5.779264214046823e-06,
+      "loss": 1.2342,
+      "step": 2400
+    },
+    {
+      "epoch": 0.23133254776655654,
+      "eval_loss": 1.113013744354248,
+      "eval_runtime": 40.1439,
+      "eval_samples_per_second": 249.104,
+      "eval_steps_per_second": 31.138,
+      "step": 2400
+    },
+    {
+      "epoch": 0.2409714039234964,
+      "grad_norm": 2.1630361080169678,
+      "learning_rate": 5.76923076923077e-06,
+      "loss": 1.21,
+      "step": 2500
+    },
+    {
+      "epoch": 0.2409714039234964,
+      "eval_loss": 1.0802332162857056,
+      "eval_runtime": 40.2724,
+      "eval_samples_per_second": 248.309,
+      "eval_steps_per_second": 31.039,
+      "step": 2500
+    },
+    {
+      "epoch": 0.25061026008043624,
+      "grad_norm": 2.5893359184265137,
+      "learning_rate": 5.759197324414716e-06,
+      "loss": 1.181,
+      "step": 2600
+    },
+    {
+      "epoch": 0.25061026008043624,
+      "eval_loss": 1.065743088722229,
+      "eval_runtime": 40.113,
+      "eval_samples_per_second": 249.296,
+      "eval_steps_per_second": 31.162,
+      "step": 2600
+    },
+    {
+      "epoch": 0.2602491162373761,
+      "grad_norm": 2.6710309982299805,
+      "learning_rate": 5.7491638795986624e-06,
+      "loss": 1.1569,
+      "step": 2700
+    },
+    {
+      "epoch": 0.2602491162373761,
+      "eval_loss": 1.0426665544509888,
+      "eval_runtime": 40.1155,
+      "eval_samples_per_second": 249.28,
+      "eval_steps_per_second": 31.16,
+      "step": 2700
+    },
+    {
+      "epoch": 0.26988797239431594,
+      "grad_norm": 2.068502902984619,
+      "learning_rate": 5.739130434782609e-06,
+      "loss": 1.1335,
+      "step": 2800
+    },
+    {
+      "epoch": 0.26988797239431594,
+      "eval_loss": 1.0142563581466675,
+      "eval_runtime": 40.1365,
+      "eval_samples_per_second": 249.15,
+      "eval_steps_per_second": 31.144,
+      "step": 2800
+    },
+    {
+      "epoch": 0.2795268285512558,
+      "grad_norm": 2.485072374343872,
+      "learning_rate": 5.729096989966555e-06,
+      "loss": 1.1049,
+      "step": 2900
+    },
+    {
+      "epoch": 0.2795268285512558,
+      "eval_loss": 0.988125205039978,
+      "eval_runtime": 40.1288,
+      "eval_samples_per_second": 249.198,
+      "eval_steps_per_second": 31.15,
+      "step": 2900
+    },
+    {
+      "epoch": 0.2891656847081957,
+      "grad_norm": 2.3873965740203857,
+      "learning_rate": 5.719063545150502e-06,
+      "loss": 1.0827,
+      "step": 3000
+    },
+    {
+      "epoch": 0.2891656847081957,
+      "eval_loss": 0.95792555809021,
+      "eval_runtime": 40.2725,
+      "eval_samples_per_second": 248.308,
+      "eval_steps_per_second": 31.039,
+      "step": 3000
+    },
+    {
+      "epoch": 0.2988045408651355,
+      "grad_norm": 2.1445956230163574,
+      "learning_rate": 5.709030100334449e-06,
+      "loss": 1.0598,
+      "step": 3100
+    },
+    {
+      "epoch": 0.2988045408651355,
+      "eval_loss": 0.9487287402153015,
+      "eval_runtime": 40.201,
+      "eval_samples_per_second": 248.75,
+      "eval_steps_per_second": 31.094,
+      "step": 3100
+    },
+    {
+      "epoch": 0.3084433970220754,
+      "grad_norm": 2.3310909271240234,
+      "learning_rate": 5.698996655518395e-06,
+      "loss": 1.041,
+      "step": 3200
+    },
+    {
+      "epoch": 0.3084433970220754,
+      "eval_loss": 0.9160116910934448,
+      "eval_runtime": 40.2197,
+      "eval_samples_per_second": 248.634,
+      "eval_steps_per_second": 31.079,
+      "step": 3200
+    },
+    {
+      "epoch": 0.31808225317901523,
+      "grad_norm": 2.3795151710510254,
+      "learning_rate": 5.688963210702341e-06,
+      "loss": 1.0179,
+      "step": 3300
+    },
+    {
+      "epoch": 0.31808225317901523,
+      "eval_loss": 0.8998765349388123,
+      "eval_runtime": 40.0659,
+      "eval_samples_per_second": 249.589,
+      "eval_steps_per_second": 31.199,
+      "step": 3300
+    },
+    {
+      "epoch": 0.3277211093359551,
+      "grad_norm": 2.6683619022369385,
+      "learning_rate": 5.678929765886288e-06,
+      "loss": 1.0022,
+      "step": 3400
+    },
+    {
+      "epoch": 0.3277211093359551,
+      "eval_loss": 0.8904856443405151,
+      "eval_runtime": 40.0602,
+      "eval_samples_per_second": 249.624,
+      "eval_steps_per_second": 31.203,
+      "step": 3400
+    },
+    {
+      "epoch": 0.33735996549289493,
+      "grad_norm": 2.24725079536438,
+      "learning_rate": 5.668896321070235e-06,
+      "loss": 0.9901,
+      "step": 3500
+    },
+    {
+      "epoch": 0.33735996549289493,
+      "eval_loss": 0.8752567172050476,
+      "eval_runtime": 40.0856,
+      "eval_samples_per_second": 249.466,
+      "eval_steps_per_second": 31.183,
+      "step": 3500
+    },
+    {
+      "epoch": 0.3469988216498348,
+      "grad_norm": 2.042285442352295,
+      "learning_rate": 5.658862876254181e-06,
+      "loss": 0.9708,
+      "step": 3600
+    },
+    {
+      "epoch": 0.3469988216498348,
+      "eval_loss": 0.8505752086639404,
+      "eval_runtime": 40.0593,
+      "eval_samples_per_second": 249.63,
+      "eval_steps_per_second": 31.204,
+      "step": 3600
+    },
+    {
+      "epoch": 0.3566376778067747,
+      "grad_norm": 1.9062834978103638,
+      "learning_rate": 5.6488294314381275e-06,
+      "loss": 0.9577,
+      "step": 3700
+    },
+    {
+      "epoch": 0.3566376778067747,
+      "eval_loss": 0.8313038945198059,
+      "eval_runtime": 40.2235,
+      "eval_samples_per_second": 248.611,
+      "eval_steps_per_second": 31.076,
+      "step": 3700
+    },
+    {
+      "epoch": 0.3662765339637145,
+      "grad_norm": 1.9753172397613525,
+      "learning_rate": 5.638795986622074e-06,
+      "loss": 0.94,
+      "step": 3800
+    },
+    {
+      "epoch": 0.3662765339637145,
+      "eval_loss": 0.8297803997993469,
+      "eval_runtime": 40.0249,
+      "eval_samples_per_second": 249.844,
+      "eval_steps_per_second": 31.231,
+      "step": 3800
+    },
+    {
+      "epoch": 0.3759153901206544,
+      "grad_norm": 2.3993759155273438,
+      "learning_rate": 5.62876254180602e-06,
+      "loss": 0.9221,
+      "step": 3900
+    },
+    {
+      "epoch": 0.3759153901206544,
+      "eval_loss": 0.8069967031478882,
+      "eval_runtime": 40.0327,
+      "eval_samples_per_second": 249.796,
+      "eval_steps_per_second": 31.224,
+      "step": 3900
+    },
+    {
+      "epoch": 0.3855542462775942,
+      "grad_norm": 2.130056858062744,
+      "learning_rate": 5.618729096989967e-06,
+      "loss": 0.9078,
+      "step": 4000
+    },
+    {
+      "epoch": 0.3855542462775942,
+      "eval_loss": 0.8021091818809509,
+      "eval_runtime": 40.0159,
+      "eval_samples_per_second": 249.901,
+      "eval_steps_per_second": 31.238,
+      "step": 4000
+    },
+    {
+      "epoch": 0.3951931024345341,
+      "grad_norm": 2.2790122032165527,
+      "learning_rate": 5.608695652173914e-06,
+      "loss": 0.8954,
+      "step": 4100
+    },
+    {
+      "epoch": 0.3951931024345341,
+      "eval_loss": 0.797188401222229,
+      "eval_runtime": 40.062,
+      "eval_samples_per_second": 249.613,
+      "eval_steps_per_second": 31.202,
+      "step": 4100
+    },
+    {
+      "epoch": 0.404831958591474,
+      "grad_norm": 2.6065945625305176,
+      "learning_rate": 5.59866220735786e-06,
+      "loss": 0.8816,
+      "step": 4200
+    },
+    {
+      "epoch": 0.404831958591474,
+      "eval_loss": 0.7731242179870605,
+      "eval_runtime": 40.1911,
+      "eval_samples_per_second": 248.811,
+      "eval_steps_per_second": 31.101,
+      "step": 4200
+    },
+    {
+      "epoch": 0.4144708147484138,
+      "grad_norm": 2.0770599842071533,
+      "learning_rate": 5.588628762541806e-06,
+      "loss": 0.869,
+      "step": 4300
+    },
+    {
+      "epoch": 0.4144708147484138,
+      "eval_loss": 0.7588244080543518,
+      "eval_runtime": 40.0263,
+      "eval_samples_per_second": 249.836,
+      "eval_steps_per_second": 31.229,
+      "step": 4300
+    },
+    {
+      "epoch": 0.4241096709053537,
+      "grad_norm": 2.3637514114379883,
+      "learning_rate": 5.578595317725753e-06,
+      "loss": 0.8572,
+      "step": 4400
+    },
+    {
+      "epoch": 0.4241096709053537,
+      "eval_loss": 0.754951000213623,
+      "eval_runtime": 40.0508,
+      "eval_samples_per_second": 249.683,
+      "eval_steps_per_second": 31.21,
+      "step": 4400
+    },
+    {
+      "epoch": 0.4337485270622935,
+      "grad_norm": 2.044037342071533,
+      "learning_rate": 5.568561872909699e-06,
+      "loss": 0.8505,
+      "step": 4500
+    },
+    {
+      "epoch": 0.4337485270622935,
+      "eval_loss": 0.7327916026115417,
+      "eval_runtime": 40.0028,
+      "eval_samples_per_second": 249.983,
+      "eval_steps_per_second": 31.248,
+      "step": 4500
+    },
+    {
+      "epoch": 0.4433873832192334,
+      "grad_norm": 1.824724793434143,
+      "learning_rate": 5.558528428093646e-06,
+      "loss": 0.8372,
+      "step": 4600
+    },
+    {
+      "epoch": 0.4433873832192334,
+      "eval_loss": 0.7417221665382385,
+      "eval_runtime": 40.1915,
+      "eval_samples_per_second": 248.809,
+      "eval_steps_per_second": 31.101,
+      "step": 4600
+    },
+    {
+      "epoch": 0.45302623937617326,
+      "grad_norm": 2.4582533836364746,
+      "learning_rate": 5.548494983277593e-06,
+      "loss": 0.8265,
+      "step": 4700
+    },
+    {
+      "epoch": 0.45302623937617326,
+      "eval_loss": 0.7247117757797241,
+      "eval_runtime": 39.9831,
+      "eval_samples_per_second": 250.105,
+      "eval_steps_per_second": 31.263,
+      "step": 4700
+    },
+    {
+      "epoch": 0.4626650955331131,
+      "grad_norm": 2.2355294227600098,
+      "learning_rate": 5.5384615384615385e-06,
+      "loss": 0.8125,
+      "step": 4800
+    },
+    {
+      "epoch": 0.4626650955331131,
+      "eval_loss": 0.7139586806297302,
+      "eval_runtime": 39.9822,
+      "eval_samples_per_second": 250.111,
+      "eval_steps_per_second": 31.264,
+      "step": 4800
+    },
+    {
+      "epoch": 0.47230395169005296,
+      "grad_norm": 2.021684169769287,
+      "learning_rate": 5.528428093645485e-06,
+      "loss": 0.8085,
+      "step": 4900
+    },
+    {
+      "epoch": 0.47230395169005296,
+      "eval_loss": 0.7031977772712708,
+      "eval_runtime": 39.9965,
+      "eval_samples_per_second": 250.022,
+      "eval_steps_per_second": 31.253,
+      "step": 4900
+    },
+    {
+      "epoch": 0.4819428078469928,
+      "grad_norm": 2.307088613510132,
+      "learning_rate": 5.518394648829432e-06,
+      "loss": 0.7937,
+      "step": 5000
+    },
+    {
+      "epoch": 0.4819428078469928,
+      "eval_loss": 0.6927512884140015,
+      "eval_runtime": 40.0327,
+      "eval_samples_per_second": 249.796,
+      "eval_steps_per_second": 31.225,
+      "step": 5000
+    },
+    {
+      "epoch": 0.49158166400393266,
+      "grad_norm": 1.824004888534546,
+      "learning_rate": 5.508361204013378e-06,
+      "loss": 0.7882,
+      "step": 5100
+    },
+    {
+      "epoch": 0.49158166400393266,
+      "eval_loss": 0.6837274432182312,
+      "eval_runtime": 40.1793,
+      "eval_samples_per_second": 248.884,
+      "eval_steps_per_second": 31.111,
+      "step": 5100
+    },
+    {
+      "epoch": 0.5012205201608725,
+      "grad_norm": 1.8306970596313477,
+      "learning_rate": 5.498327759197324e-06,
+      "loss": 0.7758,
+      "step": 5200
+    },
+    {
+      "epoch": 0.5012205201608725,
+      "eval_loss": 0.6825730204582214,
+      "eval_runtime": 40.0105,
+      "eval_samples_per_second": 249.934,
+      "eval_steps_per_second": 31.242,
+      "step": 5200
+    },
+    {
+      "epoch": 0.5108593763178124,
+      "grad_norm": 1.8359283208847046,
+      "learning_rate": 5.488294314381271e-06,
+      "loss": 0.7665,
+      "step": 5300
+    },
+    {
+      "epoch": 0.5108593763178124,
+      "eval_loss": 0.6687480211257935,
+      "eval_runtime": 40.0097,
+      "eval_samples_per_second": 249.939,
+      "eval_steps_per_second": 31.242,
+      "step": 5300
+    },
+    {
+      "epoch": 0.5204982324747522,
+      "grad_norm": 1.8846337795257568,
+      "learning_rate": 5.478260869565217e-06,
+      "loss": 0.7621,
+      "step": 5400
+    },
+    {
+      "epoch": 0.5204982324747522,
+      "eval_loss": 0.6538456082344055,
+      "eval_runtime": 39.9879,
+      "eval_samples_per_second": 250.076,
+      "eval_steps_per_second": 31.259,
+      "step": 5400
+    },
+    {
+      "epoch": 0.5301370886316921,
+      "grad_norm": 1.734246850013733,
+      "learning_rate": 5.468227424749163e-06,
+      "loss": 0.7518,
+      "step": 5500
+    },
+    {
+      "epoch": 0.5301370886316921,
+      "eval_loss": 0.6537781953811646,
+      "eval_runtime": 40.0172,
+      "eval_samples_per_second": 249.893,
+      "eval_steps_per_second": 31.237,
+      "step": 5500
+    },
+    {
+      "epoch": 0.5397759447886319,
+      "grad_norm": 1.805600881576538,
+      "learning_rate": 5.45819397993311e-06,
+      "loss": 0.7433,
+      "step": 5600
+    },
+    {
+      "epoch": 0.5397759447886319,
+      "eval_loss": 0.6553384065628052,
+      "eval_runtime": 40.1835,
+      "eval_samples_per_second": 248.858,
+      "eval_steps_per_second": 31.107,
+      "step": 5600
+    },
+    {
+      "epoch": 0.5494148009455718,
+      "grad_norm": 1.7596203088760376,
+      "learning_rate": 5.448160535117057e-06,
+      "loss": 0.7319,
+      "step": 5700
+    },
+    {
+      "epoch": 0.5494148009455718,
+      "eval_loss": 0.6483535170555115,
+      "eval_runtime": 40.012,
+      "eval_samples_per_second": 249.925,
+      "eval_steps_per_second": 31.241,
+      "step": 5700
+    },
+    {
+      "epoch": 0.5590536571025116,
+      "grad_norm": 2.0771901607513428,
+      "learning_rate": 5.438127090301003e-06,
+      "loss": 0.727,
+      "step": 5800
+    },
+    {
+      "epoch": 0.5590536571025116,
+      "eval_loss": 0.645814836025238,
+      "eval_runtime": 40.0155,
+      "eval_samples_per_second": 249.903,
+      "eval_steps_per_second": 31.238,
+      "step": 5800
+    },
+    {
+      "epoch": 0.5686925132594515,
+      "grad_norm": 1.8789221048355103,
+      "learning_rate": 5.4280936454849495e-06,
+      "loss": 0.7269,
+      "step": 5900
+    },
+    {
+      "epoch": 0.5686925132594515,
+      "eval_loss": 0.6407626271247864,
+      "eval_runtime": 40.0142,
+      "eval_samples_per_second": 249.911,
+      "eval_steps_per_second": 31.239,
+      "step": 5900
+    },
+    {
+      "epoch": 0.5783313694163914,
+      "grad_norm": 2.0252554416656494,
+      "learning_rate": 5.418060200668896e-06,
+      "loss": 0.7133,
+      "step": 6000
+    },
+    {
+      "epoch": 0.5783313694163914,
+      "eval_loss": 0.6353247761726379,
+      "eval_runtime": 40.0245,
+      "eval_samples_per_second": 249.847,
+      "eval_steps_per_second": 31.231,
+      "step": 6000
+    },
+    {
+      "epoch": 0.5879702255733312,
+      "grad_norm": 2.003438949584961,
+      "learning_rate": 5.408026755852843e-06,
+      "loss": 0.7096,
+      "step": 6100
+    },
+    {
+      "epoch": 0.5879702255733312,
+      "eval_loss": 0.6176116466522217,
+      "eval_runtime": 40.1568,
+      "eval_samples_per_second": 249.024,
+      "eval_steps_per_second": 31.128,
+      "step": 6100
+    },
+    {
+      "epoch": 0.597609081730271,
+      "grad_norm": 1.8938828706741333,
+      "learning_rate": 5.397993311036789e-06,
+      "loss": 0.7001,
+      "step": 6200
+    },
+    {
+      "epoch": 0.597609081730271,
+      "eval_loss": 0.6332681775093079,
+      "eval_runtime": 39.9982,
+      "eval_samples_per_second": 250.011,
+      "eval_steps_per_second": 31.251,
+      "step": 6200
+    },
+    {
+      "epoch": 0.6072479378872109,
+      "grad_norm": 1.6841580867767334,
+      "learning_rate": 5.387959866220736e-06,
+      "loss": 0.6938,
+      "step": 6300
+    },
+    {
+      "epoch": 0.6072479378872109,
+      "eval_loss": 0.613711953163147,
+      "eval_runtime": 40.0044,
+      "eval_samples_per_second": 249.972,
+      "eval_steps_per_second": 31.247,
+      "step": 6300
+    },
+    {
+      "epoch": 0.6168867940441508,
+      "grad_norm": 1.654695987701416,
+      "learning_rate": 5.3779264214046825e-06,
+      "loss": 0.6902,
+      "step": 6400
+    },
+    {
+      "epoch": 0.6168867940441508,
+      "eval_loss": 0.6061456799507141,
+      "eval_runtime": 39.9958,
+      "eval_samples_per_second": 250.026,
+      "eval_steps_per_second": 31.253,
+      "step": 6400
+    },
+    {
+      "epoch": 0.6265256502010906,
+      "grad_norm": 1.8732696771621704,
+      "learning_rate": 5.367892976588628e-06,
+      "loss": 0.6801,
+      "step": 6500
+    },
+    {
+      "epoch": 0.6265256502010906,
+      "eval_loss": 0.5902617573738098,
+      "eval_runtime": 40.1557,
+      "eval_samples_per_second": 249.031,
+      "eval_steps_per_second": 31.129,
+      "step": 6500
+    },
+    {
+      "epoch": 0.6361645063580305,
+      "grad_norm": 1.8029794692993164,
+      "learning_rate": 5.357859531772575e-06,
+      "loss": 0.6756,
+      "step": 6600
+    },
+    {
+      "epoch": 0.6361645063580305,
+      "eval_loss": 0.5980539321899414,
+      "eval_runtime": 39.9884,
+      "eval_samples_per_second": 250.072,
+      "eval_steps_per_second": 31.259,
+      "step": 6600
+    },
+    {
+      "epoch": 0.6458033625149704,
+      "grad_norm": 2.164788007736206,
+      "learning_rate": 5.347826086956522e-06,
+      "loss": 0.6702,
+      "step": 6700
+    },
+    {
+      "epoch": 0.6458033625149704,
+      "eval_loss": 0.5885195732116699,
+      "eval_runtime": 39.9791,
+      "eval_samples_per_second": 250.131,
+      "eval_steps_per_second": 31.266,
+      "step": 6700
+    },
+    {
+      "epoch": 0.6554422186719102,
+      "grad_norm": 1.8862236738204956,
+      "learning_rate": 5.337792642140468e-06,
+      "loss": 0.6638,
+      "step": 6800
+    },
+    {
+      "epoch": 0.6554422186719102,
+      "eval_loss": 0.5894142389297485,
+      "eval_runtime": 39.9968,
+      "eval_samples_per_second": 250.02,
+      "eval_steps_per_second": 31.252,
+      "step": 6800
+    },
+    {
+      "epoch": 0.66508107482885,
+      "grad_norm": 1.7155135869979858,
+      "learning_rate": 5.327759197324415e-06,
+      "loss": 0.6624,
+      "step": 6900
+    },
+    {
+      "epoch": 0.66508107482885,
+      "eval_loss": 0.5867804884910583,
+      "eval_runtime": 39.9878,
+      "eval_samples_per_second": 250.077,
+      "eval_steps_per_second": 31.26,
+      "step": 6900
+    },
+    {
+      "epoch": 0.6747199309857899,
+      "grad_norm": 1.592544436454773,
+      "learning_rate": 5.317725752508361e-06,
+      "loss": 0.6581,
+      "step": 7000
+    },
+    {
+      "epoch": 0.6747199309857899,
+      "eval_loss": 0.5733709931373596,
+      "eval_runtime": 40.159,
+      "eval_samples_per_second": 249.01,
+      "eval_steps_per_second": 31.126,
+      "step": 7000
+    },
+    {
+      "epoch": 0.6843587871427298,
+      "grad_norm": 1.7839319705963135,
+      "learning_rate": 5.307692307692307e-06,
+      "loss": 0.6516,
+      "step": 7100
+    },
+    {
+      "epoch": 0.6843587871427298,
+      "eval_loss": 0.5781156420707703,
+      "eval_runtime": 40.0095,
+      "eval_samples_per_second": 249.94,
+      "eval_steps_per_second": 31.243,
+      "step": 7100
+    },
+    {
+      "epoch": 0.6939976432996696,
+      "grad_norm": 1.6162323951721191,
+      "learning_rate": 5.297658862876254e-06,
+      "loss": 0.646,
+      "step": 7200
+    },
+    {
+      "epoch": 0.6939976432996696,
+      "eval_loss": 0.5711376667022705,
+      "eval_runtime": 40.0115,
+      "eval_samples_per_second": 249.928,
+      "eval_steps_per_second": 31.241,
+      "step": 7200
+    },
+    {
+      "epoch": 0.7036364994566094,
+      "grad_norm": 1.8823449611663818,
+      "learning_rate": 5.287625418060201e-06,
+      "loss": 0.6408,
+      "step": 7300
+    },
+    {
+      "epoch": 0.7036364994566094,
+      "eval_loss": 0.5678967237472534,
+      "eval_runtime": 40.0142,
+      "eval_samples_per_second": 249.912,
+      "eval_steps_per_second": 31.239,
+      "step": 7300
+    },
+    {
+      "epoch": 0.7132753556135494,
+      "grad_norm": 1.7034283876419067,
+      "learning_rate": 5.277591973244147e-06,
+      "loss": 0.6353,
+      "step": 7400
+    },
+    {
+      "epoch": 0.7132753556135494,
+      "eval_loss": 0.5637656450271606,
+      "eval_runtime": 40.0085,
+      "eval_samples_per_second": 249.947,
+      "eval_steps_per_second": 31.243,
+      "step": 7400
+    },
+    {
+      "epoch": 0.7229142117704892,
+      "grad_norm": 1.791896104812622,
+      "learning_rate": 5.2675585284280935e-06,
+      "loss": 0.6364,
+      "step": 7500
+    },
+    {
+      "epoch": 0.7229142117704892,
+      "eval_loss": 0.559363842010498,
+      "eval_runtime": 40.1708,
+      "eval_samples_per_second": 248.937,
+      "eval_steps_per_second": 31.117,
+      "step": 7500
+    },
+    {
+      "epoch": 0.732553067927429,
+      "grad_norm": 1.6761842966079712,
+      "learning_rate": 5.25752508361204e-06,
+      "loss": 0.6317,
+      "step": 7600
+    },
+    {
+      "epoch": 0.732553067927429,
+      "eval_loss": 0.5549041628837585,
+      "eval_runtime": 40.009,
+      "eval_samples_per_second": 249.944,
+      "eval_steps_per_second": 31.243,
+      "step": 7600
+    },
+    {
+      "epoch": 0.742191924084369,
+      "grad_norm": 1.7356570959091187,
+      "learning_rate": 5.247491638795986e-06,
+      "loss": 0.6226,
+      "step": 7700
+    },
+    {
+      "epoch": 0.742191924084369,
+      "eval_loss": 0.5542047619819641,
+      "eval_runtime": 40.0074,
+      "eval_samples_per_second": 249.954,
+      "eval_steps_per_second": 31.244,
+      "step": 7700
+    },
+    {
+      "epoch": 0.7518307802413088,
+      "grad_norm": 2.016263246536255,
+      "learning_rate": 5.237458193979933e-06,
+      "loss": 0.6205,
+      "step": 7800
+    },
+    {
+      "epoch": 0.7518307802413088,
+      "eval_loss": 0.5474265813827515,
+      "eval_runtime": 40.0104,
+      "eval_samples_per_second": 249.935,
+      "eval_steps_per_second": 31.242,
+      "step": 7800
+    },
+    {
+      "epoch": 0.7614696363982486,
+      "grad_norm": 1.9018930196762085,
+      "learning_rate": 5.22742474916388e-06,
+      "loss": 0.613,
+      "step": 7900
+    },
+    {
+      "epoch": 0.7614696363982486,
+      "eval_loss": 0.5480296611785889,
+      "eval_runtime": 40.0133,
+      "eval_samples_per_second": 249.917,
+      "eval_steps_per_second": 31.24,
+      "step": 7900
+    },
+    {
+      "epoch": 0.7711084925551884,
+      "grad_norm": 1.6764400005340576,
+      "learning_rate": 5.2173913043478265e-06,
+      "loss": 0.6159,
+      "step": 8000
+    },
+    {
+      "epoch": 0.7711084925551884,
+      "eval_loss": 0.5415164828300476,
+      "eval_runtime": 40.1661,
+      "eval_samples_per_second": 248.966,
+      "eval_steps_per_second": 31.121,
+      "step": 8000
+    },
+    {
+      "epoch": 0.7807473487121284,
+      "grad_norm": 1.817726492881775,
+      "learning_rate": 5.207357859531772e-06,
+      "loss": 0.6104,
+      "step": 8100
+    },
+    {
+      "epoch": 0.7807473487121284,
+      "eval_loss": 0.5427911281585693,
+      "eval_runtime": 39.9874,
+      "eval_samples_per_second": 250.079,
+      "eval_steps_per_second": 31.26,
+      "step": 8100
+    },
+    {
+      "epoch": 0.7903862048690682,
+      "grad_norm": 1.9600380659103394,
+      "learning_rate": 5.197324414715719e-06,
+      "loss": 0.6054,
+      "step": 8200
+    },
+    {
+      "epoch": 0.7903862048690682,
+      "eval_loss": 0.5321446657180786,
+      "eval_runtime": 40.0095,
+      "eval_samples_per_second": 249.94,
+      "eval_steps_per_second": 31.243,
+      "step": 8200
+    },
+    {
+      "epoch": 0.800025061026008,
+      "grad_norm": 1.6928702592849731,
+      "learning_rate": 5.187290969899666e-06,
+      "loss": 0.6019,
+      "step": 8300
+    },
+    {
+      "epoch": 0.800025061026008,
+      "eval_loss": 0.537187933921814,
+      "eval_runtime": 40.0123,
+      "eval_samples_per_second": 249.923,
+      "eval_steps_per_second": 31.24,
+      "step": 8300
+    },
+    {
+      "epoch": 0.809663917182948,
+      "grad_norm": 1.5704461336135864,
+      "learning_rate": 5.177257525083612e-06,
+      "loss": 0.6027,
+      "step": 8400
+    },
+    {
+      "epoch": 0.809663917182948,
+      "eval_loss": 0.5241678357124329,
+      "eval_runtime": 40.156,
+      "eval_samples_per_second": 249.029,
+      "eval_steps_per_second": 31.129,
+      "step": 8400
+    },
+    {
+      "epoch": 0.8193027733398878,
+      "grad_norm": 1.6146364212036133,
+      "learning_rate": 5.167224080267559e-06,
+      "loss": 0.5972,
+      "step": 8500
+    },
+    {
+      "epoch": 0.8193027733398878,
+      "eval_loss": 0.5265556573867798,
+      "eval_runtime": 40.0497,
+      "eval_samples_per_second": 249.69,
+      "eval_steps_per_second": 31.211,
+      "step": 8500
+    },
+    {
+      "epoch": 0.8289416294968276,
+      "grad_norm": 1.5260276794433594,
+      "learning_rate": 5.157190635451505e-06,
+      "loss": 0.5943,
+      "step": 8600
+    },
+    {
+      "epoch": 0.8289416294968276,
+      "eval_loss": 0.525088369846344,
+      "eval_runtime": 40.0049,
+      "eval_samples_per_second": 249.969,
+      "eval_steps_per_second": 31.246,
+      "step": 8600
+    },
+    {
+      "epoch": 0.8385804856537674,
+      "grad_norm": 1.4389654397964478,
+      "learning_rate": 5.147157190635451e-06,
+      "loss": 0.5896,
+      "step": 8700
+    },
+    {
+      "epoch": 0.8385804856537674,
+      "eval_loss": 0.5226108431816101,
+      "eval_runtime": 40.0292,
+      "eval_samples_per_second": 249.818,
+      "eval_steps_per_second": 31.227,
+      "step": 8700
+    },
+    {
+      "epoch": 0.8482193418107074,
+      "grad_norm": 1.5730172395706177,
+      "learning_rate": 5.137123745819398e-06,
+      "loss": 0.5866,
+      "step": 8800
+    },
+    {
+      "epoch": 0.8482193418107074,
+      "eval_loss": 0.5186478495597839,
+      "eval_runtime": 40.0165,
+      "eval_samples_per_second": 249.897,
+      "eval_steps_per_second": 31.237,
+      "step": 8800
+    },
+    {
+      "epoch": 0.8578581979676472,
+      "grad_norm": 1.562483310699463,
+      "learning_rate": 5.127090301003345e-06,
+      "loss": 0.5844,
+      "step": 8900
+    },
+    {
+      "epoch": 0.8578581979676472,
+      "eval_loss": 0.5097201466560364,
+      "eval_runtime": 40.1856,
+      "eval_samples_per_second": 248.845,
+      "eval_steps_per_second": 31.106,
+      "step": 8900
+    },
+    {
+      "epoch": 0.867497054124587,
+      "grad_norm": 1.6470036506652832,
+      "learning_rate": 5.117056856187291e-06,
+      "loss": 0.5773,
+      "step": 9000
+    },
+    {
+      "epoch": 0.867497054124587,
+      "eval_loss": 0.5092134475708008,
+      "eval_runtime": 40.0143,
+      "eval_samples_per_second": 249.911,
+      "eval_steps_per_second": 31.239,
+      "step": 9000
+    },
+    {
+      "epoch": 0.8771359102815269,
+      "grad_norm": 1.5788582563400269,
+      "learning_rate": 5.1070234113712375e-06,
+      "loss": 0.5797,
+      "step": 9100
+    },
+    {
+      "epoch": 0.8771359102815269,
+      "eval_loss": 0.5102807879447937,
+      "eval_runtime": 40.0303,
+      "eval_samples_per_second": 249.811,
+      "eval_steps_per_second": 31.226,
+      "step": 9100
+    },
+    {
+      "epoch": 0.8867747664384668,
+      "grad_norm": 1.756942868232727,
+      "learning_rate": 5.096989966555184e-06,
+      "loss": 0.5776,
+      "step": 9200
+    },
+    {
+      "epoch": 0.8867747664384668,
+      "eval_loss": 0.5153423547744751,
+      "eval_runtime": 40.0054,
+      "eval_samples_per_second": 249.966,
+      "eval_steps_per_second": 31.246,
+      "step": 9200
+    },
+    {
+      "epoch": 0.8964136225954066,
+      "grad_norm": 1.5298547744750977,
+      "learning_rate": 5.08695652173913e-06,
+      "loss": 0.5751,
+      "step": 9300
+    },
+    {
+      "epoch": 0.8964136225954066,
+      "eval_loss": 0.5078564286231995,
+      "eval_runtime": 40.011,
+      "eval_samples_per_second": 249.932,
+      "eval_steps_per_second": 31.241,
+      "step": 9300
+    },
+    {
+      "epoch": 0.9060524787523465,
+      "grad_norm": 1.7196496725082397,
+      "learning_rate": 5.076923076923077e-06,
+      "loss": 0.5665,
+      "step": 9400
+    },
+    {
+      "epoch": 0.9060524787523465,
+      "eval_loss": 0.5065763592720032,
+      "eval_runtime": 40.1717,
+      "eval_samples_per_second": 248.932,
+      "eval_steps_per_second": 31.116,
+      "step": 9400
+    },
+    {
+      "epoch": 0.9156913349092863,
+      "grad_norm": 1.6301361322402954,
+      "learning_rate": 5.066889632107024e-06,
+      "loss": 0.5675,
+      "step": 9500
+    },
+    {
+      "epoch": 0.9156913349092863,
+      "eval_loss": 0.49776747822761536,
+      "eval_runtime": 40.0165,
+      "eval_samples_per_second": 249.897,
+      "eval_steps_per_second": 31.237,
+      "step": 9500
+    },
+    {
+      "epoch": 0.9253301910662262,
+      "grad_norm": 1.4653514623641968,
+      "learning_rate": 5.05685618729097e-06,
+      "loss": 0.5618,
+      "step": 9600
+    },
+    {
+      "epoch": 0.9253301910662262,
+      "eval_loss": 0.49646928906440735,
+      "eval_runtime": 40.017,
+      "eval_samples_per_second": 249.894,
+      "eval_steps_per_second": 31.237,
+      "step": 9600
+    },
+    {
+      "epoch": 0.934969047223166,
+      "grad_norm": 1.6670091152191162,
+      "learning_rate": 5.046822742474916e-06,
+      "loss": 0.563,
+      "step": 9700
+    },
+    {
+      "epoch": 0.934969047223166,
+      "eval_loss": 0.49914979934692383,
+      "eval_runtime": 40.0323,
+      "eval_samples_per_second": 249.798,
+      "eval_steps_per_second": 31.225,
+      "step": 9700
+    },
+    {
+      "epoch": 0.9446079033801059,
+      "grad_norm": 1.6070232391357422,
+      "learning_rate": 5.036789297658863e-06,
+      "loss": 0.5542,
+      "step": 9800
+    },
+    {
+      "epoch": 0.9446079033801059,
+      "eval_loss": 0.4884184002876282,
+      "eval_runtime": 40.0566,
+      "eval_samples_per_second": 249.646,
+      "eval_steps_per_second": 31.206,
+      "step": 9800
+    },
+    {
+      "epoch": 0.9542467595370457,
+      "grad_norm": 1.4984960556030273,
+      "learning_rate": 5.02675585284281e-06,
+      "loss": 0.5581,
+      "step": 9900
+    },
+    {
+      "epoch": 0.9542467595370457,
+      "eval_loss": 0.4900113642215729,
+      "eval_runtime": 40.2056,
+      "eval_samples_per_second": 248.721,
+      "eval_steps_per_second": 31.09,
+      "step": 9900
+    },
+    {
+      "epoch": 0.9638856156939856,
+      "grad_norm": 1.537240743637085,
+      "learning_rate": 5.016722408026756e-06,
+      "loss": 0.5539,
+      "step": 10000
+    },
+    {
+      "epoch": 0.9638856156939856,
+      "eval_loss": 0.4897591769695282,
+      "eval_runtime": 40.0038,
+      "eval_samples_per_second": 249.976,
+      "eval_steps_per_second": 31.247,
+      "step": 10000
+    },
+    {
+      "epoch": 0.9735244718509255,
+      "grad_norm": 1.6569781303405762,
+      "learning_rate": 5.0066889632107026e-06,
+      "loss": 0.5481,
+      "step": 10100
+    },
+    {
+      "epoch": 0.9735244718509255,
+      "eval_loss": 0.4842386245727539,
+      "eval_runtime": 40.0389,
+      "eval_samples_per_second": 249.757,
+      "eval_steps_per_second": 31.22,
+      "step": 10100
+    },
+    {
+      "epoch": 0.9831633280078653,
+      "grad_norm": 1.4528193473815918,
+      "learning_rate": 4.996655518394649e-06,
+      "loss": 0.5464,
+      "step": 10200
+    },
+    {
+      "epoch": 0.9831633280078653,
+      "eval_loss": 0.4880593717098236,
+      "eval_runtime": 40.0358,
+      "eval_samples_per_second": 249.776,
+      "eval_steps_per_second": 31.222,
+      "step": 10200
+    },
+    {
+      "epoch": 0.9928021841648051,
+      "grad_norm": 1.5718955993652344,
+      "learning_rate": 4.986622073578595e-06,
+      "loss": 0.543,
+      "step": 10300
+    },
+    {
+      "epoch": 0.9928021841648051,
+      "eval_loss": 0.4782450795173645,
+      "eval_runtime": 36.7041,
+      "eval_samples_per_second": 272.449,
+      "eval_steps_per_second": 34.056,
+      "step": 10300
+    },
+    {
+      "epoch": 1.002441040321745,
+      "grad_norm": 1.647326946258545,
+      "learning_rate": 4.976588628762542e-06,
+      "loss": 0.5412,
+      "step": 10400
+    },
+    {
+      "epoch": 1.002441040321745,
+      "eval_loss": 0.486726850271225,
+      "eval_runtime": 40.2013,
+      "eval_samples_per_second": 248.748,
+      "eval_steps_per_second": 31.094,
+      "step": 10400
+    },
+    {
+      "epoch": 1.0120798964786848,
+      "grad_norm": 1.6924008131027222,
+      "learning_rate": 4.966555183946489e-06,
+      "loss": 0.5394,
+      "step": 10500
+    },
+    {
+      "epoch": 1.0120798964786848,
+      "eval_loss": 0.4839903712272644,
+      "eval_runtime": 40.0205,
+      "eval_samples_per_second": 249.872,
+      "eval_steps_per_second": 31.234,
+      "step": 10500
+    },
+    {
+      "epoch": 1.0217187526356248,
+      "grad_norm": 1.5169456005096436,
+      "learning_rate": 4.956521739130435e-06,
+      "loss": 0.5397,
+      "step": 10600
+    },
+    {
+      "epoch": 1.0217187526356248,
+      "eval_loss": 0.47214385867118835,
+      "eval_runtime": 40.1949,
+      "eval_samples_per_second": 248.788,
+      "eval_steps_per_second": 31.098,
+      "step": 10600
+    },
+    {
+      "epoch": 1.0313576087925647,
+      "grad_norm": 1.5480705499649048,
+      "learning_rate": 4.9464882943143815e-06,
+      "loss": 0.5357,
+      "step": 10700
+    },
+    {
+      "epoch": 1.0313576087925647,
+      "eval_loss": 0.48204073309898376,
+      "eval_runtime": 40.0052,
+      "eval_samples_per_second": 249.968,
+      "eval_steps_per_second": 31.246,
+      "step": 10700
+    },
+    {
+      "epoch": 1.0409964649495045,
+      "grad_norm": 1.3915138244628906,
+      "learning_rate": 4.936454849498328e-06,
+      "loss": 0.534,
+      "step": 10800
+    },
+    {
+      "epoch": 1.0409964649495045,
+      "eval_loss": 0.47229185700416565,
+      "eval_runtime": 40.0072,
+      "eval_samples_per_second": 249.955,
+      "eval_steps_per_second": 31.244,
+      "step": 10800
+    },
+    {
+      "epoch": 1.0506353211064443,
+      "grad_norm": 1.4983240365982056,
+      "learning_rate": 4.926421404682274e-06,
+      "loss": 0.5276,
+      "step": 10900
+    },
+    {
+      "epoch": 1.0506353211064443,
+      "eval_loss": 0.46871235966682434,
+      "eval_runtime": 40.0389,
+      "eval_samples_per_second": 249.757,
+      "eval_steps_per_second": 31.22,
+      "step": 10900
+    },
+    {
+      "epoch": 1.0602741772633841,
+      "grad_norm": 1.533703088760376,
+      "learning_rate": 4.916387959866221e-06,
+      "loss": 0.5296,
+      "step": 11000
+    },
+    {
+      "epoch": 1.0602741772633841,
+      "eval_loss": 0.46552249789237976,
+      "eval_runtime": 40.0022,
+      "eval_samples_per_second": 249.986,
+      "eval_steps_per_second": 31.248,
+      "step": 11000
+    },
+    {
+      "epoch": 1.069913033420324,
+      "grad_norm": 1.5167957544326782,
+      "learning_rate": 4.906354515050168e-06,
+      "loss": 0.5254,
+      "step": 11100
+    },
+    {
+      "epoch": 1.069913033420324,
+      "eval_loss": 0.4699433445930481,
+      "eval_runtime": 40.1743,
+      "eval_samples_per_second": 248.915,
+      "eval_steps_per_second": 31.114,
+      "step": 11100
+    },
+    {
+      "epoch": 1.0795518895772638,
+      "grad_norm": 1.4261205196380615,
+      "learning_rate": 4.8963210702341136e-06,
+      "loss": 0.5201,
+      "step": 11200
+    },
+    {
+      "epoch": 1.0795518895772638,
+      "eval_loss": 0.4564357399940491,
+      "eval_runtime": 40.025,
+      "eval_samples_per_second": 249.844,
+      "eval_steps_per_second": 31.23,
+      "step": 11200
+    },
+    {
+      "epoch": 1.0891907457342038,
+      "grad_norm": 1.3741745948791504,
+      "learning_rate": 4.88628762541806e-06,
+      "loss": 0.5224,
+      "step": 11300
+    },
+    {
+      "epoch": 1.0891907457342038,
+      "eval_loss": 0.46628084778785706,
+      "eval_runtime": 40.0189,
+      "eval_samples_per_second": 249.882,
+      "eval_steps_per_second": 31.235,
+      "step": 11300
+    },
+    {
+      "epoch": 1.0988296018911436,
+      "grad_norm": 1.4131207466125488,
+      "learning_rate": 4.876254180602007e-06,
+      "loss": 0.5147,
+      "step": 11400
+    },
+    {
+      "epoch": 1.0988296018911436,
+      "eval_loss": 0.46065640449523926,
+      "eval_runtime": 40.0026,
+      "eval_samples_per_second": 249.984,
+      "eval_steps_per_second": 31.248,
+      "step": 11400
+    },
+    {
+      "epoch": 1.1084684580480835,
+      "grad_norm": 1.434804081916809,
+      "learning_rate": 4.866220735785953e-06,
+      "loss": 0.5186,
+      "step": 11500
+    },
+    {
+      "epoch": 1.1084684580480835,
+      "eval_loss": 0.4589378833770752,
+      "eval_runtime": 40.0058,
+      "eval_samples_per_second": 249.964,
+      "eval_steps_per_second": 31.246,
+      "step": 11500
+    },
+    {
+      "epoch": 1.1181073142050233,
+      "grad_norm": 1.382394552230835,
+      "learning_rate": 4.8561872909699e-06,
+      "loss": 0.5138,
+      "step": 11600
+    },
+    {
+      "epoch": 1.1181073142050233,
+      "eval_loss": 0.4565480053424835,
+      "eval_runtime": 40.1633,
+      "eval_samples_per_second": 248.984,
+      "eval_steps_per_second": 31.123,
+      "step": 11600
+    },
+    {
+      "epoch": 1.1277461703619631,
+      "grad_norm": 1.5272095203399658,
+      "learning_rate": 4.8461538461538465e-06,
+      "loss": 0.5164,
+      "step": 11700
+    },
+    {
+      "epoch": 1.1277461703619631,
+      "eval_loss": 0.45689794421195984,
+      "eval_runtime": 40.0182,
+      "eval_samples_per_second": 249.887,
+      "eval_steps_per_second": 31.236,
+      "step": 11700
+    },
+    {
+      "epoch": 1.137385026518903,
+      "grad_norm": 1.5336707830429077,
+      "learning_rate": 4.8361204013377925e-06,
+      "loss": 0.5089,
+      "step": 11800
+    },
+    {
+      "epoch": 1.137385026518903,
+      "eval_loss": 0.4564231336116791,
+      "eval_runtime": 40.0203,
+      "eval_samples_per_second": 249.873,
+      "eval_steps_per_second": 31.234,
+      "step": 11800
+    },
+    {
+      "epoch": 1.1470238826758428,
+      "grad_norm": 1.5161396265029907,
+      "learning_rate": 4.826086956521739e-06,
+      "loss": 0.5106,
+      "step": 11900
+    },
+    {
+      "epoch": 1.1470238826758428,
+      "eval_loss": 0.44420644640922546,
+      "eval_runtime": 40.0196,
+      "eval_samples_per_second": 249.877,
+      "eval_steps_per_second": 31.235,
+      "step": 11900
+    },
+    {
+      "epoch": 1.1566627388327828,
+      "grad_norm": 1.5325267314910889,
+      "learning_rate": 4.816053511705686e-06,
+      "loss": 0.5098,
+      "step": 12000
+    },
+    {
+      "epoch": 1.1566627388327828,
+      "eval_loss": 0.44834986329078674,
+      "eval_runtime": 40.0681,
+      "eval_samples_per_second": 249.575,
+      "eval_steps_per_second": 31.197,
+      "step": 12000
+    },
+    {
+      "epoch": 1.1663015949897226,
+      "grad_norm": 1.4805413484573364,
+      "learning_rate": 4.806020066889633e-06,
+      "loss": 0.5055,
+      "step": 12100
+    },
+    {
+      "epoch": 1.1663015949897226,
+      "eval_loss": 0.4538011848926544,
+      "eval_runtime": 40.2028,
+      "eval_samples_per_second": 248.739,
+      "eval_steps_per_second": 31.092,
+      "step": 12100
+    },
+    {
+      "epoch": 1.1759404511466625,
+      "grad_norm": 1.5557194948196411,
+      "learning_rate": 4.795986622073579e-06,
+      "loss": 0.5058,
+      "step": 12200
+    },
+    {
+      "epoch": 1.1759404511466625,
+      "eval_loss": 0.44933125376701355,
+      "eval_runtime": 40.023,
+      "eval_samples_per_second": 249.856,
+      "eval_steps_per_second": 31.232,
+      "step": 12200
+    },
+    {
+      "epoch": 1.1855793073036023,
+      "grad_norm": 1.5012928247451782,
+      "learning_rate": 4.785953177257525e-06,
+      "loss": 0.5051,
+      "step": 12300
+    },
+    {
+      "epoch": 1.1855793073036023,
+      "eval_loss": 0.44258418679237366,
+      "eval_runtime": 40.0302,
+      "eval_samples_per_second": 249.811,
+      "eval_steps_per_second": 31.226,
+      "step": 12300
+    },
+    {
+      "epoch": 1.195218163460542,
+      "grad_norm": 1.4221644401550293,
+      "learning_rate": 4.775919732441472e-06,
+      "loss": 0.4999,
+      "step": 12400
+    },
+    {
+      "epoch": 1.195218163460542,
+      "eval_loss": 0.4362325668334961,
+      "eval_runtime": 40.0213,
+      "eval_samples_per_second": 249.867,
+      "eval_steps_per_second": 31.233,
+      "step": 12400
+    },
+    {
+      "epoch": 1.204857019617482,
+      "grad_norm": 1.415287971496582,
+      "learning_rate": 4.765886287625418e-06,
+      "loss": 0.4973,
+      "step": 12500
+    },
+    {
+      "epoch": 1.204857019617482,
+      "eval_loss": 0.44187524914741516,
+      "eval_runtime": 40.0371,
+      "eval_samples_per_second": 249.769,
+      "eval_steps_per_second": 31.221,
+      "step": 12500
+    },
+    {
+      "epoch": 1.2144958757744218,
+      "grad_norm": 1.5381195545196533,
+      "learning_rate": 4.755852842809365e-06,
+      "loss": 0.4989,
+      "step": 12600
+    },
+    {
+      "epoch": 1.2144958757744218,
+      "eval_loss": 0.4413149952888489,
+      "eval_runtime": 40.2155,
+      "eval_samples_per_second": 248.66,
+      "eval_steps_per_second": 31.083,
+      "step": 12600
+    },
+    {
+      "epoch": 1.2241347319313618,
+      "grad_norm": 1.374925971031189,
+      "learning_rate": 4.745819397993312e-06,
+      "loss": 0.4923,
+      "step": 12700
+    },
+    {
+      "epoch": 1.2241347319313618,
+      "eval_loss": 0.440106064081192,
+      "eval_runtime": 40.0356,
+      "eval_samples_per_second": 249.777,
+      "eval_steps_per_second": 31.222,
+      "step": 12700
+    },
+    {
+      "epoch": 1.2337735880883016,
+      "grad_norm": 1.6059918403625488,
+      "learning_rate": 4.7357859531772575e-06,
+      "loss": 0.4918,
+      "step": 12800
+    },
+    {
+      "epoch": 1.2337735880883016,
+      "eval_loss": 0.44359320402145386,
+      "eval_runtime": 40.0597,
+      "eval_samples_per_second": 249.628,
+      "eval_steps_per_second": 31.203,
+      "step": 12800
+    },
+    {
+      "epoch": 1.2434124442452414,
+      "grad_norm": 1.4142322540283203,
+      "learning_rate": 4.725752508361204e-06,
+      "loss": 0.4881,
+      "step": 12900
+    },
+    {
+      "epoch": 1.2434124442452414,
+      "eval_loss": 0.44244951009750366,
+      "eval_runtime": 40.061,
+      "eval_samples_per_second": 249.619,
+      "eval_steps_per_second": 31.202,
+      "step": 12900
+    },
+    {
+      "epoch": 1.2530513004021813,
+      "grad_norm": 1.446840524673462,
+      "learning_rate": 4.715719063545151e-06,
+      "loss": 0.49,
+      "step": 13000
+    },
+    {
+      "epoch": 1.2530513004021813,
+      "eval_loss": 0.42693030834198,
+      "eval_runtime": 40.2377,
+      "eval_samples_per_second": 248.523,
+      "eval_steps_per_second": 31.065,
+      "step": 13000
+    },
+    {
+      "epoch": 1.262690156559121,
+      "grad_norm": 1.363279938697815,
+      "learning_rate": 4.705685618729097e-06,
+      "loss": 0.4839,
+      "step": 13100
+    },
+    {
+      "epoch": 1.262690156559121,
+      "eval_loss": 0.43453505635261536,
+      "eval_runtime": 40.0754,
+      "eval_samples_per_second": 249.53,
+      "eval_steps_per_second": 31.191,
+      "step": 13100
+    },
+    {
+      "epoch": 1.272329012716061,
+      "grad_norm": 1.4301884174346924,
+      "learning_rate": 4.695652173913044e-06,
+      "loss": 0.4867,
+      "step": 13200
+    },
+    {
+      "epoch": 1.272329012716061,
+      "eval_loss": 0.42788979411125183,
+      "eval_runtime": 40.0379,
+      "eval_samples_per_second": 249.763,
+      "eval_steps_per_second": 31.22,
+      "step": 13200
+    },
+    {
+      "epoch": 1.281967868873001,
+      "grad_norm": 1.4869626760482788,
+      "learning_rate": 4.6856187290969905e-06,
+      "loss": 0.4822,
+      "step": 13300
+    },
+    {
+      "epoch": 1.281967868873001,
+      "eval_loss": 0.42807090282440186,
+      "eval_runtime": 40.0485,
+      "eval_samples_per_second": 249.697,
+      "eval_steps_per_second": 31.212,
+      "step": 13300
+    },
+    {
+      "epoch": 1.2916067250299408,
+      "grad_norm": 1.4084575176239014,
+      "learning_rate": 4.675585284280936e-06,
+      "loss": 0.4817,
+      "step": 13400
+    },
+    {
+      "epoch": 1.2916067250299408,
+      "eval_loss": 0.431538462638855,
+      "eval_runtime": 40.2297,
+      "eval_samples_per_second": 248.572,
+      "eval_steps_per_second": 31.072,
+      "step": 13400
+    },
+    {
+      "epoch": 1.3012455811868806,
+      "grad_norm": 1.3125011920928955,
+      "learning_rate": 4.665551839464883e-06,
+      "loss": 0.4781,
+      "step": 13500
+    },
+    {
+      "epoch": 1.3012455811868806,
+      "eval_loss": 0.4229593276977539,
+      "eval_runtime": 40.0557,
+      "eval_samples_per_second": 249.652,
+      "eval_steps_per_second": 31.207,
+      "step": 13500
+    },
+    {
+      "epoch": 1.3108844373438204,
+      "grad_norm": 1.495177984237671,
+      "learning_rate": 4.65551839464883e-06,
+      "loss": 0.4782,
+      "step": 13600
+    },
+    {
+      "epoch": 1.3108844373438204,
+      "eval_loss": 0.4288323223590851,
+      "eval_runtime": 40.0435,
+      "eval_samples_per_second": 249.728,
+      "eval_steps_per_second": 31.216,
+      "step": 13600
+    },
+    {
+      "epoch": 1.3205232935007603,
+      "grad_norm": 1.349768042564392,
+      "learning_rate": 4.645484949832776e-06,
+      "loss": 0.48,
+      "step": 13700
+    },
+    {
+      "epoch": 1.3205232935007603,
+      "eval_loss": 0.41586947441101074,
+      "eval_runtime": 40.0452,
+      "eval_samples_per_second": 249.718,
+      "eval_steps_per_second": 31.215,
+      "step": 13700
+    },
+    {
+      "epoch": 1.3301621496577,
+      "grad_norm": 1.4638071060180664,
+      "learning_rate": 4.635451505016723e-06,
+      "loss": 0.4808,
+      "step": 13800
+    },
+    {
+      "epoch": 1.3301621496577,
+      "eval_loss": 0.4116751551628113,
+      "eval_runtime": 40.0376,
+      "eval_samples_per_second": 249.765,
+      "eval_steps_per_second": 31.221,
+      "step": 13800
+    },
+    {
+      "epoch": 1.33980100581464,
+      "grad_norm": 1.4769808053970337,
+      "learning_rate": 4.625418060200669e-06,
+      "loss": 0.4769,
+      "step": 13900
+    },
+    {
+      "epoch": 1.33980100581464,
+      "eval_loss": 0.4249781370162964,
+      "eval_runtime": 40.2173,
+      "eval_samples_per_second": 248.649,
+      "eval_steps_per_second": 31.081,
+      "step": 13900
+    },
+    {
+      "epoch": 1.3494398619715797,
+      "grad_norm": 1.4002916812896729,
+      "learning_rate": 4.615384615384616e-06,
+      "loss": 0.4703,
+      "step": 14000
+    },
+    {
+      "epoch": 1.3494398619715797,
+      "eval_loss": 0.41435718536376953,
+      "eval_runtime": 40.0251,
+      "eval_samples_per_second": 249.843,
+      "eval_steps_per_second": 31.23,
+      "step": 14000
+    },
+    {
+      "epoch": 1.3590787181285198,
+      "grad_norm": 1.324873685836792,
+      "learning_rate": 4.605351170568562e-06,
+      "loss": 0.4715,
+      "step": 14100
+    },
+    {
+      "epoch": 1.3590787181285198,
+      "eval_loss": 0.41557979583740234,
+      "eval_runtime": 40.0132,
+      "eval_samples_per_second": 249.917,
+      "eval_steps_per_second": 31.24,
+      "step": 14100
+    },
+    {
+      "epoch": 1.3687175742854596,
+      "grad_norm": 1.510293960571289,
+      "learning_rate": 4.595317725752509e-06,
+      "loss": 0.4661,
+      "step": 14200
+    },
+    {
+      "epoch": 1.3687175742854596,
+      "eval_loss": 0.41991928219795227,
+      "eval_runtime": 40.0151,
+      "eval_samples_per_second": 249.906,
+      "eval_steps_per_second": 31.238,
+      "step": 14200
+    },
+    {
+      "epoch": 1.3783564304423994,
+      "grad_norm": 1.4551657438278198,
+      "learning_rate": 4.585284280936456e-06,
+      "loss": 0.466,
+      "step": 14300
+    },
+    {
+      "epoch": 1.3783564304423994,
+      "eval_loss": 0.41806551814079285,
+      "eval_runtime": 40.0069,
+      "eval_samples_per_second": 249.957,
+      "eval_steps_per_second": 31.245,
+      "step": 14300
+    },
+    {
+      "epoch": 1.3879952865993392,
+      "grad_norm": 1.5093111991882324,
+      "learning_rate": 4.5752508361204015e-06,
+      "loss": 0.4653,
+      "step": 14400
+    },
+    {
+      "epoch": 1.3879952865993392,
+      "eval_loss": 0.41728827357292175,
+      "eval_runtime": 40.1654,
+      "eval_samples_per_second": 248.97,
+      "eval_steps_per_second": 31.121,
+      "step": 14400
+    },
+    {
+      "epoch": 1.397634142756279,
+      "grad_norm": 1.4863073825836182,
+      "learning_rate": 4.565217391304348e-06,
+      "loss": 0.467,
+      "step": 14500
+    },
+    {
+      "epoch": 1.397634142756279,
+      "eval_loss": 0.41565340757369995,
+      "eval_runtime": 39.9963,
+      "eval_samples_per_second": 250.023,
+      "eval_steps_per_second": 31.253,
+      "step": 14500
+    },
+    {
+      "epoch": 1.4072729989132189,
+      "grad_norm": 1.4027386903762817,
+      "learning_rate": 4.555183946488295e-06,
+      "loss": 0.4662,
+      "step": 14600
+    },
+    {
+      "epoch": 1.4072729989132189,
+      "eval_loss": 0.4067195951938629,
+      "eval_runtime": 40.0068,
+      "eval_samples_per_second": 249.957,
+      "eval_steps_per_second": 31.245,
+      "step": 14600
+    },
+    {
+      "epoch": 1.416911855070159,
+      "grad_norm": 1.4059629440307617,
+      "learning_rate": 4.545150501672241e-06,
+      "loss": 0.4651,
+      "step": 14700
+    },
+    {
+      "epoch": 1.416911855070159,
+      "eval_loss": 0.40936049818992615,
+      "eval_runtime": 39.9887,
+      "eval_samples_per_second": 250.071,
+      "eval_steps_per_second": 31.259,
+      "step": 14700
+    },
+    {
+      "epoch": 1.4265507112270988,
+      "grad_norm": 1.2644177675247192,
+      "learning_rate": 4.535117056856188e-06,
+      "loss": 0.4622,
+      "step": 14800
+    },
+    {
+      "epoch": 1.4265507112270988,
+      "eval_loss": 0.4118250906467438,
+      "eval_runtime": 39.9915,
+      "eval_samples_per_second": 250.053,
+      "eval_steps_per_second": 31.257,
+      "step": 14800
+    },
+    {
+      "epoch": 1.4361895673840386,
+      "grad_norm": 1.2830288410186768,
+      "learning_rate": 4.5250836120401345e-06,
+      "loss": 0.4597,
+      "step": 14900
+    },
+    {
+      "epoch": 1.4361895673840386,
+      "eval_loss": 0.40810731053352356,
+      "eval_runtime": 40.2112,
+      "eval_samples_per_second": 248.687,
+      "eval_steps_per_second": 31.086,
+      "step": 14900
+    },
+    {
+      "epoch": 1.4458284235409784,
+      "grad_norm": 1.2887154817581177,
+      "learning_rate": 4.51505016722408e-06,
+      "loss": 0.4602,
+      "step": 15000
+    },
+    {
+      "epoch": 1.4458284235409784,
+      "eval_loss": 0.40011951327323914,
+      "eval_runtime": 40.0118,
+      "eval_samples_per_second": 249.926,
+      "eval_steps_per_second": 31.241,
+      "step": 15000
+    },
+    {
+      "epoch": 1.4554672796979182,
+      "grad_norm": 1.348257303237915,
+      "learning_rate": 4.505016722408027e-06,
+      "loss": 0.4574,
+      "step": 15100
+    },
+    {
+      "epoch": 1.4554672796979182,
+      "eval_loss": 0.4056689143180847,
+      "eval_runtime": 40.015,
+      "eval_samples_per_second": 249.906,
+      "eval_steps_per_second": 31.238,
+      "step": 15100
+    },
+    {
+      "epoch": 1.465106135854858,
+      "grad_norm": 1.540839433670044,
+      "learning_rate": 4.494983277591973e-06,
+      "loss": 0.4549,
+      "step": 15200
+    },
+    {
+      "epoch": 1.465106135854858,
+      "eval_loss": 0.39897704124450684,
+      "eval_runtime": 40.0174,
+      "eval_samples_per_second": 249.892,
+      "eval_steps_per_second": 31.236,
+      "step": 15200
+    },
+    {
+      "epoch": 1.4747449920117979,
+      "grad_norm": 1.4020183086395264,
+      "learning_rate": 4.48494983277592e-06,
+      "loss": 0.4552,
+      "step": 15300
+    },
+    {
+      "epoch": 1.4747449920117979,
+      "eval_loss": 0.39921414852142334,
+      "eval_runtime": 40.2019,
+      "eval_samples_per_second": 248.745,
+      "eval_steps_per_second": 31.093,
+      "step": 15300
+    },
+    {
+      "epoch": 1.4843838481687377,
+      "grad_norm": 1.4397515058517456,
+      "learning_rate": 4.474916387959866e-06,
+      "loss": 0.4524,
+      "step": 15400
+    },
+    {
+      "epoch": 1.4843838481687377,
+      "eval_loss": 0.40324947237968445,
+      "eval_runtime": 40.0276,
+      "eval_samples_per_second": 249.828,
+      "eval_steps_per_second": 31.228,
+      "step": 15400
+    },
+    {
+      "epoch": 1.4940227043256777,
+      "grad_norm": 1.327071189880371,
+      "learning_rate": 4.4648829431438125e-06,
+      "loss": 0.4501,
+      "step": 15500
+    },
+    {
+      "epoch": 1.4940227043256777,
+      "eval_loss": 0.39761507511138916,
+      "eval_runtime": 39.9927,
+      "eval_samples_per_second": 250.045,
+      "eval_steps_per_second": 31.256,
+      "step": 15500
+    },
+    {
+      "epoch": 1.5036615604826176,
+      "grad_norm": 1.4344964027404785,
+      "learning_rate": 4.454849498327759e-06,
+      "loss": 0.4482,
+      "step": 15600
+    },
+    {
+      "epoch": 1.5036615604826176,
+      "eval_loss": 0.39871275424957275,
+      "eval_runtime": 40.227,
+      "eval_samples_per_second": 248.589,
+      "eval_steps_per_second": 31.074,
+      "step": 15600
+    },
+    {
+      "epoch": 1.5133004166395574,
+      "grad_norm": 1.3496111631393433,
+      "learning_rate": 4.444816053511705e-06,
+      "loss": 0.4457,
+      "step": 15700
+    },
+    {
+      "epoch": 1.5133004166395574,
+      "eval_loss": 0.40053361654281616,
+      "eval_runtime": 40.0154,
+      "eval_samples_per_second": 249.904,
+      "eval_steps_per_second": 31.238,
+      "step": 15700
+    },
+    {
+      "epoch": 1.5229392727964972,
+      "grad_norm": 1.3155416250228882,
+      "learning_rate": 4.434782608695652e-06,
+      "loss": 0.4469,
+      "step": 15800
+    },
+    {
+      "epoch": 1.5229392727964972,
+      "eval_loss": 0.399395227432251,
+      "eval_runtime": 40.0839,
+      "eval_samples_per_second": 249.476,
+      "eval_steps_per_second": 31.185,
+      "step": 15800
+    },
+    {
+      "epoch": 1.5325781289534373,
+      "grad_norm": 1.438828945159912,
+      "learning_rate": 4.424749163879599e-06,
+      "loss": 0.4467,
+      "step": 15900
+    },
+    {
+      "epoch": 1.5325781289534373,
+      "eval_loss": 0.39455991983413696,
+      "eval_runtime": 40.1699,
+      "eval_samples_per_second": 248.943,
+      "eval_steps_per_second": 31.118,
+      "step": 15900
+    },
+    {
+      "epoch": 1.542216985110377,
+      "grad_norm": 1.3380151987075806,
+      "learning_rate": 4.414715719063545e-06,
+      "loss": 0.4479,
+      "step": 16000
+    },
+    {
+      "epoch": 1.542216985110377,
+      "eval_loss": 0.3920137584209442,
+      "eval_runtime": 40.0458,
+      "eval_samples_per_second": 249.714,
+      "eval_steps_per_second": 31.214,
+      "step": 16000
+    },
+    {
+      "epoch": 1.551855841267317,
+      "grad_norm": 1.3055827617645264,
+      "learning_rate": 4.404682274247491e-06,
+      "loss": 0.4437,
+      "step": 16100
+    },
+    {
+      "epoch": 1.551855841267317,
+      "eval_loss": 0.39583614468574524,
+      "eval_runtime": 40.0719,
+      "eval_samples_per_second": 249.551,
+      "eval_steps_per_second": 31.194,
+      "step": 16100
+    },
+    {
+      "epoch": 1.5614946974242567,
+      "grad_norm": 1.3439549207687378,
+      "learning_rate": 4.394648829431438e-06,
+      "loss": 0.4421,
+      "step": 16200
+    },
+    {
+      "epoch": 1.5614946974242567,
+      "eval_loss": 0.3982265293598175,
+      "eval_runtime": 40.0502,
+      "eval_samples_per_second": 249.687,
+      "eval_steps_per_second": 31.211,
+      "step": 16200
+    },
+    {
+      "epoch": 1.5711335535811966,
+      "grad_norm": 1.2873746156692505,
+      "learning_rate": 4.384615384615384e-06,
+      "loss": 0.4421,
+      "step": 16300
+    },
+    {
+      "epoch": 1.5711335535811966,
+      "eval_loss": 0.392922580242157,
+      "eval_runtime": 40.0623,
+      "eval_samples_per_second": 249.611,
+      "eval_steps_per_second": 31.201,
+      "step": 16300
+    },
+    {
+      "epoch": 1.5807724097381364,
+      "grad_norm": 1.4414591789245605,
+      "learning_rate": 4.374581939799331e-06,
+      "loss": 0.4388,
+      "step": 16400
+    },
+    {
+      "epoch": 1.5807724097381364,
+      "eval_loss": 0.39475396275520325,
+      "eval_runtime": 40.2477,
+      "eval_samples_per_second": 248.462,
+      "eval_steps_per_second": 31.058,
+      "step": 16400
+    },
+    {
+      "epoch": 1.5904112658950762,
+      "grad_norm": 1.3728173971176147,
+      "learning_rate": 4.364548494983278e-06,
+      "loss": 0.4414,
+      "step": 16500
+    },
+    {
+      "epoch": 1.5904112658950762,
+      "eval_loss": 0.39176425337791443,
+      "eval_runtime": 40.0569,
+      "eval_samples_per_second": 249.645,
+      "eval_steps_per_second": 31.206,
+      "step": 16500
+    },
+    {
+      "epoch": 1.600050122052016,
+      "grad_norm": 1.5120171308517456,
+      "learning_rate": 4.354515050167224e-06,
+      "loss": 0.4414,
+      "step": 16600
+    },
+    {
+      "epoch": 1.600050122052016,
+      "eval_loss": 0.39494696259498596,
+      "eval_runtime": 40.0501,
+      "eval_samples_per_second": 249.687,
+      "eval_steps_per_second": 31.211,
+      "step": 16600
+    },
+    {
+      "epoch": 1.6096889782089558,
+      "grad_norm": 1.7337701320648193,
+      "learning_rate": 4.34448160535117e-06,
+      "loss": 0.436,
+      "step": 16700
+    },
+    {
+      "epoch": 1.6096889782089558,
+      "eval_loss": 0.3927867114543915,
+      "eval_runtime": 40.0345,
+      "eval_samples_per_second": 249.785,
+      "eval_steps_per_second": 31.223,
+      "step": 16700
+    },
+    {
+      "epoch": 1.6193278343658957,
+      "grad_norm": 1.465790033340454,
+      "learning_rate": 4.334448160535117e-06,
+      "loss": 0.436,
+      "step": 16800
+    },
+    {
+      "epoch": 1.6193278343658957,
+      "eval_loss": 0.38513514399528503,
+      "eval_runtime": 40.227,
+      "eval_samples_per_second": 248.589,
+      "eval_steps_per_second": 31.074,
+      "step": 16800
+    },
+    {
+      "epoch": 1.6289666905228355,
+      "grad_norm": 1.401219129562378,
+      "learning_rate": 4.324414715719064e-06,
+      "loss": 0.4375,
+      "step": 16900
+    },
+    {
+      "epoch": 1.6289666905228355,
+      "eval_loss": 0.37909185886383057,
+      "eval_runtime": 40.1381,
+      "eval_samples_per_second": 249.14,
+      "eval_steps_per_second": 31.142,
+      "step": 16900
+    },
+    {
+      "epoch": 1.6386055466797755,
+      "grad_norm": 1.3634768724441528,
+      "learning_rate": 4.31438127090301e-06,
+      "loss": 0.4358,
+      "step": 17000
+    },
+    {
+      "epoch": 1.6386055466797755,
+      "eval_loss": 0.3808463513851166,
+      "eval_runtime": 40.0323,
+      "eval_samples_per_second": 249.798,
+      "eval_steps_per_second": 31.225,
+      "step": 17000
+    },
+    {
+      "epoch": 1.6482444028367154,
+      "grad_norm": 1.4873460531234741,
+      "learning_rate": 4.3043478260869565e-06,
+      "loss": 0.4321,
+      "step": 17100
+    },
+    {
+      "epoch": 1.6482444028367154,
+      "eval_loss": 0.3821844160556793,
+      "eval_runtime": 40.0357,
+      "eval_samples_per_second": 249.777,
+      "eval_steps_per_second": 31.222,
+      "step": 17100
+    },
+    {
+      "epoch": 1.6578832589936552,
+      "grad_norm": 1.323455572128296,
+      "learning_rate": 4.294314381270903e-06,
+      "loss": 0.431,
+      "step": 17200
+    },
+    {
+      "epoch": 1.6578832589936552,
+      "eval_loss": 0.37755951285362244,
+      "eval_runtime": 40.2365,
+      "eval_samples_per_second": 248.531,
+      "eval_steps_per_second": 31.066,
+      "step": 17200
+    },
+    {
+      "epoch": 1.6675221151505952,
+      "grad_norm": 1.3670837879180908,
+      "learning_rate": 4.284280936454849e-06,
+      "loss": 0.4302,
+      "step": 17300
+    },
+    {
+      "epoch": 1.6675221151505952,
+      "eval_loss": 0.37524840235710144,
+      "eval_runtime": 40.0419,
+      "eval_samples_per_second": 249.738,
+      "eval_steps_per_second": 31.217,
+      "step": 17300
+    },
+    {
+      "epoch": 1.677160971307535,
+      "grad_norm": 1.507103681564331,
+      "learning_rate": 4.274247491638796e-06,
+      "loss": 0.4288,
+      "step": 17400
+    },
+    {
+      "epoch": 1.677160971307535,
+      "eval_loss": 0.38275057077407837,
+      "eval_runtime": 40.0423,
+      "eval_samples_per_second": 249.736,
+      "eval_steps_per_second": 31.217,
+      "step": 17400
+    },
+    {
+      "epoch": 1.6867998274644749,
+      "grad_norm": 1.3555055856704712,
+      "learning_rate": 4.264214046822743e-06,
+      "loss": 0.4307,
+      "step": 17500
+    },
+    {
+      "epoch": 1.6867998274644749,
+      "eval_loss": 0.3769054710865021,
+      "eval_runtime": 40.045,
+      "eval_samples_per_second": 249.719,
+      "eval_steps_per_second": 31.215,
+      "step": 17500
+    },
+    {
+      "epoch": 1.6964386836214147,
+      "grad_norm": 1.2957983016967773,
+      "learning_rate": 4.254180602006689e-06,
+      "loss": 0.4287,
+      "step": 17600
+    },
+    {
+      "epoch": 1.6964386836214147,
+      "eval_loss": 0.3762720823287964,
+      "eval_runtime": 40.2206,
+      "eval_samples_per_second": 248.629,
+      "eval_steps_per_second": 31.079,
+      "step": 17600
+    },
+    {
+      "epoch": 1.7060775397783545,
+      "grad_norm": 1.4515742063522339,
+      "learning_rate": 4.244147157190635e-06,
+      "loss": 0.4248,
+      "step": 17700
+    },
+    {
+      "epoch": 1.7060775397783545,
+      "eval_loss": 0.37980878353118896,
+      "eval_runtime": 40.0419,
+      "eval_samples_per_second": 249.738,
+      "eval_steps_per_second": 31.217,
+      "step": 17700
+    },
+    {
+      "epoch": 1.7157163959352943,
+      "grad_norm": 1.3719956874847412,
+      "learning_rate": 4.234113712374582e-06,
+      "loss": 0.4237,
+      "step": 17800
+    },
+    {
+      "epoch": 1.7157163959352943,
+      "eval_loss": 0.3736819326877594,
+      "eval_runtime": 40.0386,
+      "eval_samples_per_second": 249.759,
+      "eval_steps_per_second": 31.22,
+      "step": 17800
+    },
+    {
+      "epoch": 1.7253552520922342,
+      "grad_norm": 1.2918155193328857,
+      "learning_rate": 4.224080267558528e-06,
+      "loss": 0.4234,
+      "step": 17900
+    },
+    {
+      "epoch": 1.7253552520922342,
+      "eval_loss": 0.37445351481437683,
+      "eval_runtime": 40.0158,
+      "eval_samples_per_second": 249.901,
+      "eval_steps_per_second": 31.238,
+      "step": 17900
+    },
+    {
+      "epoch": 1.734994108249174,
+      "grad_norm": 1.5635416507720947,
+      "learning_rate": 4.214046822742475e-06,
+      "loss": 0.4226,
+      "step": 18000
+    },
+    {
+      "epoch": 1.734994108249174,
+      "eval_loss": 0.37227481603622437,
+      "eval_runtime": 40.0091,
+      "eval_samples_per_second": 249.943,
+      "eval_steps_per_second": 31.243,
+      "step": 18000
+    },
+    {
+      "epoch": 1.7446329644061138,
+      "grad_norm": 1.3523764610290527,
+      "learning_rate": 4.2040133779264216e-06,
+      "loss": 0.4253,
+      "step": 18100
+    },
+    {
+      "epoch": 1.7446329644061138,
+      "eval_loss": 0.37538984417915344,
+      "eval_runtime": 40.1924,
+      "eval_samples_per_second": 248.803,
+      "eval_steps_per_second": 31.1,
+      "step": 18100
+    },
+    {
+      "epoch": 1.7542718205630536,
+      "grad_norm": 1.2665579319000244,
+      "learning_rate": 4.1939799331103675e-06,
+      "loss": 0.4228,
+      "step": 18200
+    },
+    {
+      "epoch": 1.7542718205630536,
+      "eval_loss": 0.3729080259799957,
+      "eval_runtime": 40.0153,
+      "eval_samples_per_second": 249.905,
+      "eval_steps_per_second": 31.238,
+      "step": 18200
+    },
+    {
+      "epoch": 1.7639106767199937,
+      "grad_norm": 1.2079461812973022,
+      "learning_rate": 4.183946488294314e-06,
+      "loss": 0.4184,
+      "step": 18300
+    },
+    {
+      "epoch": 1.7639106767199937,
+      "eval_loss": 0.37542304396629333,
+      "eval_runtime": 40.226,
+      "eval_samples_per_second": 248.595,
+      "eval_steps_per_second": 31.074,
+      "step": 18300
+    },
+    {
+      "epoch": 1.7735495328769335,
+      "grad_norm": 1.2203370332717896,
+      "learning_rate": 4.173913043478261e-06,
+      "loss": 0.4186,
+      "step": 18400
+    },
+    {
+      "epoch": 1.7735495328769335,
+      "eval_loss": 0.3722892701625824,
+      "eval_runtime": 40.0431,
+      "eval_samples_per_second": 249.731,
+      "eval_steps_per_second": 31.216,
+      "step": 18400
+    },
+    {
+      "epoch": 1.7831883890338733,
+      "grad_norm": 1.2480080127716064,
+      "learning_rate": 4.163879598662208e-06,
+      "loss": 0.4214,
+      "step": 18500
+    },
+    {
+      "epoch": 1.7831883890338733,
+      "eval_loss": 0.3761518597602844,
+      "eval_runtime": 40.0595,
+      "eval_samples_per_second": 249.629,
+      "eval_steps_per_second": 31.204,
+      "step": 18500
+    },
+    {
+      "epoch": 1.7928272451908132,
+      "grad_norm": 1.514190912246704,
+      "learning_rate": 4.153846153846154e-06,
+      "loss": 0.4153,
+      "step": 18600
+    },
+    {
+      "epoch": 1.7928272451908132,
+      "eval_loss": 0.3671943247318268,
+      "eval_runtime": 40.0447,
+      "eval_samples_per_second": 249.721,
+      "eval_steps_per_second": 31.215,
+      "step": 18600
+    },
+    {
+      "epoch": 1.8024661013477532,
+      "grad_norm": 1.4013168811798096,
+      "learning_rate": 4.1438127090301005e-06,
+      "loss": 0.4152,
+      "step": 18700
+    },
+    {
+      "epoch": 1.8024661013477532,
+      "eval_loss": 0.36363351345062256,
+      "eval_runtime": 40.2138,
+      "eval_samples_per_second": 248.671,
+      "eval_steps_per_second": 31.084,
+      "step": 18700
+    },
+    {
+      "epoch": 1.812104957504693,
+      "grad_norm": 1.3701732158660889,
+      "learning_rate": 4.133779264214047e-06,
+      "loss": 0.4139,
+      "step": 18800
+    },
+    {
+      "epoch": 1.812104957504693,
+      "eval_loss": 0.37101006507873535,
+      "eval_runtime": 40.0504,
+      "eval_samples_per_second": 249.685,
+      "eval_steps_per_second": 31.211,
+      "step": 18800
+    },
+    {
+      "epoch": 1.8217438136616328,
+      "grad_norm": 1.2950865030288696,
+      "learning_rate": 4.123745819397993e-06,
+      "loss": 0.4127,
+      "step": 18900
+    },
+    {
+      "epoch": 1.8217438136616328,
+      "eval_loss": 0.3611432611942291,
+      "eval_runtime": 39.9977,
+      "eval_samples_per_second": 250.014,
+      "eval_steps_per_second": 31.252,
+      "step": 18900
+    },
+    {
+      "epoch": 1.8313826698185727,
+      "grad_norm": 1.2140529155731201,
+      "learning_rate": 4.11371237458194e-06,
+      "loss": 0.4175,
+      "step": 19000
+    },
+    {
+      "epoch": 1.8313826698185727,
+      "eval_loss": 0.3688327372074127,
+      "eval_runtime": 39.9976,
+      "eval_samples_per_second": 250.015,
+      "eval_steps_per_second": 31.252,
+      "step": 19000
+    },
+    {
+      "epoch": 1.8410215259755125,
+      "grad_norm": 1.4030205011367798,
+      "learning_rate": 4.103678929765887e-06,
+      "loss": 0.4123,
+      "step": 19100
+    },
+    {
+      "epoch": 1.8410215259755125,
+      "eval_loss": 0.36343154311180115,
+      "eval_runtime": 40.2164,
+      "eval_samples_per_second": 248.655,
+      "eval_steps_per_second": 31.082,
+      "step": 19100
+    },
+    {
+      "epoch": 1.8506603821324523,
+      "grad_norm": 1.2921127080917358,
+      "learning_rate": 4.0936454849498326e-06,
+      "loss": 0.4119,
+      "step": 19200
+    },
+    {
+      "epoch": 1.8506603821324523,
+      "eval_loss": 0.36798274517059326,
+      "eval_runtime": 40.0209,
+      "eval_samples_per_second": 249.869,
+      "eval_steps_per_second": 31.234,
+      "step": 19200
+    },
+    {
+      "epoch": 1.8602992382893921,
+      "grad_norm": 1.2748223543167114,
+      "learning_rate": 4.083612040133779e-06,
+      "loss": 0.4143,
+      "step": 19300
+    },
+    {
+      "epoch": 1.8602992382893921,
+      "eval_loss": 0.36462557315826416,
+      "eval_runtime": 40.0258,
+      "eval_samples_per_second": 249.839,
+      "eval_steps_per_second": 31.23,
+      "step": 19300
+    },
+    {
+      "epoch": 1.869938094446332,
+      "grad_norm": 1.2742102146148682,
+      "learning_rate": 4.073578595317726e-06,
+      "loss": 0.4093,
+      "step": 19400
+    },
+    {
+      "epoch": 1.869938094446332,
+      "eval_loss": 0.36330220103263855,
+      "eval_runtime": 40.0305,
+      "eval_samples_per_second": 249.809,
+      "eval_steps_per_second": 31.226,
+      "step": 19400
+    },
+    {
+      "epoch": 1.8795769506032718,
+      "grad_norm": 1.3298977613449097,
+      "learning_rate": 4.063545150501672e-06,
+      "loss": 0.4094,
+      "step": 19500
+    },
+    {
+      "epoch": 1.8795769506032718,
+      "eval_loss": 0.36082392930984497,
+      "eval_runtime": 40.0284,
+      "eval_samples_per_second": 249.822,
+      "eval_steps_per_second": 31.228,
+      "step": 19500
+    },
+    {
+      "epoch": 1.8892158067602116,
+      "grad_norm": 1.4168245792388916,
+      "learning_rate": 4.053511705685619e-06,
+      "loss": 0.4081,
+      "step": 19600
+    },
+    {
+      "epoch": 1.8892158067602116,
+      "eval_loss": 0.3604665994644165,
+      "eval_runtime": 40.2162,
+      "eval_samples_per_second": 248.656,
+      "eval_steps_per_second": 31.082,
+      "step": 19600
+    },
+    {
+      "epoch": 1.8988546629171517,
+      "grad_norm": 1.3064316511154175,
+      "learning_rate": 4.0434782608695655e-06,
+      "loss": 0.4096,
+      "step": 19700
+    },
+    {
+      "epoch": 1.8988546629171517,
+      "eval_loss": 0.35935744643211365,
+      "eval_runtime": 40.019,
+      "eval_samples_per_second": 249.881,
+      "eval_steps_per_second": 31.235,
+      "step": 19700
+    },
+    {
+      "epoch": 1.9084935190740915,
+      "grad_norm": 1.380152702331543,
+      "learning_rate": 4.0334448160535115e-06,
+      "loss": 0.4097,
+      "step": 19800
+    },
+    {
+      "epoch": 1.9084935190740915,
+      "eval_loss": 0.36837923526763916,
+      "eval_runtime": 40.0097,
+      "eval_samples_per_second": 249.94,
+      "eval_steps_per_second": 31.242,
+      "step": 19800
+    },
+    {
+      "epoch": 1.9181323752310313,
+      "grad_norm": 1.3063116073608398,
+      "learning_rate": 4.023411371237458e-06,
+      "loss": 0.4066,
+      "step": 19900
+    },
+    {
+      "epoch": 1.9181323752310313,
+      "eval_loss": 0.3581005036830902,
+      "eval_runtime": 40.0395,
+      "eval_samples_per_second": 249.754,
+      "eval_steps_per_second": 31.219,
+      "step": 19900
+    },
+    {
+      "epoch": 1.9277712313879714,
+      "grad_norm": 1.276556134223938,
+      "learning_rate": 4.013377926421405e-06,
+      "loss": 0.4048,
+      "step": 20000
+    },
+    {
+      "epoch": 1.9277712313879714,
+      "eval_loss": 0.3602534234523773,
+      "eval_runtime": 40.0325,
+      "eval_samples_per_second": 249.797,
+      "eval_steps_per_second": 31.225,
+      "step": 20000
+    },
+    {
+      "epoch": 1.9374100875449112,
+      "grad_norm": 1.4124329090118408,
+      "learning_rate": 4.003344481605351e-06,
+      "loss": 0.4067,
+      "step": 20100
+    },
+    {
+      "epoch": 1.9374100875449112,
+      "eval_loss": 0.35985687375068665,
+      "eval_runtime": 40.2758,
+      "eval_samples_per_second": 248.288,
+      "eval_steps_per_second": 31.036,
+      "step": 20100
+    },
+    {
+      "epoch": 1.947048943701851,
+      "grad_norm": 1.3454973697662354,
+      "learning_rate": 3.993311036789298e-06,
+      "loss": 0.4051,
+      "step": 20200
+    },
+    {
+      "epoch": 1.947048943701851,
+      "eval_loss": 0.353756844997406,
+      "eval_runtime": 40.0628,
+      "eval_samples_per_second": 249.608,
+      "eval_steps_per_second": 31.201,
+      "step": 20200
+    },
+    {
+      "epoch": 1.9566877998587908,
+      "grad_norm": 1.326913595199585,
+      "learning_rate": 3.9832775919732444e-06,
+      "loss": 0.4017,
+      "step": 20300
+    },
+    {
+      "epoch": 1.9566877998587908,
+      "eval_loss": 0.35797813534736633,
+      "eval_runtime": 40.067,
+      "eval_samples_per_second": 249.582,
+      "eval_steps_per_second": 31.198,
+      "step": 20300
+    },
+    {
+      "epoch": 1.9663266560157306,
+      "grad_norm": 1.1353386640548706,
+      "learning_rate": 3.97324414715719e-06,
+      "loss": 0.4039,
+      "step": 20400
+    },
+    {
+      "epoch": 1.9663266560157306,
+      "eval_loss": 0.3506408631801605,
+      "eval_runtime": 40.0693,
+      "eval_samples_per_second": 249.568,
+      "eval_steps_per_second": 31.196,
+      "step": 20400
+    },
+    {
+      "epoch": 1.9759655121726705,
+      "grad_norm": 1.3176063299179077,
+      "learning_rate": 3.963210702341137e-06,
+      "loss": 0.4042,
+      "step": 20500
+    },
+    {
+      "epoch": 1.9759655121726705,
+      "eval_loss": 0.35446348786354065,
+      "eval_runtime": 40.2855,
+      "eval_samples_per_second": 248.228,
+      "eval_steps_per_second": 31.029,
+      "step": 20500
+    },
+    {
+      "epoch": 1.9856043683296103,
+      "grad_norm": 1.243143081665039,
+      "learning_rate": 3.953177257525084e-06,
+      "loss": 0.4,
+      "step": 20600
+    },
+    {
+      "epoch": 1.9856043683296103,
+      "eval_loss": 0.35905522108078003,
+      "eval_runtime": 40.1017,
+      "eval_samples_per_second": 249.366,
+      "eval_steps_per_second": 31.171,
+      "step": 20600
+    },
+    {
+      "epoch": 1.9952432244865501,
+      "grad_norm": 1.3850924968719482,
+      "learning_rate": 3.943143812709031e-06,
+      "loss": 0.4025,
+      "step": 20700
+    },
+    {
+      "epoch": 1.9952432244865501,
+      "eval_loss": 0.35461366176605225,
+      "eval_runtime": 40.0328,
+      "eval_samples_per_second": 249.795,
+      "eval_steps_per_second": 31.224,
+      "step": 20700
+    },
+    {
+      "epoch": 2.00488208064349,
+      "grad_norm": 1.2397725582122803,
+      "learning_rate": 3.9331103678929765e-06,
+      "loss": 0.3984,
+      "step": 20800
+    },
+    {
+      "epoch": 2.00488208064349,
+      "eval_loss": 0.35152626037597656,
+      "eval_runtime": 40.0186,
+      "eval_samples_per_second": 249.884,
+      "eval_steps_per_second": 31.235,
+      "step": 20800
+    },
+    {
+      "epoch": 2.0145209368004298,
+      "grad_norm": 1.2805670499801636,
+      "learning_rate": 3.923076923076923e-06,
+      "loss": 0.3973,
+      "step": 20900
+    },
+    {
+      "epoch": 2.0145209368004298,
+      "eval_loss": 0.3505365252494812,
+      "eval_runtime": 40.0175,
+      "eval_samples_per_second": 249.89,
+      "eval_steps_per_second": 31.236,
+      "step": 20900
+    },
+    {
+      "epoch": 2.0241597929573696,
+      "grad_norm": 1.4358912706375122,
+      "learning_rate": 3.91304347826087e-06,
+      "loss": 0.3955,
+      "step": 21000
+    },
+    {
+      "epoch": 2.0241597929573696,
+      "eval_loss": 0.3397263288497925,
+      "eval_runtime": 40.2153,
+      "eval_samples_per_second": 248.661,
+      "eval_steps_per_second": 31.083,
+      "step": 21000
+    },
+    {
+      "epoch": 2.0337986491143094,
+      "grad_norm": 1.312941312789917,
+      "learning_rate": 3.903010033444816e-06,
+      "loss": 0.397,
+      "step": 21100
+    },
+    {
+      "epoch": 2.0337986491143094,
+      "eval_loss": 0.35078564286231995,
+      "eval_runtime": 40.0071,
+      "eval_samples_per_second": 249.956,
+      "eval_steps_per_second": 31.244,
+      "step": 21100
+    },
+    {
+      "epoch": 2.0434375052712497,
+      "grad_norm": 1.678265929222107,
+      "learning_rate": 3.892976588628763e-06,
+      "loss": 0.3929,
+      "step": 21200
+    },
+    {
+      "epoch": 2.0434375052712497,
+      "eval_loss": 0.3554067313671112,
+      "eval_runtime": 40.0148,
+      "eval_samples_per_second": 249.907,
+      "eval_steps_per_second": 31.238,
+      "step": 21200
+    },
+    {
+      "epoch": 2.0530763614281895,
+      "grad_norm": 1.271547794342041,
+      "learning_rate": 3.8829431438127095e-06,
+      "loss": 0.3969,
+      "step": 21300
+    },
+    {
+      "epoch": 2.0530763614281895,
+      "eval_loss": 0.3503868877887726,
+      "eval_runtime": 40.0058,
+      "eval_samples_per_second": 249.964,
+      "eval_steps_per_second": 31.245,
+      "step": 21300
+    },
+    {
+      "epoch": 2.0627152175851293,
+      "grad_norm": 1.3128001689910889,
+      "learning_rate": 3.8729096989966554e-06,
+      "loss": 0.3928,
+      "step": 21400
+    },
+    {
+      "epoch": 2.0627152175851293,
+      "eval_loss": 0.35138940811157227,
+      "eval_runtime": 40.2205,
+      "eval_samples_per_second": 248.63,
+      "eval_steps_per_second": 31.079,
+      "step": 21400
+    },
+    {
+      "epoch": 2.072354073742069,
+      "grad_norm": 1.2826029062271118,
+      "learning_rate": 3.862876254180602e-06,
+      "loss": 0.3956,
+      "step": 21500
+    },
+    {
+      "epoch": 2.072354073742069,
+      "eval_loss": 0.34093034267425537,
+      "eval_runtime": 40.0778,
+      "eval_samples_per_second": 249.515,
+      "eval_steps_per_second": 31.189,
+      "step": 21500
+    },
+    {
+      "epoch": 2.081992929899009,
+      "grad_norm": 1.2576078176498413,
+      "learning_rate": 3.852842809364549e-06,
+      "loss": 0.3909,
+      "step": 21600
+    },
+    {
+      "epoch": 2.081992929899009,
+      "eval_loss": 0.3406901955604553,
+      "eval_runtime": 40.2132,
+      "eval_samples_per_second": 248.675,
+      "eval_steps_per_second": 31.084,
+      "step": 21600
+    },
+    {
+      "epoch": 2.091631786055949,
+      "grad_norm": 1.2353730201721191,
+      "learning_rate": 3.842809364548495e-06,
+      "loss": 0.3899,
+      "step": 21700
+    },
+    {
+      "epoch": 2.091631786055949,
+      "eval_loss": 0.3464890122413635,
+      "eval_runtime": 40.0356,
+      "eval_samples_per_second": 249.778,
+      "eval_steps_per_second": 31.222,
+      "step": 21700
+    },
+    {
+      "epoch": 2.1012706422128886,
+      "grad_norm": 1.2551077604293823,
+      "learning_rate": 3.832775919732442e-06,
+      "loss": 0.3885,
+      "step": 21800
+    },
+    {
+      "epoch": 2.1012706422128886,
+      "eval_loss": 0.3510824739933014,
+      "eval_runtime": 40.0097,
+      "eval_samples_per_second": 249.939,
+      "eval_steps_per_second": 31.242,
+      "step": 21800
+    },
+    {
+      "epoch": 2.1109094983698284,
+      "grad_norm": 1.2558820247650146,
+      "learning_rate": 3.822742474916388e-06,
+      "loss": 0.393,
+      "step": 21900
+    },
+    {
+      "epoch": 2.1109094983698284,
+      "eval_loss": 0.3473747968673706,
+      "eval_runtime": 40.0293,
+      "eval_samples_per_second": 249.817,
+      "eval_steps_per_second": 31.227,
+      "step": 21900
+    },
+    {
+      "epoch": 2.1205483545267683,
+      "grad_norm": 1.2809922695159912,
+      "learning_rate": 3.8127090301003347e-06,
+      "loss": 0.3943,
+      "step": 22000
+    },
+    {
+      "epoch": 2.1205483545267683,
+      "eval_loss": 0.3469405770301819,
+      "eval_runtime": 40.2871,
+      "eval_samples_per_second": 248.218,
+      "eval_steps_per_second": 31.027,
+      "step": 22000
+    },
+    {
+      "epoch": 2.130187210683708,
+      "grad_norm": 1.2375646829605103,
+      "learning_rate": 3.802675585284281e-06,
+      "loss": 0.3895,
+      "step": 22100
+    },
+    {
+      "epoch": 2.130187210683708,
+      "eval_loss": 0.34392210841178894,
+      "eval_runtime": 40.0647,
+      "eval_samples_per_second": 249.596,
+      "eval_steps_per_second": 31.2,
+      "step": 22100
+    },
+    {
+      "epoch": 2.139826066840648,
+      "grad_norm": 1.3119927644729614,
+      "learning_rate": 3.792642140468228e-06,
+      "loss": 0.3918,
+      "step": 22200
+    },
+    {
+      "epoch": 2.139826066840648,
+      "eval_loss": 0.34562018513679504,
+      "eval_runtime": 40.0596,
+      "eval_samples_per_second": 249.628,
+      "eval_steps_per_second": 31.204,
+      "step": 22200
+    },
+    {
+      "epoch": 2.1494649229975877,
+      "grad_norm": 1.2270475625991821,
+      "learning_rate": 3.782608695652174e-06,
+      "loss": 0.3882,
+      "step": 22300
+    },
+    {
+      "epoch": 2.1494649229975877,
+      "eval_loss": 0.3436911702156067,
+      "eval_runtime": 40.1832,
+      "eval_samples_per_second": 248.86,
+      "eval_steps_per_second": 31.108,
+      "step": 22300
+    },
+    {
+      "epoch": 2.1591037791545276,
+      "grad_norm": 1.2549434900283813,
+      "learning_rate": 3.7725752508361205e-06,
+      "loss": 0.3855,
+      "step": 22400
+    },
+    {
+      "epoch": 2.1591037791545276,
+      "eval_loss": 0.3425428569316864,
+      "eval_runtime": 40.2231,
+      "eval_samples_per_second": 248.613,
+      "eval_steps_per_second": 31.077,
+      "step": 22400
+    },
+    {
+      "epoch": 2.1687426353114674,
+      "grad_norm": 1.2413984537124634,
+      "learning_rate": 3.7625418060200673e-06,
+      "loss": 0.3855,
+      "step": 22500
+    },
+    {
+      "epoch": 2.1687426353114674,
+      "eval_loss": 0.34506529569625854,
+      "eval_runtime": 40.0554,
+      "eval_samples_per_second": 249.655,
+      "eval_steps_per_second": 31.207,
+      "step": 22500
+    },
+    {
+      "epoch": 2.1783814914684076,
+      "grad_norm": 1.4316176176071167,
+      "learning_rate": 3.7525083612040136e-06,
+      "loss": 0.3846,
+      "step": 22600
+    },
+    {
+      "epoch": 2.1783814914684076,
+      "eval_loss": 0.3439718186855316,
+      "eval_runtime": 40.0313,
+      "eval_samples_per_second": 249.805,
+      "eval_steps_per_second": 31.226,
+      "step": 22600
+    },
+    {
+      "epoch": 2.1880203476253475,
+      "grad_norm": 1.3857834339141846,
+      "learning_rate": 3.74247491638796e-06,
+      "loss": 0.3845,
+      "step": 22700
+    },
+    {
+      "epoch": 2.1880203476253475,
+      "eval_loss": 0.34433916211128235,
+      "eval_runtime": 40.0274,
+      "eval_samples_per_second": 249.829,
+      "eval_steps_per_second": 31.229,
+      "step": 22700
+    },
+    {
+      "epoch": 2.1976592037822873,
+      "grad_norm": 1.2255629301071167,
+      "learning_rate": 3.7324414715719067e-06,
+      "loss": 0.3847,
+      "step": 22800
+    },
+    {
+      "epoch": 2.1976592037822873,
+      "eval_loss": 0.343479722738266,
+      "eval_runtime": 40.0139,
+      "eval_samples_per_second": 249.913,
+      "eval_steps_per_second": 31.239,
+      "step": 22800
+    },
+    {
+      "epoch": 2.207298059939227,
+      "grad_norm": 1.1473214626312256,
+      "learning_rate": 3.722408026755853e-06,
+      "loss": 0.3827,
+      "step": 22900
+    },
+    {
+      "epoch": 2.207298059939227,
+      "eval_loss": 0.34567588567733765,
+      "eval_runtime": 40.2165,
+      "eval_samples_per_second": 248.654,
+      "eval_steps_per_second": 31.082,
+      "step": 22900
+    },
+    {
+      "epoch": 2.216936916096167,
+      "grad_norm": 1.4547693729400635,
+      "learning_rate": 3.7123745819398e-06,
+      "loss": 0.3815,
+      "step": 23000
+    },
+    {
+      "epoch": 2.216936916096167,
+      "eval_loss": 0.340980589389801,
+      "eval_runtime": 40.0285,
+      "eval_samples_per_second": 249.822,
+      "eval_steps_per_second": 31.228,
+      "step": 23000
+    },
+    {
+      "epoch": 2.2265757722531068,
+      "grad_norm": 1.3170710802078247,
+      "learning_rate": 3.702341137123746e-06,
+      "loss": 0.3853,
+      "step": 23100
+    },
+    {
+      "epoch": 2.2265757722531068,
+      "eval_loss": 0.3409317433834076,
+      "eval_runtime": 40.0248,
+      "eval_samples_per_second": 249.845,
+      "eval_steps_per_second": 31.231,
+      "step": 23100
+    },
+    {
+      "epoch": 2.2362146284100466,
+      "grad_norm": 1.328068733215332,
+      "learning_rate": 3.6923076923076925e-06,
+      "loss": 0.3823,
+      "step": 23200
+    },
+    {
+      "epoch": 2.2362146284100466,
+      "eval_loss": 0.33983665704727173,
+      "eval_runtime": 40.1406,
+      "eval_samples_per_second": 249.124,
+      "eval_steps_per_second": 31.141,
+      "step": 23200
+    },
+    {
+      "epoch": 2.2458534845669864,
+      "grad_norm": 1.3068214654922485,
+      "learning_rate": 3.6822742474916393e-06,
+      "loss": 0.3802,
+      "step": 23300
+    },
+    {
+      "epoch": 2.2458534845669864,
+      "eval_loss": 0.3398624658584595,
+      "eval_runtime": 40.0336,
+      "eval_samples_per_second": 249.79,
+      "eval_steps_per_second": 31.224,
+      "step": 23300
+    },
+    {
+      "epoch": 2.2554923407239262,
+      "grad_norm": 1.2860172986984253,
+      "learning_rate": 3.6722408026755856e-06,
+      "loss": 0.3852,
+      "step": 23400
+    },
+    {
+      "epoch": 2.2554923407239262,
+      "eval_loss": 0.3490051031112671,
+      "eval_runtime": 40.1888,
+      "eval_samples_per_second": 248.826,
+      "eval_steps_per_second": 31.103,
+      "step": 23400
+    },
+    {
+      "epoch": 2.265131196880866,
+      "grad_norm": 1.3662693500518799,
+      "learning_rate": 3.662207357859532e-06,
+      "loss": 0.3787,
+      "step": 23500
+    },
+    {
+      "epoch": 2.265131196880866,
+      "eval_loss": 0.33699285984039307,
+      "eval_runtime": 40.0257,
+      "eval_samples_per_second": 249.839,
+      "eval_steps_per_second": 31.23,
+      "step": 23500
+    },
+    {
+      "epoch": 2.274770053037806,
+      "grad_norm": 1.3295284509658813,
+      "learning_rate": 3.6521739130434787e-06,
+      "loss": 0.3802,
+      "step": 23600
+    },
+    {
+      "epoch": 2.274770053037806,
+      "eval_loss": 0.33331725001335144,
+      "eval_runtime": 40.0229,
+      "eval_samples_per_second": 249.857,
+      "eval_steps_per_second": 31.232,
+      "step": 23600
+    },
+    {
+      "epoch": 2.2844089091947457,
+      "grad_norm": 1.1981773376464844,
+      "learning_rate": 3.642140468227425e-06,
+      "loss": 0.3777,
+      "step": 23700
+    },
+    {
+      "epoch": 2.2844089091947457,
+      "eval_loss": 0.33419373631477356,
+      "eval_runtime": 40.0172,
+      "eval_samples_per_second": 249.893,
+      "eval_steps_per_second": 31.237,
+      "step": 23700
+    },
+    {
+      "epoch": 2.2940477653516855,
+      "grad_norm": 1.318207025527954,
+      "learning_rate": 3.6321070234113714e-06,
+      "loss": 0.374,
+      "step": 23800
+    },
+    {
+      "epoch": 2.2940477653516855,
+      "eval_loss": 0.3326634168624878,
+      "eval_runtime": 40.0247,
+      "eval_samples_per_second": 249.846,
+      "eval_steps_per_second": 31.231,
+      "step": 23800
+    },
+    {
+      "epoch": 2.303686621508626,
+      "grad_norm": 1.204872727394104,
+      "learning_rate": 3.622073578595318e-06,
+      "loss": 0.3774,
+      "step": 23900
+    },
+    {
+      "epoch": 2.303686621508626,
+      "eval_loss": 0.3303450047969818,
+      "eval_runtime": 40.1914,
+      "eval_samples_per_second": 248.809,
+      "eval_steps_per_second": 31.101,
+      "step": 23900
+    },
+    {
+      "epoch": 2.3133254776655656,
+      "grad_norm": 1.30888831615448,
+      "learning_rate": 3.6120401337792645e-06,
+      "loss": 0.3752,
+      "step": 24000
+    },
+    {
+      "epoch": 2.3133254776655656,
+      "eval_loss": 0.3322921395301819,
+      "eval_runtime": 40.0327,
+      "eval_samples_per_second": 249.796,
+      "eval_steps_per_second": 31.224,
+      "step": 24000
+    },
+    {
+      "epoch": 2.3229643338225054,
+      "grad_norm": 1.296487808227539,
+      "learning_rate": 3.6020066889632112e-06,
+      "loss": 0.3759,
+      "step": 24100
+    },
+    {
+      "epoch": 2.3229643338225054,
+      "eval_loss": 0.3311348259449005,
+      "eval_runtime": 40.1498,
+      "eval_samples_per_second": 249.067,
+      "eval_steps_per_second": 31.133,
+      "step": 24100
+    },
+    {
+      "epoch": 2.3326031899794453,
+      "grad_norm": 1.2083899974822998,
+      "learning_rate": 3.5919732441471576e-06,
+      "loss": 0.3744,
+      "step": 24200
+    },
+    {
+      "epoch": 2.3326031899794453,
+      "eval_loss": 0.3285525441169739,
+      "eval_runtime": 40.0431,
+      "eval_samples_per_second": 249.731,
+      "eval_steps_per_second": 31.216,
+      "step": 24200
+    },
+    {
+      "epoch": 2.342242046136385,
+      "grad_norm": 1.2163653373718262,
+      "learning_rate": 3.581939799331104e-06,
+      "loss": 0.3752,
+      "step": 24300
+    },
+    {
+      "epoch": 2.342242046136385,
+      "eval_loss": 0.3351224660873413,
+      "eval_runtime": 40.0491,
+      "eval_samples_per_second": 249.694,
+      "eval_steps_per_second": 31.212,
+      "step": 24300
+    },
+    {
+      "epoch": 2.351880902293325,
+      "grad_norm": 1.176680564880371,
+      "learning_rate": 3.5719063545150507e-06,
+      "loss": 0.3689,
+      "step": 24400
+    },
+    {
+      "epoch": 2.351880902293325,
+      "eval_loss": 0.3293921947479248,
+      "eval_runtime": 40.2376,
+      "eval_samples_per_second": 248.524,
+      "eval_steps_per_second": 31.065,
+      "step": 24400
+    },
+    {
+      "epoch": 2.3615197584502647,
+      "grad_norm": 1.2454367876052856,
+      "learning_rate": 3.561872909698997e-06,
+      "loss": 0.3706,
+      "step": 24500
+    },
+    {
+      "epoch": 2.3615197584502647,
+      "eval_loss": 0.3308497369289398,
+      "eval_runtime": 40.0489,
+      "eval_samples_per_second": 249.695,
+      "eval_steps_per_second": 31.212,
+      "step": 24500
+    },
+    {
+      "epoch": 2.3711586146072046,
+      "grad_norm": 1.4466218948364258,
+      "learning_rate": 3.5518394648829434e-06,
+      "loss": 0.3719,
+      "step": 24600
+    },
+    {
+      "epoch": 2.3711586146072046,
+      "eval_loss": 0.32894277572631836,
+      "eval_runtime": 40.0499,
+      "eval_samples_per_second": 249.688,
+      "eval_steps_per_second": 31.211,
+      "step": 24600
+    },
+    {
+      "epoch": 2.3807974707641444,
+      "grad_norm": 1.1407536268234253,
+      "learning_rate": 3.54180602006689e-06,
+      "loss": 0.3726,
+      "step": 24700
+    },
+    {
+      "epoch": 2.3807974707641444,
+      "eval_loss": 0.3315465450286865,
+      "eval_runtime": 40.0281,
+      "eval_samples_per_second": 249.825,
+      "eval_steps_per_second": 31.228,
+      "step": 24700
+    },
+    {
+      "epoch": 2.390436326921084,
+      "grad_norm": 1.2593247890472412,
+      "learning_rate": 3.5317725752508365e-06,
+      "loss": 0.3707,
+      "step": 24800
+    },
+    {
+      "epoch": 2.390436326921084,
+      "eval_loss": 0.32788795232772827,
+      "eval_runtime": 39.9985,
+      "eval_samples_per_second": 250.009,
+      "eval_steps_per_second": 31.251,
+      "step": 24800
+    },
+    {
+      "epoch": 2.400075183078024,
+      "grad_norm": 1.183972954750061,
+      "learning_rate": 3.521739130434783e-06,
+      "loss": 0.3725,
+      "step": 24900
+    },
+    {
+      "epoch": 2.400075183078024,
+      "eval_loss": 0.3257734477519989,
+      "eval_runtime": 40.195,
+      "eval_samples_per_second": 248.787,
+      "eval_steps_per_second": 31.098,
+      "step": 24900
+    },
+    {
+      "epoch": 2.409714039234964,
+      "grad_norm": 1.3162939548492432,
+      "learning_rate": 3.5117056856187296e-06,
+      "loss": 0.3726,
+      "step": 25000
+    },
+    {
+      "epoch": 2.409714039234964,
+      "eval_loss": 0.32726946473121643,
+      "eval_runtime": 40.1077,
+      "eval_samples_per_second": 249.329,
+      "eval_steps_per_second": 31.166,
+      "step": 25000
+    },
+    {
+      "epoch": 2.4193528953919037,
+      "grad_norm": 1.2255990505218506,
+      "learning_rate": 3.501672240802676e-06,
+      "loss": 0.3723,
+      "step": 25100
+    },
+    {
+      "epoch": 2.4193528953919037,
+      "eval_loss": 0.32688045501708984,
+      "eval_runtime": 40.0031,
+      "eval_samples_per_second": 249.98,
+      "eval_steps_per_second": 31.248,
+      "step": 25100
+    },
+    {
+      "epoch": 2.4289917515488435,
+      "grad_norm": 1.2405096292495728,
+      "learning_rate": 3.491638795986622e-06,
+      "loss": 0.3701,
+      "step": 25200
+    },
+    {
+      "epoch": 2.4289917515488435,
+      "eval_loss": 0.32833394408226013,
+      "eval_runtime": 40.0179,
+      "eval_samples_per_second": 249.888,
+      "eval_steps_per_second": 31.236,
+      "step": 25200
+    },
+    {
+      "epoch": 2.4386306077057833,
+      "grad_norm": 1.1878525018692017,
+      "learning_rate": 3.481605351170568e-06,
+      "loss": 0.371,
+      "step": 25300
+    },
+    {
+      "epoch": 2.4386306077057833,
+      "eval_loss": 0.3243899941444397,
+      "eval_runtime": 40.2224,
+      "eval_samples_per_second": 248.618,
+      "eval_steps_per_second": 31.077,
+      "step": 25300
+    },
+    {
+      "epoch": 2.4482694638627236,
+      "grad_norm": 1.160326600074768,
+      "learning_rate": 3.471571906354515e-06,
+      "loss": 0.3664,
+      "step": 25400
+    },
+    {
+      "epoch": 2.4482694638627236,
+      "eval_loss": 0.3271392583847046,
+      "eval_runtime": 40.0288,
+      "eval_samples_per_second": 249.82,
+      "eval_steps_per_second": 31.228,
+      "step": 25400
+    },
+    {
+      "epoch": 2.4579083200196634,
+      "grad_norm": 1.1720784902572632,
+      "learning_rate": 3.4615384615384613e-06,
+      "loss": 0.3703,
+      "step": 25500
+    },
+    {
+      "epoch": 2.4579083200196634,
+      "eval_loss": 0.3221685588359833,
+      "eval_runtime": 40.0382,
+      "eval_samples_per_second": 249.761,
+      "eval_steps_per_second": 31.22,
+      "step": 25500
+    },
+    {
+      "epoch": 2.4675471761766032,
+      "grad_norm": 1.195279836654663,
+      "learning_rate": 3.4515050167224076e-06,
+      "loss": 0.368,
+      "step": 25600
+    },
+    {
+      "epoch": 2.4675471761766032,
+      "eval_loss": 0.3238082528114319,
+      "eval_runtime": 40.0341,
+      "eval_samples_per_second": 249.787,
+      "eval_steps_per_second": 31.223,
+      "step": 25600
+    },
+    {
+      "epoch": 2.477186032333543,
+      "grad_norm": 1.1934449672698975,
+      "learning_rate": 3.4414715719063544e-06,
+      "loss": 0.3661,
+      "step": 25700
+    },
+    {
+      "epoch": 2.477186032333543,
+      "eval_loss": 0.3191760778427124,
+      "eval_runtime": 40.2569,
+      "eval_samples_per_second": 248.404,
+      "eval_steps_per_second": 31.051,
+      "step": 25700
+    },
+    {
+      "epoch": 2.486824888490483,
+      "grad_norm": 1.185477375984192,
+      "learning_rate": 3.4314381270903007e-06,
+      "loss": 0.365,
+      "step": 25800
+    },
+    {
+      "epoch": 2.486824888490483,
+      "eval_loss": 0.32491880655288696,
+      "eval_runtime": 40.0528,
+      "eval_samples_per_second": 249.67,
+      "eval_steps_per_second": 31.209,
+      "step": 25800
+    },
+    {
+      "epoch": 2.4964637446474227,
+      "grad_norm": 1.2137432098388672,
+      "learning_rate": 3.4214046822742475e-06,
+      "loss": 0.3641,
+      "step": 25900
+    },
+    {
+      "epoch": 2.4964637446474227,
+      "eval_loss": 0.3219539523124695,
+      "eval_runtime": 40.0329,
+      "eval_samples_per_second": 249.795,
+      "eval_steps_per_second": 31.224,
+      "step": 25900
+    },
+    {
+      "epoch": 2.5061026008043625,
+      "grad_norm": 1.2091009616851807,
+      "learning_rate": 3.411371237458194e-06,
+      "loss": 0.3675,
+      "step": 26000
+    },
+    {
+      "epoch": 2.5061026008043625,
+      "eval_loss": 0.32177218794822693,
+      "eval_runtime": 40.0916,
+      "eval_samples_per_second": 249.429,
+      "eval_steps_per_second": 31.179,
+      "step": 26000
+    },
+    {
+      "epoch": 2.5157414569613024,
+      "grad_norm": 1.2672072649002075,
+      "learning_rate": 3.40133779264214e-06,
+      "loss": 0.3662,
+      "step": 26100
+    },
+    {
+      "epoch": 2.5157414569613024,
+      "eval_loss": 0.32613876461982727,
+      "eval_runtime": 40.0426,
+      "eval_samples_per_second": 249.734,
+      "eval_steps_per_second": 31.217,
+      "step": 26100
+    },
+    {
+      "epoch": 2.525380313118242,
+      "grad_norm": 1.0901987552642822,
+      "learning_rate": 3.391304347826087e-06,
+      "loss": 0.3665,
+      "step": 26200
+    },
+    {
+      "epoch": 2.525380313118242,
+      "eval_loss": 0.3274555206298828,
+      "eval_runtime": 40.2412,
+      "eval_samples_per_second": 248.501,
+      "eval_steps_per_second": 31.063,
+      "step": 26200
+    },
+    {
+      "epoch": 2.535019169275182,
+      "grad_norm": 1.1770433187484741,
+      "learning_rate": 3.3812709030100333e-06,
+      "loss": 0.3643,
+      "step": 26300
+    },
+    {
+      "epoch": 2.535019169275182,
+      "eval_loss": 0.32851478457450867,
+      "eval_runtime": 40.0516,
+      "eval_samples_per_second": 249.678,
+      "eval_steps_per_second": 31.21,
+      "step": 26300
+    },
+    {
+      "epoch": 2.544658025432122,
+      "grad_norm": 1.2189823389053345,
+      "learning_rate": 3.3712374581939796e-06,
+      "loss": 0.3603,
+      "step": 26400
+    },
+    {
+      "epoch": 2.544658025432122,
+      "eval_loss": 0.3176610767841339,
+      "eval_runtime": 40.0452,
+      "eval_samples_per_second": 249.718,
+      "eval_steps_per_second": 31.215,
+      "step": 26400
+    },
+    {
+      "epoch": 2.554296881589062,
+      "grad_norm": 1.1684179306030273,
+      "learning_rate": 3.3612040133779264e-06,
+      "loss": 0.3639,
+      "step": 26500
+    },
+    {
+      "epoch": 2.554296881589062,
+      "eval_loss": 0.3156176805496216,
+      "eval_runtime": 40.0435,
+      "eval_samples_per_second": 249.728,
+      "eval_steps_per_second": 31.216,
+      "step": 26500
+    },
+    {
+      "epoch": 2.563935737746002,
+      "grad_norm": 1.2573341131210327,
+      "learning_rate": 3.3511705685618727e-06,
+      "loss": 0.3586,
+      "step": 26600
+    },
+    {
+      "epoch": 2.563935737746002,
+      "eval_loss": 0.32222485542297363,
+      "eval_runtime": 40.199,
+      "eval_samples_per_second": 248.762,
+      "eval_steps_per_second": 31.095,
+      "step": 26600
+    },
+    {
+      "epoch": 2.5735745939029417,
+      "grad_norm": 1.2410520315170288,
+      "learning_rate": 3.3411371237458195e-06,
+      "loss": 0.3626,
+      "step": 26700
+    },
+    {
+      "epoch": 2.5735745939029417,
+      "eval_loss": 0.31266114115715027,
+      "eval_runtime": 40.0104,
+      "eval_samples_per_second": 249.935,
+      "eval_steps_per_second": 31.242,
+      "step": 26700
+    },
+    {
+      "epoch": 2.5832134500598816,
+      "grad_norm": 1.2394918203353882,
+      "learning_rate": 3.331103678929766e-06,
+      "loss": 0.3606,
+      "step": 26800
+    },
+    {
+      "epoch": 2.5832134500598816,
+      "eval_loss": 0.32011231780052185,
+      "eval_runtime": 40.1084,
+      "eval_samples_per_second": 249.324,
+      "eval_steps_per_second": 31.166,
+      "step": 26800
+    },
+    {
+      "epoch": 2.5928523062168214,
+      "grad_norm": 1.2512286901474,
+      "learning_rate": 3.321070234113712e-06,
+      "loss": 0.3614,
+      "step": 26900
+    },
+    {
+      "epoch": 2.5928523062168214,
+      "eval_loss": 0.31560882925987244,
+      "eval_runtime": 40.027,
+      "eval_samples_per_second": 249.831,
+      "eval_steps_per_second": 31.229,
+      "step": 26900
+    },
+    {
+      "epoch": 2.602491162373761,
+      "grad_norm": 1.3661274909973145,
+      "learning_rate": 3.311036789297659e-06,
+      "loss": 0.3601,
+      "step": 27000
+    },
+    {
+      "epoch": 2.602491162373761,
+      "eval_loss": 0.3139115869998932,
+      "eval_runtime": 40.2328,
+      "eval_samples_per_second": 248.554,
+      "eval_steps_per_second": 31.069,
+      "step": 27000
+    },
+    {
+      "epoch": 2.612130018530701,
+      "grad_norm": 1.293144702911377,
+      "learning_rate": 3.3010033444816052e-06,
+      "loss": 0.3602,
+      "step": 27100
+    },
+    {
+      "epoch": 2.612130018530701,
+      "eval_loss": 0.31472310423851013,
+      "eval_runtime": 40.0369,
+      "eval_samples_per_second": 249.77,
+      "eval_steps_per_second": 31.221,
+      "step": 27100
+    },
+    {
+      "epoch": 2.621768874687641,
+      "grad_norm": 1.1856138706207275,
+      "learning_rate": 3.2909698996655516e-06,
+      "loss": 0.3593,
+      "step": 27200
+    },
+    {
+      "epoch": 2.621768874687641,
+      "eval_loss": 0.3115543723106384,
+      "eval_runtime": 40.2262,
+      "eval_samples_per_second": 248.594,
+      "eval_steps_per_second": 31.074,
+      "step": 27200
+    },
+    {
+      "epoch": 2.6314077308445807,
+      "grad_norm": 1.2642232179641724,
+      "learning_rate": 3.2809364548494983e-06,
+      "loss": 0.3577,
+      "step": 27300
+    },
+    {
+      "epoch": 2.6314077308445807,
+      "eval_loss": 0.31309688091278076,
+      "eval_runtime": 40.0885,
+      "eval_samples_per_second": 249.448,
+      "eval_steps_per_second": 31.181,
+      "step": 27300
+    },
+    {
+      "epoch": 2.6410465870015205,
+      "grad_norm": 1.2292590141296387,
+      "learning_rate": 3.2709030100334447e-06,
+      "loss": 0.357,
+      "step": 27400
+    },
+    {
+      "epoch": 2.6410465870015205,
+      "eval_loss": 0.3183058202266693,
+      "eval_runtime": 40.0953,
+      "eval_samples_per_second": 249.406,
+      "eval_steps_per_second": 31.176,
+      "step": 27400
+    },
+    {
+      "epoch": 2.6506854431584603,
+      "grad_norm": 1.255295753479004,
+      "learning_rate": 3.260869565217391e-06,
+      "loss": 0.3552,
+      "step": 27500
+    },
+    {
+      "epoch": 2.6506854431584603,
+      "eval_loss": 0.3173259198665619,
+      "eval_runtime": 40.1197,
+      "eval_samples_per_second": 249.254,
+      "eval_steps_per_second": 31.157,
+      "step": 27500
+    },
+    {
+      "epoch": 2.6603242993154,
+      "grad_norm": 1.2543479204177856,
+      "learning_rate": 3.2508361204013378e-06,
+      "loss": 0.3562,
+      "step": 27600
+    },
+    {
+      "epoch": 2.6603242993154,
+      "eval_loss": 0.31650179624557495,
+      "eval_runtime": 40.2725,
+      "eval_samples_per_second": 248.308,
+      "eval_steps_per_second": 31.039,
+      "step": 27600
+    },
+    {
+      "epoch": 2.66996315547234,
+      "grad_norm": 1.1937369108200073,
+      "learning_rate": 3.240802675585284e-06,
+      "loss": 0.3548,
+      "step": 27700
+    },
+    {
+      "epoch": 2.66996315547234,
+      "eval_loss": 0.3191297948360443,
+      "eval_runtime": 40.0849,
+      "eval_samples_per_second": 249.47,
+      "eval_steps_per_second": 31.184,
+      "step": 27700
+    },
+    {
+      "epoch": 2.67960201162928,
+      "grad_norm": 1.1102817058563232,
+      "learning_rate": 3.230769230769231e-06,
+      "loss": 0.3558,
+      "step": 27800
+    },
+    {
+      "epoch": 2.67960201162928,
+      "eval_loss": 0.31787580251693726,
+      "eval_runtime": 40.045,
+      "eval_samples_per_second": 249.719,
+      "eval_steps_per_second": 31.215,
+      "step": 27800
+    },
+    {
+      "epoch": 2.6892408677862196,
+      "grad_norm": 1.1875417232513428,
+      "learning_rate": 3.2207357859531772e-06,
+      "loss": 0.3561,
+      "step": 27900
+    },
+    {
+      "epoch": 2.6892408677862196,
+      "eval_loss": 0.3153333067893982,
+      "eval_runtime": 40.0433,
+      "eval_samples_per_second": 249.73,
+      "eval_steps_per_second": 31.216,
+      "step": 27900
+    },
+    {
+      "epoch": 2.6988797239431594,
+      "grad_norm": 1.203171968460083,
+      "learning_rate": 3.2107023411371236e-06,
+      "loss": 0.3557,
+      "step": 28000
+    },
+    {
+      "epoch": 2.6988797239431594,
+      "eval_loss": 0.3114122450351715,
+      "eval_runtime": 40.207,
+      "eval_samples_per_second": 248.713,
+      "eval_steps_per_second": 31.089,
+      "step": 28000
+    },
+    {
+      "epoch": 2.7085185801000993,
+      "grad_norm": 1.1906991004943848,
+      "learning_rate": 3.2006688963210703e-06,
+      "loss": 0.3507,
+      "step": 28100
+    },
+    {
+      "epoch": 2.7085185801000993,
+      "eval_loss": 0.31228846311569214,
+      "eval_runtime": 40.2407,
+      "eval_samples_per_second": 248.504,
+      "eval_steps_per_second": 31.063,
+      "step": 28100
+    },
+    {
+      "epoch": 2.7181574362570395,
+      "grad_norm": 1.0473979711532593,
+      "learning_rate": 3.1906354515050167e-06,
+      "loss": 0.3547,
+      "step": 28200
+    },
+    {
+      "epoch": 2.7181574362570395,
+      "eval_loss": 0.31403276324272156,
+      "eval_runtime": 40.052,
+      "eval_samples_per_second": 249.675,
+      "eval_steps_per_second": 31.209,
+      "step": 28200
+    },
+    {
+      "epoch": 2.7277962924139794,
+      "grad_norm": 1.3040709495544434,
+      "learning_rate": 3.180602006688963e-06,
+      "loss": 0.3549,
+      "step": 28300
+    },
+    {
+      "epoch": 2.7277962924139794,
+      "eval_loss": 0.31078001856803894,
+      "eval_runtime": 40.0565,
+      "eval_samples_per_second": 249.648,
+      "eval_steps_per_second": 31.206,
+      "step": 28300
+    },
+    {
+      "epoch": 2.737435148570919,
+      "grad_norm": 1.1051684617996216,
+      "learning_rate": 3.1705685618729098e-06,
+      "loss": 0.3527,
+      "step": 28400
+    },
+    {
+      "epoch": 2.737435148570919,
+      "eval_loss": 0.311916321516037,
+      "eval_runtime": 40.0592,
+      "eval_samples_per_second": 249.63,
+      "eval_steps_per_second": 31.204,
+      "step": 28400
+    },
+    {
+      "epoch": 2.747074004727859,
+      "grad_norm": 1.2204347848892212,
+      "learning_rate": 3.160535117056856e-06,
+      "loss": 0.3529,
+      "step": 28500
+    },
+    {
+      "epoch": 2.747074004727859,
+      "eval_loss": 0.31446966528892517,
+      "eval_runtime": 40.0552,
+      "eval_samples_per_second": 249.655,
+      "eval_steps_per_second": 31.207,
+      "step": 28500
+    },
+    {
+      "epoch": 2.756712860884799,
+      "grad_norm": 1.1805849075317383,
+      "learning_rate": 3.1505016722408024e-06,
+      "loss": 0.3514,
+      "step": 28600
+    },
+    {
+      "epoch": 2.756712860884799,
+      "eval_loss": 0.305553138256073,
+      "eval_runtime": 40.2179,
+      "eval_samples_per_second": 248.645,
+      "eval_steps_per_second": 31.081,
+      "step": 28600
+    },
+    {
+      "epoch": 2.7663517170417387,
+      "grad_norm": 1.2930576801300049,
+      "learning_rate": 3.140468227424749e-06,
+      "loss": 0.3508,
+      "step": 28700
+    },
+    {
+      "epoch": 2.7663517170417387,
+      "eval_loss": 0.31795772910118103,
+      "eval_runtime": 40.0252,
+      "eval_samples_per_second": 249.842,
+      "eval_steps_per_second": 31.23,
+      "step": 28700
+    },
+    {
+      "epoch": 2.7759905731986785,
+      "grad_norm": 1.2266051769256592,
+      "learning_rate": 3.1304347826086955e-06,
+      "loss": 0.3513,
+      "step": 28800
+    },
+    {
+      "epoch": 2.7759905731986785,
+      "eval_loss": 0.31347090005874634,
+      "eval_runtime": 40.0594,
+      "eval_samples_per_second": 249.629,
+      "eval_steps_per_second": 31.204,
+      "step": 28800
+    },
+    {
+      "epoch": 2.7856294293556183,
+      "grad_norm": 1.1644967794418335,
+      "learning_rate": 3.1204013377926423e-06,
+      "loss": 0.3477,
+      "step": 28900
+    },
+    {
+      "epoch": 2.7856294293556183,
+      "eval_loss": 0.3088667392730713,
+      "eval_runtime": 40.0626,
+      "eval_samples_per_second": 249.609,
+      "eval_steps_per_second": 31.201,
+      "step": 28900
+    },
+    {
+      "epoch": 2.795268285512558,
+      "grad_norm": 1.1567316055297852,
+      "learning_rate": 3.1103678929765886e-06,
+      "loss": 0.3512,
+      "step": 29000
+    },
+    {
+      "epoch": 2.795268285512558,
+      "eval_loss": 0.30603039264678955,
+      "eval_runtime": 40.0801,
+      "eval_samples_per_second": 249.5,
+      "eval_steps_per_second": 31.188,
+      "step": 29000
+    },
+    {
+      "epoch": 2.804907141669498,
+      "grad_norm": 1.324166178703308,
+      "learning_rate": 3.100334448160535e-06,
+      "loss": 0.3508,
+      "step": 29100
+    },
+    {
+      "epoch": 2.804907141669498,
+      "eval_loss": 0.3052707612514496,
+      "eval_runtime": 40.1749,
+      "eval_samples_per_second": 248.912,
+      "eval_steps_per_second": 31.114,
+      "step": 29100
+    },
+    {
+      "epoch": 2.8145459978264378,
+      "grad_norm": 1.2231703996658325,
+      "learning_rate": 3.0903010033444818e-06,
+      "loss": 0.3491,
+      "step": 29200
+    },
+    {
+      "epoch": 2.8145459978264378,
+      "eval_loss": 0.3067169487476349,
+      "eval_runtime": 40.0129,
+      "eval_samples_per_second": 249.92,
+      "eval_steps_per_second": 31.24,
+      "step": 29200
+    },
+    {
+      "epoch": 2.824184853983378,
+      "grad_norm": 1.203022837638855,
+      "learning_rate": 3.080267558528428e-06,
+      "loss": 0.3476,
+      "step": 29300
+    },
+    {
+      "epoch": 2.824184853983378,
+      "eval_loss": 0.31380364298820496,
+      "eval_runtime": 39.9974,
+      "eval_samples_per_second": 250.016,
+      "eval_steps_per_second": 31.252,
+      "step": 29300
+    },
+    {
+      "epoch": 2.833823710140318,
+      "grad_norm": 1.1473413705825806,
+      "learning_rate": 3.0702341137123744e-06,
+      "loss": 0.3497,
+      "step": 29400
+    },
+    {
+      "epoch": 2.833823710140318,
+      "eval_loss": 0.30883774161338806,
+      "eval_runtime": 40.0159,
+      "eval_samples_per_second": 249.901,
+      "eval_steps_per_second": 31.238,
+      "step": 29400
+    },
+    {
+      "epoch": 2.8434625662972577,
+      "grad_norm": 1.0918190479278564,
+      "learning_rate": 3.060200668896321e-06,
+      "loss": 0.3505,
+      "step": 29500
+    },
+    {
+      "epoch": 2.8434625662972577,
+      "eval_loss": 0.3049847185611725,
+      "eval_runtime": 40.2816,
+      "eval_samples_per_second": 248.252,
+      "eval_steps_per_second": 31.032,
+      "step": 29500
+    },
+    {
+      "epoch": 2.8531014224541975,
+      "grad_norm": 1.121775507926941,
+      "learning_rate": 3.0501672240802675e-06,
+      "loss": 0.348,
+      "step": 29600
+    },
+    {
+      "epoch": 2.8531014224541975,
+      "eval_loss": 0.304568886756897,
+      "eval_runtime": 40.1199,
+      "eval_samples_per_second": 249.253,
+      "eval_steps_per_second": 31.157,
+      "step": 29600
+    },
+    {
+      "epoch": 2.8627402786111373,
+      "grad_norm": 1.1627501249313354,
+      "learning_rate": 3.0401337792642143e-06,
+      "loss": 0.3466,
+      "step": 29700
+    },
+    {
+      "epoch": 2.8627402786111373,
+      "eval_loss": 0.3042547404766083,
+      "eval_runtime": 40.2234,
+      "eval_samples_per_second": 248.611,
+      "eval_steps_per_second": 31.076,
+      "step": 29700
+    },
+    {
+      "epoch": 2.872379134768077,
+      "grad_norm": 1.2211036682128906,
+      "learning_rate": 3.0301003344481606e-06,
+      "loss": 0.3481,
+      "step": 29800
+    },
+    {
+      "epoch": 2.872379134768077,
+      "eval_loss": 0.3090800642967224,
+      "eval_runtime": 40.0403,
+      "eval_samples_per_second": 249.748,
+      "eval_steps_per_second": 31.219,
+      "step": 29800
+    },
+    {
+      "epoch": 2.882017990925017,
+      "grad_norm": 1.2113616466522217,
+      "learning_rate": 3.020066889632107e-06,
+      "loss": 0.3473,
+      "step": 29900
+    },
+    {
+      "epoch": 2.882017990925017,
+      "eval_loss": 0.30445191264152527,
+      "eval_runtime": 40.0405,
+      "eval_samples_per_second": 249.747,
+      "eval_steps_per_second": 31.218,
+      "step": 29900
+    },
+    {
+      "epoch": 2.891656847081957,
+      "grad_norm": 1.2371474504470825,
+      "learning_rate": 3.0100334448160537e-06,
+      "loss": 0.3465,
+      "step": 30000
+    },
+    {
+      "epoch": 2.891656847081957,
+      "eval_loss": 0.30408233404159546,
+      "eval_runtime": 40.0474,
+      "eval_samples_per_second": 249.704,
+      "eval_steps_per_second": 31.213,
+      "step": 30000
+    },
+    {
+      "epoch": 2.9012957032388966,
+      "grad_norm": 1.1639801263809204,
+      "learning_rate": 3e-06,
+      "loss": 0.3454,
+      "step": 30100
+    },
+    {
+      "epoch": 2.9012957032388966,
+      "eval_loss": 0.3157171905040741,
+      "eval_runtime": 40.2277,
+      "eval_samples_per_second": 248.585,
+      "eval_steps_per_second": 31.073,
+      "step": 30100
+    },
+    {
+      "epoch": 2.9109345593958365,
+      "grad_norm": 1.1769859790802002,
+      "learning_rate": 2.9899665551839464e-06,
+      "loss": 0.344,
+      "step": 30200
+    },
+    {
+      "epoch": 2.9109345593958365,
+      "eval_loss": 0.3092867136001587,
+      "eval_runtime": 40.029,
+      "eval_samples_per_second": 249.819,
+      "eval_steps_per_second": 31.227,
+      "step": 30200
+    },
+    {
+      "epoch": 2.9205734155527763,
+      "grad_norm": 1.1697938442230225,
+      "learning_rate": 2.979933110367893e-06,
+      "loss": 0.3455,
+      "step": 30300
+    },
+    {
+      "epoch": 2.9205734155527763,
+      "eval_loss": 0.30681154131889343,
+      "eval_runtime": 40.0152,
+      "eval_samples_per_second": 249.905,
+      "eval_steps_per_second": 31.238,
+      "step": 30300
+    },
+    {
+      "epoch": 2.930212271709716,
+      "grad_norm": 1.064261555671692,
+      "learning_rate": 2.9698996655518395e-06,
+      "loss": 0.3436,
+      "step": 30400
+    },
+    {
+      "epoch": 2.930212271709716,
+      "eval_loss": 0.30829107761383057,
+      "eval_runtime": 40.1179,
+      "eval_samples_per_second": 249.265,
+      "eval_steps_per_second": 31.158,
+      "step": 30400
+    },
+    {
+      "epoch": 2.939851127866656,
+      "grad_norm": 1.2219582796096802,
+      "learning_rate": 2.959866220735786e-06,
+      "loss": 0.3432,
+      "step": 30500
+    },
+    {
+      "epoch": 2.939851127866656,
+      "eval_loss": 0.30284854769706726,
+      "eval_runtime": 40.2058,
+      "eval_samples_per_second": 248.72,
+      "eval_steps_per_second": 31.09,
+      "step": 30500
+    },
+    {
+      "epoch": 2.9494899840235957,
+      "grad_norm": 1.1802637577056885,
+      "learning_rate": 2.9498327759197326e-06,
+      "loss": 0.3479,
+      "step": 30600
+    },
+    {
+      "epoch": 2.9494899840235957,
+      "eval_loss": 0.30530956387519836,
+      "eval_runtime": 40.0579,
+      "eval_samples_per_second": 249.638,
+      "eval_steps_per_second": 31.205,
+      "step": 30600
+    },
+    {
+      "epoch": 2.9591288401805356,
+      "grad_norm": 1.1790562868118286,
+      "learning_rate": 2.939799331103679e-06,
+      "loss": 0.3443,
+      "step": 30700
+    },
+    {
+      "epoch": 2.9591288401805356,
+      "eval_loss": 0.3033536672592163,
+      "eval_runtime": 40.0211,
+      "eval_samples_per_second": 249.868,
+      "eval_steps_per_second": 31.234,
+      "step": 30700
+    },
+    {
+      "epoch": 2.9687676963374754,
+      "grad_norm": 1.3000251054763794,
+      "learning_rate": 2.9297658862876257e-06,
+      "loss": 0.3445,
+      "step": 30800
+    },
+    {
+      "epoch": 2.9687676963374754,
+      "eval_loss": 0.30252954363822937,
+      "eval_runtime": 40.0123,
+      "eval_samples_per_second": 249.923,
+      "eval_steps_per_second": 31.24,
+      "step": 30800
+    },
+    {
+      "epoch": 2.978406552494415,
+      "grad_norm": 1.2018824815750122,
+      "learning_rate": 2.919732441471572e-06,
+      "loss": 0.3419,
+      "step": 30900
+    },
+    {
+      "epoch": 2.978406552494415,
+      "eval_loss": 0.30083614587783813,
+      "eval_runtime": 40.0481,
+      "eval_samples_per_second": 249.7,
+      "eval_steps_per_second": 31.212,
+      "step": 30900
+    },
+    {
+      "epoch": 2.9880454086513555,
+      "grad_norm": 1.2023593187332153,
+      "learning_rate": 2.9096989966555184e-06,
+      "loss": 0.3434,
+      "step": 31000
+    },
+    {
+      "epoch": 2.9880454086513555,
+      "eval_loss": 0.30430230498313904,
+      "eval_runtime": 40.2329,
+      "eval_samples_per_second": 248.553,
+      "eval_steps_per_second": 31.069,
+      "step": 31000
+    },
+    {
+      "epoch": 2.9976842648082953,
+      "grad_norm": 1.164267659187317,
+      "learning_rate": 2.899665551839465e-06,
+      "loss": 0.3416,
+      "step": 31100
+    },
+    {
+      "epoch": 2.9976842648082953,
+      "eval_loss": 0.301971435546875,
+      "eval_runtime": 40.0594,
+      "eval_samples_per_second": 249.629,
+      "eval_steps_per_second": 31.204,
+      "step": 31100
+    },
+    {
+      "epoch": 3.007323120965235,
+      "grad_norm": 1.1940221786499023,
+      "learning_rate": 2.8896321070234115e-06,
+      "loss": 0.3403,
+      "step": 31200
+    },
+    {
+      "epoch": 3.007323120965235,
+      "eval_loss": 0.2989708483219147,
+      "eval_runtime": 40.053,
+      "eval_samples_per_second": 249.669,
+      "eval_steps_per_second": 31.209,
+      "step": 31200
+    },
+    {
+      "epoch": 3.016961977122175,
+      "grad_norm": 1.1385316848754883,
+      "learning_rate": 2.879598662207358e-06,
+      "loss": 0.3412,
+      "step": 31300
+    },
+    {
+      "epoch": 3.016961977122175,
+      "eval_loss": 0.3076593279838562,
+      "eval_runtime": 40.0512,
+      "eval_samples_per_second": 249.68,
+      "eval_steps_per_second": 31.21,
+      "step": 31300
+    },
+    {
+      "epoch": 3.0266008332791148,
+      "grad_norm": 1.1974833011627197,
+      "learning_rate": 2.8695652173913046e-06,
+      "loss": 0.3404,
+      "step": 31400
+    },
+    {
+      "epoch": 3.0266008332791148,
+      "eval_loss": 0.29814204573631287,
+      "eval_runtime": 40.2335,
+      "eval_samples_per_second": 248.549,
+      "eval_steps_per_second": 31.069,
+      "step": 31400
+    },
+    {
+      "epoch": 3.0362396894360546,
+      "grad_norm": 1.170362114906311,
+      "learning_rate": 2.859531772575251e-06,
+      "loss": 0.3393,
+      "step": 31500
+    },
+    {
+      "epoch": 3.0362396894360546,
+      "eval_loss": 0.29789820313453674,
+      "eval_runtime": 40.0606,
+      "eval_samples_per_second": 249.622,
+      "eval_steps_per_second": 31.203,
+      "step": 31500
+    },
+    {
+      "epoch": 3.0458785455929944,
+      "grad_norm": 1.2427632808685303,
+      "learning_rate": 2.8494983277591977e-06,
+      "loss": 0.3425,
+      "step": 31600
+    },
+    {
+      "epoch": 3.0458785455929944,
+      "eval_loss": 0.30032023787498474,
+      "eval_runtime": 40.0293,
+      "eval_samples_per_second": 249.817,
+      "eval_steps_per_second": 31.227,
+      "step": 31600
+    },
+    {
+      "epoch": 3.0555174017499342,
+      "grad_norm": 1.2110202312469482,
+      "learning_rate": 2.839464882943144e-06,
+      "loss": 0.3411,
+      "step": 31700
+    },
+    {
+      "epoch": 3.0555174017499342,
+      "eval_loss": 0.3062838912010193,
+      "eval_runtime": 40.0201,
+      "eval_samples_per_second": 249.875,
+      "eval_steps_per_second": 31.234,
+      "step": 31700
+    },
+    {
+      "epoch": 3.065156257906874,
+      "grad_norm": 1.2322800159454346,
+      "learning_rate": 2.8294314381270904e-06,
+      "loss": 0.3378,
+      "step": 31800
+    },
+    {
+      "epoch": 3.065156257906874,
+      "eval_loss": 0.2991129755973816,
+      "eval_runtime": 40.0041,
+      "eval_samples_per_second": 249.974,
+      "eval_steps_per_second": 31.247,
+      "step": 31800
+    },
+    {
+      "epoch": 3.074795114063814,
+      "grad_norm": 1.1711872816085815,
+      "learning_rate": 2.819397993311037e-06,
+      "loss": 0.3386,
+      "step": 31900
+    },
+    {
+      "epoch": 3.074795114063814,
+      "eval_loss": 0.2999269962310791,
+      "eval_runtime": 40.2184,
+      "eval_samples_per_second": 248.642,
+      "eval_steps_per_second": 31.08,
+      "step": 31900
+    },
+    {
+      "epoch": 3.0844339702207537,
+      "grad_norm": 1.1980561017990112,
+      "learning_rate": 2.8093645484949835e-06,
+      "loss": 0.3391,
+      "step": 32000
+    },
+    {
+      "epoch": 3.0844339702207537,
+      "eval_loss": 0.3014710545539856,
+      "eval_runtime": 40.0414,
+      "eval_samples_per_second": 249.741,
+      "eval_steps_per_second": 31.218,
+      "step": 32000
+    },
+    {
+      "epoch": 3.094072826377694,
+      "grad_norm": 1.2509229183197021,
+      "learning_rate": 2.79933110367893e-06,
+      "loss": 0.3357,
+      "step": 32100
+    },
+    {
+      "epoch": 3.094072826377694,
+      "eval_loss": 0.29611119627952576,
+      "eval_runtime": 40.0618,
+      "eval_samples_per_second": 249.614,
+      "eval_steps_per_second": 31.202,
+      "step": 32100
+    },
+    {
+      "epoch": 3.103711682534634,
+      "grad_norm": 1.1109113693237305,
+      "learning_rate": 2.7892976588628766e-06,
+      "loss": 0.3377,
+      "step": 32200
+    },
+    {
+      "epoch": 3.103711682534634,
+      "eval_loss": 0.2999332547187805,
+      "eval_runtime": 40.0641,
+      "eval_samples_per_second": 249.6,
+      "eval_steps_per_second": 31.2,
+      "step": 32200
+    },
+    {
+      "epoch": 3.1133505386915736,
+      "grad_norm": 1.1124744415283203,
+      "learning_rate": 2.779264214046823e-06,
+      "loss": 0.3379,
+      "step": 32300
+    },
+    {
+      "epoch": 3.1133505386915736,
+      "eval_loss": 0.29913681745529175,
+      "eval_runtime": 40.0747,
+      "eval_samples_per_second": 249.534,
+      "eval_steps_per_second": 31.192,
+      "step": 32300
+    },
+    {
+      "epoch": 3.1229893948485135,
+      "grad_norm": 1.1505558490753174,
+      "learning_rate": 2.7692307692307693e-06,
+      "loss": 0.3377,
+      "step": 32400
+    },
+    {
+      "epoch": 3.1229893948485135,
+      "eval_loss": 0.29324981570243835,
+      "eval_runtime": 40.2233,
+      "eval_samples_per_second": 248.612,
+      "eval_steps_per_second": 31.077,
+      "step": 32400
+    },
+    {
+      "epoch": 3.1326282510054533,
+      "grad_norm": 1.2245198488235474,
+      "learning_rate": 2.759197324414716e-06,
+      "loss": 0.3364,
+      "step": 32500
+    },
+    {
+      "epoch": 3.1326282510054533,
+      "eval_loss": 0.2940920293331146,
+      "eval_runtime": 40.0385,
+      "eval_samples_per_second": 249.76,
+      "eval_steps_per_second": 31.22,
+      "step": 32500
+    },
+    {
+      "epoch": 3.142267107162393,
+      "grad_norm": 1.1837635040283203,
+      "learning_rate": 2.749163879598662e-06,
+      "loss": 0.3411,
+      "step": 32600
+    },
+    {
+      "epoch": 3.142267107162393,
+      "eval_loss": 0.29967695474624634,
+      "eval_runtime": 40.0599,
+      "eval_samples_per_second": 249.626,
+      "eval_steps_per_second": 31.203,
+      "step": 32600
+    },
+    {
+      "epoch": 3.151905963319333,
+      "grad_norm": 1.1685434579849243,
+      "learning_rate": 2.7391304347826087e-06,
+      "loss": 0.3395,
+      "step": 32700
+    },
+    {
+      "epoch": 3.151905963319333,
+      "eval_loss": 0.2951061427593231,
+      "eval_runtime": 40.0528,
+      "eval_samples_per_second": 249.67,
+      "eval_steps_per_second": 31.209,
+      "step": 32700
+    },
+    {
+      "epoch": 3.1615448194762727,
+      "grad_norm": 1.0835115909576416,
+      "learning_rate": 2.729096989966555e-06,
+      "loss": 0.3377,
+      "step": 32800
+    },
+    {
+      "epoch": 3.1615448194762727,
+      "eval_loss": 0.29684531688690186,
+      "eval_runtime": 40.2629,
+      "eval_samples_per_second": 248.367,
+      "eval_steps_per_second": 31.046,
+      "step": 32800
+    },
+    {
+      "epoch": 3.1711836756332126,
+      "grad_norm": 1.2234493494033813,
+      "learning_rate": 2.7190635451505014e-06,
+      "loss": 0.3361,
+      "step": 32900
+    },
+    {
+      "epoch": 3.1711836756332126,
+      "eval_loss": 0.29388144612312317,
+      "eval_runtime": 40.0533,
+      "eval_samples_per_second": 249.667,
+      "eval_steps_per_second": 31.208,
+      "step": 32900
+    },
+    {
+      "epoch": 3.1808225317901524,
+      "grad_norm": 1.1904512643814087,
+      "learning_rate": 2.709030100334448e-06,
+      "loss": 0.3339,
+      "step": 33000
+    },
+    {
+      "epoch": 3.1808225317901524,
+      "eval_loss": 0.2957378327846527,
+      "eval_runtime": 40.0266,
+      "eval_samples_per_second": 249.834,
+      "eval_steps_per_second": 31.229,
+      "step": 33000
+    },
+    {
+      "epoch": 3.190461387947092,
+      "grad_norm": 1.1379910707473755,
+      "learning_rate": 2.6989966555183945e-06,
+      "loss": 0.334,
+      "step": 33100
+    },
+    {
+      "epoch": 3.190461387947092,
+      "eval_loss": 0.2910006046295166,
+      "eval_runtime": 40.0239,
+      "eval_samples_per_second": 249.851,
+      "eval_steps_per_second": 31.231,
+      "step": 33100
+    },
+    {
+      "epoch": 3.200100244104032,
+      "grad_norm": 1.106467366218567,
+      "learning_rate": 2.6889632107023413e-06,
+      "loss": 0.3348,
+      "step": 33200
+    },
+    {
+      "epoch": 3.200100244104032,
+      "eval_loss": 0.29840487241744995,
+      "eval_runtime": 40.0152,
+      "eval_samples_per_second": 249.905,
+      "eval_steps_per_second": 31.238,
+      "step": 33200
+    },
+    {
+      "epoch": 3.209739100260972,
+      "grad_norm": 1.0919976234436035,
+      "learning_rate": 2.6789297658862876e-06,
+      "loss": 0.3342,
+      "step": 33300
+    },
+    {
+      "epoch": 3.209739100260972,
+      "eval_loss": 0.2929946184158325,
+      "eval_runtime": 40.2471,
+      "eval_samples_per_second": 248.465,
+      "eval_steps_per_second": 31.058,
+      "step": 33300
+    },
+    {
+      "epoch": 3.2193779564179117,
+      "grad_norm": 1.1051640510559082,
+      "learning_rate": 2.668896321070234e-06,
+      "loss": 0.3356,
+      "step": 33400
+    },
+    {
+      "epoch": 3.2193779564179117,
+      "eval_loss": 0.29255038499832153,
+      "eval_runtime": 40.0458,
+      "eval_samples_per_second": 249.714,
+      "eval_steps_per_second": 31.214,
+      "step": 33400
+    },
+    {
+      "epoch": 3.2290168125748515,
+      "grad_norm": 1.156020164489746,
+      "learning_rate": 2.6588628762541807e-06,
+      "loss": 0.3326,
+      "step": 33500
+    },
+    {
+      "epoch": 3.2290168125748515,
+      "eval_loss": 0.29639294743537903,
+      "eval_runtime": 40.0474,
+      "eval_samples_per_second": 249.704,
+      "eval_steps_per_second": 31.213,
+      "step": 33500
+    },
+    {
+      "epoch": 3.238655668731792,
+      "grad_norm": 1.1010196208953857,
+      "learning_rate": 2.648829431438127e-06,
+      "loss": 0.3321,
+      "step": 33600
+    },
+    {
+      "epoch": 3.238655668731792,
+      "eval_loss": 0.28783321380615234,
+      "eval_runtime": 40.0521,
+      "eval_samples_per_second": 249.675,
+      "eval_steps_per_second": 31.209,
+      "step": 33600
+    },
+    {
+      "epoch": 3.2482945248887316,
+      "grad_norm": 1.0766925811767578,
+      "learning_rate": 2.6387959866220734e-06,
+      "loss": 0.3346,
+      "step": 33700
+    },
+    {
+      "epoch": 3.2482945248887316,
+      "eval_loss": 0.2899346351623535,
+      "eval_runtime": 40.0356,
+      "eval_samples_per_second": 249.778,
+      "eval_steps_per_second": 31.222,
+      "step": 33700
+    },
+    {
+      "epoch": 3.2579333810456714,
+      "grad_norm": 1.1832084655761719,
+      "learning_rate": 2.62876254180602e-06,
+      "loss": 0.3323,
+      "step": 33800
+    },
+    {
+      "epoch": 3.2579333810456714,
+      "eval_loss": 0.29377686977386475,
+      "eval_runtime": 40.1913,
+      "eval_samples_per_second": 248.81,
+      "eval_steps_per_second": 31.101,
+      "step": 33800
+    },
+    {
+      "epoch": 3.2675722372026113,
+      "grad_norm": 1.1712465286254883,
+      "learning_rate": 2.6187290969899665e-06,
+      "loss": 0.3334,
+      "step": 33900
+    },
+    {
+      "epoch": 3.2675722372026113,
+      "eval_loss": 0.29734405875205994,
+      "eval_runtime": 40.0053,
+      "eval_samples_per_second": 249.967,
+      "eval_steps_per_second": 31.246,
+      "step": 33900
+    },
+    {
+      "epoch": 3.277211093359551,
+      "grad_norm": 1.2784638404846191,
+      "learning_rate": 2.6086956521739132e-06,
+      "loss": 0.3318,
+      "step": 34000
+    },
+    {
+      "epoch": 3.277211093359551,
+      "eval_loss": 0.29785603284835815,
+      "eval_runtime": 40.0215,
+      "eval_samples_per_second": 249.866,
+      "eval_steps_per_second": 31.233,
+      "step": 34000
+    },
+    {
+      "epoch": 3.286849949516491,
+      "grad_norm": 1.2473056316375732,
+      "learning_rate": 2.5986622073578596e-06,
+      "loss": 0.3315,
+      "step": 34100
+    },
+    {
+      "epoch": 3.286849949516491,
+      "eval_loss": 0.30169832706451416,
+      "eval_runtime": 40.0368,
+      "eval_samples_per_second": 249.77,
+      "eval_steps_per_second": 31.221,
+      "step": 34100
+    },
+    {
+      "epoch": 3.2964888056734307,
+      "grad_norm": 1.1944090127944946,
+      "learning_rate": 2.588628762541806e-06,
+      "loss": 0.3319,
+      "step": 34200
+    },
+    {
+      "epoch": 3.2964888056734307,
+      "eval_loss": 0.2977898120880127,
+      "eval_runtime": 40.0376,
+      "eval_samples_per_second": 249.765,
+      "eval_steps_per_second": 31.221,
+      "step": 34200
+    },
+    {
+      "epoch": 3.3061276618303705,
+      "grad_norm": 1.2597849369049072,
+      "learning_rate": 2.5785953177257527e-06,
+      "loss": 0.3323,
+      "step": 34300
+    },
+    {
+      "epoch": 3.3061276618303705,
+      "eval_loss": 0.29248785972595215,
+      "eval_runtime": 40.1905,
+      "eval_samples_per_second": 248.815,
+      "eval_steps_per_second": 31.102,
+      "step": 34300
+    },
+    {
+      "epoch": 3.3157665179873104,
+      "grad_norm": 1.139083981513977,
+      "learning_rate": 2.568561872909699e-06,
+      "loss": 0.3286,
+      "step": 34400
+    },
+    {
+      "epoch": 3.3157665179873104,
+      "eval_loss": 0.289288729429245,
+      "eval_runtime": 40.0409,
+      "eval_samples_per_second": 249.745,
+      "eval_steps_per_second": 31.218,
+      "step": 34400
+    },
+    {
+      "epoch": 3.32540537414425,
+      "grad_norm": 1.1464444398880005,
+      "learning_rate": 2.5585284280936454e-06,
+      "loss": 0.3347,
+      "step": 34500
+    },
+    {
+      "epoch": 3.32540537414425,
+      "eval_loss": 0.2892674207687378,
+      "eval_runtime": 40.0307,
+      "eval_samples_per_second": 249.808,
+      "eval_steps_per_second": 31.226,
+      "step": 34500
+    },
+    {
+      "epoch": 3.33504423030119,
+      "grad_norm": 1.17436945438385,
+      "learning_rate": 2.548494983277592e-06,
+      "loss": 0.3275,
+      "step": 34600
+    },
+    {
+      "epoch": 3.33504423030119,
+      "eval_loss": 0.2948303818702698,
+      "eval_runtime": 40.0039,
+      "eval_samples_per_second": 249.976,
+      "eval_steps_per_second": 31.247,
+      "step": 34600
+    },
+    {
+      "epoch": 3.34468308645813,
+      "grad_norm": 1.2058557271957397,
+      "learning_rate": 2.5384615384615385e-06,
+      "loss": 0.3324,
+      "step": 34700
+    },
+    {
+      "epoch": 3.34468308645813,
+      "eval_loss": 0.2879999577999115,
+      "eval_runtime": 40.215,
+      "eval_samples_per_second": 248.663,
+      "eval_steps_per_second": 31.083,
+      "step": 34700
+    },
+    {
+      "epoch": 3.35432194261507,
+      "grad_norm": 1.1361554861068726,
+      "learning_rate": 2.528428093645485e-06,
+      "loss": 0.3283,
+      "step": 34800
+    },
+    {
+      "epoch": 3.35432194261507,
+      "eval_loss": 0.29396873712539673,
+      "eval_runtime": 40.0234,
+      "eval_samples_per_second": 249.854,
+      "eval_steps_per_second": 31.232,
+      "step": 34800
+    },
+    {
+      "epoch": 3.36396079877201,
+      "grad_norm": 1.1167203187942505,
+      "learning_rate": 2.5183946488294316e-06,
+      "loss": 0.3291,
+      "step": 34900
+    },
+    {
+      "epoch": 3.36396079877201,
+      "eval_loss": 0.28754281997680664,
+      "eval_runtime": 40.0006,
+      "eval_samples_per_second": 249.996,
+      "eval_steps_per_second": 31.25,
+      "step": 34900
+    },
+    {
+      "epoch": 3.3735996549289498,
+      "grad_norm": 1.1369761228561401,
+      "learning_rate": 2.508361204013378e-06,
+      "loss": 0.3324,
+      "step": 35000
+    },
+    {
+      "epoch": 3.3735996549289498,
+      "eval_loss": 0.2906251847743988,
+      "eval_runtime": 40.0165,
+      "eval_samples_per_second": 249.897,
+      "eval_steps_per_second": 31.237,
+      "step": 35000
+    },
+    {
+      "epoch": 3.3832385110858896,
+      "grad_norm": 1.1644196510314941,
+      "learning_rate": 2.4983277591973247e-06,
+      "loss": 0.33,
+      "step": 35100
+    },
+    {
+      "epoch": 3.3832385110858896,
+      "eval_loss": 0.29271912574768066,
+      "eval_runtime": 40.0328,
+      "eval_samples_per_second": 249.795,
+      "eval_steps_per_second": 31.224,
+      "step": 35100
+    },
+    {
+      "epoch": 3.3928773672428294,
+      "grad_norm": 1.2349485158920288,
+      "learning_rate": 2.488294314381271e-06,
+      "loss": 0.3296,
+      "step": 35200
+    },
+    {
+      "epoch": 3.3928773672428294,
+      "eval_loss": 0.2844071686267853,
+      "eval_runtime": 40.2281,
+      "eval_samples_per_second": 248.582,
+      "eval_steps_per_second": 31.073,
+      "step": 35200
+    },
+    {
+      "epoch": 3.4025162233997692,
+      "grad_norm": 1.0957030057907104,
+      "learning_rate": 2.4782608695652173e-06,
+      "loss": 0.3259,
+      "step": 35300
+    },
+    {
+      "epoch": 3.4025162233997692,
+      "eval_loss": 0.28942251205444336,
+      "eval_runtime": 40.049,
+      "eval_samples_per_second": 249.694,
+      "eval_steps_per_second": 31.212,
+      "step": 35300
+    },
+    {
+      "epoch": 3.412155079556709,
+      "grad_norm": 1.2255046367645264,
+      "learning_rate": 2.468227424749164e-06,
+      "loss": 0.3278,
+      "step": 35400
+    },
+    {
+      "epoch": 3.412155079556709,
+      "eval_loss": 0.2880132794380188,
+      "eval_runtime": 40.0318,
+      "eval_samples_per_second": 249.802,
+      "eval_steps_per_second": 31.225,
+      "step": 35400
+    },
+    {
+      "epoch": 3.421793935713649,
+      "grad_norm": 1.0865519046783447,
+      "learning_rate": 2.4581939799331104e-06,
+      "loss": 0.3264,
+      "step": 35500
+    },
+    {
+      "epoch": 3.421793935713649,
+      "eval_loss": 0.2906026840209961,
+      "eval_runtime": 40.0258,
+      "eval_samples_per_second": 249.839,
+      "eval_steps_per_second": 31.23,
+      "step": 35500
+    },
+    {
+      "epoch": 3.4314327918705887,
+      "grad_norm": 1.1252706050872803,
+      "learning_rate": 2.4481605351170568e-06,
+      "loss": 0.3272,
+      "step": 35600
+    },
+    {
+      "epoch": 3.4314327918705887,
+      "eval_loss": 0.28184667229652405,
+      "eval_runtime": 40.2117,
+      "eval_samples_per_second": 248.684,
+      "eval_steps_per_second": 31.085,
+      "step": 35600
+    },
+    {
+      "epoch": 3.4410716480275285,
+      "grad_norm": 1.1938196420669556,
+      "learning_rate": 2.4381270903010035e-06,
+      "loss": 0.3273,
+      "step": 35700
+    },
+    {
+      "epoch": 3.4410716480275285,
+      "eval_loss": 0.28648287057876587,
+      "eval_runtime": 40.2043,
+      "eval_samples_per_second": 248.73,
+      "eval_steps_per_second": 31.091,
+      "step": 35700
+    },
+    {
+      "epoch": 3.4507105041844683,
+      "grad_norm": 1.2802789211273193,
+      "learning_rate": 2.42809364548495e-06,
+      "loss": 0.3293,
+      "step": 35800
+    },
+    {
+      "epoch": 3.4507105041844683,
+      "eval_loss": 0.2890487313270569,
+      "eval_runtime": 40.0181,
+      "eval_samples_per_second": 249.887,
+      "eval_steps_per_second": 31.236,
+      "step": 35800
+    },
+    {
+      "epoch": 3.460349360341408,
+      "grad_norm": 1.2290120124816895,
+      "learning_rate": 2.4180602006688962e-06,
+      "loss": 0.3279,
+      "step": 35900
+    },
+    {
+      "epoch": 3.460349360341408,
+      "eval_loss": 0.2872641086578369,
+      "eval_runtime": 40.013,
+      "eval_samples_per_second": 249.919,
+      "eval_steps_per_second": 31.24,
+      "step": 35900
+    },
+    {
+      "epoch": 3.469988216498348,
+      "grad_norm": 1.1569044589996338,
+      "learning_rate": 2.408026755852843e-06,
+      "loss": 0.3263,
+      "step": 36000
+    },
+    {
+      "epoch": 3.469988216498348,
+      "eval_loss": 0.2899464964866638,
+      "eval_runtime": 40.0117,
+      "eval_samples_per_second": 249.927,
+      "eval_steps_per_second": 31.241,
+      "step": 36000
+    },
+    {
+      "epoch": 3.479627072655288,
+      "grad_norm": 1.1881693601608276,
+      "learning_rate": 2.3979933110367893e-06,
+      "loss": 0.3271,
+      "step": 36100
+    },
+    {
+      "epoch": 3.479627072655288,
+      "eval_loss": 0.29130667448043823,
+      "eval_runtime": 40.0206,
+      "eval_samples_per_second": 249.871,
+      "eval_steps_per_second": 31.234,
+      "step": 36100
+    },
+    {
+      "epoch": 3.4892659288122276,
+      "grad_norm": 1.3138039112091064,
+      "learning_rate": 2.387959866220736e-06,
+      "loss": 0.3239,
+      "step": 36200
+    },
+    {
+      "epoch": 3.4892659288122276,
+      "eval_loss": 0.2865622341632843,
+      "eval_runtime": 40.1762,
+      "eval_samples_per_second": 248.904,
+      "eval_steps_per_second": 31.113,
+      "step": 36200
+    },
+    {
+      "epoch": 3.498904784969168,
+      "grad_norm": 1.1673741340637207,
+      "learning_rate": 2.3779264214046824e-06,
+      "loss": 0.3277,
+      "step": 36300
+    },
+    {
+      "epoch": 3.498904784969168,
+      "eval_loss": 0.2852942645549774,
+      "eval_runtime": 40.0125,
+      "eval_samples_per_second": 249.922,
+      "eval_steps_per_second": 31.24,
+      "step": 36300
+    },
+    {
+      "epoch": 3.5085436411261077,
+      "grad_norm": 1.1211373805999756,
+      "learning_rate": 2.3678929765886288e-06,
+      "loss": 0.3237,
+      "step": 36400
+    },
+    {
+      "epoch": 3.5085436411261077,
+      "eval_loss": 0.290414422750473,
+      "eval_runtime": 40.0075,
+      "eval_samples_per_second": 249.953,
+      "eval_steps_per_second": 31.244,
+      "step": 36400
+    },
+    {
+      "epoch": 3.5181824972830475,
+      "grad_norm": 1.1853524446487427,
+      "learning_rate": 2.3578595317725755e-06,
+      "loss": 0.324,
+      "step": 36500
+    },
+    {
+      "epoch": 3.5181824972830475,
+      "eval_loss": 0.2909868657588959,
+      "eval_runtime": 40.0112,
+      "eval_samples_per_second": 249.93,
+      "eval_steps_per_second": 31.241,
+      "step": 36500
+    },
+    {
+      "epoch": 3.5278213534399874,
+      "grad_norm": 1.2360116243362427,
+      "learning_rate": 2.347826086956522e-06,
+      "loss": 0.3261,
+      "step": 36600
+    },
+    {
+      "epoch": 3.5278213534399874,
+      "eval_loss": 0.28790682554244995,
+      "eval_runtime": 40.0256,
+      "eval_samples_per_second": 249.84,
+      "eval_steps_per_second": 31.23,
+      "step": 36600
+    },
+    {
+      "epoch": 3.537460209596927,
+      "grad_norm": 1.0577596426010132,
+      "learning_rate": 2.337792642140468e-06,
+      "loss": 0.3236,
+      "step": 36700
+    },
+    {
+      "epoch": 3.537460209596927,
+      "eval_loss": 0.28858110308647156,
+      "eval_runtime": 40.2009,
+      "eval_samples_per_second": 248.751,
+      "eval_steps_per_second": 31.094,
+      "step": 36700
+    },
+    {
+      "epoch": 3.547099065753867,
+      "grad_norm": 1.0951604843139648,
+      "learning_rate": 2.327759197324415e-06,
+      "loss": 0.3216,
+      "step": 36800
+    },
+    {
+      "epoch": 3.547099065753867,
+      "eval_loss": 0.28311312198638916,
+      "eval_runtime": 40.0195,
+      "eval_samples_per_second": 249.878,
+      "eval_steps_per_second": 31.235,
+      "step": 36800
+    },
+    {
+      "epoch": 3.556737921910807,
+      "grad_norm": 1.11006498336792,
+      "learning_rate": 2.3177257525083613e-06,
+      "loss": 0.3243,
+      "step": 36900
+    },
+    {
+      "epoch": 3.556737921910807,
+      "eval_loss": 0.28506574034690857,
+      "eval_runtime": 40.033,
+      "eval_samples_per_second": 249.794,
+      "eval_steps_per_second": 31.224,
+      "step": 36900
+    },
+    {
+      "epoch": 3.5663767780677467,
+      "grad_norm": 1.2008014917373657,
+      "learning_rate": 2.307692307692308e-06,
+      "loss": 0.3211,
+      "step": 37000
+    },
+    {
+      "epoch": 3.5663767780677467,
+      "eval_loss": 0.2800537049770355,
+      "eval_runtime": 40.0317,
+      "eval_samples_per_second": 249.802,
+      "eval_steps_per_second": 31.225,
+      "step": 37000
+    },
+    {
+      "epoch": 3.5760156342246865,
+      "grad_norm": 1.1588464975357056,
+      "learning_rate": 2.2976588628762544e-06,
+      "loss": 0.3225,
+      "step": 37100
+    },
+    {
+      "epoch": 3.5760156342246865,
+      "eval_loss": 0.2877085506916046,
+      "eval_runtime": 40.0153,
+      "eval_samples_per_second": 249.904,
+      "eval_steps_per_second": 31.238,
+      "step": 37100
+    },
+    {
+      "epoch": 3.5856544903816263,
+      "grad_norm": 1.095901608467102,
+      "learning_rate": 2.2876254180602008e-06,
+      "loss": 0.3233,
+      "step": 37200
+    },
+    {
+      "epoch": 3.5856544903816263,
+      "eval_loss": 0.290816992521286,
+      "eval_runtime": 40.1774,
+      "eval_samples_per_second": 248.896,
+      "eval_steps_per_second": 31.112,
+      "step": 37200
+    },
+    {
+      "epoch": 3.595293346538566,
+      "grad_norm": 1.2038841247558594,
+      "learning_rate": 2.2775919732441475e-06,
+      "loss": 0.322,
+      "step": 37300
+    },
+    {
+      "epoch": 3.595293346538566,
+      "eval_loss": 0.2854418456554413,
+      "eval_runtime": 40.067,
+      "eval_samples_per_second": 249.582,
+      "eval_steps_per_second": 31.198,
+      "step": 37300
+    },
+    {
+      "epoch": 3.604932202695506,
+      "grad_norm": 1.1646316051483154,
+      "learning_rate": 2.267558528428094e-06,
+      "loss": 0.3224,
+      "step": 37400
+    },
+    {
+      "epoch": 3.604932202695506,
+      "eval_loss": 0.2838365137577057,
+      "eval_runtime": 40.055,
+      "eval_samples_per_second": 249.657,
+      "eval_steps_per_second": 31.207,
+      "step": 37400
+    },
+    {
+      "epoch": 3.6145710588524462,
+      "grad_norm": 1.166131615638733,
+      "learning_rate": 2.25752508361204e-06,
+      "loss": 0.3247,
+      "step": 37500
+    },
+    {
+      "epoch": 3.6145710588524462,
+      "eval_loss": 0.2809997797012329,
+      "eval_runtime": 40.2416,
+      "eval_samples_per_second": 248.499,
+      "eval_steps_per_second": 31.062,
+      "step": 37500
+    },
+    {
+      "epoch": 3.624209915009386,
+      "grad_norm": 1.1528489589691162,
+      "learning_rate": 2.2474916387959865e-06,
+      "loss": 0.3251,
+      "step": 37600
+    },
+    {
+      "epoch": 3.624209915009386,
+      "eval_loss": 0.2795826196670532,
+      "eval_runtime": 40.1013,
+      "eval_samples_per_second": 249.368,
+      "eval_steps_per_second": 31.171,
+      "step": 37600
+    },
+    {
+      "epoch": 3.633848771166326,
+      "grad_norm": 1.1351743936538696,
+      "learning_rate": 2.237458193979933e-06,
+      "loss": 0.319,
+      "step": 37700
+    },
+    {
+      "epoch": 3.633848771166326,
+      "eval_loss": 0.2884068489074707,
+      "eval_runtime": 40.254,
+      "eval_samples_per_second": 248.423,
+      "eval_steps_per_second": 31.053,
+      "step": 37700
+    },
+    {
+      "epoch": 3.6434876273232657,
+      "grad_norm": 1.2123479843139648,
+      "learning_rate": 2.2274247491638796e-06,
+      "loss": 0.3216,
+      "step": 37800
+    },
+    {
+      "epoch": 3.6434876273232657,
+      "eval_loss": 0.2808574140071869,
+      "eval_runtime": 40.0346,
+      "eval_samples_per_second": 249.784,
+      "eval_steps_per_second": 31.223,
+      "step": 37800
+    },
+    {
+      "epoch": 3.6531264834802055,
+      "grad_norm": 1.1580688953399658,
+      "learning_rate": 2.217391304347826e-06,
+      "loss": 0.3182,
+      "step": 37900
+    },
+    {
+      "epoch": 3.6531264834802055,
+      "eval_loss": 0.28800562024116516,
+      "eval_runtime": 40.0302,
+      "eval_samples_per_second": 249.811,
+      "eval_steps_per_second": 31.226,
+      "step": 37900
+    },
+    {
+      "epoch": 3.6627653396371453,
+      "grad_norm": 1.1823816299438477,
+      "learning_rate": 2.2073578595317723e-06,
+      "loss": 0.3208,
+      "step": 38000
+    },
+    {
+      "epoch": 3.6627653396371453,
+      "eval_loss": 0.28609785437583923,
+      "eval_runtime": 40.0272,
+      "eval_samples_per_second": 249.83,
+      "eval_steps_per_second": 31.229,
+      "step": 38000
+    },
+    {
+      "epoch": 3.672404195794085,
+      "grad_norm": 1.1534825563430786,
+      "learning_rate": 2.197324414715719e-06,
+      "loss": 0.3165,
+      "step": 38100
+    },
+    {
+      "epoch": 3.672404195794085,
+      "eval_loss": 0.28174835443496704,
+      "eval_runtime": 40.198,
+      "eval_samples_per_second": 248.769,
+      "eval_steps_per_second": 31.096,
+      "step": 38100
+    },
+    {
+      "epoch": 3.682043051951025,
+      "grad_norm": 1.160309076309204,
+      "learning_rate": 2.1872909698996654e-06,
+      "loss": 0.3203,
+      "step": 38200
+    },
+    {
+      "epoch": 3.682043051951025,
+      "eval_loss": 0.2865579128265381,
+      "eval_runtime": 40.0605,
+      "eval_samples_per_second": 249.622,
+      "eval_steps_per_second": 31.203,
+      "step": 38200
+    },
+    {
+      "epoch": 3.691681908107965,
+      "grad_norm": 1.1397422552108765,
+      "learning_rate": 2.177257525083612e-06,
+      "loss": 0.3189,
+      "step": 38300
+    },
+    {
+      "epoch": 3.691681908107965,
+      "eval_loss": 0.2845374643802643,
+      "eval_runtime": 40.0285,
+      "eval_samples_per_second": 249.822,
+      "eval_steps_per_second": 31.228,
+      "step": 38300
+    },
+    {
+      "epoch": 3.7013207642649046,
+      "grad_norm": 1.1168709993362427,
+      "learning_rate": 2.1672240802675585e-06,
+      "loss": 0.32,
+      "step": 38400
+    },
+    {
+      "epoch": 3.7013207642649046,
+      "eval_loss": 0.2816096246242523,
+      "eval_runtime": 40.0255,
+      "eval_samples_per_second": 249.841,
+      "eval_steps_per_second": 31.23,
+      "step": 38400
+    },
+    {
+      "epoch": 3.7109596204218445,
+      "grad_norm": 1.221742868423462,
+      "learning_rate": 2.157190635451505e-06,
+      "loss": 0.3167,
+      "step": 38500
+    },
+    {
+      "epoch": 3.7109596204218445,
+      "eval_loss": 0.283882737159729,
+      "eval_runtime": 40.2082,
+      "eval_samples_per_second": 248.706,
+      "eval_steps_per_second": 31.088,
+      "step": 38500
+    },
+    {
+      "epoch": 3.7205984765787843,
+      "grad_norm": 1.132840633392334,
+      "learning_rate": 2.1471571906354516e-06,
+      "loss": 0.3189,
+      "step": 38600
+    },
+    {
+      "epoch": 3.7205984765787843,
+      "eval_loss": 0.28340187668800354,
+      "eval_runtime": 40.0209,
+      "eval_samples_per_second": 249.869,
+      "eval_steps_per_second": 31.234,
+      "step": 38600
+    },
+    {
+      "epoch": 3.730237332735724,
+      "grad_norm": 1.145599365234375,
+      "learning_rate": 2.137123745819398e-06,
+      "loss": 0.3179,
+      "step": 38700
+    },
+    {
+      "epoch": 3.730237332735724,
+      "eval_loss": 0.2839237451553345,
+      "eval_runtime": 40.0677,
+      "eval_samples_per_second": 249.578,
+      "eval_steps_per_second": 31.197,
+      "step": 38700
+    },
+    {
+      "epoch": 3.739876188892664,
+      "grad_norm": 1.1449469327926636,
+      "learning_rate": 2.1270903010033443e-06,
+      "loss": 0.3209,
+      "step": 38800
+    },
+    {
+      "epoch": 3.739876188892664,
+      "eval_loss": 0.28125128149986267,
+      "eval_runtime": 40.0685,
+      "eval_samples_per_second": 249.573,
+      "eval_steps_per_second": 31.197,
+      "step": 38800
+    },
+    {
+      "epoch": 3.7495150450496038,
+      "grad_norm": 1.155439853668213,
+      "learning_rate": 2.117056856187291e-06,
+      "loss": 0.3182,
+      "step": 38900
+    },
+    {
+      "epoch": 3.7495150450496038,
+      "eval_loss": 0.2792927622795105,
+      "eval_runtime": 40.048,
+      "eval_samples_per_second": 249.7,
+      "eval_steps_per_second": 31.213,
+      "step": 38900
+    },
+    {
+      "epoch": 3.7591539012065436,
+      "grad_norm": 1.1573885679244995,
+      "learning_rate": 2.1070234113712374e-06,
+      "loss": 0.3204,
+      "step": 39000
+    },
+    {
+      "epoch": 3.7591539012065436,
+      "eval_loss": 0.27629750967025757,
+      "eval_runtime": 40.2171,
+      "eval_samples_per_second": 248.651,
+      "eval_steps_per_second": 31.081,
+      "step": 39000
+    },
+    {
+      "epoch": 3.7687927573634834,
+      "grad_norm": 1.1162937879562378,
+      "learning_rate": 2.0969899665551837e-06,
+      "loss": 0.3183,
+      "step": 39100
+    },
+    {
+      "epoch": 3.7687927573634834,
+      "eval_loss": 0.2790432274341583,
+      "eval_runtime": 40.0376,
+      "eval_samples_per_second": 249.765,
+      "eval_steps_per_second": 31.221,
+      "step": 39100
+    },
+    {
+      "epoch": 3.7784316135204237,
+      "grad_norm": 1.0601564645767212,
+      "learning_rate": 2.0869565217391305e-06,
+      "loss": 0.3182,
+      "step": 39200
+    },
+    {
+      "epoch": 3.7784316135204237,
+      "eval_loss": 0.2791585326194763,
+      "eval_runtime": 40.037,
+      "eval_samples_per_second": 249.769,
+      "eval_steps_per_second": 31.221,
+      "step": 39200
+    },
+    {
+      "epoch": 3.7880704696773635,
+      "grad_norm": 1.131667137145996,
+      "learning_rate": 2.076923076923077e-06,
+      "loss": 0.3175,
+      "step": 39300
+    },
+    {
+      "epoch": 3.7880704696773635,
+      "eval_loss": 0.27618080377578735,
+      "eval_runtime": 40.0503,
+      "eval_samples_per_second": 249.686,
+      "eval_steps_per_second": 31.211,
+      "step": 39300
+    },
+    {
+      "epoch": 3.7977093258343033,
+      "grad_norm": 1.1320117712020874,
+      "learning_rate": 2.0668896321070236e-06,
+      "loss": 0.3189,
+      "step": 39400
+    },
+    {
+      "epoch": 3.7977093258343033,
+      "eval_loss": 0.2805337607860565,
+      "eval_runtime": 40.0425,
+      "eval_samples_per_second": 249.735,
+      "eval_steps_per_second": 31.217,
+      "step": 39400
+    },
+    {
+      "epoch": 3.807348181991243,
+      "grad_norm": 1.154017448425293,
+      "learning_rate": 2.05685618729097e-06,
+      "loss": 0.3189,
+      "step": 39500
+    },
+    {
+      "epoch": 3.807348181991243,
+      "eval_loss": 0.2768949866294861,
+      "eval_runtime": 40.2119,
+      "eval_samples_per_second": 248.683,
+      "eval_steps_per_second": 31.085,
+      "step": 39500
+    },
+    {
+      "epoch": 3.816987038148183,
+      "grad_norm": 1.1326124668121338,
+      "learning_rate": 2.0468227424749163e-06,
+      "loss": 0.3176,
+      "step": 39600
+    },
+    {
+      "epoch": 3.816987038148183,
+      "eval_loss": 0.28353628516197205,
+      "eval_runtime": 40.027,
+      "eval_samples_per_second": 249.832,
+      "eval_steps_per_second": 31.229,
+      "step": 39600
+    },
+    {
+      "epoch": 3.826625894305123,
+      "grad_norm": 1.0950291156768799,
+      "learning_rate": 2.036789297658863e-06,
+      "loss": 0.3152,
+      "step": 39700
+    },
+    {
+      "epoch": 3.826625894305123,
+      "eval_loss": 0.27837592363357544,
+      "eval_runtime": 40.0294,
+      "eval_samples_per_second": 249.817,
+      "eval_steps_per_second": 31.227,
+      "step": 39700
+    },
+    {
+      "epoch": 3.8362647504620626,
+      "grad_norm": 1.0399600267410278,
+      "learning_rate": 2.0267558528428094e-06,
+      "loss": 0.3162,
+      "step": 39800
+    },
+    {
+      "epoch": 3.8362647504620626,
+      "eval_loss": 0.279599666595459,
+      "eval_runtime": 40.03,
+      "eval_samples_per_second": 249.813,
+      "eval_steps_per_second": 31.227,
+      "step": 39800
+    },
+    {
+      "epoch": 3.8459036066190024,
+      "grad_norm": 1.1865919828414917,
+      "learning_rate": 2.0167224080267557e-06,
+      "loss": 0.312,
+      "step": 39900
+    },
+    {
+      "epoch": 3.8459036066190024,
+      "eval_loss": 0.2764998972415924,
+      "eval_runtime": 40.2252,
+      "eval_samples_per_second": 248.6,
+      "eval_steps_per_second": 31.075,
+      "step": 39900
+    },
+    {
+      "epoch": 3.8555424627759423,
+      "grad_norm": 1.1612026691436768,
+      "learning_rate": 2.0066889632107025e-06,
+      "loss": 0.3176,
+      "step": 40000
+    },
+    {
+      "epoch": 3.8555424627759423,
+      "eval_loss": 0.28109410405158997,
+      "eval_runtime": 40.0865,
+      "eval_samples_per_second": 249.461,
+      "eval_steps_per_second": 31.183,
+      "step": 40000
+    },
+    {
+      "epoch": 3.865181318932882,
+      "grad_norm": 1.1188021898269653,
+      "learning_rate": 1.996655518394649e-06,
+      "loss": 0.3128,
+      "step": 40100
+    },
+    {
+      "epoch": 3.865181318932882,
+      "eval_loss": 0.276141881942749,
+      "eval_runtime": 40.0338,
+      "eval_samples_per_second": 249.789,
+      "eval_steps_per_second": 31.224,
+      "step": 40100
+    },
+    {
+      "epoch": 3.8748201750898223,
+      "grad_norm": 1.0921941995620728,
+      "learning_rate": 1.986622073578595e-06,
+      "loss": 0.3163,
+      "step": 40200
+    },
+    {
+      "epoch": 3.8748201750898223,
+      "eval_loss": 0.27618375420570374,
+      "eval_runtime": 40.0354,
+      "eval_samples_per_second": 249.779,
+      "eval_steps_per_second": 31.222,
+      "step": 40200
+    },
+    {
+      "epoch": 3.884459031246762,
+      "grad_norm": 1.1041548252105713,
+      "learning_rate": 1.976588628762542e-06,
+      "loss": 0.312,
+      "step": 40300
+    },
+    {
+      "epoch": 3.884459031246762,
+      "eval_loss": 0.2808937728404999,
+      "eval_runtime": 40.0468,
+      "eval_samples_per_second": 249.708,
+      "eval_steps_per_second": 31.214,
+      "step": 40300
+    },
+    {
+      "epoch": 3.894097887403702,
+      "grad_norm": 1.1126019954681396,
+      "learning_rate": 1.9665551839464883e-06,
+      "loss": 0.3147,
+      "step": 40400
+    },
+    {
+      "epoch": 3.894097887403702,
+      "eval_loss": 0.2763468623161316,
+      "eval_runtime": 43.5695,
+      "eval_samples_per_second": 229.519,
+      "eval_steps_per_second": 28.69,
+      "step": 40400
+    },
+    {
+      "epoch": 3.903736743560642,
+      "grad_norm": 1.0650842189788818,
+      "learning_rate": 1.956521739130435e-06,
+      "loss": 0.3143,
+      "step": 40500
+    },
+    {
+      "epoch": 3.903736743560642,
+      "eval_loss": 0.27691203355789185,
+      "eval_runtime": 40.0111,
+      "eval_samples_per_second": 249.93,
+      "eval_steps_per_second": 31.241,
+      "step": 40500
+    },
+    {
+      "epoch": 3.9133755997175816,
+      "grad_norm": 1.2413527965545654,
+      "learning_rate": 1.9464882943143814e-06,
+      "loss": 0.3144,
+      "step": 40600
+    },
+    {
+      "epoch": 3.9133755997175816,
+      "eval_loss": 0.27241915464401245,
+      "eval_runtime": 40.027,
+      "eval_samples_per_second": 249.831,
+      "eval_steps_per_second": 31.229,
+      "step": 40600
+    },
+    {
+      "epoch": 3.9230144558745215,
+      "grad_norm": 1.2129082679748535,
+      "learning_rate": 1.9364548494983277e-06,
+      "loss": 0.3148,
+      "step": 40700
+    },
+    {
+      "epoch": 3.9230144558745215,
+      "eval_loss": 0.27262434363365173,
+      "eval_runtime": 40.0193,
+      "eval_samples_per_second": 249.88,
+      "eval_steps_per_second": 31.235,
+      "step": 40700
+    },
+    {
+      "epoch": 3.9326533120314613,
+      "grad_norm": 1.2504247426986694,
+      "learning_rate": 1.9264214046822745e-06,
+      "loss": 0.3138,
+      "step": 40800
+    },
+    {
+      "epoch": 3.9326533120314613,
+      "eval_loss": 0.27585116028785706,
+      "eval_runtime": 40.042,
+      "eval_samples_per_second": 249.738,
+      "eval_steps_per_second": 31.217,
+      "step": 40800
+    },
+    {
+      "epoch": 3.942292168188401,
+      "grad_norm": 1.1798555850982666,
+      "learning_rate": 1.916387959866221e-06,
+      "loss": 0.3143,
+      "step": 40900
+    },
+    {
+      "epoch": 3.942292168188401,
+      "eval_loss": 0.27270910143852234,
+      "eval_runtime": 40.2588,
+      "eval_samples_per_second": 248.393,
+      "eval_steps_per_second": 31.049,
+      "step": 40900
+    },
+    {
+      "epoch": 3.951931024345341,
+      "grad_norm": 0.9959877133369446,
+      "learning_rate": 1.9063545150501674e-06,
+      "loss": 0.3126,
+      "step": 41000
+    },
+    {
+      "epoch": 3.951931024345341,
+      "eval_loss": 0.2772971987724304,
+      "eval_runtime": 40.0712,
+      "eval_samples_per_second": 249.555,
+      "eval_steps_per_second": 31.194,
+      "step": 41000
+    },
+    {
+      "epoch": 3.9615698805022808,
+      "grad_norm": 1.1942039728164673,
+      "learning_rate": 1.896321070234114e-06,
+      "loss": 0.3137,
+      "step": 41100
+    },
+    {
+      "epoch": 3.9615698805022808,
+      "eval_loss": 0.2713697552680969,
+      "eval_runtime": 40.052,
+      "eval_samples_per_second": 249.675,
+      "eval_steps_per_second": 31.209,
+      "step": 41100
+    },
+    {
+      "epoch": 3.9712087366592206,
+      "grad_norm": 1.1946539878845215,
+      "learning_rate": 1.8862876254180603e-06,
+      "loss": 0.3143,
+      "step": 41200
+    },
+    {
+      "epoch": 3.9712087366592206,
+      "eval_loss": 0.27480974793434143,
+      "eval_runtime": 40.0696,
+      "eval_samples_per_second": 249.566,
+      "eval_steps_per_second": 31.196,
+      "step": 41200
+    },
+    {
+      "epoch": 3.9808475928161604,
+      "grad_norm": 1.168750286102295,
+      "learning_rate": 1.8762541806020068e-06,
+      "loss": 0.311,
+      "step": 41300
+    },
+    {
+      "epoch": 3.9808475928161604,
+      "eval_loss": 0.2778546214103699,
+      "eval_runtime": 40.2246,
+      "eval_samples_per_second": 248.604,
+      "eval_steps_per_second": 31.076,
+      "step": 41300
+    },
+    {
+      "epoch": 3.9904864489731002,
+      "grad_norm": 1.2502557039260864,
+      "learning_rate": 1.8662207357859534e-06,
+      "loss": 0.3174,
+      "step": 41400
+    },
+    {
+      "epoch": 3.9904864489731002,
+      "eval_loss": 0.2807627320289612,
+      "eval_runtime": 40.07,
+      "eval_samples_per_second": 249.563,
+      "eval_steps_per_second": 31.195,
+      "step": 41400
+    },
+    {
+      "epoch": 4.00012530513004,
+      "grad_norm": 1.09615159034729,
+      "learning_rate": 1.8561872909699e-06,
+      "loss": 0.3124,
+      "step": 41500
+    },
+    {
+      "epoch": 4.00012530513004,
+      "eval_loss": 0.27661824226379395,
+      "eval_runtime": 40.0399,
+      "eval_samples_per_second": 249.751,
+      "eval_steps_per_second": 31.219,
+      "step": 41500
+    },
+    {
+      "epoch": 4.00976416128698,
+      "grad_norm": 1.100216031074524,
+      "learning_rate": 1.8461538461538462e-06,
+      "loss": 0.3136,
+      "step": 41600
+    },
+    {
+      "epoch": 4.00976416128698,
+      "eval_loss": 0.2801878750324249,
+      "eval_runtime": 40.0624,
+      "eval_samples_per_second": 249.611,
+      "eval_steps_per_second": 31.201,
+      "step": 41600
+    },
+    {
+      "epoch": 4.01940301744392,
+      "grad_norm": 1.0246331691741943,
+      "learning_rate": 1.8361204013377928e-06,
+      "loss": 0.3146,
+      "step": 41700
+    },
+    {
+      "epoch": 4.01940301744392,
+      "eval_loss": 0.2788015604019165,
+      "eval_runtime": 40.0597,
+      "eval_samples_per_second": 249.628,
+      "eval_steps_per_second": 31.203,
+      "step": 41700
+    },
+    {
+      "epoch": 4.0290418736008595,
+      "grad_norm": 1.300437092781067,
+      "learning_rate": 1.8260869565217394e-06,
+      "loss": 0.3119,
+      "step": 41800
+    },
+    {
+      "epoch": 4.0290418736008595,
+      "eval_loss": 0.272819459438324,
+      "eval_runtime": 40.2236,
+      "eval_samples_per_second": 248.61,
+      "eval_steps_per_second": 31.076,
+      "step": 41800
+    },
+    {
+      "epoch": 4.038680729757799,
+      "grad_norm": 1.1649296283721924,
+      "learning_rate": 1.8160535117056857e-06,
+      "loss": 0.312,
+      "step": 41900
+    },
+    {
+      "epoch": 4.038680729757799,
+      "eval_loss": 0.2787322402000427,
+      "eval_runtime": 40.0348,
+      "eval_samples_per_second": 249.783,
+      "eval_steps_per_second": 31.223,
+      "step": 41900
+    },
+    {
+      "epoch": 4.048319585914739,
+      "grad_norm": 1.1247646808624268,
+      "learning_rate": 1.8060200668896322e-06,
+      "loss": 0.3124,
+      "step": 42000
+    },
+    {
+      "epoch": 4.048319585914739,
+      "eval_loss": 0.2764863669872284,
+      "eval_runtime": 40.0195,
+      "eval_samples_per_second": 249.878,
+      "eval_steps_per_second": 31.235,
+      "step": 42000
+    },
+    {
+      "epoch": 4.057958442071679,
+      "grad_norm": 1.21001398563385,
+      "learning_rate": 1.7959866220735788e-06,
+      "loss": 0.3099,
+      "step": 42100
+    },
+    {
+      "epoch": 4.057958442071679,
+      "eval_loss": 0.275383859872818,
+      "eval_runtime": 40.0245,
+      "eval_samples_per_second": 249.847,
+      "eval_steps_per_second": 31.231,
+      "step": 42100
+    },
+    {
+      "epoch": 4.067597298228619,
+      "grad_norm": 1.1185123920440674,
+      "learning_rate": 1.7859531772575253e-06,
+      "loss": 0.3094,
+      "step": 42200
+    },
+    {
+      "epoch": 4.067597298228619,
+      "eval_loss": 0.27758368849754333,
+      "eval_runtime": 40.2446,
+      "eval_samples_per_second": 248.481,
+      "eval_steps_per_second": 31.06,
+      "step": 42200
+    },
+    {
+      "epoch": 4.077236154385559,
+      "grad_norm": 1.0860310792922974,
+      "learning_rate": 1.7759197324414717e-06,
+      "loss": 0.3077,
+      "step": 42300
+    },
+    {
+      "epoch": 4.077236154385559,
+      "eval_loss": 0.2693646550178528,
+      "eval_runtime": 40.0602,
+      "eval_samples_per_second": 249.624,
+      "eval_steps_per_second": 31.203,
+      "step": 42300
+    },
+    {
+      "epoch": 4.086875010542499,
+      "grad_norm": 1.1427111625671387,
+      "learning_rate": 1.7658862876254182e-06,
+      "loss": 0.3119,
+      "step": 42400
+    },
+    {
+      "epoch": 4.086875010542499,
+      "eval_loss": 0.27336445450782776,
+      "eval_runtime": 40.0655,
+      "eval_samples_per_second": 249.592,
+      "eval_steps_per_second": 31.199,
+      "step": 42400
+    },
+    {
+      "epoch": 4.096513866699439,
+      "grad_norm": 1.027213215827942,
+      "learning_rate": 1.7558528428093648e-06,
+      "loss": 0.3103,
+      "step": 42500
+    },
+    {
+      "epoch": 4.096513866699439,
+      "eval_loss": 0.27709463238716125,
+      "eval_runtime": 40.0391,
+      "eval_samples_per_second": 249.756,
+      "eval_steps_per_second": 31.22,
+      "step": 42500
+    },
+    {
+      "epoch": 4.106152722856379,
+      "grad_norm": 1.1721312999725342,
+      "learning_rate": 1.745819397993311e-06,
+      "loss": 0.3108,
+      "step": 42600
+    },
+    {
+      "epoch": 4.106152722856379,
+      "eval_loss": 0.275216281414032,
+      "eval_runtime": 40.0433,
+      "eval_samples_per_second": 249.729,
+      "eval_steps_per_second": 31.216,
+      "step": 42600
+    },
+    {
+      "epoch": 4.115791579013319,
+      "grad_norm": 1.1721031665802002,
+      "learning_rate": 1.7357859531772575e-06,
+      "loss": 0.3111,
+      "step": 42700
+    },
+    {
+      "epoch": 4.115791579013319,
+      "eval_loss": 0.27178844809532166,
+      "eval_runtime": 40.2332,
+      "eval_samples_per_second": 248.551,
+      "eval_steps_per_second": 31.069,
+      "step": 42700
+    },
+    {
+      "epoch": 4.125430435170259,
+      "grad_norm": 1.1004197597503662,
+      "learning_rate": 1.7257525083612038e-06,
+      "loss": 0.3103,
+      "step": 42800
+    },
+    {
+      "epoch": 4.125430435170259,
+      "eval_loss": 0.278656542301178,
+      "eval_runtime": 40.0331,
+      "eval_samples_per_second": 249.793,
+      "eval_steps_per_second": 31.224,
+      "step": 42800
+    },
+    {
+      "epoch": 4.1350692913271985,
+      "grad_norm": 1.1854575872421265,
+      "learning_rate": 1.7157190635451504e-06,
+      "loss": 0.3124,
+      "step": 42900
+    },
+    {
+      "epoch": 4.1350692913271985,
+      "eval_loss": 0.27243393659591675,
+      "eval_runtime": 40.0307,
+      "eval_samples_per_second": 249.808,
+      "eval_steps_per_second": 31.226,
+      "step": 42900
+    },
+    {
+      "epoch": 4.144708147484138,
+      "grad_norm": 1.1487313508987427,
+      "learning_rate": 1.705685618729097e-06,
+      "loss": 0.3111,
+      "step": 43000
+    },
+    {
+      "epoch": 4.144708147484138,
+      "eval_loss": 0.27582091093063354,
+      "eval_runtime": 40.0545,
+      "eval_samples_per_second": 249.66,
+      "eval_steps_per_second": 31.207,
+      "step": 43000
+    },
+    {
+      "epoch": 4.154347003641078,
+      "grad_norm": 1.1287168264389038,
+      "learning_rate": 1.6956521739130435e-06,
+      "loss": 0.3088,
+      "step": 43100
+    },
+    {
+      "epoch": 4.154347003641078,
+      "eval_loss": 0.2760840952396393,
+      "eval_runtime": 40.0864,
+      "eval_samples_per_second": 249.461,
+      "eval_steps_per_second": 31.183,
+      "step": 43100
+    },
+    {
+      "epoch": 4.163985859798018,
+      "grad_norm": 1.1330616474151611,
+      "learning_rate": 1.6856187290969898e-06,
+      "loss": 0.3111,
+      "step": 43200
+    },
+    {
+      "epoch": 4.163985859798018,
+      "eval_loss": 0.2729859948158264,
+      "eval_runtime": 40.1988,
+      "eval_samples_per_second": 248.764,
+      "eval_steps_per_second": 31.095,
+      "step": 43200
+    },
+    {
+      "epoch": 4.173624715954958,
+      "grad_norm": 1.1357420682907104,
+      "learning_rate": 1.6755852842809363e-06,
+      "loss": 0.3092,
+      "step": 43300
+    },
+    {
+      "epoch": 4.173624715954958,
+      "eval_loss": 0.26835474371910095,
+      "eval_runtime": 40.0242,
+      "eval_samples_per_second": 249.849,
+      "eval_steps_per_second": 31.231,
+      "step": 43300
+    },
+    {
+      "epoch": 4.183263572111898,
+      "grad_norm": 1.0758098363876343,
+      "learning_rate": 1.665551839464883e-06,
+      "loss": 0.3091,
+      "step": 43400
+    },
+    {
+      "epoch": 4.183263572111898,
+      "eval_loss": 0.271380215883255,
+      "eval_runtime": 40.039,
+      "eval_samples_per_second": 249.757,
+      "eval_steps_per_second": 31.22,
+      "step": 43400
+    },
+    {
+      "epoch": 4.192902428268837,
+      "grad_norm": 1.1093837022781372,
+      "learning_rate": 1.6555183946488294e-06,
+      "loss": 0.3092,
+      "step": 43500
+    },
+    {
+      "epoch": 4.192902428268837,
+      "eval_loss": 0.276606947183609,
+      "eval_runtime": 40.058,
+      "eval_samples_per_second": 249.638,
+      "eval_steps_per_second": 31.205,
+      "step": 43500
+    },
+    {
+      "epoch": 4.202541284425777,
+      "grad_norm": 1.1726568937301636,
+      "learning_rate": 1.6454849498327758e-06,
+      "loss": 0.3089,
+      "step": 43600
+    },
+    {
+      "epoch": 4.202541284425777,
+      "eval_loss": 0.27440130710601807,
+      "eval_runtime": 40.2425,
+      "eval_samples_per_second": 248.493,
+      "eval_steps_per_second": 31.062,
+      "step": 43600
+    },
+    {
+      "epoch": 4.212180140582717,
+      "grad_norm": 1.137256145477295,
+      "learning_rate": 1.6354515050167223e-06,
+      "loss": 0.3082,
+      "step": 43700
+    },
+    {
+      "epoch": 4.212180140582717,
+      "eval_loss": 0.268880695104599,
+      "eval_runtime": 40.1041,
+      "eval_samples_per_second": 249.351,
+      "eval_steps_per_second": 31.169,
+      "step": 43700
+    },
+    {
+      "epoch": 4.221818996739657,
+      "grad_norm": 1.194308876991272,
+      "learning_rate": 1.6254180602006689e-06,
+      "loss": 0.3089,
+      "step": 43800
+    },
+    {
+      "epoch": 4.221818996739657,
+      "eval_loss": 0.2654968202114105,
+      "eval_runtime": 40.0755,
+      "eval_samples_per_second": 249.529,
+      "eval_steps_per_second": 31.191,
+      "step": 43800
+    },
+    {
+      "epoch": 4.231457852896597,
+      "grad_norm": 1.1237385272979736,
+      "learning_rate": 1.6153846153846154e-06,
+      "loss": 0.3096,
+      "step": 43900
+    },
+    {
+      "epoch": 4.231457852896597,
+      "eval_loss": 0.2736607491970062,
+      "eval_runtime": 40.0782,
+      "eval_samples_per_second": 249.512,
+      "eval_steps_per_second": 31.189,
+      "step": 43900
+    },
+    {
+      "epoch": 4.2410967090535365,
+      "grad_norm": 1.0342357158660889,
+      "learning_rate": 1.6053511705685618e-06,
+      "loss": 0.3088,
+      "step": 44000
+    },
+    {
+      "epoch": 4.2410967090535365,
+      "eval_loss": 0.27211064100265503,
+      "eval_runtime": 40.0588,
+      "eval_samples_per_second": 249.633,
+      "eval_steps_per_second": 31.204,
+      "step": 44000
+    },
+    {
+      "epoch": 4.250735565210476,
+      "grad_norm": 1.190117597579956,
+      "learning_rate": 1.5953177257525083e-06,
+      "loss": 0.3092,
+      "step": 44100
+    },
+    {
+      "epoch": 4.250735565210476,
+      "eval_loss": 0.2710612416267395,
+      "eval_runtime": 40.227,
+      "eval_samples_per_second": 248.589,
+      "eval_steps_per_second": 31.074,
+      "step": 44100
+    },
+    {
+      "epoch": 4.260374421367416,
+      "grad_norm": 1.1709164381027222,
+      "learning_rate": 1.5852842809364549e-06,
+      "loss": 0.3081,
+      "step": 44200
+    },
+    {
+      "epoch": 4.260374421367416,
+      "eval_loss": 0.272073358297348,
+      "eval_runtime": 40.0531,
+      "eval_samples_per_second": 249.669,
+      "eval_steps_per_second": 31.209,
+      "step": 44200
+    },
+    {
+      "epoch": 4.270013277524356,
+      "grad_norm": 1.2000877857208252,
+      "learning_rate": 1.5752508361204012e-06,
+      "loss": 0.3082,
+      "step": 44300
+    },
+    {
+      "epoch": 4.270013277524356,
+      "eval_loss": 0.2707884907722473,
+      "eval_runtime": 40.0511,
+      "eval_samples_per_second": 249.681,
+      "eval_steps_per_second": 31.21,
+      "step": 44300
+    },
+    {
+      "epoch": 4.279652133681296,
+      "grad_norm": 1.084593415260315,
+      "learning_rate": 1.5652173913043478e-06,
+      "loss": 0.308,
+      "step": 44400
+    },
+    {
+      "epoch": 4.279652133681296,
+      "eval_loss": 0.26928406953811646,
+      "eval_runtime": 40.0322,
+      "eval_samples_per_second": 249.799,
+      "eval_steps_per_second": 31.225,
+      "step": 44400
+    },
+    {
+      "epoch": 4.289290989838236,
+      "grad_norm": 1.0698468685150146,
+      "learning_rate": 1.5551839464882943e-06,
+      "loss": 0.3061,
+      "step": 44500
+    },
+    {
+      "epoch": 4.289290989838236,
+      "eval_loss": 0.2769457697868347,
+      "eval_runtime": 40.2095,
+      "eval_samples_per_second": 248.697,
+      "eval_steps_per_second": 31.087,
+      "step": 44500
+    },
+    {
+      "epoch": 4.2989298459951755,
+      "grad_norm": 1.0958250761032104,
+      "learning_rate": 1.5451505016722409e-06,
+      "loss": 0.3097,
+      "step": 44600
+    },
+    {
+      "epoch": 4.2989298459951755,
+      "eval_loss": 0.26975879073143005,
+      "eval_runtime": 40.0708,
+      "eval_samples_per_second": 249.558,
+      "eval_steps_per_second": 31.195,
+      "step": 44600
+    },
+    {
+      "epoch": 4.308568702152115,
+      "grad_norm": 1.27885103225708,
+      "learning_rate": 1.5351170568561872e-06,
+      "loss": 0.3063,
+      "step": 44700
+    },
+    {
+      "epoch": 4.308568702152115,
+      "eval_loss": 0.2665342688560486,
+      "eval_runtime": 40.2308,
+      "eval_samples_per_second": 248.566,
+      "eval_steps_per_second": 31.071,
+      "step": 44700
+    },
+    {
+      "epoch": 4.318207558309055,
+      "grad_norm": 1.1225556135177612,
+      "learning_rate": 1.5250836120401338e-06,
+      "loss": 0.3084,
+      "step": 44800
+    },
+    {
+      "epoch": 4.318207558309055,
+      "eval_loss": 0.266510546207428,
+      "eval_runtime": 40.0617,
+      "eval_samples_per_second": 249.615,
+      "eval_steps_per_second": 31.202,
+      "step": 44800
+    },
+    {
+      "epoch": 4.327846414465995,
+      "grad_norm": 1.1451354026794434,
+      "learning_rate": 1.5150501672240803e-06,
+      "loss": 0.3093,
+      "step": 44900
+    },
+    {
+      "epoch": 4.327846414465995,
+      "eval_loss": 0.26877492666244507,
+      "eval_runtime": 40.0203,
+      "eval_samples_per_second": 249.873,
+      "eval_steps_per_second": 31.234,
+      "step": 44900
+    },
+    {
+      "epoch": 4.337485270622935,
+      "grad_norm": 1.1867423057556152,
+      "learning_rate": 1.5050167224080269e-06,
+      "loss": 0.3064,
+      "step": 45000
+    },
+    {
+      "epoch": 4.337485270622935,
+      "eval_loss": 0.27599042654037476,
+      "eval_runtime": 40.0219,
+      "eval_samples_per_second": 249.863,
+      "eval_steps_per_second": 31.233,
+      "step": 45000
+    },
+    {
+      "epoch": 4.347124126779875,
+      "grad_norm": 1.1630709171295166,
+      "learning_rate": 1.4949832775919732e-06,
+      "loss": 0.3078,
+      "step": 45100
+    },
+    {
+      "epoch": 4.347124126779875,
+      "eval_loss": 0.27023693919181824,
+      "eval_runtime": 40.0226,
+      "eval_samples_per_second": 249.859,
+      "eval_steps_per_second": 31.232,
+      "step": 45100
+    },
+    {
+      "epoch": 4.356762982936815,
+      "grad_norm": 1.017093300819397,
+      "learning_rate": 1.4849498327759198e-06,
+      "loss": 0.3058,
+      "step": 45200
+    },
+    {
+      "epoch": 4.356762982936815,
+      "eval_loss": 0.26991012692451477,
+      "eval_runtime": 40.1764,
+      "eval_samples_per_second": 248.902,
+      "eval_steps_per_second": 31.113,
+      "step": 45200
+    },
+    {
+      "epoch": 4.366401839093755,
+      "grad_norm": 1.2074520587921143,
+      "learning_rate": 1.4749163879598663e-06,
+      "loss": 0.3068,
+      "step": 45300
+    },
+    {
+      "epoch": 4.366401839093755,
+      "eval_loss": 0.26959505677223206,
+      "eval_runtime": 40.0291,
+      "eval_samples_per_second": 249.819,
+      "eval_steps_per_second": 31.227,
+      "step": 45300
+    },
+    {
+      "epoch": 4.376040695250695,
+      "grad_norm": 1.1239268779754639,
+      "learning_rate": 1.4648829431438129e-06,
+      "loss": 0.3055,
+      "step": 45400
+    },
+    {
+      "epoch": 4.376040695250695,
+      "eval_loss": 0.26887837052345276,
+      "eval_runtime": 40.0551,
+      "eval_samples_per_second": 249.656,
+      "eval_steps_per_second": 31.207,
+      "step": 45400
+    },
+    {
+      "epoch": 4.385679551407635,
+      "grad_norm": 1.1952552795410156,
+      "learning_rate": 1.4548494983277592e-06,
+      "loss": 0.3069,
+      "step": 45500
+    },
+    {
+      "epoch": 4.385679551407635,
+      "eval_loss": 0.268917053937912,
+      "eval_runtime": 40.0378,
+      "eval_samples_per_second": 249.764,
+      "eval_steps_per_second": 31.22,
+      "step": 45500
+    },
+    {
+      "epoch": 4.395318407564575,
+      "grad_norm": 1.2636454105377197,
+      "learning_rate": 1.4448160535117058e-06,
+      "loss": 0.3047,
+      "step": 45600
+    },
+    {
+      "epoch": 4.395318407564575,
+      "eval_loss": 0.2656193673610687,
+      "eval_runtime": 40.2451,
+      "eval_samples_per_second": 248.478,
+      "eval_steps_per_second": 31.06,
+      "step": 45600
+    },
+    {
+      "epoch": 4.404957263721514,
+      "grad_norm": 1.2565031051635742,
+      "learning_rate": 1.4347826086956523e-06,
+      "loss": 0.3032,
+      "step": 45700
+    },
+    {
+      "epoch": 4.404957263721514,
+      "eval_loss": 0.2656501531600952,
+      "eval_runtime": 40.0216,
+      "eval_samples_per_second": 249.865,
+      "eval_steps_per_second": 31.233,
+      "step": 45700
+    },
+    {
+      "epoch": 4.414596119878454,
+      "grad_norm": 1.1796543598175049,
+      "learning_rate": 1.4247491638795989e-06,
+      "loss": 0.305,
+      "step": 45800
+    },
+    {
+      "epoch": 4.414596119878454,
+      "eval_loss": 0.26977530121803284,
+      "eval_runtime": 40.0013,
+      "eval_samples_per_second": 249.992,
+      "eval_steps_per_second": 31.249,
+      "step": 45800
+    },
+    {
+      "epoch": 4.424234976035394,
+      "grad_norm": 1.1382367610931396,
+      "learning_rate": 1.4147157190635452e-06,
+      "loss": 0.3073,
+      "step": 45900
+    },
+    {
+      "epoch": 4.424234976035394,
+      "eval_loss": 0.2679605185985565,
+      "eval_runtime": 40.0148,
+      "eval_samples_per_second": 249.908,
+      "eval_steps_per_second": 31.238,
+      "step": 45900
+    },
+    {
+      "epoch": 4.433873832192334,
+      "grad_norm": 1.094008445739746,
+      "learning_rate": 1.4046822742474917e-06,
+      "loss": 0.3051,
+      "step": 46000
+    },
+    {
+      "epoch": 4.433873832192334,
+      "eval_loss": 0.26570913195610046,
+      "eval_runtime": 40.0087,
+      "eval_samples_per_second": 249.946,
+      "eval_steps_per_second": 31.243,
+      "step": 46000
+    },
+    {
+      "epoch": 4.443512688349274,
+      "grad_norm": 0.9775266051292419,
+      "learning_rate": 1.3946488294314383e-06,
+      "loss": 0.3026,
+      "step": 46100
+    },
+    {
+      "epoch": 4.443512688349274,
+      "eval_loss": 0.2708422541618347,
+      "eval_runtime": 40.2084,
+      "eval_samples_per_second": 248.705,
+      "eval_steps_per_second": 31.088,
+      "step": 46100
+    },
+    {
+      "epoch": 4.4531515445062135,
+      "grad_norm": 1.1069421768188477,
+      "learning_rate": 1.3846153846153846e-06,
+      "loss": 0.3025,
+      "step": 46200
+    },
+    {
+      "epoch": 4.4531515445062135,
+      "eval_loss": 0.2691509425640106,
+      "eval_runtime": 40.0314,
+      "eval_samples_per_second": 249.804,
+      "eval_steps_per_second": 31.226,
+      "step": 46200
+    },
+    {
+      "epoch": 4.462790400663153,
+      "grad_norm": 1.1647151708602905,
+      "learning_rate": 1.374581939799331e-06,
+      "loss": 0.3027,
+      "step": 46300
+    },
+    {
+      "epoch": 4.462790400663153,
+      "eval_loss": 0.26704952120780945,
+      "eval_runtime": 40.0318,
+      "eval_samples_per_second": 249.802,
+      "eval_steps_per_second": 31.225,
+      "step": 46300
+    },
+    {
+      "epoch": 4.472429256820093,
+      "grad_norm": 1.1714391708374023,
+      "learning_rate": 1.3645484949832775e-06,
+      "loss": 0.304,
+      "step": 46400
+    },
+    {
+      "epoch": 4.472429256820093,
+      "eval_loss": 0.2679431140422821,
+      "eval_runtime": 40.04,
+      "eval_samples_per_second": 249.75,
+      "eval_steps_per_second": 31.219,
+      "step": 46400
+    },
+    {
+      "epoch": 4.482068112977033,
+      "grad_norm": 1.1067938804626465,
+      "learning_rate": 1.354515050167224e-06,
+      "loss": 0.3024,
+      "step": 46500
+    },
+    {
+      "epoch": 4.482068112977033,
+      "eval_loss": 0.26625776290893555,
+      "eval_runtime": 40.0422,
+      "eval_samples_per_second": 249.737,
+      "eval_steps_per_second": 31.217,
+      "step": 46500
+    },
+    {
+      "epoch": 4.491706969133973,
+      "grad_norm": 1.1131701469421387,
+      "learning_rate": 1.3444816053511706e-06,
+      "loss": 0.3026,
+      "step": 46600
+    },
+    {
+      "epoch": 4.491706969133973,
+      "eval_loss": 0.2679072916507721,
+      "eval_runtime": 40.2222,
+      "eval_samples_per_second": 248.619,
+      "eval_steps_per_second": 31.077,
+      "step": 46600
+    },
+    {
+      "epoch": 4.501345825290913,
+      "grad_norm": 1.0876435041427612,
+      "learning_rate": 1.334448160535117e-06,
+      "loss": 0.306,
+      "step": 46700
+    },
+    {
+      "epoch": 4.501345825290913,
+      "eval_loss": 0.2695472836494446,
+      "eval_runtime": 40.0345,
+      "eval_samples_per_second": 249.785,
+      "eval_steps_per_second": 31.223,
+      "step": 46700
+    },
+    {
+      "epoch": 4.5109846814478525,
+      "grad_norm": 1.0998533964157104,
+      "learning_rate": 1.3244147157190635e-06,
+      "loss": 0.2985,
+      "step": 46800
+    },
+    {
+      "epoch": 4.5109846814478525,
+      "eval_loss": 0.2631019949913025,
+      "eval_runtime": 40.039,
+      "eval_samples_per_second": 249.756,
+      "eval_steps_per_second": 31.22,
+      "step": 46800
+    },
+    {
+      "epoch": 4.520623537604792,
+      "grad_norm": 1.0389171838760376,
+      "learning_rate": 1.31438127090301e-06,
+      "loss": 0.3037,
+      "step": 46900
+    },
+    {
+      "epoch": 4.520623537604792,
+      "eval_loss": 0.2689988315105438,
+      "eval_runtime": 40.0346,
+      "eval_samples_per_second": 249.784,
+      "eval_steps_per_second": 31.223,
+      "step": 46900
+    },
+    {
+      "epoch": 4.530262393761732,
+      "grad_norm": 1.1371253728866577,
+      "learning_rate": 1.3043478260869566e-06,
+      "loss": 0.3016,
+      "step": 47000
+    },
+    {
+      "epoch": 4.530262393761732,
+      "eval_loss": 0.2646733820438385,
+      "eval_runtime": 40.2249,
+      "eval_samples_per_second": 248.602,
+      "eval_steps_per_second": 31.075,
+      "step": 47000
+    },
+    {
+      "epoch": 4.539901249918672,
+      "grad_norm": 1.1179282665252686,
+      "learning_rate": 1.294314381270903e-06,
+      "loss": 0.3005,
+      "step": 47100
+    },
+    {
+      "epoch": 4.539901249918672,
+      "eval_loss": 0.2690794765949249,
+      "eval_runtime": 40.071,
+      "eval_samples_per_second": 249.557,
+      "eval_steps_per_second": 31.195,
+      "step": 47100
+    },
+    {
+      "epoch": 4.549540106075612,
+      "grad_norm": 1.0551939010620117,
+      "learning_rate": 1.2842809364548495e-06,
+      "loss": 0.3041,
+      "step": 47200
+    },
+    {
+      "epoch": 4.549540106075612,
+      "eval_loss": 0.26324746012687683,
+      "eval_runtime": 40.2273,
+      "eval_samples_per_second": 248.587,
+      "eval_steps_per_second": 31.073,
+      "step": 47200
+    },
+    {
+      "epoch": 4.559178962232552,
+      "grad_norm": 1.0626869201660156,
+      "learning_rate": 1.274247491638796e-06,
+      "loss": 0.3038,
+      "step": 47300
+    },
+    {
+      "epoch": 4.559178962232552,
+      "eval_loss": 0.26168400049209595,
+      "eval_runtime": 40.2137,
+      "eval_samples_per_second": 248.672,
+      "eval_steps_per_second": 31.084,
+      "step": 47300
+    },
+    {
+      "epoch": 4.568817818389491,
+      "grad_norm": 1.1616398096084595,
+      "learning_rate": 1.2642140468227424e-06,
+      "loss": 0.3006,
+      "step": 47400
+    },
+    {
+      "epoch": 4.568817818389491,
+      "eval_loss": 0.2649966776371002,
+      "eval_runtime": 40.0842,
+      "eval_samples_per_second": 249.475,
+      "eval_steps_per_second": 31.184,
+      "step": 47400
+    },
+    {
+      "epoch": 4.578456674546431,
+      "grad_norm": 1.1772770881652832,
+      "learning_rate": 1.254180602006689e-06,
+      "loss": 0.303,
+      "step": 47500
+    },
+    {
+      "epoch": 4.578456674546431,
+      "eval_loss": 0.2650017738342285,
+      "eval_runtime": 40.0723,
+      "eval_samples_per_second": 249.549,
+      "eval_steps_per_second": 31.194,
+      "step": 47500
+    },
+    {
+      "epoch": 4.588095530703371,
+      "grad_norm": 1.1673014163970947,
+      "learning_rate": 1.2441471571906355e-06,
+      "loss": 0.3009,
+      "step": 47600
+    },
+    {
+      "epoch": 4.588095530703371,
+      "eval_loss": 0.2697572112083435,
+      "eval_runtime": 40.0689,
+      "eval_samples_per_second": 249.57,
+      "eval_steps_per_second": 31.196,
+      "step": 47600
+    },
+    {
+      "epoch": 4.597734386860312,
+      "grad_norm": 1.1624304056167603,
+      "learning_rate": 1.234113712374582e-06,
+      "loss": 0.3049,
+      "step": 47700
+    },
+    {
+      "epoch": 4.597734386860312,
+      "eval_loss": 0.26627808809280396,
+      "eval_runtime": 40.1927,
+      "eval_samples_per_second": 248.801,
+      "eval_steps_per_second": 31.1,
+      "step": 47700
+    },
+    {
+      "epoch": 4.607373243017252,
+      "grad_norm": 1.0767208337783813,
+      "learning_rate": 1.2240802675585284e-06,
+      "loss": 0.3021,
+      "step": 47800
+    },
+    {
+      "epoch": 4.607373243017252,
+      "eval_loss": 0.2637022137641907,
+      "eval_runtime": 40.0396,
+      "eval_samples_per_second": 249.753,
+      "eval_steps_per_second": 31.219,
+      "step": 47800
+    },
+    {
+      "epoch": 4.617012099174191,
+      "grad_norm": 1.1391677856445312,
+      "learning_rate": 1.214046822742475e-06,
+      "loss": 0.3013,
+      "step": 47900
+    },
+    {
+      "epoch": 4.617012099174191,
+      "eval_loss": 0.2667087912559509,
+      "eval_runtime": 40.0721,
+      "eval_samples_per_second": 249.55,
+      "eval_steps_per_second": 31.194,
+      "step": 47900
+    },
+    {
+      "epoch": 4.626650955331131,
+      "grad_norm": 1.1649397611618042,
+      "learning_rate": 1.2040133779264215e-06,
+      "loss": 0.3002,
+      "step": 48000
+    },
+    {
+      "epoch": 4.626650955331131,
+      "eval_loss": 0.27446261048316956,
+      "eval_runtime": 40.0365,
+      "eval_samples_per_second": 249.772,
+      "eval_steps_per_second": 31.222,
+      "step": 48000
+    },
+    {
+      "epoch": 4.636289811488071,
+      "grad_norm": 1.074158787727356,
+      "learning_rate": 1.193979933110368e-06,
+      "loss": 0.3015,
+      "step": 48100
+    },
+    {
+      "epoch": 4.636289811488071,
+      "eval_loss": 0.2660681903362274,
+      "eval_runtime": 40.0402,
+      "eval_samples_per_second": 249.749,
+      "eval_steps_per_second": 31.219,
+      "step": 48100
+    },
+    {
+      "epoch": 4.645928667645011,
+      "grad_norm": 1.1402333974838257,
+      "learning_rate": 1.1839464882943144e-06,
+      "loss": 0.3013,
+      "step": 48200
+    },
+    {
+      "epoch": 4.645928667645011,
+      "eval_loss": 0.2649831175804138,
+      "eval_runtime": 40.1953,
+      "eval_samples_per_second": 248.785,
+      "eval_steps_per_second": 31.098,
+      "step": 48200
+    },
+    {
+      "epoch": 4.655567523801951,
+      "grad_norm": 1.1394840478897095,
+      "learning_rate": 1.173913043478261e-06,
+      "loss": 0.2998,
+      "step": 48300
+    },
+    {
+      "epoch": 4.655567523801951,
+      "eval_loss": 0.26467251777648926,
+      "eval_runtime": 40.0301,
+      "eval_samples_per_second": 249.812,
+      "eval_steps_per_second": 31.226,
+      "step": 48300
+    },
+    {
+      "epoch": 4.6652063799588905,
+      "grad_norm": 1.151784896850586,
+      "learning_rate": 1.1638795986622075e-06,
+      "loss": 0.2998,
+      "step": 48400
+    },
+    {
+      "epoch": 4.6652063799588905,
+      "eval_loss": 0.2676987946033478,
+      "eval_runtime": 40.0304,
+      "eval_samples_per_second": 249.81,
+      "eval_steps_per_second": 31.226,
+      "step": 48400
+    },
+    {
+      "epoch": 4.67484523611583,
+      "grad_norm": 1.134567379951477,
+      "learning_rate": 1.153846153846154e-06,
+      "loss": 0.3026,
+      "step": 48500
+    },
+    {
+      "epoch": 4.67484523611583,
+      "eval_loss": 0.2631213068962097,
+      "eval_runtime": 40.0346,
+      "eval_samples_per_second": 249.784,
+      "eval_steps_per_second": 31.223,
+      "step": 48500
+    },
+    {
+      "epoch": 4.68448409227277,
+      "grad_norm": 1.1139414310455322,
+      "learning_rate": 1.1438127090301004e-06,
+      "loss": 0.2987,
+      "step": 48600
+    },
+    {
+      "epoch": 4.68448409227277,
+      "eval_loss": 0.2675442397594452,
+      "eval_runtime": 40.2414,
+      "eval_samples_per_second": 248.5,
+      "eval_steps_per_second": 31.063,
+      "step": 48600
+    },
+    {
+      "epoch": 4.69412294842971,
+      "grad_norm": 1.140663504600525,
+      "learning_rate": 1.133779264214047e-06,
+      "loss": 0.3005,
+      "step": 48700
+    },
+    {
+      "epoch": 4.69412294842971,
+      "eval_loss": 0.26258584856987,
+      "eval_runtime": 40.0562,
+      "eval_samples_per_second": 249.649,
+      "eval_steps_per_second": 31.206,
+      "step": 48700
+    },
+    {
+      "epoch": 4.70376180458665,
+      "grad_norm": 1.1587011814117432,
+      "learning_rate": 1.1237458193979933e-06,
+      "loss": 0.2998,
+      "step": 48800
+    },
+    {
+      "epoch": 4.70376180458665,
+      "eval_loss": 0.2626008987426758,
+      "eval_runtime": 40.0476,
+      "eval_samples_per_second": 249.703,
+      "eval_steps_per_second": 31.213,
+      "step": 48800
+    },
+    {
+      "epoch": 4.71340066074359,
+      "grad_norm": 1.2201813459396362,
+      "learning_rate": 1.1137123745819398e-06,
+      "loss": 0.2994,
+      "step": 48900
+    },
+    {
+      "epoch": 4.71340066074359,
+      "eval_loss": 0.26159748435020447,
+      "eval_runtime": 40.0689,
+      "eval_samples_per_second": 249.57,
+      "eval_steps_per_second": 31.196,
+      "step": 48900
+    },
+    {
+      "epoch": 4.7230395169005295,
+      "grad_norm": 1.0543116331100464,
+      "learning_rate": 1.1036789297658862e-06,
+      "loss": 0.3009,
+      "step": 49000
+    },
+    {
+      "epoch": 4.7230395169005295,
+      "eval_loss": 0.2661503553390503,
+      "eval_runtime": 40.0833,
+      "eval_samples_per_second": 249.481,
+      "eval_steps_per_second": 31.185,
+      "step": 49000
+    },
+    {
+      "epoch": 4.732678373057469,
+      "grad_norm": 1.0388127565383911,
+      "learning_rate": 1.0936454849498327e-06,
+      "loss": 0.3005,
+      "step": 49100
+    },
+    {
+      "epoch": 4.732678373057469,
+      "eval_loss": 0.2635759711265564,
+      "eval_runtime": 40.2341,
+      "eval_samples_per_second": 248.545,
+      "eval_steps_per_second": 31.068,
+      "step": 49100
+    },
+    {
+      "epoch": 4.742317229214409,
+      "grad_norm": 1.133090853691101,
+      "learning_rate": 1.0836120401337793e-06,
+      "loss": 0.298,
+      "step": 49200
+    },
+    {
+      "epoch": 4.742317229214409,
+      "eval_loss": 0.25813475251197815,
+      "eval_runtime": 40.0481,
+      "eval_samples_per_second": 249.7,
+      "eval_steps_per_second": 31.212,
+      "step": 49200
+    },
+    {
+      "epoch": 4.751956085371349,
+      "grad_norm": 1.149713158607483,
+      "learning_rate": 1.0735785953177258e-06,
+      "loss": 0.3011,
+      "step": 49300
+    },
+    {
+      "epoch": 4.751956085371349,
+      "eval_loss": 0.26733270287513733,
+      "eval_runtime": 40.0514,
+      "eval_samples_per_second": 249.679,
+      "eval_steps_per_second": 31.21,
+      "step": 49300
+    },
+    {
+      "epoch": 4.761594941528289,
+      "grad_norm": 1.1020680665969849,
+      "learning_rate": 1.0635451505016722e-06,
+      "loss": 0.3003,
+      "step": 49400
+    },
+    {
+      "epoch": 4.761594941528289,
+      "eval_loss": 0.26301661133766174,
+      "eval_runtime": 40.0244,
+      "eval_samples_per_second": 249.848,
+      "eval_steps_per_second": 31.231,
+      "step": 49400
+    },
+    {
+      "epoch": 4.771233797685229,
+      "grad_norm": 1.1501930952072144,
+      "learning_rate": 1.0535117056856187e-06,
+      "loss": 0.2994,
+      "step": 49500
+    },
+    {
+      "epoch": 4.771233797685229,
+      "eval_loss": 0.2652057111263275,
+      "eval_runtime": 40.203,
+      "eval_samples_per_second": 248.737,
+      "eval_steps_per_second": 31.092,
+      "step": 49500
+    },
+    {
+      "epoch": 4.780872653842168,
+      "grad_norm": 1.0680407285690308,
+      "learning_rate": 1.0434782608695653e-06,
+      "loss": 0.3026,
+      "step": 49600
+    },
+    {
+      "epoch": 4.780872653842168,
+      "eval_loss": 0.26839718222618103,
+      "eval_runtime": 40.0443,
+      "eval_samples_per_second": 249.723,
+      "eval_steps_per_second": 31.215,
+      "step": 49600
+    },
+    {
+      "epoch": 4.790511509999108,
+      "grad_norm": 1.1714107990264893,
+      "learning_rate": 1.0334448160535118e-06,
+      "loss": 0.3,
+      "step": 49700
+    },
+    {
+      "epoch": 4.790511509999108,
+      "eval_loss": 0.26893192529678345,
+      "eval_runtime": 40.0425,
+      "eval_samples_per_second": 249.735,
+      "eval_steps_per_second": 31.217,
+      "step": 49700
+    },
+    {
+      "epoch": 4.800150366156048,
+      "grad_norm": 1.216985821723938,
+      "learning_rate": 1.0234113712374581e-06,
+      "loss": 0.2996,
+      "step": 49800
+    },
+    {
+      "epoch": 4.800150366156048,
+      "eval_loss": 0.2666034698486328,
+      "eval_runtime": 40.0734,
+      "eval_samples_per_second": 249.542,
+      "eval_steps_per_second": 31.193,
+      "step": 49800
+    },
+    {
+      "epoch": 4.809789222312988,
+      "grad_norm": 1.0444952249526978,
+      "learning_rate": 1.0133779264214047e-06,
+      "loss": 0.2991,
+      "step": 49900
+    },
+    {
+      "epoch": 4.809789222312988,
+      "eval_loss": 0.26440900564193726,
+      "eval_runtime": 40.0587,
+      "eval_samples_per_second": 249.634,
+      "eval_steps_per_second": 31.204,
+      "step": 49900
+    },
+    {
+      "epoch": 4.819428078469928,
+      "grad_norm": 1.002389907836914,
+      "learning_rate": 1.0033444816053512e-06,
+      "loss": 0.3,
+      "step": 50000
+    },
+    {
+      "epoch": 4.819428078469928,
+      "eval_loss": 0.2678601145744324,
+      "eval_runtime": 40.2234,
+      "eval_samples_per_second": 248.612,
+      "eval_steps_per_second": 31.076,
+      "step": 50000
+    },
+    {
+      "epoch": 4.8290669346268675,
+      "grad_norm": 1.083066701889038,
+      "learning_rate": 9.933110367892976e-07,
+      "loss": 0.3003,
+      "step": 50100
+    },
+    {
+      "epoch": 4.8290669346268675,
+      "eval_loss": 0.26301416754722595,
+      "eval_runtime": 40.0381,
+      "eval_samples_per_second": 249.762,
+      "eval_steps_per_second": 31.22,
+      "step": 50100
+    },
+    {
+      "epoch": 4.838705790783807,
+      "grad_norm": 1.0669959783554077,
+      "learning_rate": 9.832775919732441e-07,
+      "loss": 0.2988,
+      "step": 50200
+    },
+    {
+      "epoch": 4.838705790783807,
+      "eval_loss": 0.2668324112892151,
+      "eval_runtime": 40.0364,
+      "eval_samples_per_second": 249.772,
+      "eval_steps_per_second": 31.222,
+      "step": 50200
+    },
+    {
+      "epoch": 4.848344646940747,
+      "grad_norm": 1.253390908241272,
+      "learning_rate": 9.732441471571907e-07,
+      "loss": 0.2984,
+      "step": 50300
+    },
+    {
+      "epoch": 4.848344646940747,
+      "eval_loss": 0.2622229754924774,
+      "eval_runtime": 40.0375,
+      "eval_samples_per_second": 249.766,
+      "eval_steps_per_second": 31.221,
+      "step": 50300
+    },
+    {
+      "epoch": 4.857983503097687,
+      "grad_norm": 1.0932843685150146,
+      "learning_rate": 9.632107023411372e-07,
+      "loss": 0.2997,
+      "step": 50400
+    },
+    {
+      "epoch": 4.857983503097687,
+      "eval_loss": 0.2585309147834778,
+      "eval_runtime": 40.0358,
+      "eval_samples_per_second": 249.776,
+      "eval_steps_per_second": 31.222,
+      "step": 50400
+    },
+    {
+      "epoch": 4.867622359254627,
+      "grad_norm": 1.089406967163086,
+      "learning_rate": 9.531772575250837e-07,
+      "loss": 0.2995,
+      "step": 50500
+    },
+    {
+      "epoch": 4.867622359254627,
+      "eval_loss": 0.26007798314094543,
+      "eval_runtime": 40.2316,
+      "eval_samples_per_second": 248.561,
+      "eval_steps_per_second": 31.07,
+      "step": 50500
+    },
+    {
+      "epoch": 4.877261215411567,
+      "grad_norm": 1.0903793573379517,
+      "learning_rate": 9.431438127090301e-07,
+      "loss": 0.2966,
+      "step": 50600
+    },
+    {
+      "epoch": 4.877261215411567,
+      "eval_loss": 0.26448899507522583,
+      "eval_runtime": 40.0444,
+      "eval_samples_per_second": 249.723,
+      "eval_steps_per_second": 31.215,
+      "step": 50600
+    },
+    {
+      "epoch": 4.8869000715685065,
+      "grad_norm": 1.1804698705673218,
+      "learning_rate": 9.331103678929767e-07,
+      "loss": 0.2975,
+      "step": 50700
+    },
+    {
+      "epoch": 4.8869000715685065,
+      "eval_loss": 0.2629201114177704,
+      "eval_runtime": 40.056,
+      "eval_samples_per_second": 249.65,
+      "eval_steps_per_second": 31.206,
+      "step": 50700
+    },
+    {
+      "epoch": 4.896538927725447,
+      "grad_norm": 1.0072447061538696,
+      "learning_rate": 9.230769230769231e-07,
+      "loss": 0.2971,
+      "step": 50800
+    },
+    {
+      "epoch": 4.896538927725447,
+      "eval_loss": 0.2699527442455292,
+      "eval_runtime": 40.2322,
+      "eval_samples_per_second": 248.557,
+      "eval_steps_per_second": 31.07,
+      "step": 50800
+    },
+    {
+      "epoch": 4.906177783882387,
+      "grad_norm": 1.1153916120529175,
+      "learning_rate": 9.130434782608697e-07,
+      "loss": 0.2991,
+      "step": 50900
+    },
+    {
+      "epoch": 4.906177783882387,
+      "eval_loss": 0.2644394040107727,
+      "eval_runtime": 40.0519,
+      "eval_samples_per_second": 249.676,
+      "eval_steps_per_second": 31.209,
+      "step": 50900
+    },
+    {
+      "epoch": 4.915816640039327,
+      "grad_norm": 1.1166541576385498,
+      "learning_rate": 9.030100334448161e-07,
+      "loss": 0.2985,
+      "step": 51000
+    },
+    {
+      "epoch": 4.915816640039327,
+      "eval_loss": 0.2613272964954376,
+      "eval_runtime": 40.0716,
+      "eval_samples_per_second": 249.553,
+      "eval_steps_per_second": 31.194,
+      "step": 51000
+    },
+    {
+      "epoch": 4.925455496196267,
+      "grad_norm": 1.062063455581665,
+      "learning_rate": 8.929765886287627e-07,
+      "loss": 0.2959,
+      "step": 51100
+    },
+    {
+      "epoch": 4.925455496196267,
+      "eval_loss": 0.2620924413204193,
+      "eval_runtime": 40.0632,
+      "eval_samples_per_second": 249.605,
+      "eval_steps_per_second": 31.201,
+      "step": 51100
+    },
+    {
+      "epoch": 4.9350943523532065,
+      "grad_norm": 1.0991202592849731,
+      "learning_rate": 8.829431438127091e-07,
+      "loss": 0.2963,
+      "step": 51200
+    },
+    {
+      "epoch": 4.9350943523532065,
+      "eval_loss": 0.2621663510799408,
+      "eval_runtime": 40.2344,
+      "eval_samples_per_second": 248.544,
+      "eval_steps_per_second": 31.068,
+      "step": 51200
+    },
+    {
+      "epoch": 4.944733208510146,
+      "grad_norm": 1.2747081518173218,
+      "learning_rate": 8.729096989966555e-07,
+      "loss": 0.2981,
+      "step": 51300
+    },
+    {
+      "epoch": 4.944733208510146,
+      "eval_loss": 0.2616971433162689,
+      "eval_runtime": 40.0357,
+      "eval_samples_per_second": 249.777,
+      "eval_steps_per_second": 31.222,
+      "step": 51300
+    },
+    {
+      "epoch": 4.954372064667086,
+      "grad_norm": 1.046461820602417,
+      "learning_rate": 8.628762541806019e-07,
+      "loss": 0.2988,
+      "step": 51400
+    },
+    {
+      "epoch": 4.954372064667086,
+      "eval_loss": 0.26591652631759644,
+      "eval_runtime": 40.0483,
+      "eval_samples_per_second": 249.699,
+      "eval_steps_per_second": 31.212,
+      "step": 51400
+    },
+    {
+      "epoch": 4.964010920824026,
+      "grad_norm": 1.3319487571716309,
+      "learning_rate": 8.528428093645485e-07,
+      "loss": 0.2971,
+      "step": 51500
+    },
+    {
+      "epoch": 4.964010920824026,
+      "eval_loss": 0.2656097412109375,
+      "eval_runtime": 40.0468,
+      "eval_samples_per_second": 249.708,
+      "eval_steps_per_second": 31.214,
+      "step": 51500
+    },
+    {
+      "epoch": 4.973649776980966,
+      "grad_norm": 1.2510554790496826,
+      "learning_rate": 8.428093645484949e-07,
+      "loss": 0.3005,
+      "step": 51600
+    },
+    {
+      "epoch": 4.973649776980966,
+      "eval_loss": 0.2569749653339386,
+      "eval_runtime": 40.0994,
+      "eval_samples_per_second": 249.38,
+      "eval_steps_per_second": 31.173,
+      "step": 51600
+    },
+    {
+      "epoch": 4.983288633137906,
+      "grad_norm": 1.045254111289978,
+      "learning_rate": 8.327759197324414e-07,
+      "loss": 0.2978,
+      "step": 51700
+    },
+    {
+      "epoch": 4.983288633137906,
+      "eval_loss": 0.26062557101249695,
+      "eval_runtime": 40.2488,
+      "eval_samples_per_second": 248.454,
+      "eval_steps_per_second": 31.057,
+      "step": 51700
+    },
+    {
+      "epoch": 4.992927489294845,
+      "grad_norm": 1.1483185291290283,
+      "learning_rate": 8.227424749163879e-07,
+      "loss": 0.2976,
+      "step": 51800
+    },
+    {
+      "epoch": 4.992927489294845,
+      "eval_loss": 0.2605762183666229,
+      "eval_runtime": 40.0482,
+      "eval_samples_per_second": 249.699,
+      "eval_steps_per_second": 31.212,
+      "step": 51800
+    },
+    {
+      "epoch": 5.002566345451785,
+      "grad_norm": 1.193968653678894,
+      "learning_rate": 8.127090301003344e-07,
+      "loss": 0.2958,
+      "step": 51900
+    },
+    {
+      "epoch": 5.002566345451785,
+      "eval_loss": 0.26294004917144775,
+      "eval_runtime": 40.0625,
+      "eval_samples_per_second": 249.61,
+      "eval_steps_per_second": 31.201,
+      "step": 51900
+    },
+    {
+      "epoch": 5.012205201608725,
+      "grad_norm": 1.3349761962890625,
+      "learning_rate": 8.026755852842809e-07,
+      "loss": 0.2966,
+      "step": 52000
+    },
+    {
+      "epoch": 5.012205201608725,
+      "eval_loss": 0.26609280705451965,
+      "eval_runtime": 40.059,
+      "eval_samples_per_second": 249.632,
+      "eval_steps_per_second": 31.204,
+      "step": 52000
+    },
+    {
+      "epoch": 5.021844057765665,
+      "grad_norm": 1.1316077709197998,
+      "learning_rate": 7.926421404682274e-07,
+      "loss": 0.2988,
+      "step": 52100
+    },
+    {
+      "epoch": 5.021844057765665,
+      "eval_loss": 0.26353102922439575,
+      "eval_runtime": 40.0857,
+      "eval_samples_per_second": 249.465,
+      "eval_steps_per_second": 31.183,
+      "step": 52100
+    },
+    {
+      "epoch": 5.031482913922605,
+      "grad_norm": 1.1322309970855713,
+      "learning_rate": 7.826086956521739e-07,
+      "loss": 0.2979,
+      "step": 52200
+    },
+    {
+      "epoch": 5.031482913922605,
+      "eval_loss": 0.2596350908279419,
+      "eval_runtime": 40.2317,
+      "eval_samples_per_second": 248.56,
+      "eval_steps_per_second": 31.07,
+      "step": 52200
+    },
+    {
+      "epoch": 5.0411217700795445,
+      "grad_norm": 1.1747742891311646,
+      "learning_rate": 7.725752508361204e-07,
+      "loss": 0.2946,
+      "step": 52300
+    },
+    {
+      "epoch": 5.0411217700795445,
+      "eval_loss": 0.2601070702075958,
+      "eval_runtime": 40.0484,
+      "eval_samples_per_second": 249.698,
+      "eval_steps_per_second": 31.212,
+      "step": 52300
+    },
+    {
+      "epoch": 5.050760626236484,
+      "grad_norm": 1.1011712551116943,
+      "learning_rate": 7.625418060200669e-07,
+      "loss": 0.2953,
+      "step": 52400
+    },
+    {
+      "epoch": 5.050760626236484,
+      "eval_loss": 0.25822940468788147,
+      "eval_runtime": 40.0407,
+      "eval_samples_per_second": 249.746,
+      "eval_steps_per_second": 31.218,
+      "step": 52400
+    },
+    {
+      "epoch": 5.060399482393424,
+      "grad_norm": 1.0672869682312012,
+      "learning_rate": 7.525083612040134e-07,
+      "loss": 0.2986,
+      "step": 52500
+    },
+    {
+      "epoch": 5.060399482393424,
+      "eval_loss": 0.25989893078804016,
+      "eval_runtime": 40.0772,
+      "eval_samples_per_second": 249.518,
+      "eval_steps_per_second": 31.19,
+      "step": 52500
+    },
+    {
+      "epoch": 5.070038338550364,
+      "grad_norm": 1.1346361637115479,
+      "learning_rate": 7.424749163879599e-07,
+      "loss": 0.2952,
+      "step": 52600
+    },
+    {
+      "epoch": 5.070038338550364,
+      "eval_loss": 0.26356619596481323,
+      "eval_runtime": 40.0484,
+      "eval_samples_per_second": 249.698,
+      "eval_steps_per_second": 31.212,
+      "step": 52600
+    },
+    {
+      "epoch": 5.079677194707304,
+      "grad_norm": 1.152971625328064,
+      "learning_rate": 7.324414715719064e-07,
+      "loss": 0.2967,
+      "step": 52700
+    },
+    {
+      "epoch": 5.079677194707304,
+      "eval_loss": 0.26063260436058044,
+      "eval_runtime": 40.2268,
+      "eval_samples_per_second": 248.591,
+      "eval_steps_per_second": 31.074,
+      "step": 52700
+    },
+    {
+      "epoch": 5.089316050864244,
+      "grad_norm": 1.0548993349075317,
+      "learning_rate": 7.224080267558529e-07,
+      "loss": 0.2974,
+      "step": 52800
+    },
+    {
+      "epoch": 5.089316050864244,
+      "eval_loss": 0.2643623650074005,
+      "eval_runtime": 40.0545,
+      "eval_samples_per_second": 249.66,
+      "eval_steps_per_second": 31.208,
+      "step": 52800
+    },
+    {
+      "epoch": 5.0989549070211835,
+      "grad_norm": 1.1734200716018677,
+      "learning_rate": 7.123745819397994e-07,
+      "loss": 0.2983,
+      "step": 52900
+    },
+    {
+      "epoch": 5.0989549070211835,
+      "eval_loss": 0.2603817284107208,
+      "eval_runtime": 40.0535,
+      "eval_samples_per_second": 249.666,
+      "eval_steps_per_second": 31.208,
+      "step": 52900
+    },
+    {
+      "epoch": 5.108593763178123,
+      "grad_norm": 1.1038990020751953,
+      "learning_rate": 7.023411371237459e-07,
+      "loss": 0.2958,
+      "step": 53000
+    },
+    {
+      "epoch": 5.108593763178123,
+      "eval_loss": 0.25843310356140137,
+      "eval_runtime": 40.062,
+      "eval_samples_per_second": 249.613,
+      "eval_steps_per_second": 31.202,
+      "step": 53000
+    },
+    {
+      "epoch": 5.118232619335063,
+      "grad_norm": 1.1007676124572754,
+      "learning_rate": 6.923076923076923e-07,
+      "loss": 0.2946,
+      "step": 53100
+    },
+    {
+      "epoch": 5.118232619335063,
+      "eval_loss": 0.2619819939136505,
+      "eval_runtime": 40.3148,
+      "eval_samples_per_second": 248.048,
+      "eval_steps_per_second": 31.006,
+      "step": 53100
+    },
+    {
+      "epoch": 5.127871475492003,
+      "grad_norm": 1.149030089378357,
+      "learning_rate": 6.822742474916388e-07,
+      "loss": 0.2974,
+      "step": 53200
+    },
+    {
+      "epoch": 5.127871475492003,
+      "eval_loss": 0.25853490829467773,
+      "eval_runtime": 40.1284,
+      "eval_samples_per_second": 249.2,
+      "eval_steps_per_second": 31.15,
+      "step": 53200
+    },
+    {
+      "epoch": 5.137510331648944,
+      "grad_norm": 1.102474331855774,
+      "learning_rate": 6.722408026755853e-07,
+      "loss": 0.2974,
+      "step": 53300
+    },
+    {
+      "epoch": 5.137510331648944,
+      "eval_loss": 0.26003292202949524,
+      "eval_runtime": 40.0545,
+      "eval_samples_per_second": 249.66,
+      "eval_steps_per_second": 31.207,
+      "step": 53300
+    },
+    {
+      "epoch": 5.1471491878058835,
+      "grad_norm": 1.1578916311264038,
+      "learning_rate": 6.622073578595318e-07,
+      "loss": 0.2969,
+      "step": 53400
+    },
+    {
+      "epoch": 5.1471491878058835,
+      "eval_loss": 0.2637874186038971,
+      "eval_runtime": 40.0602,
+      "eval_samples_per_second": 249.625,
+      "eval_steps_per_second": 31.203,
+      "step": 53400
+    },
+    {
+      "epoch": 5.156788043962823,
+      "grad_norm": 1.0911157131195068,
+      "learning_rate": 6.521739130434783e-07,
+      "loss": 0.2973,
+      "step": 53500
+    },
+    {
+      "epoch": 5.156788043962823,
+      "eval_loss": 0.26259294152259827,
+      "eval_runtime": 40.0592,
+      "eval_samples_per_second": 249.631,
+      "eval_steps_per_second": 31.204,
+      "step": 53500
+    },
+    {
+      "epoch": 5.166426900119763,
+      "grad_norm": 1.0731229782104492,
+      "learning_rate": 6.421404682274248e-07,
+      "loss": 0.2985,
+      "step": 53600
+    },
+    {
+      "epoch": 5.166426900119763,
+      "eval_loss": 0.2570517659187317,
+      "eval_runtime": 40.0808,
+      "eval_samples_per_second": 249.496,
+      "eval_steps_per_second": 31.187,
+      "step": 53600
+    },
+    {
+      "epoch": 5.176065756276703,
+      "grad_norm": 1.133985996246338,
+      "learning_rate": 6.321070234113712e-07,
+      "loss": 0.298,
+      "step": 53700
+    },
+    {
+      "epoch": 5.176065756276703,
+      "eval_loss": 0.2622462511062622,
+      "eval_runtime": 40.2293,
+      "eval_samples_per_second": 248.575,
+      "eval_steps_per_second": 31.072,
+      "step": 53700
+    },
+    {
+      "epoch": 5.185704612433643,
+      "grad_norm": 1.1667418479919434,
+      "learning_rate": 6.220735785953178e-07,
+      "loss": 0.2951,
+      "step": 53800
+    },
+    {
+      "epoch": 5.185704612433643,
+      "eval_loss": 0.2598443031311035,
+      "eval_runtime": 40.0513,
+      "eval_samples_per_second": 249.68,
+      "eval_steps_per_second": 31.21,
+      "step": 53800
+    },
+    {
+      "epoch": 5.195343468590583,
+      "grad_norm": 1.1579954624176025,
+      "learning_rate": 6.120401337792642e-07,
+      "loss": 0.2949,
+      "step": 53900
+    },
+    {
+      "epoch": 5.195343468590583,
+      "eval_loss": 0.25939348340034485,
+      "eval_runtime": 40.0758,
+      "eval_samples_per_second": 249.527,
+      "eval_steps_per_second": 31.191,
+      "step": 53900
+    },
+    {
+      "epoch": 5.204982324747522,
+      "grad_norm": 1.0994336605072021,
+      "learning_rate": 6.020066889632107e-07,
+      "loss": 0.2955,
+      "step": 54000
+    },
+    {
+      "epoch": 5.204982324747522,
+      "eval_loss": 0.2639375925064087,
+      "eval_runtime": 40.0757,
+      "eval_samples_per_second": 249.528,
+      "eval_steps_per_second": 31.191,
+      "step": 54000
+    },
+    {
+      "epoch": 5.214621180904462,
+      "grad_norm": 1.2090504169464111,
+      "learning_rate": 5.919732441471572e-07,
+      "loss": 0.2934,
+      "step": 54100
+    },
+    {
+      "epoch": 5.214621180904462,
+      "eval_loss": 0.2580001652240753,
+      "eval_runtime": 40.2509,
+      "eval_samples_per_second": 248.442,
+      "eval_steps_per_second": 31.055,
+      "step": 54100
+    },
+    {
+      "epoch": 5.224260037061402,
+      "grad_norm": 1.0528264045715332,
+      "learning_rate": 5.819397993311037e-07,
+      "loss": 0.2975,
+      "step": 54200
+    },
+    {
+      "epoch": 5.224260037061402,
+      "eval_loss": 0.2598421275615692,
+      "eval_runtime": 40.0571,
+      "eval_samples_per_second": 249.644,
+      "eval_steps_per_second": 31.205,
+      "step": 54200
+    },
+    {
+      "epoch": 5.233898893218342,
+      "grad_norm": 1.1567331552505493,
+      "learning_rate": 5.719063545150502e-07,
+      "loss": 0.2978,
+      "step": 54300
+    },
+    {
+      "epoch": 5.233898893218342,
+      "eval_loss": 0.26616182923316956,
+      "eval_runtime": 40.2149,
+      "eval_samples_per_second": 248.664,
+      "eval_steps_per_second": 31.083,
+      "step": 54300
+    },
+    {
+      "epoch": 5.243537749375282,
+      "grad_norm": 1.0571229457855225,
+      "learning_rate": 5.618729096989966e-07,
+      "loss": 0.2942,
+      "step": 54400
+    },
+    {
+      "epoch": 5.243537749375282,
+      "eval_loss": 0.26477307081222534,
+      "eval_runtime": 40.0949,
+      "eval_samples_per_second": 249.408,
+      "eval_steps_per_second": 31.176,
+      "step": 54400
+    },
+    {
+      "epoch": 5.2531766055322215,
+      "grad_norm": 1.148074984550476,
+      "learning_rate": 5.518394648829431e-07,
+      "loss": 0.295,
+      "step": 54500
+    },
+    {
+      "epoch": 5.2531766055322215,
+      "eval_loss": 0.2539403736591339,
+      "eval_runtime": 40.2349,
+      "eval_samples_per_second": 248.54,
+      "eval_steps_per_second": 31.068,
+      "step": 54500
+    },
+    {
+      "epoch": 5.262815461689161,
+      "grad_norm": 1.0529351234436035,
+      "learning_rate": 5.418060200668896e-07,
+      "loss": 0.2948,
+      "step": 54600
+    },
+    {
+      "epoch": 5.262815461689161,
+      "eval_loss": 0.26054322719573975,
+      "eval_runtime": 40.1935,
+      "eval_samples_per_second": 248.796,
+      "eval_steps_per_second": 31.1,
+      "step": 54600
+    },
+    {
+      "epoch": 5.272454317846101,
+      "grad_norm": 1.1665048599243164,
+      "learning_rate": 5.317725752508361e-07,
+      "loss": 0.2939,
+      "step": 54700
+    },
+    {
+      "epoch": 5.272454317846101,
+      "eval_loss": 0.2621845304965973,
+      "eval_runtime": 40.2294,
+      "eval_samples_per_second": 248.575,
+      "eval_steps_per_second": 31.072,
+      "step": 54700
+    },
+    {
+      "epoch": 5.282093174003041,
+      "grad_norm": 1.1923383474349976,
+      "learning_rate": 5.217391304347826e-07,
+      "loss": 0.2943,
+      "step": 54800
+    },
+    {
+      "epoch": 5.282093174003041,
+      "eval_loss": 0.2575650215148926,
+      "eval_runtime": 40.1924,
+      "eval_samples_per_second": 248.803,
+      "eval_steps_per_second": 31.1,
+      "step": 54800
+    },
+    {
+      "epoch": 5.291732030159981,
+      "grad_norm": 1.0694471597671509,
+      "learning_rate": 5.117056856187291e-07,
+      "loss": 0.2941,
+      "step": 54900
+    },
+    {
+      "epoch": 5.291732030159981,
+      "eval_loss": 0.25641068816185,
+      "eval_runtime": 40.2068,
+      "eval_samples_per_second": 248.714,
+      "eval_steps_per_second": 31.089,
+      "step": 54900
+    },
+    {
+      "epoch": 5.301370886316921,
+      "grad_norm": 1.207488775253296,
+      "learning_rate": 5.016722408026756e-07,
+      "loss": 0.2945,
+      "step": 55000
+    },
+    {
+      "epoch": 5.301370886316921,
+      "eval_loss": 0.2628210186958313,
+      "eval_runtime": 40.2419,
+      "eval_samples_per_second": 248.497,
+      "eval_steps_per_second": 31.062,
+      "step": 55000
+    },
+    {
+      "epoch": 5.3110097424738605,
+      "grad_norm": 1.0832284688949585,
+      "learning_rate": 4.916387959866221e-07,
+      "loss": 0.2946,
+      "step": 55100
+    },
+    {
+      "epoch": 5.3110097424738605,
+      "eval_loss": 0.26105597615242004,
+      "eval_runtime": 40.1107,
+      "eval_samples_per_second": 249.31,
+      "eval_steps_per_second": 31.164,
+      "step": 55100
+    },
+    {
+      "epoch": 5.3206485986308,
+      "grad_norm": 1.1050950288772583,
+      "learning_rate": 4.816053511705686e-07,
+      "loss": 0.2957,
+      "step": 55200
+    },
+    {
+      "epoch": 5.3206485986308,
+      "eval_loss": 0.2594417333602905,
+      "eval_runtime": 40.2651,
+      "eval_samples_per_second": 248.354,
+      "eval_steps_per_second": 31.044,
+      "step": 55200
+    },
+    {
+      "epoch": 5.33028745478774,
+      "grad_norm": 1.088828206062317,
+      "learning_rate": 4.7157190635451506e-07,
+      "loss": 0.2924,
+      "step": 55300
+    },
+    {
+      "epoch": 5.33028745478774,
+      "eval_loss": 0.26388004422187805,
+      "eval_runtime": 40.2575,
+      "eval_samples_per_second": 248.401,
+      "eval_steps_per_second": 31.05,
+      "step": 55300
+    },
+    {
+      "epoch": 5.33992631094468,
+      "grad_norm": 1.1768171787261963,
+      "learning_rate": 4.6153846153846156e-07,
+      "loss": 0.2951,
+      "step": 55400
+    },
+    {
+      "epoch": 5.33992631094468,
+      "eval_loss": 0.2630363702774048,
+      "eval_runtime": 40.2627,
+      "eval_samples_per_second": 248.369,
+      "eval_steps_per_second": 31.046,
+      "step": 55400
+    },
+    {
+      "epoch": 5.34956516710162,
+      "grad_norm": 1.1391232013702393,
+      "learning_rate": 4.5150501672240806e-07,
+      "loss": 0.2952,
+      "step": 55500
+    },
+    {
+      "epoch": 5.34956516710162,
+      "eval_loss": 0.25834527611732483,
+      "eval_runtime": 40.1693,
+      "eval_samples_per_second": 248.946,
+      "eval_steps_per_second": 31.118,
+      "step": 55500
+    },
+    {
+      "epoch": 5.35920402325856,
+      "grad_norm": 1.1385185718536377,
+      "learning_rate": 4.4147157190635456e-07,
+      "loss": 0.2935,
+      "step": 55600
+    },
+    {
+      "epoch": 5.35920402325856,
+      "eval_loss": 0.25431880354881287,
+      "eval_runtime": 40.1701,
+      "eval_samples_per_second": 248.941,
+      "eval_steps_per_second": 31.118,
+      "step": 55600
+    },
+    {
+      "epoch": 5.368842879415499,
+      "grad_norm": 1.0914981365203857,
+      "learning_rate": 4.3143812709030095e-07,
+      "loss": 0.2934,
+      "step": 55700
+    },
+    {
+      "epoch": 5.368842879415499,
+      "eval_loss": 0.25752854347229004,
+      "eval_runtime": 40.1548,
+      "eval_samples_per_second": 249.037,
+      "eval_steps_per_second": 31.13,
+      "step": 55700
+    },
+    {
+      "epoch": 5.378481735572439,
+      "grad_norm": 1.0367567539215088,
+      "learning_rate": 4.2140468227424745e-07,
+      "loss": 0.2929,
+      "step": 55800
+    },
+    {
+      "epoch": 5.378481735572439,
+      "eval_loss": 0.2595641613006592,
+      "eval_runtime": 40.1628,
+      "eval_samples_per_second": 248.986,
+      "eval_steps_per_second": 31.123,
+      "step": 55800
+    },
+    {
+      "epoch": 5.388120591729379,
+      "grad_norm": 1.072582483291626,
+      "learning_rate": 4.1137123745819395e-07,
+      "loss": 0.2947,
+      "step": 55900
+    },
+    {
+      "epoch": 5.388120591729379,
+      "eval_loss": 0.26283690333366394,
+      "eval_runtime": 40.1998,
+      "eval_samples_per_second": 248.758,
+      "eval_steps_per_second": 31.095,
+      "step": 55900
+    },
+    {
+      "epoch": 5.397759447886319,
+      "grad_norm": 1.0787990093231201,
+      "learning_rate": 4.0133779264214045e-07,
+      "loss": 0.2941,
+      "step": 56000
+    },
+    {
+      "epoch": 5.397759447886319,
+      "eval_loss": 0.2607322335243225,
+      "eval_runtime": 40.2136,
+      "eval_samples_per_second": 248.672,
+      "eval_steps_per_second": 31.084,
+      "step": 56000
+    },
+    {
+      "epoch": 5.407398304043259,
+      "grad_norm": 1.1289974451065063,
+      "learning_rate": 3.9130434782608694e-07,
+      "loss": 0.2951,
+      "step": 56100
+    },
+    {
+      "epoch": 5.407398304043259,
+      "eval_loss": 0.2661938965320587,
+      "eval_runtime": 40.0629,
+      "eval_samples_per_second": 249.608,
+      "eval_steps_per_second": 31.201,
+      "step": 56100
+    },
+    {
+      "epoch": 5.417037160200199,
+      "grad_norm": 1.1110820770263672,
+      "learning_rate": 3.8127090301003344e-07,
+      "loss": 0.2947,
+      "step": 56200
+    },
+    {
+      "epoch": 5.417037160200199,
+      "eval_loss": 0.26097166538238525,
+      "eval_runtime": 40.0727,
+      "eval_samples_per_second": 249.546,
+      "eval_steps_per_second": 31.193,
+      "step": 56200
+    },
+    {
+      "epoch": 5.426676016357139,
+      "grad_norm": 1.101787805557251,
+      "learning_rate": 3.7123745819397994e-07,
+      "loss": 0.2934,
+      "step": 56300
+    },
+    {
+      "epoch": 5.426676016357139,
+      "eval_loss": 0.2626558542251587,
+      "eval_runtime": 40.0699,
+      "eval_samples_per_second": 249.564,
+      "eval_steps_per_second": 31.195,
+      "step": 56300
+    },
+    {
+      "epoch": 5.436314872514079,
+      "grad_norm": 1.143552541732788,
+      "learning_rate": 3.6120401337792644e-07,
+      "loss": 0.2943,
+      "step": 56400
+    },
+    {
+      "epoch": 5.436314872514079,
+      "eval_loss": 0.2595503032207489,
+      "eval_runtime": 40.0704,
+      "eval_samples_per_second": 249.561,
+      "eval_steps_per_second": 31.195,
+      "step": 56400
+    },
+    {
+      "epoch": 5.445953728671019,
+      "grad_norm": 1.152707576751709,
+      "learning_rate": 3.5117056856187294e-07,
+      "loss": 0.2931,
+      "step": 56500
+    },
+    {
+      "epoch": 5.445953728671019,
+      "eval_loss": 0.2601962685585022,
+      "eval_runtime": 40.1774,
+      "eval_samples_per_second": 248.896,
+      "eval_steps_per_second": 31.112,
+      "step": 56500
+    },
+    {
+      "epoch": 5.455592584827959,
+      "grad_norm": 1.1379551887512207,
+      "learning_rate": 3.411371237458194e-07,
+      "loss": 0.2934,
+      "step": 56600
+    },
+    {
+      "epoch": 5.455592584827959,
+      "eval_loss": 0.2626563608646393,
+      "eval_runtime": 40.008,
+      "eval_samples_per_second": 249.95,
+      "eval_steps_per_second": 31.244,
+      "step": 56600
+    },
+    {
+      "epoch": 5.4652314409848985,
+      "grad_norm": 1.171043038368225,
+      "learning_rate": 3.311036789297659e-07,
+      "loss": 0.2935,
+      "step": 56700
+    },
+    {
+      "epoch": 5.4652314409848985,
+      "eval_loss": 0.2636509835720062,
+      "eval_runtime": 40.0145,
+      "eval_samples_per_second": 249.91,
+      "eval_steps_per_second": 31.239,
+      "step": 56700
+    },
+    {
+      "epoch": 5.474870297141838,
+      "grad_norm": 1.0666399002075195,
+      "learning_rate": 3.210702341137124e-07,
+      "loss": 0.2932,
+      "step": 56800
+    },
+    {
+      "epoch": 5.474870297141838,
+      "eval_loss": 0.26089200377464294,
+      "eval_runtime": 40.0108,
+      "eval_samples_per_second": 249.932,
+      "eval_steps_per_second": 31.242,
+      "step": 56800
+    },
+    {
+      "epoch": 5.484509153298778,
+      "grad_norm": 1.118462324142456,
+      "learning_rate": 3.110367892976589e-07,
+      "loss": 0.2953,
+      "step": 56900
+    },
+    {
+      "epoch": 5.484509153298778,
+      "eval_loss": 0.26198524236679077,
+      "eval_runtime": 40.1717,
+      "eval_samples_per_second": 248.931,
+      "eval_steps_per_second": 31.116,
+      "step": 56900
+    },
+    {
+      "epoch": 5.494148009455718,
+      "grad_norm": 1.1037251949310303,
+      "learning_rate": 3.010033444816054e-07,
+      "loss": 0.2926,
+      "step": 57000
+    },
+    {
+      "epoch": 5.494148009455718,
+      "eval_loss": 0.2536342144012451,
+      "eval_runtime": 40.0426,
+      "eval_samples_per_second": 249.734,
+      "eval_steps_per_second": 31.217,
+      "step": 57000
+    },
+    {
+      "epoch": 5.503786865612658,
+      "grad_norm": 1.0776835680007935,
+      "learning_rate": 2.9096989966555187e-07,
+      "loss": 0.291,
+      "step": 57100
+    },
+    {
+      "epoch": 5.503786865612658,
+      "eval_loss": 0.2602877914905548,
+      "eval_runtime": 40.0401,
+      "eval_samples_per_second": 249.75,
+      "eval_steps_per_second": 31.219,
+      "step": 57100
+    },
+    {
+      "epoch": 5.513425721769598,
+      "grad_norm": 1.1141369342803955,
+      "learning_rate": 2.809364548494983e-07,
+      "loss": 0.2932,
+      "step": 57200
+    },
+    {
+      "epoch": 5.513425721769598,
+      "eval_loss": 0.258954793214798,
+      "eval_runtime": 40.2256,
+      "eval_samples_per_second": 248.598,
+      "eval_steps_per_second": 31.075,
+      "step": 57200
+    },
+    {
+      "epoch": 5.5230645779265375,
+      "grad_norm": 1.111812710762024,
+      "learning_rate": 2.709030100334448e-07,
+      "loss": 0.2919,
+      "step": 57300
+    },
+    {
+      "epoch": 5.5230645779265375,
+      "eval_loss": 0.2630453407764435,
+      "eval_runtime": 40.0543,
+      "eval_samples_per_second": 249.661,
+      "eval_steps_per_second": 31.208,
+      "step": 57300
+    },
+    {
+      "epoch": 5.532703434083477,
+      "grad_norm": 1.165526032447815,
+      "learning_rate": 2.608695652173913e-07,
+      "loss": 0.2926,
+      "step": 57400
+    },
+    {
+      "epoch": 5.532703434083477,
+      "eval_loss": 0.25891971588134766,
+      "eval_runtime": 40.055,
+      "eval_samples_per_second": 249.657,
+      "eval_steps_per_second": 31.207,
+      "step": 57400
+    },
+    {
+      "epoch": 5.542342290240417,
+      "grad_norm": 1.124131679534912,
+      "learning_rate": 2.508361204013378e-07,
+      "loss": 0.2927,
+      "step": 57500
+    },
+    {
+      "epoch": 5.542342290240417,
+      "eval_loss": 0.2566971480846405,
+      "eval_runtime": 40.0829,
+      "eval_samples_per_second": 249.483,
+      "eval_steps_per_second": 31.185,
+      "step": 57500
+    },
+    {
+      "epoch": 5.551981146397357,
+      "grad_norm": 1.0840249061584473,
+      "learning_rate": 2.408026755852843e-07,
+      "loss": 0.2944,
+      "step": 57600
+    },
+    {
+      "epoch": 5.551981146397357,
+      "eval_loss": 0.25784093141555786,
+      "eval_runtime": 40.2031,
+      "eval_samples_per_second": 248.737,
+      "eval_steps_per_second": 31.092,
+      "step": 57600
+    },
+    {
+      "epoch": 5.561620002554297,
+      "grad_norm": 1.1037781238555908,
+      "learning_rate": 2.3076923076923078e-07,
+      "loss": 0.2934,
+      "step": 57700
+    },
+    {
+      "epoch": 5.561620002554297,
+      "eval_loss": 0.251878559589386,
+      "eval_runtime": 40.0523,
+      "eval_samples_per_second": 249.674,
+      "eval_steps_per_second": 31.209,
+      "step": 57700
+    },
+    {
+      "epoch": 5.571258858711237,
+      "grad_norm": 1.1695195436477661,
+      "learning_rate": 2.2073578595317728e-07,
+      "loss": 0.294,
+      "step": 57800
+    },
+    {
+      "epoch": 5.571258858711237,
+      "eval_loss": 0.2621340751647949,
+      "eval_runtime": 40.0487,
+      "eval_samples_per_second": 249.696,
+      "eval_steps_per_second": 31.212,
+      "step": 57800
+    },
+    {
+      "epoch": 5.580897714868176,
+      "grad_norm": 1.1314599514007568,
+      "learning_rate": 2.1070234113712372e-07,
+      "loss": 0.2925,
+      "step": 57900
+    },
+    {
+      "epoch": 5.580897714868176,
+      "eval_loss": 0.2625725567340851,
+      "eval_runtime": 40.0655,
+      "eval_samples_per_second": 249.592,
+      "eval_steps_per_second": 31.199,
+      "step": 57900
+    },
+    {
+      "epoch": 5.590536571025116,
+      "grad_norm": 1.0930743217468262,
+      "learning_rate": 2.0066889632107022e-07,
+      "loss": 0.2933,
+      "step": 58000
+    },
+    {
+      "epoch": 5.590536571025116,
+      "eval_loss": 0.25974881649017334,
+      "eval_runtime": 40.0392,
+      "eval_samples_per_second": 249.755,
+      "eval_steps_per_second": 31.219,
+      "step": 58000
+    },
+    {
+      "epoch": 5.600175427182056,
+      "grad_norm": 1.1561408042907715,
+      "learning_rate": 1.9063545150501672e-07,
+      "loss": 0.2942,
+      "step": 58100
+    },
+    {
+      "epoch": 5.600175427182056,
+      "eval_loss": 0.2584717869758606,
+      "eval_runtime": 40.2075,
+      "eval_samples_per_second": 248.71,
+      "eval_steps_per_second": 31.089,
+      "step": 58100
+    },
+    {
+      "epoch": 5.609814283338996,
+      "grad_norm": 1.1130927801132202,
+      "learning_rate": 1.8060200668896322e-07,
+      "loss": 0.2938,
+      "step": 58200
+    },
+    {
+      "epoch": 5.609814283338996,
+      "eval_loss": 0.2553929090499878,
+      "eval_runtime": 40.0363,
+      "eval_samples_per_second": 249.774,
+      "eval_steps_per_second": 31.222,
+      "step": 58200
+    },
+    {
+      "epoch": 5.619453139495936,
+      "grad_norm": 1.2237213850021362,
+      "learning_rate": 1.705685618729097e-07,
+      "loss": 0.2942,
+      "step": 58300
+    },
+    {
+      "epoch": 5.619453139495936,
+      "eval_loss": 0.24946899712085724,
+      "eval_runtime": 40.0454,
+      "eval_samples_per_second": 249.716,
+      "eval_steps_per_second": 31.215,
+      "step": 58300
+    },
+    {
+      "epoch": 5.6290919956528755,
+      "grad_norm": 1.1548402309417725,
+      "learning_rate": 1.605351170568562e-07,
+      "loss": 0.2948,
+      "step": 58400
+    },
+    {
+      "epoch": 5.6290919956528755,
+      "eval_loss": 0.25441214442253113,
+      "eval_runtime": 40.0484,
+      "eval_samples_per_second": 249.698,
+      "eval_steps_per_second": 31.212,
+      "step": 58400
+    },
+    {
+      "epoch": 5.638730851809815,
+      "grad_norm": 1.1330386400222778,
+      "learning_rate": 1.505016722408027e-07,
+      "loss": 0.2913,
+      "step": 58500
+    },
+    {
+      "epoch": 5.638730851809815,
+      "eval_loss": 0.2585105895996094,
+      "eval_runtime": 40.2225,
+      "eval_samples_per_second": 248.617,
+      "eval_steps_per_second": 31.077,
+      "step": 58500
+    },
+    {
+      "epoch": 5.648369707966756,
+      "grad_norm": 1.1387040615081787,
+      "learning_rate": 1.4046822742474916e-07,
+      "loss": 0.2938,
+      "step": 58600
+    },
+    {
+      "epoch": 5.648369707966756,
+      "eval_loss": 0.2615092098712921,
+      "eval_runtime": 40.0294,
+      "eval_samples_per_second": 249.816,
+      "eval_steps_per_second": 31.227,
+      "step": 58600
+    },
+    {
+      "epoch": 5.658008564123696,
+      "grad_norm": 1.1653701066970825,
+      "learning_rate": 1.3043478260869566e-07,
+      "loss": 0.2947,
+      "step": 58700
+    },
+    {
+      "epoch": 5.658008564123696,
+      "eval_loss": 0.2623739242553711,
+      "eval_runtime": 40.0354,
+      "eval_samples_per_second": 249.779,
+      "eval_steps_per_second": 31.222,
+      "step": 58700
+    },
+    {
+      "epoch": 5.667647420280636,
+      "grad_norm": 1.1270488500595093,
+      "learning_rate": 1.2040133779264215e-07,
+      "loss": 0.2928,
+      "step": 58800
+    },
+    {
+      "epoch": 5.667647420280636,
+      "eval_loss": 0.26067566871643066,
+      "eval_runtime": 40.0307,
+      "eval_samples_per_second": 249.809,
+      "eval_steps_per_second": 31.226,
+      "step": 58800
+    },
+    {
+      "epoch": 5.6772862764375756,
+      "grad_norm": 1.117533802986145,
+      "learning_rate": 1.1036789297658864e-07,
+      "loss": 0.2908,
+      "step": 58900
+    },
+    {
+      "epoch": 5.6772862764375756,
+      "eval_loss": 0.2551814019680023,
+      "eval_runtime": 40.0838,
+      "eval_samples_per_second": 249.478,
+      "eval_steps_per_second": 31.185,
+      "step": 58900
+    },
+    {
+      "epoch": 5.686925132594515,
+      "grad_norm": 1.0880376100540161,
+      "learning_rate": 1.0033444816053511e-07,
+      "loss": 0.2912,
+      "step": 59000
+    },
+    {
+      "epoch": 5.686925132594515,
+      "eval_loss": 0.25541287660598755,
+      "eval_runtime": 40.2323,
+      "eval_samples_per_second": 248.556,
+      "eval_steps_per_second": 31.07,
+      "step": 59000
+    },
+    {
+      "epoch": 5.696563988751455,
+      "grad_norm": 1.0031907558441162,
+      "learning_rate": 9.030100334448161e-08,
+      "loss": 0.2935,
+      "step": 59100
+    },
+    {
+      "epoch": 5.696563988751455,
+      "eval_loss": 0.2621237337589264,
+      "eval_runtime": 40.0531,
+      "eval_samples_per_second": 249.669,
+      "eval_steps_per_second": 31.209,
+      "step": 59100
+    },
+    {
+      "epoch": 5.706202844908395,
+      "grad_norm": 1.104634165763855,
+      "learning_rate": 8.02675585284281e-08,
+      "loss": 0.2951,
+      "step": 59200
+    },
+    {
+      "epoch": 5.706202844908395,
+      "eval_loss": 0.2626442015171051,
+      "eval_runtime": 40.0642,
+      "eval_samples_per_second": 249.599,
+      "eval_steps_per_second": 31.2,
+      "step": 59200
+    },
+    {
+      "epoch": 5.715841701065335,
+      "grad_norm": 1.1673884391784668,
+      "learning_rate": 7.023411371237458e-08,
+      "loss": 0.2927,
+      "step": 59300
+    },
+    {
+      "epoch": 5.715841701065335,
+      "eval_loss": 0.25724858045578003,
+      "eval_runtime": 40.2787,
+      "eval_samples_per_second": 248.27,
+      "eval_steps_per_second": 31.034,
+      "step": 59300
+    },
+    {
+      "epoch": 5.725480557222275,
+      "grad_norm": 1.0944865942001343,
+      "learning_rate": 6.020066889632108e-08,
+      "loss": 0.2926,
+      "step": 59400
+    },
+    {
+      "epoch": 5.725480557222275,
+      "eval_loss": 0.2603595554828644,
+      "eval_runtime": 40.0311,
+      "eval_samples_per_second": 249.806,
+      "eval_steps_per_second": 31.226,
+      "step": 59400
+    },
+    {
+      "epoch": 5.7351194133792145,
+      "grad_norm": 1.142904281616211,
+      "learning_rate": 5.0167224080267556e-08,
+      "loss": 0.2935,
+      "step": 59500
+    },
+    {
+      "epoch": 5.7351194133792145,
+      "eval_loss": 0.25873175263404846,
+      "eval_runtime": 40.0183,
+      "eval_samples_per_second": 249.886,
+      "eval_steps_per_second": 31.236,
+      "step": 59500
+    },
+    {
+      "epoch": 5.744758269536154,
+      "grad_norm": 1.1131585836410522,
+      "learning_rate": 4.013377926421405e-08,
+      "loss": 0.2913,
+      "step": 59600
+    },
+    {
+      "epoch": 5.744758269536154,
+      "eval_loss": 0.25963783264160156,
+      "eval_runtime": 40.0082,
+      "eval_samples_per_second": 249.949,
+      "eval_steps_per_second": 31.244,
+      "step": 59600
+    },
+    {
+      "epoch": 5.754397125693094,
+      "grad_norm": 1.1289507150650024,
+      "learning_rate": 3.010033444816054e-08,
+      "loss": 0.2934,
+      "step": 59700
+    },
+    {
+      "epoch": 5.754397125693094,
+      "eval_loss": 0.2582632005214691,
+      "eval_runtime": 40.0014,
+      "eval_samples_per_second": 249.991,
+      "eval_steps_per_second": 31.249,
+      "step": 59700
+    },
+    {
+      "epoch": 5.764035981850034,
+      "grad_norm": 1.0394152402877808,
+      "learning_rate": 2.0066889632107024e-08,
+      "loss": 0.2915,
+      "step": 59800
+    },
+    {
+      "epoch": 5.764035981850034,
+      "eval_loss": 0.2560790181159973,
+      "eval_runtime": 40.2031,
+      "eval_samples_per_second": 248.737,
+      "eval_steps_per_second": 31.092,
+      "step": 59800
+    },
+    {
+      "epoch": 5.773674838006974,
+      "grad_norm": 1.1087628602981567,
+      "learning_rate": 1.0033444816053512e-08,
+      "loss": 0.2915,
+      "step": 59900
+    },
+    {
+      "epoch": 5.773674838006974,
+      "eval_loss": 0.2614119350910187,
+      "eval_runtime": 40.0216,
+      "eval_samples_per_second": 249.865,
+      "eval_steps_per_second": 31.233,
+      "step": 59900
+    },
+    {
+      "epoch": 5.783313694163914,
+      "grad_norm": 1.0176420211791992,
+      "learning_rate": 0.0,
+      "loss": 0.2935,
+      "step": 60000
+    },
+    {
+      "epoch": 5.783313694163914,
+      "eval_loss": 0.2631940245628357,
+      "eval_runtime": 40.0188,
+      "eval_samples_per_second": 249.882,
+      "eval_steps_per_second": 31.235,
+      "step": 60000
+    }
+  ],
+  "logging_steps": 100,
+  "max_steps": 60000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 6,
+  "save_steps": 1000,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7893111401591828e+19,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}