diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4331 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.983416252072969,
+  "eval_steps": 500,
+  "global_step": 3010,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.003316749585406302,
+      "grad_norm": 330.0,
+      "learning_rate": 6.644518272425249e-07,
+      "loss": 34.0644,
+      "step": 1
+    },
+    {
+      "epoch": 0.01658374792703151,
+      "grad_norm": 328.0,
+      "learning_rate": 3.322259136212625e-06,
+      "loss": 34.4226,
+      "step": 5
+    },
+    {
+      "epoch": 0.03316749585406302,
+      "grad_norm": 189.0,
+      "learning_rate": 6.64451827242525e-06,
+      "loss": 33.946,
+      "step": 10
+    },
+    {
+      "epoch": 0.04975124378109453,
+      "grad_norm": 117.5,
+      "learning_rate": 9.966777408637874e-06,
+      "loss": 28.2669,
+      "step": 15
+    },
+    {
+      "epoch": 0.06633499170812604,
+      "grad_norm": 55.25,
+      "learning_rate": 1.32890365448505e-05,
+      "loss": 23.3646,
+      "step": 20
+    },
+    {
+      "epoch": 0.08291873963515754,
+      "grad_norm": 19.125,
+      "learning_rate": 1.6611295681063124e-05,
+      "loss": 20.0723,
+      "step": 25
+    },
+    {
+      "epoch": 0.09950248756218906,
+      "grad_norm": 16.25,
+      "learning_rate": 1.9933554817275747e-05,
+      "loss": 18.4564,
+      "step": 30
+    },
+    {
+      "epoch": 0.11608623548922056,
+      "grad_norm": 12.1875,
+      "learning_rate": 2.3255813953488374e-05,
+      "loss": 16.9883,
+      "step": 35
+    },
+    {
+      "epoch": 0.13266998341625208,
+      "grad_norm": 7.15625,
+      "learning_rate": 2.6578073089701e-05,
+      "loss": 15.7277,
+      "step": 40
+    },
+    {
+      "epoch": 0.14925373134328357,
+      "grad_norm": 4.15625,
+      "learning_rate": 2.9900332225913624e-05,
+      "loss": 14.9339,
+      "step": 45
+    },
+    {
+      "epoch": 0.16583747927031509,
+      "grad_norm": 3.359375,
+      "learning_rate": 3.322259136212625e-05,
+      "loss": 14.2643,
+      "step": 50
+    },
+    {
+      "epoch": 0.1824212271973466,
+      "grad_norm": 3.40625,
+      "learning_rate": 3.654485049833887e-05,
+      "loss": 13.907,
+      "step": 55
+    },
+    {
+      "epoch": 0.19900497512437812,
+      "grad_norm": 4.53125,
+      "learning_rate": 3.9867109634551495e-05,
+      "loss": 13.3731,
+      "step": 60
+    },
+    {
+      "epoch": 0.2155887230514096,
+      "grad_norm": 5.34375,
+      "learning_rate": 4.3189368770764125e-05,
+      "loss": 12.8995,
+      "step": 65
+    },
+    {
+      "epoch": 0.23217247097844113,
+      "grad_norm": 7.75,
+      "learning_rate": 4.651162790697675e-05,
+      "loss": 12.1993,
+      "step": 70
+    },
+    {
+      "epoch": 0.24875621890547264,
+      "grad_norm": 12.5,
+      "learning_rate": 4.983388704318937e-05,
+      "loss": 11.2401,
+      "step": 75
+    },
+    {
+      "epoch": 0.26533996683250416,
+      "grad_norm": 18.75,
+      "learning_rate": 5.3156146179402e-05,
+      "loss": 9.4202,
+      "step": 80
+    },
+    {
+      "epoch": 0.28192371475953565,
+      "grad_norm": 21.375,
+      "learning_rate": 5.6478405315614625e-05,
+      "loss": 6.5708,
+      "step": 85
+    },
+    {
+      "epoch": 0.29850746268656714,
+      "grad_norm": 12.125,
+      "learning_rate": 5.980066445182725e-05,
+      "loss": 3.6952,
+      "step": 90
+    },
+    {
+      "epoch": 0.3150912106135987,
+      "grad_norm": 3.9375,
+      "learning_rate": 6.312292358803988e-05,
+      "loss": 2.1631,
+      "step": 95
+    },
+    {
+      "epoch": 0.33167495854063017,
+      "grad_norm": 1.6015625,
+      "learning_rate": 6.64451827242525e-05,
+      "loss": 1.8363,
+      "step": 100
+    },
+    {
+      "epoch": 0.3482587064676617,
+      "grad_norm": 1.2265625,
+      "learning_rate": 6.976744186046513e-05,
+      "loss": 1.6362,
+      "step": 105
+    },
+    {
+      "epoch": 0.3648424543946932,
+      "grad_norm": 0.734375,
+      "learning_rate": 7.308970099667774e-05,
+      "loss": 1.5453,
+      "step": 110
+    },
+    {
+      "epoch": 0.3814262023217247,
+      "grad_norm": 1.015625,
+      "learning_rate": 7.641196013289037e-05,
+      "loss": 1.4575,
+      "step": 115
+    },
+    {
+      "epoch": 0.39800995024875624,
+      "grad_norm": 1.046875,
+      "learning_rate": 7.973421926910299e-05,
+      "loss": 1.3978,
+      "step": 120
+    },
+    {
+      "epoch": 0.41459369817578773,
+      "grad_norm": 0.76171875,
+      "learning_rate": 8.305647840531562e-05,
+      "loss": 1.3648,
+      "step": 125
+    },
+    {
+      "epoch": 0.4311774461028192,
+      "grad_norm": 0.8359375,
+      "learning_rate": 8.637873754152825e-05,
+      "loss": 1.3133,
+      "step": 130
+    },
+    {
+      "epoch": 0.44776119402985076,
+      "grad_norm": 0.95703125,
+      "learning_rate": 8.970099667774087e-05,
+      "loss": 1.2923,
+      "step": 135
+    },
+    {
+      "epoch": 0.46434494195688225,
+      "grad_norm": 0.96484375,
+      "learning_rate": 9.30232558139535e-05,
+      "loss": 1.2604,
+      "step": 140
+    },
+    {
+      "epoch": 0.48092868988391374,
+      "grad_norm": 1.2734375,
+      "learning_rate": 9.634551495016611e-05,
+      "loss": 1.2395,
+      "step": 145
+    },
+    {
+      "epoch": 0.4975124378109453,
+      "grad_norm": 0.78515625,
+      "learning_rate": 9.966777408637874e-05,
+      "loss": 1.227,
+      "step": 150
+    },
+    {
+      "epoch": 0.5140961857379768,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00010299003322259137,
+      "loss": 1.1989,
+      "step": 155
+    },
+    {
+      "epoch": 0.5306799336650083,
+      "grad_norm": 1.234375,
+      "learning_rate": 0.000106312292358804,
+      "loss": 1.1822,
+      "step": 160
+    },
+    {
+      "epoch": 0.5472636815920398,
+      "grad_norm": 0.7890625,
+      "learning_rate": 0.00010963455149501662,
+      "loss": 1.1642,
+      "step": 165
+    },
+    {
+      "epoch": 0.5638474295190713,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.00011295681063122925,
+      "loss": 1.1605,
+      "step": 170
+    },
+    {
+      "epoch": 0.5804311774461028,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00011627906976744187,
+      "loss": 1.1318,
+      "step": 175
+    },
+    {
+      "epoch": 0.5970149253731343,
+      "grad_norm": 2.1875,
+      "learning_rate": 0.0001196013289036545,
+      "loss": 1.1453,
+      "step": 180
+    },
+    {
+      "epoch": 0.6135986733001658,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.00012292358803986711,
+      "loss": 1.1316,
+      "step": 185
+    },
+    {
+      "epoch": 0.6301824212271974,
+      "grad_norm": 0.98046875,
+      "learning_rate": 0.00012624584717607976,
+      "loss": 1.1202,
+      "step": 190
+    },
+    {
+      "epoch": 0.6467661691542289,
+      "grad_norm": 0.90625,
+      "learning_rate": 0.00012956810631229237,
+      "loss": 1.1183,
+      "step": 195
+    },
+    {
+      "epoch": 0.6633499170812603,
+      "grad_norm": 0.8828125,
+      "learning_rate": 0.000132890365448505,
+      "loss": 1.1174,
+      "step": 200
+    },
+    {
+      "epoch": 0.6799336650082919,
+      "grad_norm": 2.9375,
+      "learning_rate": 0.0001362126245847176,
+      "loss": 1.1185,
+      "step": 205
+    },
+    {
+      "epoch": 0.6965174129353234,
+      "grad_norm": 3.390625,
+      "learning_rate": 0.00013953488372093025,
+      "loss": 1.1048,
+      "step": 210
+    },
+    {
+      "epoch": 0.7131011608623549,
+      "grad_norm": 2.765625,
+      "learning_rate": 0.00014285714285714287,
+      "loss": 1.1048,
+      "step": 215
+    },
+    {
+      "epoch": 0.7296849087893864,
+      "grad_norm": 0.953125,
+      "learning_rate": 0.00014617940199335548,
+      "loss": 1.0892,
+      "step": 220
+    },
+    {
+      "epoch": 0.746268656716418,
+      "grad_norm": 1.5,
+      "learning_rate": 0.00014950166112956813,
+      "loss": 1.0847,
+      "step": 225
+    },
+    {
+      "epoch": 0.7628524046434494,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.00015282392026578074,
+      "loss": 1.074,
+      "step": 230
+    },
+    {
+      "epoch": 0.7794361525704809,
+      "grad_norm": 1.96875,
+      "learning_rate": 0.00015614617940199336,
+      "loss": 1.0749,
+      "step": 235
+    },
+    {
+      "epoch": 0.7960199004975125,
+      "grad_norm": 6.53125,
+      "learning_rate": 0.00015946843853820598,
+      "loss": 1.0725,
+      "step": 240
+    },
+    {
+      "epoch": 0.8126036484245439,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.00016279069767441862,
+      "loss": 1.0739,
+      "step": 245
+    },
+    {
+      "epoch": 0.8291873963515755,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.00016611295681063124,
+      "loss": 1.0684,
+      "step": 250
+    },
+    {
+      "epoch": 0.845771144278607,
+      "grad_norm": 2.421875,
+      "learning_rate": 0.00016943521594684386,
+      "loss": 1.0621,
+      "step": 255
+    },
+    {
+      "epoch": 0.8623548922056384,
+      "grad_norm": 1.375,
+      "learning_rate": 0.0001727574750830565,
+      "loss": 1.0597,
+      "step": 260
+    },
+    {
+      "epoch": 0.87893864013267,
+      "grad_norm": 1.65625,
+      "learning_rate": 0.00017607973421926912,
+      "loss": 1.0527,
+      "step": 265
+    },
+    {
+      "epoch": 0.8955223880597015,
+      "grad_norm": 1.40625,
+      "learning_rate": 0.00017940199335548173,
+      "loss": 1.0494,
+      "step": 270
+    },
+    {
+      "epoch": 0.912106135986733,
+      "grad_norm": 1.3671875,
+      "learning_rate": 0.00018272425249169435,
+      "loss": 1.0473,
+      "step": 275
+    },
+    {
+      "epoch": 0.9286898839137645,
+      "grad_norm": 1.7265625,
+      "learning_rate": 0.000186046511627907,
+      "loss": 1.0262,
+      "step": 280
+    },
+    {
+      "epoch": 0.945273631840796,
+      "grad_norm": 0.8203125,
+      "learning_rate": 0.0001893687707641196,
+      "loss": 1.0243,
+      "step": 285
+    },
+    {
+      "epoch": 0.9618573797678275,
+      "grad_norm": 2.65625,
+      "learning_rate": 0.00019269102990033223,
+      "loss": 1.038,
+      "step": 290
+    },
+    {
+      "epoch": 0.978441127694859,
+      "grad_norm": 1.8671875,
+      "learning_rate": 0.00019601328903654484,
+      "loss": 1.0341,
+      "step": 295
+    },
+    {
+      "epoch": 0.9950248756218906,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.00019933554817275749,
+      "loss": 1.0247,
+      "step": 300
+    },
+    {
+      "epoch": 0.9983416252072969,
+      "eval_loss": 2.53955078125,
+      "eval_runtime": 0.2726,
+      "eval_samples_per_second": 36.685,
+      "eval_steps_per_second": 3.668,
+      "step": 301
+    },
+    {
+      "epoch": 1.0116086235489221,
+      "grad_norm": 1.2109375,
+      "learning_rate": 0.00019999892410238807,
+      "loss": 1.0242,
+      "step": 305
+    },
+    {
+      "epoch": 1.0281923714759535,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.0001999945533080175,
+      "loss": 1.0062,
+      "step": 310
+    },
+    {
+      "epoch": 1.044776119402985,
+      "grad_norm": 1.3359375,
+      "learning_rate": 0.00019998682052012838,
+      "loss": 0.9926,
+      "step": 315
+    },
+    {
+      "epoch": 1.0613598673300166,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.0001999757259987106,
+      "loss": 1.0042,
+      "step": 320
+    },
+    {
+      "epoch": 1.077943615257048,
+      "grad_norm": 2.203125,
+      "learning_rate": 0.00019996127011678166,
+      "loss": 1.0115,
+      "step": 325
+    },
+    {
+      "epoch": 1.0945273631840795,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.0001999434533603738,
+      "loss": 1.0115,
+      "step": 330
+    },
+    {
+      "epoch": 1.1111111111111112,
+      "grad_norm": 1.7734375,
+      "learning_rate": 0.00019992227632851788,
+      "loss": 1.0111,
+      "step": 335
+    },
+    {
+      "epoch": 1.1276948590381426,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.00019989773973322307,
+      "loss": 0.9926,
+      "step": 340
+    },
+    {
+      "epoch": 1.144278606965174,
+      "grad_norm": 8.0,
+      "learning_rate": 0.00019986984439945306,
+      "loss": 1.003,
+      "step": 345
+    },
+    {
+      "epoch": 1.1608623548922057,
+      "grad_norm": 2.40625,
+      "learning_rate": 0.00019983859126509828,
+      "loss": 1.0011,
+      "step": 350
+    },
+    {
+      "epoch": 1.1774461028192371,
+      "grad_norm": 1.859375,
+      "learning_rate": 0.00019980398138094428,
+      "loss": 1.0071,
+      "step": 355
+    },
+    {
+      "epoch": 1.1940298507462686,
+      "grad_norm": 1.3828125,
+      "learning_rate": 0.00019976601591063654,
+      "loss": 0.9914,
+      "step": 360
+    },
+    {
+      "epoch": 1.2106135986733002,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.00019972469613064117,
+      "loss": 0.9791,
+      "step": 365
+    },
+    {
+      "epoch": 1.2271973466003316,
+      "grad_norm": 1.6484375,
+      "learning_rate": 0.00019968002343020224,
+      "loss": 0.9874,
+      "step": 370
+    },
+    {
+      "epoch": 1.243781094527363,
+      "grad_norm": 2.40625,
+      "learning_rate": 0.0001996319993112948,
+      "loss": 0.9944,
+      "step": 375
+    },
+    {
+      "epoch": 1.2603648424543947,
+      "grad_norm": 0.609375,
+      "learning_rate": 0.0001995806253885746,
+      "loss": 0.9769,
+      "step": 380
+    },
+    {
+      "epoch": 1.2769485903814262,
+      "grad_norm": 1.6640625,
+      "learning_rate": 0.00019952590338932356,
+      "loss": 0.981,
+      "step": 385
+    },
+    {
+      "epoch": 1.2935323383084576,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.00019946783515339215,
+      "loss": 0.9871,
+      "step": 390
+    },
+    {
+      "epoch": 1.3101160862354893,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.00019940642263313688,
+      "loss": 0.9796,
+      "step": 395
+    },
+    {
+      "epoch": 1.3266998341625207,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00019934166789335525,
+      "loss": 0.9704,
+      "step": 400
+    },
+    {
+      "epoch": 1.3432835820895521,
+      "grad_norm": 3.796875,
+      "learning_rate": 0.00019927357311121596,
+      "loss": 0.9874,
+      "step": 405
+    },
+    {
+      "epoch": 1.3598673300165838,
+      "grad_norm": 1.5625,
+      "learning_rate": 0.00019920214057618586,
+      "loss": 0.9816,
+      "step": 410
+    },
+    {
+      "epoch": 1.3764510779436152,
+      "grad_norm": 7.625,
+      "learning_rate": 0.00019912737268995296,
+      "loss": 0.9708,
+      "step": 415
+    },
+    {
+      "epoch": 1.3930348258706466,
+      "grad_norm": 2.921875,
+      "learning_rate": 0.0001990492719663457,
+      "loss": 0.9613,
+      "step": 420
+    },
+    {
+      "epoch": 1.4096185737976783,
+      "grad_norm": 2.421875,
+      "learning_rate": 0.0001989678410312484,
+      "loss": 0.9652,
+      "step": 425
+    },
+    {
+      "epoch": 1.4262023217247097,
+      "grad_norm": 2.078125,
+      "learning_rate": 0.00019888308262251285,
+      "loss": 0.9506,
+      "step": 430
+    },
+    {
+      "epoch": 1.4427860696517412,
+      "grad_norm": 1.3515625,
+      "learning_rate": 0.00019879499958986655,
+      "loss": 0.9535,
+      "step": 435
+    },
+    {
+      "epoch": 1.4593698175787728,
+      "grad_norm": 4.96875,
+      "learning_rate": 0.00019870359489481665,
+      "loss": 0.9652,
+      "step": 440
+    },
+    {
+      "epoch": 1.4759535655058043,
+      "grad_norm": 1.15625,
+      "learning_rate": 0.00019860887161055038,
+      "loss": 0.9615,
+      "step": 445
+    },
+    {
+      "epoch": 1.4925373134328357,
+      "grad_norm": 8.1875,
+      "learning_rate": 0.00019851083292183198,
+      "loss": 0.956,
+      "step": 450
+    },
+    {
+      "epoch": 1.5091210613598673,
+      "grad_norm": 1.71875,
+      "learning_rate": 0.00019840948212489527,
+      "loss": 0.9687,
+      "step": 455
+    },
+    {
+      "epoch": 1.525704809286899,
+      "grad_norm": 1.0625,
+      "learning_rate": 0.00019830482262733315,
+      "loss": 0.9636,
+      "step": 460
+    },
+    {
+      "epoch": 1.5422885572139302,
+      "grad_norm": 1.375,
+      "learning_rate": 0.00019819685794798274,
+      "loss": 0.9527,
+      "step": 465
+    },
+    {
+      "epoch": 1.5588723051409619,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0001980855917168073,
+      "loss": 0.9558,
+      "step": 470
+    },
+    {
+      "epoch": 1.5754560530679935,
+      "grad_norm": 2.6875,
+      "learning_rate": 0.0001979710276747741,
+      "loss": 0.9477,
+      "step": 475
+    },
+    {
+      "epoch": 1.5920398009950247,
+      "grad_norm": 4.0625,
+      "learning_rate": 0.00019785316967372855,
+      "loss": 0.9663,
+      "step": 480
+    },
+    {
+      "epoch": 1.6086235489220564,
+      "grad_norm": 1.125,
+      "learning_rate": 0.00019773202167626487,
+      "loss": 0.962,
+      "step": 485
+    },
+    {
+      "epoch": 1.625207296849088,
+      "grad_norm": 1.1328125,
+      "learning_rate": 0.00019760758775559274,
+      "loss": 0.9535,
+      "step": 490
+    },
+    {
+      "epoch": 1.6417910447761193,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.00019747987209540034,
+      "loss": 0.9362,
+      "step": 495
+    },
+    {
+      "epoch": 1.658374792703151,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00019734887898971383,
+      "loss": 0.933,
+      "step": 500
+    },
+    {
+      "epoch": 1.6749585406301826,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.00019721461284275278,
+      "loss": 0.924,
+      "step": 505
+    },
+    {
+      "epoch": 1.6915422885572138,
+      "grad_norm": 1.8515625,
+      "learning_rate": 0.0001970770781687822,
+      "loss": 0.9412,
+      "step": 510
+    },
+    {
+      "epoch": 1.7081260364842454,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.00019693627959196087,
+      "loss": 0.9391,
+      "step": 515
+    },
+    {
+      "epoch": 1.724709784411277,
+      "grad_norm": 1.5625,
+      "learning_rate": 0.00019679222184618554,
+      "loss": 0.9398,
+      "step": 520
+    },
+    {
+      "epoch": 1.7412935323383083,
+      "grad_norm": 1.34375,
+      "learning_rate": 0.00019664490977493223,
+      "loss": 0.9401,
+      "step": 525
+    },
+    {
+      "epoch": 1.75787728026534,
+      "grad_norm": 4.65625,
+      "learning_rate": 0.00019649434833109286,
+      "loss": 0.9347,
+      "step": 530
+    },
+    {
+      "epoch": 1.7744610281923716,
+      "grad_norm": 1.671875,
+      "learning_rate": 0.00019634054257680923,
+      "loss": 0.941,
+      "step": 535
+    },
+    {
+      "epoch": 1.7910447761194028,
+      "grad_norm": 3.234375,
+      "learning_rate": 0.00019618349768330245,
+      "loss": 0.9353,
+      "step": 540
+    },
+    {
+      "epoch": 1.8076285240464345,
+      "grad_norm": 0.97265625,
+      "learning_rate": 0.00019602321893069923,
+      "loss": 0.9432,
+      "step": 545
+    },
+    {
+      "epoch": 1.8242122719734661,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.00019585971170785443,
+      "loss": 0.9353,
+      "step": 550
+    },
+    {
+      "epoch": 1.8407960199004973,
+      "grad_norm": 2.5,
+      "learning_rate": 0.00019569298151216963,
+      "loss": 0.9368,
+      "step": 555
+    },
+    {
+      "epoch": 1.857379767827529,
+      "grad_norm": 4.09375,
+      "learning_rate": 0.0001955230339494086,
+      "loss": 0.9263,
+      "step": 560
+    },
+    {
+      "epoch": 1.8739635157545607,
+      "grad_norm": 2.359375,
+      "learning_rate": 0.00019534987473350857,
+      "loss": 0.9375,
+      "step": 565
+    },
+    {
+      "epoch": 1.890547263681592,
+      "grad_norm": 1.0234375,
+      "learning_rate": 0.00019517350968638827,
+      "loss": 0.9222,
+      "step": 570
+    },
+    {
+      "epoch": 1.9071310116086235,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.0001949939447377521,
+      "loss": 0.9327,
+      "step": 575
+    },
+    {
+      "epoch": 1.9237147595356552,
+      "grad_norm": 1.125,
+      "learning_rate": 0.00019481118592489086,
+      "loss": 0.9252,
+      "step": 580
+    },
+    {
+      "epoch": 1.9402985074626866,
+      "grad_norm": 2.0,
+      "learning_rate": 0.00019462523939247863,
+      "loss": 0.9288,
+      "step": 585
+    },
+    {
+      "epoch": 1.956882255389718,
+      "grad_norm": 1.4921875,
+      "learning_rate": 0.00019443611139236634,
+      "loss": 0.9254,
+      "step": 590
+    },
+    {
+      "epoch": 1.9734660033167497,
+      "grad_norm": 0.7578125,
+      "learning_rate": 0.00019424380828337144,
+      "loss": 0.9156,
+      "step": 595
+    },
+    {
+      "epoch": 1.9900497512437811,
+      "grad_norm": 1.0859375,
+      "learning_rate": 0.00019404833653106413,
+      "loss": 0.9144,
+      "step": 600
+    },
+    {
+      "epoch": 2.0,
+      "eval_loss": 2.461941957473755,
+      "eval_runtime": 0.2391,
+      "eval_samples_per_second": 41.827,
+      "eval_steps_per_second": 4.183,
+      "step": 603
+    },
+    {
+      "epoch": 2.0066334991708126,
+      "grad_norm": 1.625,
+      "learning_rate": 0.00019384970270755006,
+      "loss": 0.9144,
+      "step": 605
+    },
+    {
+      "epoch": 2.0232172470978442,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.00019364791349124917,
+      "loss": 0.9027,
+      "step": 610
+    },
+    {
+      "epoch": 2.0398009950248754,
+      "grad_norm": 1.328125,
+      "learning_rate": 0.00019344297566667145,
+      "loss": 0.9047,
+      "step": 615
+    },
+    {
+      "epoch": 2.056384742951907,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.0001932348961241885,
+      "loss": 0.8985,
+      "step": 620
+    },
+    {
+      "epoch": 2.0729684908789388,
+      "grad_norm": 2.0625,
+      "learning_rate": 0.00019302368185980217,
+      "loss": 0.8889,
+      "step": 625
+    },
+    {
+      "epoch": 2.08955223880597,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00019280933997490912,
+      "loss": 0.8964,
+      "step": 630
+    },
+    {
+      "epoch": 2.1061359867330016,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.0001925918776760621,
+      "loss": 0.911,
+      "step": 635
+    },
+    {
+      "epoch": 2.1227197346600333,
+      "grad_norm": 0.95703125,
+      "learning_rate": 0.00019237130227472778,
+      "loss": 0.8997,
+      "step": 640
+    },
+    {
+      "epoch": 2.1393034825870645,
+      "grad_norm": 1.5859375,
+      "learning_rate": 0.00019214762118704076,
+      "loss": 0.8944,
+      "step": 645
+    },
+    {
+      "epoch": 2.155887230514096,
+      "grad_norm": 1.546875,
+      "learning_rate": 0.0001919208419335544,
+      "loss": 0.8983,
+      "step": 650
+    },
+    {
+      "epoch": 2.172470978441128,
+      "grad_norm": 1.609375,
+      "learning_rate": 0.00019169097213898767,
+      "loss": 0.8932,
+      "step": 655
+    },
+    {
+      "epoch": 2.189054726368159,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00019145801953196926,
+      "loss": 0.8847,
+      "step": 660
+    },
+    {
+      "epoch": 2.2056384742951907,
+      "grad_norm": 1.796875,
+      "learning_rate": 0.0001912219919447772,
+      "loss": 0.8974,
+      "step": 665
+    },
+    {
+      "epoch": 2.2222222222222223,
+      "grad_norm": 1.359375,
+      "learning_rate": 0.000190982897313076,
+      "loss": 0.8943,
+      "step": 670
+    },
+    {
+      "epoch": 2.2388059701492535,
+      "grad_norm": 1.8125,
+      "learning_rate": 0.00019074074367564947,
+      "loss": 0.8926,
+      "step": 675
+    },
+    {
+      "epoch": 2.255389718076285,
+      "grad_norm": 1.1171875,
+      "learning_rate": 0.0001904955391741307,
+      "loss": 0.8991,
+      "step": 680
+    },
+    {
+      "epoch": 2.271973466003317,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.0001902472920527281,
+      "loss": 0.9024,
+      "step": 685
+    },
+    {
+      "epoch": 2.288557213930348,
+      "grad_norm": 1.2421875,
+      "learning_rate": 0.0001899960106579485,
+      "loss": 0.8939,
+      "step": 690
+    },
+    {
+      "epoch": 2.3051409618573797,
+      "grad_norm": 1.09375,
+      "learning_rate": 0.00018974170343831622,
+      "loss": 0.8837,
+      "step": 695
+    },
+    {
+      "epoch": 2.3217247097844114,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.00018948437894408918,
+      "loss": 0.8919,
+      "step": 700
+    },
+    {
+      "epoch": 2.3383084577114426,
+      "grad_norm": 4.03125,
+      "learning_rate": 0.00018922404582697152,
+      "loss": 0.9034,
+      "step": 705
+    },
+    {
+      "epoch": 2.3548922056384742,
+      "grad_norm": 0.7421875,
+      "learning_rate": 0.00018896071283982241,
+      "loss": 0.8907,
+      "step": 710
+    },
+    {
+      "epoch": 2.371475953565506,
+      "grad_norm": 0.86328125,
+      "learning_rate": 0.00018869438883636214,
+      "loss": 0.8935,
+      "step": 715
+    },
+    {
+      "epoch": 2.388059701492537,
+      "grad_norm": 1.9140625,
+      "learning_rate": 0.00018842508277087404,
+      "loss": 0.8882,
+      "step": 720
+    },
+    {
+      "epoch": 2.4046434494195688,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.00018815280369790385,
+      "loss": 0.8947,
+      "step": 725
+    },
+    {
+      "epoch": 2.4212271973466004,
+      "grad_norm": 4.4375,
+      "learning_rate": 0.00018787756077195494,
+      "loss": 0.9034,
+      "step": 730
+    },
+    {
+      "epoch": 2.4378109452736316,
+      "grad_norm": 2.875,
+      "learning_rate": 0.00018759936324718067,
+      "loss": 0.9,
+      "step": 735
+    },
+    {
+      "epoch": 2.4543946932006633,
+      "grad_norm": 1.4609375,
+      "learning_rate": 0.00018731822047707327,
+      "loss": 0.8933,
+      "step": 740
+    },
+    {
+      "epoch": 2.470978441127695,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.0001870341419141493,
+      "loss": 0.8957,
+      "step": 745
+    },
+    {
+      "epoch": 2.487562189054726,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.00018674713710963183,
+      "loss": 0.8884,
+      "step": 750
+    },
+    {
+      "epoch": 2.504145936981758,
+      "grad_norm": 8.625,
+      "learning_rate": 0.0001864572157131294,
+      "loss": 0.9041,
+      "step": 755
+    },
+    {
+      "epoch": 2.5207296849087895,
+      "grad_norm": 2.875,
+      "learning_rate": 0.00018616438747231148,
+      "loss": 0.9185,
+      "step": 760
+    },
+    {
+      "epoch": 2.5373134328358207,
+      "grad_norm": 0.9140625,
+      "learning_rate": 0.0001858686622325808,
+      "loss": 0.8923,
+      "step": 765
+    },
+    {
+      "epoch": 2.5538971807628523,
+      "grad_norm": 1.8984375,
+      "learning_rate": 0.00018557004993674232,
+      "loss": 0.8744,
+      "step": 770
+    },
+    {
+      "epoch": 2.570480928689884,
+      "grad_norm": 0.78125,
+      "learning_rate": 0.00018526856062466886,
+      "loss": 0.8707,
+      "step": 775
+    },
+    {
+      "epoch": 2.587064676616915,
+      "grad_norm": 2.234375,
+      "learning_rate": 0.00018496420443296375,
+      "loss": 0.8833,
+      "step": 780
+    },
+    {
+      "epoch": 2.603648424543947,
+      "grad_norm": 0.8046875,
+      "learning_rate": 0.00018465699159461968,
+      "loss": 0.8999,
+      "step": 785
+    },
+    {
+      "epoch": 2.6202321724709785,
+      "grad_norm": 6.5625,
+      "learning_rate": 0.00018434693243867507,
+      "loss": 0.8843,
+      "step": 790
+    },
+    {
+      "epoch": 2.6368159203980097,
+      "grad_norm": 1.078125,
+      "learning_rate": 0.00018403403738986634,
+      "loss": 0.8776,
+      "step": 795
+    },
+    {
+      "epoch": 2.6533996683250414,
+      "grad_norm": 1.8984375,
+      "learning_rate": 0.00018371831696827786,
+      "loss": 0.8755,
+      "step": 800
+    },
+    {
+      "epoch": 2.669983416252073,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.0001833997817889878,
+      "loss": 0.8787,
+      "step": 805
+    },
+    {
+      "epoch": 2.6865671641791042,
+      "grad_norm": 10.125,
+      "learning_rate": 0.00018307844256171163,
+      "loss": 0.8804,
+      "step": 810
+    },
+    {
+      "epoch": 2.703150912106136,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.00018275431009044183,
+      "loss": 0.874,
+      "step": 815
+    },
+    {
+      "epoch": 2.7197346600331676,
+      "grad_norm": 2.03125,
+      "learning_rate": 0.00018242739527308456,
+      "loss": 0.8792,
+      "step": 820
+    },
+    {
+      "epoch": 2.7363184079601988,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.00018209770910109353,
+      "loss": 0.9012,
+      "step": 825
+    },
+    {
+      "epoch": 2.7529021558872304,
+      "grad_norm": 2.625,
+      "learning_rate": 0.00018176526265910022,
+      "loss": 0.8855,
+      "step": 830
+    },
+    {
+      "epoch": 2.769485903814262,
+      "grad_norm": 3.171875,
+      "learning_rate": 0.00018143006712454125,
+      "loss": 0.8863,
+      "step": 835
+    },
+    {
+      "epoch": 2.7860696517412933,
+      "grad_norm": 1.03125,
+      "learning_rate": 0.0001810921337672826,
+      "loss": 0.875,
+      "step": 840
+    },
+    {
+      "epoch": 2.802653399668325,
+      "grad_norm": 3.109375,
+      "learning_rate": 0.00018075147394924063,
+      "loss": 0.8712,
+      "step": 845
+    },
+    {
+      "epoch": 2.8192371475953566,
+      "grad_norm": 1.1640625,
+      "learning_rate": 0.0001804080991240003,
+      "loss": 0.8635,
+      "step": 850
+    },
+    {
+      "epoch": 2.835820895522388,
+      "grad_norm": 2.828125,
+      "learning_rate": 0.00018006202083642973,
+      "loss": 0.8759,
+      "step": 855
+    },
+    {
+      "epoch": 2.8524046434494195,
+      "grad_norm": 5.28125,
+      "learning_rate": 0.00017971325072229226,
+      "loss": 0.872,
+      "step": 860
+    },
+    {
+      "epoch": 2.868988391376451,
+      "grad_norm": 1.6875,
+      "learning_rate": 0.00017936180050785524,
+      "loss": 0.8586,
+      "step": 865
+    },
+    {
+      "epoch": 2.8855721393034823,
+      "grad_norm": 3.71875,
+      "learning_rate": 0.00017900768200949574,
+      "loss": 0.8725,
+      "step": 870
+    },
+    {
+      "epoch": 2.902155887230514,
+      "grad_norm": 0.96875,
+      "learning_rate": 0.00017865090713330313,
+      "loss": 0.87,
+      "step": 875
+    },
+    {
+      "epoch": 2.9187396351575456,
+      "grad_norm": 3.484375,
+      "learning_rate": 0.00017829148787467904,
+      "loss": 0.8607,
+      "step": 880
+    },
+    {
+      "epoch": 2.935323383084577,
+      "grad_norm": 1.109375,
+      "learning_rate": 0.0001779294363179338,
+      "loss": 0.8702,
+      "step": 885
+    },
+    {
+      "epoch": 2.9519071310116085,
+      "grad_norm": 0.75390625,
+      "learning_rate": 0.00017756476463588037,
+      "loss": 0.8609,
+      "step": 890
+    },
+    {
+      "epoch": 2.96849087893864,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.0001771974850894248,
+      "loss": 0.8667,
+      "step": 895
+    },
+    {
+      "epoch": 2.9850746268656714,
+      "grad_norm": 8.0,
+      "learning_rate": 0.00017682761002715423,
+      "loss": 0.8616,
+      "step": 900
+    },
+    {
+      "epoch": 2.998341625207297,
+      "eval_loss": 2.453416347503662,
+      "eval_runtime": 0.2591,
+      "eval_samples_per_second": 38.597,
+      "eval_steps_per_second": 3.86,
+      "step": 904
+    },
+    {
+      "epoch": 3.001658374792703,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.0001764551518849217,
+      "loss": 0.8514,
+      "step": 905
+    },
+    {
+      "epoch": 3.0182421227197347,
+      "grad_norm": 0.90234375,
+      "learning_rate": 0.0001760801231854278,
+      "loss": 0.839,
+      "step": 910
+    },
+    {
+      "epoch": 3.0348258706467663,
+      "grad_norm": 6.375,
+      "learning_rate": 0.00017570253653779994,
+      "loss": 0.8316,
+      "step": 915
+    },
+    {
+      "epoch": 3.0514096185737976,
+      "grad_norm": 0.6953125,
+      "learning_rate": 0.00017532240463716815,
+      "loss": 0.8412,
+      "step": 920
+    },
+    {
+      "epoch": 3.067993366500829,
+      "grad_norm": 9.125,
+      "learning_rate": 0.00017493974026423848,
+      "loss": 0.8423,
+      "step": 925
+    },
+    {
+      "epoch": 3.084577114427861,
+      "grad_norm": 1.90625,
+      "learning_rate": 0.00017455455628486303,
+      "loss": 0.8468,
+      "step": 930
+    },
+    {
+      "epoch": 3.101160862354892,
+      "grad_norm": 1.3984375,
+      "learning_rate": 0.0001741668656496077,
+      "loss": 0.8545,
+      "step": 935
+    },
+    {
+      "epoch": 3.1177446102819237,
+      "grad_norm": 8.125,
+      "learning_rate": 0.0001737766813933164,
+      "loss": 0.8576,
+      "step": 940
+    },
+    {
+      "epoch": 3.1343283582089554,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0001733840166346731,
+      "loss": 0.8469,
+      "step": 945
+    },
+    {
+      "epoch": 3.1509121061359866,
+      "grad_norm": 1.5703125,
+      "learning_rate": 0.0001729888845757606,
+      "loss": 0.8537,
+      "step": 950
+    },
+    {
+      "epoch": 3.1674958540630183,
+      "grad_norm": 1.1953125,
+      "learning_rate": 0.00017259129850161675,
+      "loss": 0.8543,
+      "step": 955
+    },
+    {
+      "epoch": 3.18407960199005,
+      "grad_norm": 1.0078125,
+      "learning_rate": 0.00017219127177978765,
+      "loss": 0.8281,
+      "step": 960
+    },
+    {
+      "epoch": 3.200663349917081,
+      "grad_norm": 1.296875,
+      "learning_rate": 0.0001717888178598783,
+      "loss": 0.8428,
+      "step": 965
+    },
+    {
+      "epoch": 3.217247097844113,
+      "grad_norm": 1.1796875,
+      "learning_rate": 0.00017138395027310044,
+      "loss": 0.8425,
+      "step": 970
+    },
+    {
+      "epoch": 3.2338308457711444,
+      "grad_norm": 1.203125,
+      "learning_rate": 0.00017097668263181748,
+      "loss": 0.846,
+      "step": 975
+    },
+    {
+      "epoch": 3.2504145936981756,
+      "grad_norm": 1.2578125,
+      "learning_rate": 0.00017056702862908702,
+      "loss": 0.8257,
+      "step": 980
+    },
+    {
+      "epoch": 3.2669983416252073,
+      "grad_norm": 14.3125,
+      "learning_rate": 0.00017015500203820022,
+      "loss": 0.8441,
+      "step": 985
+    },
+    {
+      "epoch": 3.283582089552239,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00016974061671221885,
+      "loss": 0.8499,
+      "step": 990
+    },
+    {
+      "epoch": 3.30016583747927,
+      "grad_norm": 1.421875,
+      "learning_rate": 0.00016932388658350963,
+      "loss": 0.8586,
+      "step": 995
+    },
+    {
+      "epoch": 3.316749585406302,
+      "grad_norm": 0.890625,
+      "learning_rate": 0.0001689048256632756,
+      "loss": 0.8364,
+      "step": 1000
+    },
+    {
+      "epoch": 3.3333333333333335,
+      "grad_norm": 0.94140625,
+      "learning_rate": 0.00016848344804108517,
+      "loss": 0.826,
+      "step": 1005
+    },
+    {
+      "epoch": 3.3499170812603647,
+      "grad_norm": 0.73046875,
+      "learning_rate": 0.0001680597678843983,
+      "loss": 0.8269,
+      "step": 1010
+    },
+    {
+      "epoch": 3.3665008291873963,
+      "grad_norm": 2.796875,
+      "learning_rate": 0.00016763379943809028,
+      "loss": 0.8436,
+      "step": 1015
+    },
+    {
+      "epoch": 3.383084577114428,
+      "grad_norm": 2.84375,
+      "learning_rate": 0.00016720555702397281,
+      "loss": 0.8422,
+      "step": 1020
+    },
+    {
+      "epoch": 3.399668325041459,
+      "grad_norm": 8.5625,
+      "learning_rate": 0.00016677505504031223,
+      "loss": 0.8417,
+      "step": 1025
+    },
+    {
+      "epoch": 3.416252072968491,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00016634230796134576,
+      "loss": 0.8262,
+      "step": 1030
+    },
+    {
+      "epoch": 3.4328358208955225,
+      "grad_norm": 0.9921875,
+      "learning_rate": 0.00016590733033679456,
+      "loss": 0.8463,
+      "step": 1035
+    },
+    {
+      "epoch": 3.4494195688225537,
+      "grad_norm": 1.796875,
+      "learning_rate": 0.00016547013679137483,
+      "loss": 0.8434,
+      "step": 1040
+    },
+    {
+      "epoch": 3.4660033167495854,
+      "grad_norm": 1.0546875,
+      "learning_rate": 0.00016503074202430575,
+      "loss": 0.8328,
+      "step": 1045
+    },
+    {
+      "epoch": 3.482587064676617,
+      "grad_norm": 0.546875,
+      "learning_rate": 0.00016458916080881565,
+      "loss": 0.8209,
+      "step": 1050
+    },
+    {
+      "epoch": 3.4991708126036483,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.00016414540799164495,
+      "loss": 0.8171,
+      "step": 1055
+    },
+    {
+      "epoch": 3.51575456053068,
+      "grad_norm": 5.65625,
+      "learning_rate": 0.0001636994984925473,
+      "loss": 0.8226,
+      "step": 1060
+    },
+    {
+      "epoch": 3.5323383084577116,
+      "grad_norm": 0.6328125,
+      "learning_rate": 0.00016325144730378773,
+      "loss": 0.8126,
+      "step": 1065
+    },
+    {
+      "epoch": 3.5489220563847432,
+      "grad_norm": 0.5546875,
+      "learning_rate": 0.00016280126948963863,
+      "loss": 0.807,
+      "step": 1070
+    },
+    {
+      "epoch": 3.5655058043117744,
+      "grad_norm": 0.5859375,
+      "learning_rate": 0.00016234898018587337,
+      "loss": 0.8255,
+      "step": 1075
+    },
+    {
+      "epoch": 3.582089552238806,
+      "grad_norm": 0.89453125,
+      "learning_rate": 0.0001618945945992573,
+      "loss": 0.8317,
+      "step": 1080
+    },
+    {
+      "epoch": 3.5986733001658378,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.00016143812800703644,
+      "loss": 0.8194,
+      "step": 1085
+    },
+    {
+      "epoch": 3.615257048092869,
+      "grad_norm": 0.5703125,
+      "learning_rate": 0.00016097959575642402,
+      "loss": 0.8146,
+      "step": 1090
+    },
+    {
+      "epoch": 3.6318407960199006,
+      "grad_norm": 0.6796875,
+      "learning_rate": 0.00016051901326408422,
+      "loss": 0.8105,
+      "step": 1095
+    },
+    {
+      "epoch": 3.6484245439469323,
+      "grad_norm": 0.52734375,
+      "learning_rate": 0.0001600563960156141,
+      "loss": 0.8126,
+      "step": 1100
+    },
+    {
+      "epoch": 3.6650082918739635,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.0001595917595650227,
+      "loss": 0.812,
+      "step": 1105
+    },
+    {
+      "epoch": 3.681592039800995,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.0001591251195342083,
+      "loss": 0.8064,
+      "step": 1110
+    },
+    {
+      "epoch": 3.698175787728027,
+      "grad_norm": 0.7109375,
+      "learning_rate": 0.00015865649161243297,
+      "loss": 0.8133,
+      "step": 1115
+    },
+    {
+      "epoch": 3.714759535655058,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.0001581858915557953,
+      "loss": 0.8223,
+      "step": 1120
+    },
+    {
+      "epoch": 3.7313432835820897,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00015771333518670047,
+      "loss": 0.8138,
+      "step": 1125
+    },
+    {
+      "epoch": 3.7479270315091213,
+      "grad_norm": 1.890625,
+      "learning_rate": 0.00015723883839332837,
+      "loss": 0.8132,
+      "step": 1130
+    },
+    {
+      "epoch": 3.7645107794361525,
+      "grad_norm": 1.28125,
+      "learning_rate": 0.00015676241712909937,
+      "loss": 0.812,
+      "step": 1135
+    },
+    {
+      "epoch": 3.781094527363184,
+      "grad_norm": 2.28125,
+      "learning_rate": 0.00015628408741213794,
+      "loss": 0.8052,
+      "step": 1140
+    },
+    {
+      "epoch": 3.797678275290216,
+      "grad_norm": 0.875,
+      "learning_rate": 0.00015580386532473413,
+      "loss": 0.8115,
+      "step": 1145
+    },
+    {
+      "epoch": 3.814262023217247,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.0001553217670128028,
+      "loss": 0.808,
+      "step": 1150
+    },
+    {
+      "epoch": 3.8308457711442787,
+      "grad_norm": 0.90234375,
+      "learning_rate": 0.00015483780868534087,
+      "loss": 0.8034,
+      "step": 1155
+    },
+    {
+      "epoch": 3.8474295190713104,
+      "grad_norm": 0.55859375,
+      "learning_rate": 0.00015435200661388207,
+      "loss": 0.8123,
+      "step": 1160
+    },
+    {
+      "epoch": 3.8640132669983416,
+      "grad_norm": 0.5078125,
+      "learning_rate": 0.0001538643771319503,
+      "loss": 0.8004,
+      "step": 1165
+    },
+    {
+      "epoch": 3.8805970149253732,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.0001533749366345101,
+      "loss": 0.7991,
+      "step": 1170
+    },
+    {
+      "epoch": 3.897180762852405,
+      "grad_norm": 0.451171875,
+      "learning_rate": 0.00015288370157741553,
+      "loss": 0.8091,
+      "step": 1175
+    },
+    {
+      "epoch": 3.913764510779436,
+      "grad_norm": 0.43359375,
+      "learning_rate": 0.00015239068847685704,
+      "loss": 0.7937,
+      "step": 1180
+    },
+    {
+      "epoch": 3.9303482587064678,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.00015189591390880582,
+      "loss": 0.8036,
+      "step": 1185
+    },
+    {
+      "epoch": 3.9469320066334994,
+      "grad_norm": 0.59375,
+      "learning_rate": 0.000151399394508457,
+      "loss": 0.8098,
+      "step": 1190
+    },
+    {
+      "epoch": 3.9635157545605306,
+      "grad_norm": 0.486328125,
+      "learning_rate": 0.0001509011469696698,
+      "loss": 0.8046,
+      "step": 1195
+    },
+    {
+      "epoch": 3.9800995024875623,
+      "grad_norm": 0.54296875,
+      "learning_rate": 0.00015040118804440666,
+      "loss": 0.8051,
+      "step": 1200
+    },
+    {
+      "epoch": 3.996683250414594,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.00014989953454216975,
+      "loss": 0.8075,
+      "step": 1205
+    },
+    {
+      "epoch": 4.0,
+      "eval_loss": 2.4530348777770996,
+      "eval_runtime": 0.237,
+      "eval_samples_per_second": 42.199,
+      "eval_steps_per_second": 4.22,
+      "step": 1206
+    },
+    {
+      "epoch": 4.013266998341625,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.00014939620332943604,
+      "loss": 0.7894,
+      "step": 1210
+    },
+    {
+      "epoch": 4.029850746268656,
+      "grad_norm": 0.5,
+      "learning_rate": 0.00014889121132908995,
+      "loss": 0.7843,
+      "step": 1215
+    },
+    {
+      "epoch": 4.0464344941956885,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.00014838457551985464,
+      "loss": 0.7768,
+      "step": 1220
+    },
+    {
+      "epoch": 4.06301824212272,
+      "grad_norm": 0.60546875,
+      "learning_rate": 0.00014787631293572092,
+      "loss": 0.7876,
+      "step": 1225
+    },
+    {
+      "epoch": 4.079601990049751,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.0001473664406653747,
+      "loss": 0.7907,
+      "step": 1230
+    },
+    {
+      "epoch": 4.096185737976783,
+      "grad_norm": 0.6171875,
+      "learning_rate": 0.0001468549758516224,
+      "loss": 0.7897,
+      "step": 1235
+    },
+    {
+      "epoch": 4.112769485903814,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00014634193569081457,
+      "loss": 0.7859,
+      "step": 1240
+    },
+    {
+      "epoch": 4.129353233830845,
+      "grad_norm": 0.62109375,
+      "learning_rate": 0.00014582733743226758,
+      "loss": 0.7942,
+      "step": 1245
+    },
+    {
+      "epoch": 4.1459369817578775,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00014531119837768398,
+      "loss": 0.7878,
+      "step": 1250
+    },
+    {
+      "epoch": 4.162520729684909,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00014479353588057052,
+      "loss": 0.7883,
+      "step": 1255
+    },
+    {
+      "epoch": 4.17910447761194,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00014427436734565474,
+      "loss": 0.7871,
+      "step": 1260
+    },
+    {
+      "epoch": 4.195688225538972,
+      "grad_norm": 0.6640625,
+      "learning_rate": 0.0001437537102282999,
+      "loss": 0.7785,
+      "step": 1265
+    },
+    {
+      "epoch": 4.212271973466003,
+      "grad_norm": 2.984375,
+      "learning_rate": 0.00014323158203391804,
+      "loss": 0.7883,
+      "step": 1270
+    },
+    {
+      "epoch": 4.2288557213930345,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.0001427080003173813,
+      "loss": 0.7945,
+      "step": 1275
+    },
+    {
+      "epoch": 4.2454394693200666,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.00014218298268243196,
+      "loss": 0.778,
+      "step": 1280
+    },
+    {
+      "epoch": 4.262023217247098,
+      "grad_norm": 0.91796875,
+      "learning_rate": 0.00014165654678109025,
+      "loss": 0.7901,
+      "step": 1285
+    },
+    {
+      "epoch": 4.278606965174129,
+      "grad_norm": 0.63671875,
+      "learning_rate": 0.00014112871031306119,
+      "loss": 0.7764,
+      "step": 1290
+    },
+    {
+      "epoch": 4.295190713101161,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.00014059949102513913,
+      "loss": 0.7881,
+      "step": 1295
+    },
+    {
+      "epoch": 4.311774461028192,
+      "grad_norm": 0.44921875,
+      "learning_rate": 0.00014006890671061143,
+      "loss": 0.7803,
+      "step": 1300
+    },
+    {
+      "epoch": 4.3283582089552235,
+      "grad_norm": 0.62890625,
+      "learning_rate": 0.00013953697520866003,
+      "loss": 0.7756,
+      "step": 1305
+    },
+    {
+      "epoch": 4.344941956882256,
+      "grad_norm": 0.4921875,
+      "learning_rate": 0.0001390037144037616,
+      "loss": 0.782,
+      "step": 1310
+    },
+    {
+      "epoch": 4.361525704809287,
+      "grad_norm": 0.4140625,
+      "learning_rate": 0.00013846914222508645,
+      "loss": 0.7779,
+      "step": 1315
+    },
+    {
+      "epoch": 4.378109452736318,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00013793327664589545,
+      "loss": 0.7739,
+      "step": 1320
+    },
+    {
+      "epoch": 4.39469320066335,
+      "grad_norm": 0.443359375,
+      "learning_rate": 0.00013739613568293605,
+      "loss": 0.7778,
+      "step": 1325
+    },
+    {
+      "epoch": 4.411276948590381,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00013685773739583617,
+      "loss": 0.7845,
+      "step": 1330
+    },
+    {
+      "epoch": 4.4278606965174125,
+      "grad_norm": 0.47265625,
+      "learning_rate": 0.00013631809988649734,
+      "loss": 0.7889,
+      "step": 1335
+    },
+    {
+      "epoch": 4.444444444444445,
+      "grad_norm": 0.5234375,
+      "learning_rate": 0.00013577724129848587,
+      "loss": 0.7835,
+      "step": 1340
+    },
+    {
+      "epoch": 4.461028192371476,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.00013523517981642286,
+      "loss": 0.7868,
+      "step": 1345
+    },
+    {
+      "epoch": 4.477611940298507,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.00013469193366537292,
+      "loss": 0.774,
+      "step": 1350
+    },
+    {
+      "epoch": 4.494195688225539,
+      "grad_norm": 0.484375,
+      "learning_rate": 0.00013414752111023118,
+      "loss": 0.7808,
+      "step": 1355
+    },
+    {
+      "epoch": 4.51077943615257,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.00013360196045510955,
+      "loss": 0.7855,
+      "step": 1360
+    },
+    {
+      "epoch": 4.5273631840796025,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.00013305527004272086,
+      "loss": 0.7752,
+      "step": 1365
+    },
+    {
+      "epoch": 4.543946932006634,
+      "grad_norm": 0.65625,
+      "learning_rate": 0.00013250746825376256,
+      "loss": 0.78,
+      "step": 1370
+    },
+    {
+      "epoch": 4.560530679933665,
+      "grad_norm": 0.46875,
+      "learning_rate": 0.00013195857350629836,
+      "loss": 0.7777,
+      "step": 1375
+    },
+    {
+      "epoch": 4.577114427860696,
+      "grad_norm": 1.046875,
+      "learning_rate": 0.0001314086042551394,
+      "loss": 0.7771,
+      "step": 1380
+    },
+    {
+      "epoch": 4.593698175787728,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00013085757899122333,
+      "loss": 0.7768,
+      "step": 1385
+    },
+    {
+      "epoch": 4.610281923714759,
+      "grad_norm": 0.482421875,
+      "learning_rate": 0.00013030551624099287,
+      "loss": 0.7795,
+      "step": 1390
+    },
+    {
+      "epoch": 4.6268656716417915,
+      "grad_norm": 0.76171875,
+      "learning_rate": 0.00012975243456577292,
+      "loss": 0.7815,
+      "step": 1395
+    },
+    {
+      "epoch": 4.643449419568823,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.0001291983525611464,
+      "loss": 0.7815,
+      "step": 1400
+    },
+    {
+      "epoch": 4.660033167495854,
+      "grad_norm": 0.58984375,
+      "learning_rate": 0.00012864328885632903,
+      "loss": 0.7843,
+      "step": 1405
+    },
+    {
+      "epoch": 4.676616915422885,
+      "grad_norm": 0.61328125,
+      "learning_rate": 0.00012808726211354306,
+      "loss": 0.7826,
+      "step": 1410
+    },
+    {
+      "epoch": 4.693200663349917,
+      "grad_norm": 0.94921875,
+      "learning_rate": 0.0001275302910273899,
+      "loss": 0.7757,
+      "step": 1415
+    },
+    {
+      "epoch": 4.7097844112769485,
+      "grad_norm": 0.765625,
+      "learning_rate": 0.00012697239432422116,
+      "loss": 0.7811,
+      "step": 1420
+    },
+    {
+      "epoch": 4.726368159203981,
+      "grad_norm": 0.6875,
+      "learning_rate": 0.00012641359076150961,
+      "loss": 0.7787,
+      "step": 1425
+    },
+    {
+      "epoch": 4.742951907131012,
+      "grad_norm": 0.4609375,
+      "learning_rate": 0.00012585389912721812,
+      "loss": 0.7804,
+      "step": 1430
+    },
+    {
+      "epoch": 4.759535655058043,
+      "grad_norm": 0.6015625,
+      "learning_rate": 0.00012529333823916807,
+      "loss": 0.783,
+      "step": 1435
+    },
+    {
+      "epoch": 4.776119402985074,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.00012473192694440678,
+      "loss": 0.7787,
+      "step": 1440
+    },
+    {
+      "epoch": 4.792703150912106,
+      "grad_norm": 0.423828125,
+      "learning_rate": 0.00012416968411857365,
+      "loss": 0.7786,
+      "step": 1445
+    },
+    {
+      "epoch": 4.8092868988391375,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00012360662866526566,
+      "loss": 0.7786,
+      "step": 1450
+    },
+    {
+      "epoch": 4.82587064676617,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.0001230427795154018,
+      "loss": 0.7767,
+      "step": 1455
+    },
+    {
+      "epoch": 4.842454394693201,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00012247815562658646,
+      "loss": 0.7854,
+      "step": 1460
+    },
+    {
+      "epoch": 4.859038142620232,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.00012191277598247219,
+      "loss": 0.7787,
+      "step": 1465
+    },
+    {
+      "epoch": 4.875621890547263,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00012134665959212137,
+      "loss": 0.7781,
+      "step": 1470
+    },
+    {
+      "epoch": 4.892205638474295,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00012077982548936704,
+      "loss": 0.7735,
+      "step": 1475
+    },
+    {
+      "epoch": 4.908789386401327,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.00012021229273217302,
+      "loss": 0.7743,
+      "step": 1480
+    },
+    {
+      "epoch": 4.925373134328359,
+      "grad_norm": 0.416015625,
+      "learning_rate": 0.00011964408040199315,
+      "loss": 0.7717,
+      "step": 1485
+    },
+    {
+      "epoch": 4.94195688225539,
+      "grad_norm": 0.44140625,
+      "learning_rate": 0.00011907520760312967,
+      "loss": 0.7756,
+      "step": 1490
+    },
+    {
+      "epoch": 4.958540630182421,
+      "grad_norm": 0.498046875,
+      "learning_rate": 0.00011850569346209099,
+      "loss": 0.7672,
+      "step": 1495
+    },
+    {
+      "epoch": 4.975124378109452,
+      "grad_norm": 0.39453125,
+      "learning_rate": 0.00011793555712694855,
+      "loss": 0.7704,
+      "step": 1500
+    },
+    {
+      "epoch": 4.991708126036484,
+      "grad_norm": 0.421875,
+      "learning_rate": 0.00011736481776669306,
+      "loss": 0.7724,
+      "step": 1505
+    },
+    {
+      "epoch": 4.9983416252072965,
+      "eval_loss": 2.467383623123169,
+      "eval_runtime": 0.2562,
+      "eval_samples_per_second": 39.032,
+      "eval_steps_per_second": 3.903,
+      "step": 1507
+    },
+    {
+      "epoch": 5.008291873963516,
+      "grad_norm": 0.40234375,
+      "learning_rate": 0.00011679349457058997,
+      "loss": 0.7691,
+      "step": 1510
+    },
+    {
+      "epoch": 5.024875621890548,
+      "grad_norm": 0.51171875,
+      "learning_rate": 0.0001162216067475344,
+      "loss": 0.759,
+      "step": 1515
+    },
+    {
+      "epoch": 5.041459369817579,
+      "grad_norm": 0.470703125,
+      "learning_rate": 0.00011564917352540519,
+      "loss": 0.7557,
+      "step": 1520
+    },
+    {
+      "epoch": 5.05804311774461,
+      "grad_norm": 0.53125,
+      "learning_rate": 0.00011507621415041837,
+      "loss": 0.7592,
+      "step": 1525
+    },
+    {
+      "epoch": 5.074626865671641,
+      "grad_norm": 0.458984375,
+      "learning_rate": 0.0001145027478864803,
+      "loss": 0.752,
+      "step": 1530
+    },
+    {
+      "epoch": 5.091210613598673,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00011392879401453977,
+      "loss": 0.7546,
+      "step": 1535
+    },
+    {
+      "epoch": 5.107794361525705,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.0001133543718319398,
+      "loss": 0.7628,
+      "step": 1540
+    },
+    {
+      "epoch": 5.124378109452737,
+      "grad_norm": 0.8125,
+      "learning_rate": 0.00011277950065176887,
+      "loss": 0.7526,
+      "step": 1545
+    },
+    {
+      "epoch": 5.140961857379768,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.0001122041998022116,
+      "loss": 0.7634,
+      "step": 1550
+    },
+    {
+      "epoch": 5.157545605306799,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.00011162848862589872,
+      "loss": 0.7646,
+      "step": 1555
+    },
+    {
+      "epoch": 5.174129353233831,
+      "grad_norm": 0.50390625,
+      "learning_rate": 0.00011105238647925696,
+      "loss": 0.754,
+      "step": 1560
+    },
+    {
+      "epoch": 5.1907131011608625,
+      "grad_norm": 0.58203125,
+      "learning_rate": 0.00011047591273185818,
+      "loss": 0.756,
+      "step": 1565
+    },
+    {
+      "epoch": 5.207296849087894,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.00010989908676576807,
+      "loss": 0.7602,
+      "step": 1570
+    },
+    {
+      "epoch": 5.223880597014926,
+      "grad_norm": 0.41796875,
+      "learning_rate": 0.0001093219279748946,
+      "loss": 0.7649,
+      "step": 1575
+    },
+    {
+      "epoch": 5.240464344941957,
+      "grad_norm": 0.431640625,
+      "learning_rate": 0.00010874445576433574,
+      "loss": 0.7522,
+      "step": 1580
+    },
+    {
+      "epoch": 5.257048092868988,
+      "grad_norm": 0.435546875,
+      "learning_rate": 0.00010816668954972737,
+      "loss": 0.7545,
+      "step": 1585
+    },
+    {
+      "epoch": 5.273631840796019,
+      "grad_norm": 0.4375,
+      "learning_rate": 0.00010758864875659022,
+      "loss": 0.7539,
+      "step": 1590
+    },
+    {
+      "epoch": 5.2902155887230515,
+      "grad_norm": 0.46484375,
+      "learning_rate": 0.00010701035281967682,
+      "loss": 0.7522,
+      "step": 1595
+    },
+    {
+      "epoch": 5.306799336650083,
+      "grad_norm": 0.474609375,
+      "learning_rate": 0.00010643182118231812,
+      "loss": 0.7639,
+      "step": 1600
+    },
+    {
+      "epoch": 5.323383084577115,
+      "grad_norm": 0.515625,
+      "learning_rate": 0.0001058530732957697,
+      "loss": 0.752,
+      "step": 1605
+    },
+    {
+      "epoch": 5.339966832504146,
+      "grad_norm": 0.53515625,
+      "learning_rate": 0.0001052741286185579,
+      "loss": 0.7648,
+      "step": 1610
+    },
+    {
+      "epoch": 5.356550580431177,
+      "grad_norm": 0.45703125,
+      "learning_rate": 0.00010469500661582536,
+      "loss": 0.7596,
+      "step": 1615
+    },
+    {
+      "epoch": 5.373134328358209,
+      "grad_norm": 0.42578125,
+      "learning_rate": 0.00010411572675867687,
+      "loss": 0.7512,
+      "step": 1620
+    },
+    {
+      "epoch": 5.389718076285241,
+      "grad_norm": 0.494140625,
+      "learning_rate": 0.0001035363085235245,
+      "loss": 0.761,
+      "step": 1625
+    },
+    {
+      "epoch": 5.406301824212272,
+      "grad_norm": 0.48046875,
+      "learning_rate": 0.00010295677139143282,
+      "loss": 0.7581,
+      "step": 1630
+    },
+    {
+      "epoch": 5.422885572139304,
+      "grad_norm": 0.447265625,
+      "learning_rate": 0.0001023771348474639,
+      "loss": 0.7573,
+      "step": 1635
+    },
+    {
+      "epoch": 5.439469320066335,
+      "grad_norm": 0.4453125,
+      "learning_rate": 0.00010179741838002232,
+      "loss": 0.7548,
+      "step": 1640
+    },
+    {
+      "epoch": 5.456053067993366,
+      "grad_norm": 0.453125,
+      "learning_rate": 0.00010121764148019976,
+      "loss": 0.7505,
+      "step": 1645
+    },
+    {
+      "epoch": 5.472636815920398,
+      "grad_norm": 0.466796875,
+      "learning_rate": 0.00010063782364111971,
+      "loss": 0.7499,
+      "step": 1650
+    },
+    {
+      "epoch": 5.48922056384743,
+      "grad_norm": 0.51953125,
+      "learning_rate": 0.00010005798435728215,
+      "loss": 0.7544,
+      "step": 1655
+    },
+    {
+      "epoch": 5.505804311774461,
+      "grad_norm": 0.50390625,
+      "learning_rate": 9.947814312390808e-05,
+      "loss": 0.7569,
+      "step": 1660
+    },
+    {
+      "epoch": 5.522388059701493,
+      "grad_norm": 0.67578125,
+      "learning_rate": 9.889831943628401e-05,
+      "loss": 0.7528,
+      "step": 1665
+    },
+    {
+      "epoch": 5.538971807628524,
+      "grad_norm": 0.60546875,
+      "learning_rate": 9.831853278910656e-05,
+      "loss": 0.7541,
+      "step": 1670
+    },
+    {
+      "epoch": 5.555555555555555,
+      "grad_norm": 0.5859375,
+      "learning_rate": 9.773880267582697e-05,
+      "loss": 0.7611,
+      "step": 1675
+    },
+    {
+      "epoch": 5.572139303482587,
+      "grad_norm": 0.640625,
+      "learning_rate": 9.715914858799576e-05,
+      "loss": 0.7516,
+      "step": 1680
+    },
+    {
+      "epoch": 5.588723051409619,
+      "grad_norm": 0.93359375,
+      "learning_rate": 9.657959001460732e-05,
+      "loss": 0.7539,
+      "step": 1685
+    },
+    {
+      "epoch": 5.60530679933665,
+      "grad_norm": 0.69921875,
+      "learning_rate": 9.600014644144453e-05,
+      "loss": 0.7526,
+      "step": 1690
+    },
+    {
+      "epoch": 5.621890547263682,
+      "grad_norm": 0.5,
+      "learning_rate": 9.542083735042398e-05,
+      "loss": 0.7533,
+      "step": 1695
+    },
+    {
+      "epoch": 5.638474295190713,
+      "grad_norm": 0.47265625,
+      "learning_rate": 9.484168221894062e-05,
+      "loss": 0.7569,
+      "step": 1700
+    },
+    {
+      "epoch": 5.655058043117744,
+      "grad_norm": 0.64453125,
+      "learning_rate": 9.426270051921304e-05,
+      "loss": 0.7549,
+      "step": 1705
+    },
+    {
+      "epoch": 5.6716417910447765,
+      "grad_norm": 0.435546875,
+      "learning_rate": 9.36839117176287e-05,
+      "loss": 0.7527,
+      "step": 1710
+    },
+    {
+      "epoch": 5.688225538971808,
+      "grad_norm": 0.4140625,
+      "learning_rate": 9.310533527408952e-05,
+      "loss": 0.7564,
+      "step": 1715
+    },
+    {
+      "epoch": 5.704809286898839,
+      "grad_norm": 0.447265625,
+      "learning_rate": 9.252699064135758e-05,
+      "loss": 0.7598,
+      "step": 1720
+    },
+    {
+      "epoch": 5.721393034825871,
+      "grad_norm": 0.6171875,
+      "learning_rate": 9.194889726440105e-05,
+      "loss": 0.7567,
+      "step": 1725
+    },
+    {
+      "epoch": 5.737976782752902,
+      "grad_norm": 0.419921875,
+      "learning_rate": 9.137107457974046e-05,
+      "loss": 0.7579,
+      "step": 1730
+    },
+    {
+      "epoch": 5.7545605306799334,
+      "grad_norm": 0.5078125,
+      "learning_rate": 9.079354201479513e-05,
+      "loss": 0.7465,
+      "step": 1735
+    },
+    {
+      "epoch": 5.7711442786069655,
+      "grad_norm": 0.49609375,
+      "learning_rate": 9.02163189872301e-05,
+      "loss": 0.7597,
+      "step": 1740
+    },
+    {
+      "epoch": 5.787728026533997,
+      "grad_norm": 0.4453125,
+      "learning_rate": 8.96394249043032e-05,
+      "loss": 0.7522,
+      "step": 1745
+    },
+    {
+      "epoch": 5.804311774461028,
+      "grad_norm": 0.4375,
+      "learning_rate": 8.906287916221259e-05,
+      "loss": 0.7494,
+      "step": 1750
+    },
+    {
+      "epoch": 5.82089552238806,
+      "grad_norm": 0.416015625,
+      "learning_rate": 8.84867011454445e-05,
+      "loss": 0.7393,
+      "step": 1755
+    },
+    {
+      "epoch": 5.837479270315091,
+      "grad_norm": 0.431640625,
+      "learning_rate": 8.791091022612168e-05,
+      "loss": 0.7463,
+      "step": 1760
+    },
+    {
+      "epoch": 5.8540630182421225,
+      "grad_norm": 0.4375,
+      "learning_rate": 8.733552576335195e-05,
+      "loss": 0.749,
+      "step": 1765
+    },
+    {
+      "epoch": 5.870646766169155,
+      "grad_norm": 0.421875,
+      "learning_rate": 8.676056710257734e-05,
+      "loss": 0.7518,
+      "step": 1770
+    },
+    {
+      "epoch": 5.887230514096186,
+      "grad_norm": 0.447265625,
+      "learning_rate": 8.618605357492372e-05,
+      "loss": 0.7579,
+      "step": 1775
+    },
+    {
+      "epoch": 5.903814262023217,
+      "grad_norm": 0.48046875,
+      "learning_rate": 8.561200449655064e-05,
+      "loss": 0.7557,
+      "step": 1780
+    },
+    {
+      "epoch": 5.920398009950249,
+      "grad_norm": 0.58984375,
+      "learning_rate": 8.50384391680022e-05,
+      "loss": 0.7443,
+      "step": 1785
+    },
+    {
+      "epoch": 5.93698175787728,
+      "grad_norm": 0.455078125,
+      "learning_rate": 8.446537687355795e-05,
+      "loss": 0.7502,
+      "step": 1790
+    },
+    {
+      "epoch": 5.9535655058043115,
+      "grad_norm": 0.51953125,
+      "learning_rate": 8.389283688058454e-05,
+      "loss": 0.7499,
+      "step": 1795
+    },
+    {
+      "epoch": 5.970149253731344,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.33208384388878e-05,
+      "loss": 0.7567,
+      "step": 1800
+    },
+    {
+      "epoch": 5.986733001658375,
+      "grad_norm": 0.44921875,
+      "learning_rate": 8.274940078006585e-05,
+      "loss": 0.7571,
+      "step": 1805
+    },
+    {
+      "epoch": 6.0,
+      "eval_loss": 2.4824671745300293,
+      "eval_runtime": 0.2383,
+      "eval_samples_per_second": 41.959,
+      "eval_steps_per_second": 4.196,
+      "step": 1809
+    },
+    {
+      "epoch": 6.003316749585406,
+      "grad_norm": 0.48828125,
+      "learning_rate": 8.217854311686215e-05,
+      "loss": 0.7459,
+      "step": 1810
+    },
+    {
+      "epoch": 6.019900497512438,
+      "grad_norm": 0.486328125,
+      "learning_rate": 8.160828464251975e-05,
+      "loss": 0.7313,
+      "step": 1815
+    },
+    {
+      "epoch": 6.036484245439469,
+      "grad_norm": 0.466796875,
+      "learning_rate": 8.103864453013593e-05,
+      "loss": 0.7401,
+      "step": 1820
+    },
+    {
+      "epoch": 6.053067993366501,
+      "grad_norm": 0.51953125,
+      "learning_rate": 8.046964193201746e-05,
+      "loss": 0.7409,
+      "step": 1825
+    },
+    {
+      "epoch": 6.069651741293533,
+      "grad_norm": 0.47265625,
+      "learning_rate": 7.990129597903682e-05,
+      "loss": 0.7354,
+      "step": 1830
+    },
+    {
+      "epoch": 6.086235489220564,
+      "grad_norm": 0.416015625,
+      "learning_rate": 7.933362577998892e-05,
+      "loss": 0.7373,
+      "step": 1835
+    },
+    {
+      "epoch": 6.102819237147595,
+      "grad_norm": 0.470703125,
+      "learning_rate": 7.876665042094867e-05,
+      "loss": 0.744,
+      "step": 1840
+    },
+    {
+      "epoch": 6.119402985074627,
+      "grad_norm": 0.41796875,
+      "learning_rate": 7.820038896462906e-05,
+      "loss": 0.7344,
+      "step": 1845
+    },
+    {
+      "epoch": 6.135986733001658,
+      "grad_norm": 0.4453125,
+      "learning_rate": 7.763486044974064e-05,
+      "loss": 0.7336,
+      "step": 1850
+    },
+    {
+      "epoch": 6.15257048092869,
+      "grad_norm": 0.42578125,
+      "learning_rate": 7.707008389035101e-05,
+      "loss": 0.7481,
+      "step": 1855
+    },
+    {
+      "epoch": 6.169154228855722,
+      "grad_norm": 0.4453125,
+      "learning_rate": 7.650607827524576e-05,
+      "loss": 0.7288,
+      "step": 1860
+    },
+    {
+      "epoch": 6.185737976782753,
+      "grad_norm": 0.455078125,
+      "learning_rate": 7.594286256729e-05,
+      "loss": 0.7398,
+      "step": 1865
+    },
+    {
+      "epoch": 6.202321724709784,
+      "grad_norm": 0.44921875,
+      "learning_rate": 7.538045570279067e-05,
+      "loss": 0.7315,
+      "step": 1870
+    },
+    {
+      "epoch": 6.218905472636816,
+      "grad_norm": 0.439453125,
+      "learning_rate": 7.481887659086005e-05,
+      "loss": 0.7452,
+      "step": 1875
+    },
+    {
+      "epoch": 6.2354892205638475,
+      "grad_norm": 0.478515625,
+      "learning_rate": 7.425814411277988e-05,
+      "loss": 0.7219,
+      "step": 1880
+    },
+    {
+      "epoch": 6.252072968490879,
+      "grad_norm": 0.53125,
+      "learning_rate": 7.369827712136661e-05,
+      "loss": 0.7438,
+      "step": 1885
+    },
+    {
+      "epoch": 6.268656716417911,
+      "grad_norm": 0.546875,
+      "learning_rate": 7.31392944403375e-05,
+      "loss": 0.7322,
+      "step": 1890
+    },
+    {
+      "epoch": 6.285240464344942,
+      "grad_norm": 0.42578125,
+      "learning_rate": 7.258121486367762e-05,
+      "loss": 0.7318,
+      "step": 1895
+    },
+    {
+      "epoch": 6.301824212271973,
+      "grad_norm": 0.458984375,
+      "learning_rate": 7.202405715500821e-05,
+      "loss": 0.7323,
+      "step": 1900
+    },
+    {
+      "epoch": 6.318407960199005,
+      "grad_norm": 0.451171875,
+      "learning_rate": 7.146784004695563e-05,
+      "loss": 0.7336,
+      "step": 1905
+    },
+    {
+      "epoch": 6.3349917081260365,
+      "grad_norm": 0.43359375,
+      "learning_rate": 7.091258224052164e-05,
+      "loss": 0.7351,
+      "step": 1910
+    },
+    {
+      "epoch": 6.351575456053068,
+      "grad_norm": 0.46484375,
+      "learning_rate": 7.035830240445443e-05,
+      "loss": 0.7341,
+      "step": 1915
+    },
+    {
+      "epoch": 6.3681592039801,
+      "grad_norm": 0.474609375,
+      "learning_rate": 6.980501917462127e-05,
+      "loss": 0.7363,
+      "step": 1920
+    },
+    {
+      "epoch": 6.384742951907131,
+      "grad_norm": 0.453125,
+      "learning_rate": 6.925275115338168e-05,
+      "loss": 0.7388,
+      "step": 1925
+    },
+    {
+      "epoch": 6.401326699834162,
+      "grad_norm": 0.4296875,
+      "learning_rate": 6.870151690896209e-05,
+      "loss": 0.7325,
+      "step": 1930
+    },
+    {
+      "epoch": 6.417910447761194,
+      "grad_norm": 0.53125,
+      "learning_rate": 6.815133497483157e-05,
+      "loss": 0.7326,
+      "step": 1935
+    },
+    {
+      "epoch": 6.434494195688226,
+      "grad_norm": 0.4296875,
+      "learning_rate": 6.760222384907858e-05,
+      "loss": 0.7337,
+      "step": 1940
+    },
+    {
+      "epoch": 6.451077943615257,
+      "grad_norm": 0.42578125,
+      "learning_rate": 6.705420199378914e-05,
+      "loss": 0.7416,
+      "step": 1945
+    },
+    {
+      "epoch": 6.467661691542289,
+      "grad_norm": 0.470703125,
+      "learning_rate": 6.65072878344261e-05,
+      "loss": 0.7323,
+      "step": 1950
+    },
+    {
+      "epoch": 6.48424543946932,
+      "grad_norm": 0.435546875,
+      "learning_rate": 6.596149975920962e-05,
+      "loss": 0.7321,
+      "step": 1955
+    },
+    {
+      "epoch": 6.500829187396351,
+      "grad_norm": 0.4609375,
+      "learning_rate": 6.541685611849887e-05,
+      "loss": 0.7343,
+      "step": 1960
+    },
+    {
+      "epoch": 6.517412935323383,
+      "grad_norm": 0.43359375,
+      "learning_rate": 6.487337522417516e-05,
+      "loss": 0.7385,
+      "step": 1965
+    },
+    {
+      "epoch": 6.533996683250415,
+      "grad_norm": 0.46484375,
+      "learning_rate": 6.433107534902618e-05,
+      "loss": 0.7337,
+      "step": 1970
+    },
+    {
+      "epoch": 6.550580431177446,
+      "grad_norm": 0.42578125,
+      "learning_rate": 6.378997472613169e-05,
+      "loss": 0.7375,
+      "step": 1975
+    },
+    {
+      "epoch": 6.567164179104478,
+      "grad_norm": 0.431640625,
+      "learning_rate": 6.32500915482505e-05,
+      "loss": 0.7303,
+      "step": 1980
+    },
+    {
+      "epoch": 6.583747927031509,
+      "grad_norm": 0.435546875,
+      "learning_rate": 6.271144396720869e-05,
+      "loss": 0.7404,
+      "step": 1985
+    },
+    {
+      "epoch": 6.60033167495854,
+      "grad_norm": 0.484375,
+      "learning_rate": 6.21740500932894e-05,
+      "loss": 0.739,
+      "step": 1990
+    },
+    {
+      "epoch": 6.616915422885572,
+      "grad_norm": 0.435546875,
+      "learning_rate": 6.163792799462404e-05,
+      "loss": 0.7348,
+      "step": 1995
+    },
+    {
+      "epoch": 6.633499170812604,
+      "grad_norm": 0.4765625,
+      "learning_rate": 6.11030956965846e-05,
+      "loss": 0.7481,
+      "step": 2000
+    },
+    {
+      "epoch": 6.650082918739635,
+      "grad_norm": 0.4296875,
+      "learning_rate": 6.056957118117761e-05,
+      "loss": 0.7387,
+      "step": 2005
+    },
+    {
+      "epoch": 6.666666666666667,
+      "grad_norm": 0.416015625,
+      "learning_rate": 6.0037372386439774e-05,
+      "loss": 0.7216,
+      "step": 2010
+    },
+    {
+      "epoch": 6.683250414593698,
+      "grad_norm": 0.431640625,
+      "learning_rate": 5.950651720583471e-05,
+      "loss": 0.7365,
+      "step": 2015
+    },
+    {
+      "epoch": 6.699834162520729,
+      "grad_norm": 0.435546875,
+      "learning_rate": 5.897702348765129e-05,
+      "loss": 0.7414,
+      "step": 2020
+    },
+    {
+      "epoch": 6.7164179104477615,
+      "grad_norm": 0.421875,
+      "learning_rate": 5.84489090344037e-05,
+      "loss": 0.7323,
+      "step": 2025
+    },
+    {
+      "epoch": 6.733001658374793,
+      "grad_norm": 0.41796875,
+      "learning_rate": 5.792219160223268e-05,
+      "loss": 0.7329,
+      "step": 2030
+    },
+    {
+      "epoch": 6.749585406301824,
+      "grad_norm": 0.439453125,
+      "learning_rate": 5.7396888900308785e-05,
+      "loss": 0.7412,
+      "step": 2035
+    },
+    {
+      "epoch": 6.766169154228856,
+      "grad_norm": 0.458984375,
+      "learning_rate": 5.6873018590236707e-05,
+      "loss": 0.7376,
+      "step": 2040
+    },
+    {
+      "epoch": 6.782752902155887,
+      "grad_norm": 0.466796875,
+      "learning_rate": 5.635059828546174e-05,
+      "loss": 0.7354,
+      "step": 2045
+    },
+    {
+      "epoch": 6.799336650082918,
+      "grad_norm": 0.419921875,
+      "learning_rate": 5.582964555067736e-05,
+      "loss": 0.7364,
+      "step": 2050
+    },
+    {
+      "epoch": 6.8159203980099505,
+      "grad_norm": 0.41015625,
+      "learning_rate": 5.5310177901234675e-05,
+      "loss": 0.7312,
+      "step": 2055
+    },
+    {
+      "epoch": 6.832504145936982,
+      "grad_norm": 0.427734375,
+      "learning_rate": 5.4792212802553736e-05,
+      "loss": 0.7446,
+      "step": 2060
+    },
+    {
+      "epoch": 6.849087893864013,
+      "grad_norm": 0.431640625,
+      "learning_rate": 5.4275767669536146e-05,
+      "loss": 0.7396,
+      "step": 2065
+    },
+    {
+      "epoch": 6.865671641791045,
+      "grad_norm": 0.421875,
+      "learning_rate": 5.376085986597955e-05,
+      "loss": 0.7438,
+      "step": 2070
+    },
+    {
+      "epoch": 6.882255389718076,
+      "grad_norm": 0.421875,
+      "learning_rate": 5.3247506703993814e-05,
+      "loss": 0.7369,
+      "step": 2075
+    },
+    {
+      "epoch": 6.8988391376451075,
+      "grad_norm": 0.4140625,
+      "learning_rate": 5.273572544341912e-05,
+      "loss": 0.7339,
+      "step": 2080
+    },
+    {
+      "epoch": 6.91542288557214,
+      "grad_norm": 0.427734375,
+      "learning_rate": 5.222553329124544e-05,
+      "loss": 0.7381,
+      "step": 2085
+    },
+    {
+      "epoch": 6.932006633499171,
+      "grad_norm": 0.447265625,
+      "learning_rate": 5.1716947401034235e-05,
+      "loss": 0.7303,
+      "step": 2090
+    },
+    {
+      "epoch": 6.948590381426202,
+      "grad_norm": 0.484375,
+      "learning_rate": 5.1209984872341474e-05,
+      "loss": 0.7369,
+      "step": 2095
+    },
+    {
+      "epoch": 6.965174129353234,
+      "grad_norm": 0.4296875,
+      "learning_rate": 5.070466275014287e-05,
+      "loss": 0.7454,
+      "step": 2100
+    },
+    {
+      "epoch": 6.981757877280265,
+      "grad_norm": 0.48828125,
+      "learning_rate": 5.0200998024260905e-05,
+      "loss": 0.7428,
+      "step": 2105
+    },
+    {
+      "epoch": 6.9983416252072965,
+      "grad_norm": 0.466796875,
+      "learning_rate": 4.969900762879325e-05,
+      "loss": 0.7389,
+      "step": 2110
+    },
+    {
+      "epoch": 6.9983416252072965,
+      "eval_loss": 2.51469087600708,
+      "eval_runtime": 0.2574,
+      "eval_samples_per_second": 38.853,
+      "eval_steps_per_second": 3.885,
+      "step": 2110
+    },
+    {
+      "epoch": 7.014925373134329,
+      "grad_norm": 0.51953125,
+      "learning_rate": 4.919870844154385e-05,
+      "loss": 0.722,
+      "step": 2115
+    },
+    {
+      "epoch": 7.03150912106136,
+      "grad_norm": 0.54296875,
+      "learning_rate": 4.8700117283455095e-05,
+      "loss": 0.7287,
+      "step": 2120
+    },
+    {
+      "epoch": 7.048092868988391,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.820325091804242e-05,
+      "loss": 0.7271,
+      "step": 2125
+    },
+    {
+      "epoch": 7.064676616915423,
+      "grad_norm": 0.439453125,
+      "learning_rate": 4.77081260508308e-05,
+      "loss": 0.7317,
+      "step": 2130
+    },
+    {
+      "epoch": 7.081260364842454,
+      "grad_norm": 0.45703125,
+      "learning_rate": 4.721475932879282e-05,
+      "loss": 0.7211,
+      "step": 2135
+    },
+    {
+      "epoch": 7.097844112769486,
+      "grad_norm": 0.453125,
+      "learning_rate": 4.672316733978926e-05,
+      "loss": 0.7222,
+      "step": 2140
+    },
+    {
+      "epoch": 7.114427860696518,
+      "grad_norm": 0.515625,
+      "learning_rate": 4.623336661201113e-05,
+      "loss": 0.7266,
+      "step": 2145
+    },
+    {
+      "epoch": 7.131011608623549,
+      "grad_norm": 0.435546875,
+      "learning_rate": 4.574537361342407e-05,
+      "loss": 0.7268,
+      "step": 2150
+    },
+    {
+      "epoch": 7.14759535655058,
+      "grad_norm": 0.43359375,
+      "learning_rate": 4.5259204751214743e-05,
+      "loss": 0.7198,
+      "step": 2155
+    },
+    {
+      "epoch": 7.164179104477612,
+      "grad_norm": 0.408203125,
+      "learning_rate": 4.4774876371239184e-05,
+      "loss": 0.7317,
+      "step": 2160
+    },
+    {
+      "epoch": 7.180762852404643,
+      "grad_norm": 0.4140625,
+      "learning_rate": 4.429240475747293e-05,
+      "loss": 0.7266,
+      "step": 2165
+    },
+    {
+      "epoch": 7.197346600331675,
+      "grad_norm": 0.498046875,
+      "learning_rate": 4.381180613146395e-05,
+      "loss": 0.7158,
+      "step": 2170
+    },
+    {
+      "epoch": 7.213930348258707,
+      "grad_norm": 0.486328125,
+      "learning_rate": 4.3333096651787065e-05,
+      "loss": 0.7346,
+      "step": 2175
+    },
+    {
+      "epoch": 7.230514096185738,
+      "grad_norm": 0.451171875,
+      "learning_rate": 4.285629241350053e-05,
+      "loss": 0.7367,
+      "step": 2180
+    },
+    {
+      "epoch": 7.247097844112769,
+      "grad_norm": 0.48828125,
+      "learning_rate": 4.238140944760517e-05,
+      "loss": 0.715,
+      "step": 2185
+    },
+    {
+      "epoch": 7.263681592039801,
+      "grad_norm": 0.458984375,
+      "learning_rate": 4.190846372050511e-05,
+      "loss": 0.7205,
+      "step": 2190
+    },
+    {
+      "epoch": 7.2802653399668324,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.143747113347109e-05,
+      "loss": 0.7265,
+      "step": 2195
+    },
+    {
+      "epoch": 7.296849087893864,
+      "grad_norm": 0.478515625,
+      "learning_rate": 4.096844752210598e-05,
+      "loss": 0.7204,
+      "step": 2200
+    },
+    {
+      "epoch": 7.313432835820896,
+      "grad_norm": 0.4375,
+      "learning_rate": 4.050140865581205e-05,
+      "loss": 0.7229,
+      "step": 2205
+    },
+    {
+      "epoch": 7.330016583747927,
+      "grad_norm": 0.43359375,
+      "learning_rate": 4.003637023726095e-05,
+      "loss": 0.7272,
+      "step": 2210
+    },
+    {
+      "epoch": 7.346600331674958,
+      "grad_norm": 0.458984375,
+      "learning_rate": 3.9573347901865853e-05,
+      "loss": 0.7186,
+      "step": 2215
+    },
+    {
+      "epoch": 7.36318407960199,
+      "grad_norm": 0.443359375,
+      "learning_rate": 3.9112357217255526e-05,
+      "loss": 0.7189,
+      "step": 2220
+    },
+    {
+      "epoch": 7.3797678275290215,
+      "grad_norm": 0.4375,
+      "learning_rate": 3.8653413682751185e-05,
+      "loss": 0.7232,
+      "step": 2225
+    },
+    {
+      "epoch": 7.396351575456053,
+      "grad_norm": 0.4453125,
+      "learning_rate": 3.8196532728845105e-05,
+      "loss": 0.7243,
+      "step": 2230
+    },
+    {
+      "epoch": 7.412935323383085,
+      "grad_norm": 0.4140625,
+      "learning_rate": 3.774172971668201e-05,
+      "loss": 0.7345,
+      "step": 2235
+    },
+    {
+      "epoch": 7.429519071310116,
+      "grad_norm": 0.41796875,
+      "learning_rate": 3.72890199375426e-05,
+      "loss": 0.7258,
+      "step": 2240
+    },
+    {
+      "epoch": 7.446102819237147,
+      "grad_norm": 0.41796875,
+      "learning_rate": 3.6838418612329305e-05,
+      "loss": 0.726,
+      "step": 2245
+    },
+    {
+      "epoch": 7.462686567164179,
+      "grad_norm": 0.42578125,
+      "learning_rate": 3.63899408910547e-05,
+      "loss": 0.7278,
+      "step": 2250
+    },
+    {
+      "epoch": 7.4792703150912105,
+      "grad_norm": 0.427734375,
+      "learning_rate": 3.5943601852331953e-05,
+      "loss": 0.7214,
+      "step": 2255
+    },
+    {
+      "epoch": 7.495854063018242,
+      "grad_norm": 0.431640625,
+      "learning_rate": 3.549941650286797e-05,
+      "loss": 0.7172,
+      "step": 2260
+    },
+    {
+      "epoch": 7.512437810945274,
+      "grad_norm": 0.416015625,
+      "learning_rate": 3.505739977695885e-05,
+      "loss": 0.7265,
+      "step": 2265
+    },
+    {
+      "epoch": 7.529021558872305,
+      "grad_norm": 0.421875,
+      "learning_rate": 3.461756653598778e-05,
+      "loss": 0.724,
+      "step": 2270
+    },
+    {
+      "epoch": 7.545605306799336,
+      "grad_norm": 0.453125,
+      "learning_rate": 3.4179931567925216e-05,
+      "loss": 0.728,
+      "step": 2275
+    },
+    {
+      "epoch": 7.562189054726368,
+      "grad_norm": 0.4140625,
+      "learning_rate": 3.3744509586831805e-05,
+      "loss": 0.7226,
+      "step": 2280
+    },
+    {
+      "epoch": 7.5787728026534,
+      "grad_norm": 0.443359375,
+      "learning_rate": 3.331131523236381e-05,
+      "loss": 0.7272,
+      "step": 2285
+    },
+    {
+      "epoch": 7.595356550580431,
+      "grad_norm": 0.416015625,
+      "learning_rate": 3.288036306928055e-05,
+      "loss": 0.7304,
+      "step": 2290
+    },
+    {
+      "epoch": 7.611940298507463,
+      "grad_norm": 0.443359375,
+      "learning_rate": 3.2451667586955095e-05,
+      "loss": 0.7303,
+      "step": 2295
+    },
+    {
+      "epoch": 7.628524046434494,
+      "grad_norm": 0.47265625,
+      "learning_rate": 3.202524319888679e-05,
+      "loss": 0.7236,
+      "step": 2300
+    },
+    {
+      "epoch": 7.645107794361525,
+      "grad_norm": 0.451171875,
+      "learning_rate": 3.1601104242216785e-05,
+      "loss": 0.7228,
+      "step": 2305
+    },
+    {
+      "epoch": 7.661691542288557,
+      "grad_norm": 0.44921875,
+      "learning_rate": 3.1179264977246106e-05,
+      "loss": 0.7319,
+      "step": 2310
+    },
+    {
+      "epoch": 7.678275290215589,
+      "grad_norm": 0.435546875,
+      "learning_rate": 3.075973958695593e-05,
+      "loss": 0.7222,
+      "step": 2315
+    },
+    {
+      "epoch": 7.69485903814262,
+      "grad_norm": 0.423828125,
+      "learning_rate": 3.034254217653102e-05,
+      "loss": 0.726,
+      "step": 2320
+    },
+    {
+      "epoch": 7.711442786069652,
+      "grad_norm": 0.41015625,
+      "learning_rate": 2.9927686772885255e-05,
+      "loss": 0.716,
+      "step": 2325
+    },
+    {
+      "epoch": 7.728026533996683,
+      "grad_norm": 0.421875,
+      "learning_rate": 2.9515187324190076e-05,
+      "loss": 0.7215,
+      "step": 2330
+    },
+    {
+      "epoch": 7.744610281923714,
+      "grad_norm": 0.4765625,
+      "learning_rate": 2.9105057699405702e-05,
+      "loss": 0.7278,
+      "step": 2335
+    },
+    {
+      "epoch": 7.7611940298507465,
+      "grad_norm": 0.4140625,
+      "learning_rate": 2.8697311687814542e-05,
+      "loss": 0.7191,
+      "step": 2340
+    },
+    {
+      "epoch": 7.777777777777778,
+      "grad_norm": 0.423828125,
+      "learning_rate": 2.8291962998557752e-05,
+      "loss": 0.7301,
+      "step": 2345
+    },
+    {
+      "epoch": 7.79436152570481,
+      "grad_norm": 0.421875,
+      "learning_rate": 2.788902526017436e-05,
+      "loss": 0.7183,
+      "step": 2350
+    },
+    {
+      "epoch": 7.810945273631841,
+      "grad_norm": 0.41796875,
+      "learning_rate": 2.7488512020142842e-05,
+      "loss": 0.7225,
+      "step": 2355
+    },
+    {
+      "epoch": 7.827529021558872,
+      "grad_norm": 0.423828125,
+      "learning_rate": 2.7090436744425852e-05,
+      "loss": 0.7263,
+      "step": 2360
+    },
+    {
+      "epoch": 7.844112769485903,
+      "grad_norm": 0.408203125,
+      "learning_rate": 2.669481281701739e-05,
+      "loss": 0.7276,
+      "step": 2365
+    },
+    {
+      "epoch": 7.8606965174129355,
+      "grad_norm": 0.400390625,
+      "learning_rate": 2.6301653539492642e-05,
+      "loss": 0.7224,
+      "step": 2370
+    },
+    {
+      "epoch": 7.877280265339967,
+      "grad_norm": 0.42578125,
+      "learning_rate": 2.5910972130561086e-05,
+      "loss": 0.7269,
+      "step": 2375
+    },
+    {
+      "epoch": 7.893864013266999,
+      "grad_norm": 0.419921875,
+      "learning_rate": 2.5522781725621813e-05,
+      "loss": 0.7237,
+      "step": 2380
+    },
+    {
+      "epoch": 7.91044776119403,
+      "grad_norm": 0.4453125,
+      "learning_rate": 2.513709537632194e-05,
+      "loss": 0.7225,
+      "step": 2385
+    },
+    {
+      "epoch": 7.927031509121061,
+      "grad_norm": 0.412109375,
+      "learning_rate": 2.475392605011785e-05,
+      "loss": 0.7183,
+      "step": 2390
+    },
+    {
+      "epoch": 7.9436152570480925,
+      "grad_norm": 0.4609375,
+      "learning_rate": 2.4373286629839142e-05,
+      "loss": 0.7267,
+      "step": 2395
+    },
+    {
+      "epoch": 7.960199004975125,
+      "grad_norm": 0.423828125,
+      "learning_rate": 2.3995189913255467e-05,
+      "loss": 0.7158,
+      "step": 2400
+    },
+    {
+      "epoch": 7.976782752902156,
+      "grad_norm": 0.4453125,
+      "learning_rate": 2.3619648612646396e-05,
+      "loss": 0.7305,
+      "step": 2405
+    },
+    {
+      "epoch": 7.993366500829188,
+      "grad_norm": 0.44140625,
+      "learning_rate": 2.3246675354373805e-05,
+      "loss": 0.7223,
+      "step": 2410
+    },
+    {
+      "epoch": 8.0,
+      "eval_loss": 2.5134639739990234,
+      "eval_runtime": 0.2372,
+      "eval_samples_per_second": 42.157,
+      "eval_steps_per_second": 4.216,
+      "step": 2412
+    },
+    {
+      "epoch": 8.009950248756219,
+      "grad_norm": 0.423828125,
+      "learning_rate": 2.287628267845744e-05,
+      "loss": 0.7212,
+      "step": 2415
+    },
+    {
+      "epoch": 8.02653399668325,
+      "grad_norm": 0.40625,
+      "learning_rate": 2.2508483038153417e-05,
+      "loss": 0.7154,
+      "step": 2420
+    },
+    {
+      "epoch": 8.043117744610282,
+      "grad_norm": 0.41796875,
+      "learning_rate": 2.214328879953528e-05,
+      "loss": 0.7264,
+      "step": 2425
+    },
+    {
+      "epoch": 8.059701492537313,
+      "grad_norm": 0.41796875,
+      "learning_rate": 2.1780712241078495e-05,
+      "loss": 0.7231,
+      "step": 2430
+    },
+    {
+      "epoch": 8.076285240464346,
+      "grad_norm": 0.42578125,
+      "learning_rate": 2.1420765553247414e-05,
+      "loss": 0.7108,
+      "step": 2435
+    },
+    {
+      "epoch": 8.092868988391377,
+      "grad_norm": 0.431640625,
+      "learning_rate": 2.106346083808547e-05,
+      "loss": 0.719,
+      "step": 2440
+    },
+    {
+      "epoch": 8.109452736318408,
+      "grad_norm": 0.41015625,
+      "learning_rate": 2.0708810108808407e-05,
+      "loss": 0.7275,
+      "step": 2445
+    },
+    {
+      "epoch": 8.12603648424544,
+      "grad_norm": 0.408203125,
+      "learning_rate": 2.0356825289400183e-05,
+      "loss": 0.7213,
+      "step": 2450
+    },
+    {
+      "epoch": 8.14262023217247,
+      "grad_norm": 0.416015625,
+      "learning_rate": 2.0007518214212227e-05,
+      "loss": 0.7172,
+      "step": 2455
+    },
+    {
+      "epoch": 8.159203980099502,
+      "grad_norm": 0.427734375,
+      "learning_rate": 1.9660900627565437e-05,
+      "loss": 0.7232,
+      "step": 2460
+    },
+    {
+      "epoch": 8.175787728026535,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.9316984183355312e-05,
+      "loss": 0.7243,
+      "step": 2465
+    },
+    {
+      "epoch": 8.192371475953566,
+      "grad_norm": 0.42578125,
+      "learning_rate": 1.8975780444660273e-05,
+      "loss": 0.7186,
+      "step": 2470
+    },
+    {
+      "epoch": 8.208955223880597,
+      "grad_norm": 0.4296875,
+      "learning_rate": 1.863730088335275e-05,
+      "loss": 0.7164,
+      "step": 2475
+    },
+    {
+      "epoch": 8.225538971807628,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.8301556879713465e-05,
+      "loss": 0.7186,
+      "step": 2480
+    },
+    {
+      "epoch": 8.24212271973466,
+      "grad_norm": 0.41796875,
+      "learning_rate": 1.7968559722048906e-05,
+      "loss": 0.7129,
+      "step": 2485
+    },
+    {
+      "epoch": 8.25870646766169,
+      "grad_norm": 0.4296875,
+      "learning_rate": 1.7638320606311752e-05,
+      "loss": 0.7169,
+      "step": 2490
+    },
+    {
+      "epoch": 8.275290215588724,
+      "grad_norm": 0.41796875,
+      "learning_rate": 1.7310850635724418e-05,
+      "loss": 0.7118,
+      "step": 2495
+    },
+    {
+      "epoch": 8.291873963515755,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.6986160820405827e-05,
+      "loss": 0.7151,
+      "step": 2500
+    },
+    {
+      "epoch": 8.308457711442786,
+      "grad_norm": 0.427734375,
+      "learning_rate": 1.6664262077001026e-05,
+      "loss": 0.7183,
+      "step": 2505
+    },
+    {
+      "epoch": 8.325041459369817,
+      "grad_norm": 0.412109375,
+      "learning_rate": 1.634516522831444e-05,
+      "loss": 0.7195,
+      "step": 2510
+    },
+    {
+      "epoch": 8.341625207296849,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.6028881002945818e-05,
+      "loss": 0.7256,
+      "step": 2515
+    },
+    {
+      "epoch": 8.35820895522388,
+      "grad_norm": 0.431640625,
+      "learning_rate": 1.5715420034929474e-05,
+      "loss": 0.7148,
+      "step": 2520
+    },
+    {
+      "epoch": 8.374792703150913,
+      "grad_norm": 0.41015625,
+      "learning_rate": 1.5404792863376916e-05,
+      "loss": 0.7156,
+      "step": 2525
+    },
+    {
+      "epoch": 8.391376451077944,
+      "grad_norm": 0.4140625,
+      "learning_rate": 1.5097009932122341e-05,
+      "loss": 0.725,
+      "step": 2530
+    },
+    {
+      "epoch": 8.407960199004975,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.4792081589371564e-05,
+      "loss": 0.7227,
+      "step": 2535
+    },
+    {
+      "epoch": 8.424543946932006,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.449001808735414e-05,
+      "loss": 0.7208,
+      "step": 2540
+    },
+    {
+      "epoch": 8.441127694859038,
+      "grad_norm": 0.439453125,
+      "learning_rate": 1.4190829581978582e-05,
+      "loss": 0.7204,
+      "step": 2545
+    },
+    {
+      "epoch": 8.457711442786069,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.3894526132490925e-05,
+      "loss": 0.7177,
+      "step": 2550
+    },
+    {
+      "epoch": 8.474295190713102,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.360111770113659e-05,
+      "loss": 0.7192,
+      "step": 2555
+    },
+    {
+      "epoch": 8.490878938640133,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.3310614152825273e-05,
+      "loss": 0.7187,
+      "step": 2560
+    },
+    {
+      "epoch": 8.507462686567164,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.3023025254799481e-05,
+      "loss": 0.7169,
+      "step": 2565
+    },
+    {
+      "epoch": 8.524046434494196,
+      "grad_norm": 0.439453125,
+      "learning_rate": 1.2738360676306015e-05,
+      "loss": 0.7166,
+      "step": 2570
+    },
+    {
+      "epoch": 8.540630182421227,
+      "grad_norm": 0.41796875,
+      "learning_rate": 1.2456629988270751e-05,
+      "loss": 0.7185,
+      "step": 2575
+    },
+    {
+      "epoch": 8.557213930348258,
+      "grad_norm": 0.4296875,
+      "learning_rate": 1.2177842662977135e-05,
+      "loss": 0.7232,
+      "step": 2580
+    },
+    {
+      "epoch": 8.573797678275291,
+      "grad_norm": 0.421875,
+      "learning_rate": 1.1902008073747551e-05,
+      "loss": 0.7158,
+      "step": 2585
+    },
+    {
+      "epoch": 8.590381426202322,
+      "grad_norm": 0.400390625,
+      "learning_rate": 1.1629135494628096e-05,
+      "loss": 0.7216,
+      "step": 2590
+    },
+    {
+      "epoch": 8.606965174129353,
+      "grad_norm": 0.427734375,
+      "learning_rate": 1.1359234100076898e-05,
+      "loss": 0.7236,
+      "step": 2595
+    },
+    {
+      "epoch": 8.623548922056385,
+      "grad_norm": 0.421875,
+      "learning_rate": 1.1092312964655615e-05,
+      "loss": 0.7225,
+      "step": 2600
+    },
+    {
+      "epoch": 8.640132669983416,
+      "grad_norm": 0.40625,
+      "learning_rate": 1.0828381062724324e-05,
+      "loss": 0.7374,
+      "step": 2605
+    },
+    {
+      "epoch": 8.656716417910447,
+      "grad_norm": 0.41015625,
+      "learning_rate": 1.0567447268139818e-05,
+      "loss": 0.7183,
+      "step": 2610
+    },
+    {
+      "epoch": 8.67330016583748,
+      "grad_norm": 0.404296875,
+      "learning_rate": 1.0309520353957181e-05,
+      "loss": 0.7208,
+      "step": 2615
+    },
+    {
+      "epoch": 8.689883913764511,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.0054608992134862e-05,
+      "loss": 0.7198,
+      "step": 2620
+    },
+    {
+      "epoch": 8.706467661691542,
+      "grad_norm": 0.453125,
+      "learning_rate": 9.802721753243194e-06,
+      "loss": 0.7149,
+      "step": 2625
+    },
+    {
+      "epoch": 8.723051409618574,
+      "grad_norm": 0.41796875,
+      "learning_rate": 9.553867106176029e-06,
+      "loss": 0.7333,
+      "step": 2630
+    },
+    {
+      "epoch": 8.739635157545605,
+      "grad_norm": 0.40625,
+      "learning_rate": 9.308053417866259e-06,
+      "loss": 0.7118,
+      "step": 2635
+    },
+    {
+      "epoch": 8.756218905472636,
+      "grad_norm": 0.408203125,
+      "learning_rate": 9.065288953004258e-06,
+      "loss": 0.7224,
+      "step": 2640
+    },
+    {
+      "epoch": 8.772802653399669,
+      "grad_norm": 0.419921875,
+      "learning_rate": 8.82558187376018e-06,
+      "loss": 0.7159,
+      "step": 2645
+    },
+    {
+      "epoch": 8.7893864013267,
+      "grad_norm": 0.51171875,
+      "learning_rate": 8.58894023950948e-06,
+      "loss": 0.7213,
+      "step": 2650
+    },
+    {
+      "epoch": 8.805970149253731,
+      "grad_norm": 0.423828125,
+      "learning_rate": 8.35537200656188e-06,
+      "loss": 0.7248,
+      "step": 2655
+    },
+    {
+      "epoch": 8.822553897180763,
+      "grad_norm": 0.3984375,
+      "learning_rate": 8.124885027894014e-06,
+      "loss": 0.7112,
+      "step": 2660
+    },
+    {
+      "epoch": 8.839137645107794,
+      "grad_norm": 0.421875,
+      "learning_rate": 7.897487052885212e-06,
+      "loss": 0.7162,
+      "step": 2665
+    },
+    {
+      "epoch": 8.855721393034825,
+      "grad_norm": 0.42578125,
+      "learning_rate": 7.673185727057087e-06,
+      "loss": 0.7124,
+      "step": 2670
+    },
+    {
+      "epoch": 8.872305140961858,
+      "grad_norm": 0.408203125,
+      "learning_rate": 7.451988591816439e-06,
+      "loss": 0.7174,
+      "step": 2675
+    },
+    {
+      "epoch": 8.88888888888889,
+      "grad_norm": 0.412109375,
+      "learning_rate": 7.233903084201699e-06,
+      "loss": 0.7119,
+      "step": 2680
+    },
+    {
+      "epoch": 8.90547263681592,
+      "grad_norm": 0.419921875,
+      "learning_rate": 7.018936536632869e-06,
+      "loss": 0.7266,
+      "step": 2685
+    },
+    {
+      "epoch": 8.922056384742952,
+      "grad_norm": 0.421875,
+      "learning_rate": 6.8070961766650065e-06,
+      "loss": 0.7122,
+      "step": 2690
+    },
+    {
+      "epoch": 8.938640132669983,
+      "grad_norm": 0.404296875,
+      "learning_rate": 6.598389126745208e-06,
+      "loss": 0.7161,
+      "step": 2695
+    },
+    {
+      "epoch": 8.955223880597014,
+      "grad_norm": 0.423828125,
+      "learning_rate": 6.392822403973165e-06,
+      "loss": 0.7175,
+      "step": 2700
+    },
+    {
+      "epoch": 8.971807628524047,
+      "grad_norm": 0.4140625,
+      "learning_rate": 6.190402919865268e-06,
+      "loss": 0.724,
+      "step": 2705
+    },
+    {
+      "epoch": 8.988391376451078,
+      "grad_norm": 0.421875,
+      "learning_rate": 5.9911374801220665e-06,
+      "loss": 0.7178,
+      "step": 2710
+    },
+    {
+      "epoch": 8.998341625207297,
+      "eval_loss": 2.5209028720855713,
+      "eval_runtime": 0.2555,
+      "eval_samples_per_second": 39.14,
+      "eval_steps_per_second": 3.914,
+      "step": 2713
+    },
+    {
+      "epoch": 9.00497512437811,
+      "grad_norm": 0.416015625,
+      "learning_rate": 5.795032784399679e-06,
+      "loss": 0.7202,
+      "step": 2715
+    },
+    {
+      "epoch": 9.02155887230514,
+      "grad_norm": 0.400390625,
+      "learning_rate": 5.602095426084408e-06,
+      "loss": 0.7167,
+      "step": 2720
+    },
+    {
+      "epoch": 9.038142620232172,
+      "grad_norm": 0.423828125,
+      "learning_rate": 5.41233189207101e-06,
+      "loss": 0.714,
+      "step": 2725
+    },
+    {
+      "epoch": 9.054726368159203,
+      "grad_norm": 0.40625,
+      "learning_rate": 5.225748562544741e-06,
+      "loss": 0.7183,
+      "step": 2730
+    },
+    {
+      "epoch": 9.071310116086236,
+      "grad_norm": 0.421875,
+      "learning_rate": 5.042351710766724e-06,
+      "loss": 0.716,
+      "step": 2735
+    },
+    {
+      "epoch": 9.087893864013267,
+      "grad_norm": 0.42578125,
+      "learning_rate": 4.862147502863057e-06,
+      "loss": 0.7196,
+      "step": 2740
+    },
+    {
+      "epoch": 9.104477611940299,
+      "grad_norm": 0.419921875,
+      "learning_rate": 4.68514199761757e-06,
+      "loss": 0.7204,
+      "step": 2745
+    },
+    {
+      "epoch": 9.12106135986733,
+      "grad_norm": 0.427734375,
+      "learning_rate": 4.511341146268e-06,
+      "loss": 0.7146,
+      "step": 2750
+    },
+    {
+      "epoch": 9.137645107794361,
+      "grad_norm": 0.40625,
+      "learning_rate": 4.340750792305981e-06,
+      "loss": 0.7258,
+      "step": 2755
+    },
+    {
+      "epoch": 9.154228855721392,
+      "grad_norm": 0.412109375,
+      "learning_rate": 4.17337667128056e-06,
+      "loss": 0.7193,
+      "step": 2760
+    },
+    {
+      "epoch": 9.170812603648425,
+      "grad_norm": 0.4140625,
+      "learning_rate": 4.009224410605339e-06,
+      "loss": 0.727,
+      "step": 2765
+    },
+    {
+      "epoch": 9.187396351575456,
+      "grad_norm": 0.419921875,
+      "learning_rate": 3.848299529369292e-06,
+      "loss": 0.7228,
+      "step": 2770
+    },
+    {
+      "epoch": 9.203980099502488,
+      "grad_norm": 0.41015625,
+      "learning_rate": 3.690607438151228e-06,
+      "loss": 0.7174,
+      "step": 2775
+    },
+    {
+      "epoch": 9.220563847429519,
+      "grad_norm": 0.412109375,
+      "learning_rate": 3.5361534388377524e-06,
+      "loss": 0.7166,
+      "step": 2780
+    },
+    {
+      "epoch": 9.23714759535655,
+      "grad_norm": 0.4140625,
+      "learning_rate": 3.384942724445195e-06,
+      "loss": 0.7203,
+      "step": 2785
+    },
+    {
+      "epoch": 9.253731343283581,
+      "grad_norm": 0.41015625,
+      "learning_rate": 3.236980378944865e-06,
+      "loss": 0.714,
+      "step": 2790
+    },
+    {
+      "epoch": 9.270315091210614,
+      "grad_norm": 0.412109375,
+      "learning_rate": 3.092271377092215e-06,
+      "loss": 0.7176,
+      "step": 2795
+    },
+    {
+      "epoch": 9.286898839137645,
+      "grad_norm": 0.412109375,
+      "learning_rate": 2.9508205842594728e-06,
+      "loss": 0.7224,
+      "step": 2800
+    },
+    {
+      "epoch": 9.303482587064677,
+      "grad_norm": 0.408203125,
+      "learning_rate": 2.812632756272182e-06,
+      "loss": 0.7193,
+      "step": 2805
+    },
+    {
+      "epoch": 9.320066334991708,
+      "grad_norm": 0.408203125,
+      "learning_rate": 2.6777125392492107e-06,
+      "loss": 0.7217,
+      "step": 2810
+    },
+    {
+      "epoch": 9.336650082918739,
+      "grad_norm": 0.416015625,
+      "learning_rate": 2.5460644694466073e-06,
+      "loss": 0.7113,
+      "step": 2815
+    },
+    {
+      "epoch": 9.35323383084577,
+      "grad_norm": 0.431640625,
+      "learning_rate": 2.4176929731050236e-06,
+      "loss": 0.7217,
+      "step": 2820
+    },
+    {
+      "epoch": 9.369817578772803,
+      "grad_norm": 0.41796875,
+      "learning_rate": 2.2926023663009e-06,
+      "loss": 0.7262,
+      "step": 2825
+    },
+    {
+      "epoch": 9.386401326699835,
+      "grad_norm": 0.423828125,
+      "learning_rate": 2.170796854801449e-06,
+      "loss": 0.7119,
+      "step": 2830
+    },
+    {
+      "epoch": 9.402985074626866,
+      "grad_norm": 0.408203125,
+      "learning_rate": 2.052280533923101e-06,
+      "loss": 0.7195,
+      "step": 2835
+    },
+    {
+      "epoch": 9.419568822553897,
+      "grad_norm": 0.42578125,
+      "learning_rate": 1.937057388393926e-06,
+      "loss": 0.7189,
+      "step": 2840
+    },
+    {
+      "epoch": 9.436152570480928,
+      "grad_norm": 0.421875,
+      "learning_rate": 1.8251312922196284e-06,
+      "loss": 0.7144,
+      "step": 2845
+    },
+    {
+      "epoch": 9.45273631840796,
+      "grad_norm": 0.408203125,
+      "learning_rate": 1.7165060085532425e-06,
+      "loss": 0.7257,
+      "step": 2850
+    },
+    {
+      "epoch": 9.469320066334992,
+      "grad_norm": 0.40625,
+      "learning_rate": 1.6111851895687202e-06,
+      "loss": 0.7104,
+      "step": 2855
+    },
+    {
+      "epoch": 9.485903814262024,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.5091723763380193e-06,
+      "loss": 0.7229,
+      "step": 2860
+    },
+    {
+      "epoch": 9.502487562189055,
+      "grad_norm": 0.42578125,
+      "learning_rate": 1.4104709987121433e-06,
+      "loss": 0.7157,
+      "step": 2865
+    },
+    {
+      "epoch": 9.519071310116086,
+      "grad_norm": 0.419921875,
+      "learning_rate": 1.315084375205744e-06,
+      "loss": 0.7234,
+      "step": 2870
+    },
+    {
+      "epoch": 9.535655058043117,
+      "grad_norm": 0.416015625,
+      "learning_rate": 1.223015712885589e-06,
+      "loss": 0.7158,
+      "step": 2875
+    },
+    {
+      "epoch": 9.552238805970148,
+      "grad_norm": 0.447265625,
+      "learning_rate": 1.1342681072627703e-06,
+      "loss": 0.7115,
+      "step": 2880
+    },
+    {
+      "epoch": 9.568822553897181,
+      "grad_norm": 0.423828125,
+      "learning_rate": 1.0488445421885318e-06,
+      "loss": 0.7131,
+      "step": 2885
+    },
+    {
+      "epoch": 9.585406301824213,
+      "grad_norm": 0.408203125,
+      "learning_rate": 9.667478897540494e-07,
+      "loss": 0.7124,
+      "step": 2890
+    },
+    {
+      "epoch": 9.601990049751244,
+      "grad_norm": 0.41796875,
+      "learning_rate": 8.879809101937864e-07,
+      "loss": 0.7133,
+      "step": 2895
+    },
+    {
+      "epoch": 9.618573797678275,
+      "grad_norm": 0.404296875,
+      "learning_rate": 8.12546251792734e-07,
+      "loss": 0.709,
+      "step": 2900
+    },
+    {
+      "epoch": 9.635157545605306,
+      "grad_norm": 0.421875,
+      "learning_rate": 7.404464507973608e-07,
+      "loss": 0.7187,
+      "step": 2905
+    },
+    {
+      "epoch": 9.65174129353234,
+      "grad_norm": 0.412109375,
+      "learning_rate": 6.716839313303469e-07,
+      "loss": 0.7304,
+      "step": 2910
+    },
+    {
+      "epoch": 9.66832504145937,
+      "grad_norm": 0.41015625,
+      "learning_rate": 6.0626100530905e-07,
+      "loss": 0.7186,
+      "step": 2915
+    },
+    {
+      "epoch": 9.684908789386402,
+      "grad_norm": 0.419921875,
+      "learning_rate": 5.441798723678115e-07,
+      "loss": 0.7282,
+      "step": 2920
+    },
+    {
+      "epoch": 9.701492537313433,
+      "grad_norm": 0.4140625,
+      "learning_rate": 4.854426197839934e-07,
+      "loss": 0.7248,
+      "step": 2925
+    },
+    {
+      "epoch": 9.718076285240464,
+      "grad_norm": 0.4140625,
+      "learning_rate": 4.300512224077791e-07,
+      "loss": 0.7223,
+      "step": 2930
+    },
+    {
+      "epoch": 9.734660033167495,
+      "grad_norm": 0.400390625,
+      "learning_rate": 3.7800754259580405e-07,
+      "loss": 0.7143,
+      "step": 2935
+    },
+    {
+      "epoch": 9.751243781094526,
+      "grad_norm": 0.419921875,
+      "learning_rate": 3.293133301485063e-07,
+      "loss": 0.7091,
+      "step": 2940
+    },
+    {
+      "epoch": 9.76782752902156,
+      "grad_norm": 0.412109375,
+      "learning_rate": 2.839702222513174e-07,
+      "loss": 0.7174,
+      "step": 2945
+    },
+    {
+      "epoch": 9.78441127694859,
+      "grad_norm": 0.419921875,
+      "learning_rate": 2.4197974341964026e-07,
+      "loss": 0.7199,
+      "step": 2950
+    },
+    {
+      "epoch": 9.800995024875622,
+      "grad_norm": 0.40625,
+      "learning_rate": 2.0334330544753422e-07,
+      "loss": 0.7116,
+      "step": 2955
+    },
+    {
+      "epoch": 9.817578772802653,
+      "grad_norm": 0.466796875,
+      "learning_rate": 1.6806220736029777e-07,
+      "loss": 0.708,
+      "step": 2960
+    },
+    {
+      "epoch": 9.834162520729684,
+      "grad_norm": 0.443359375,
+      "learning_rate": 1.3613763537078105e-07,
+      "loss": 0.7248,
+      "step": 2965
+    },
+    {
+      "epoch": 9.850746268656717,
+      "grad_norm": 0.40625,
+      "learning_rate": 1.0757066283946237e-07,
+      "loss": 0.7182,
+      "step": 2970
+    },
+    {
+      "epoch": 9.867330016583749,
+      "grad_norm": 0.4140625,
+      "learning_rate": 8.236225023844357e-08,
+      "loss": 0.7213,
+      "step": 2975
+    },
+    {
+      "epoch": 9.88391376451078,
+      "grad_norm": 0.41015625,
+      "learning_rate": 6.051324511908707e-08,
+      "loss": 0.7214,
+      "step": 2980
+    },
+    {
+      "epoch": 9.900497512437811,
+      "grad_norm": 0.4140625,
+      "learning_rate": 4.202438208352755e-08,
+      "loss": 0.7165,
+      "step": 2985
+    },
+    {
+      "epoch": 9.917081260364842,
+      "grad_norm": 0.427734375,
+      "learning_rate": 2.689628276000278e-08,
+      "loss": 0.722,
+      "step": 2990
+    },
+    {
+      "epoch": 9.933665008291873,
+      "grad_norm": 0.41015625,
+      "learning_rate": 1.5129455781937028e-08,
+      "loss": 0.7155,
+      "step": 2995
+    },
+    {
+      "epoch": 9.950248756218905,
+      "grad_norm": 0.41015625,
+      "learning_rate": 6.724296770832528e-09,
+      "loss": 0.7144,
+      "step": 3000
+    },
+    {
+      "epoch": 9.966832504145938,
+      "grad_norm": 0.408203125,
+      "learning_rate": 1.6810883230022888e-09,
+      "loss": 0.7196,
+      "step": 3005
+    },
+    {
+      "epoch": 9.983416252072969,
+      "grad_norm": 0.427734375,
+      "learning_rate": 0.0,
+      "loss": 0.7148,
+      "step": 3010
+    },
+    {
+      "epoch": 9.983416252072969,
+      "eval_loss": 2.5221571922302246,
+      "eval_runtime": 0.2369,
+      "eval_samples_per_second": 42.213,
+      "eval_steps_per_second": 4.221,
+      "step": 3010
+    },
+    {
+      "epoch": 9.983416252072969,
+      "step": 3010,
+      "total_flos": 9.189427803286667e+18,
+      "train_loss": 1.299621812608155,
+      "train_runtime": 7345.2542,
+      "train_samples_per_second": 26.255,
+      "train_steps_per_second": 0.41
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3010,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 100,
+  "total_flos": 9.189427803286667e+18,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}