diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4543 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 3219,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.004659832246039142,
+      "grad_norm": 55.73911101293821,
+      "learning_rate": 7.763975155279503e-07,
+      "loss": 11.0693,
+      "step": 5
+    },
+    {
+      "epoch": 0.009319664492078284,
+      "grad_norm": 57.1718581665509,
+      "learning_rate": 1.5527950310559006e-06,
+      "loss": 10.8839,
+      "step": 10
+    },
+    {
+      "epoch": 0.013979496738117428,
+      "grad_norm": 98.21065880054624,
+      "learning_rate": 2.329192546583851e-06,
+      "loss": 9.4422,
+      "step": 15
+    },
+    {
+      "epoch": 0.01863932898415657,
+      "grad_norm": 32.82367751669906,
+      "learning_rate": 3.1055900621118013e-06,
+      "loss": 3.2146,
+      "step": 20
+    },
+    {
+      "epoch": 0.023299161230195712,
+      "grad_norm": 3.30697302187253,
+      "learning_rate": 3.881987577639752e-06,
+      "loss": 1.347,
+      "step": 25
+    },
+    {
+      "epoch": 0.027958993476234855,
+      "grad_norm": 1.2985174034570703,
+      "learning_rate": 4.658385093167702e-06,
+      "loss": 1.0061,
+      "step": 30
+    },
+    {
+      "epoch": 0.032618825722273995,
+      "grad_norm": 0.8623629897688332,
+      "learning_rate": 5.4347826086956525e-06,
+      "loss": 0.8411,
+      "step": 35
+    },
+    {
+      "epoch": 0.03727865796831314,
+      "grad_norm": 0.6188703104605264,
+      "learning_rate": 6.2111801242236025e-06,
+      "loss": 0.7755,
+      "step": 40
+    },
+    {
+      "epoch": 0.04193849021435228,
+      "grad_norm": 0.461009679090195,
+      "learning_rate": 6.9875776397515525e-06,
+      "loss": 0.7136,
+      "step": 45
+    },
+    {
+      "epoch": 0.046598322460391424,
+      "grad_norm": 0.4286257732967241,
+      "learning_rate": 7.763975155279503e-06,
+      "loss": 0.6831,
+      "step": 50
+    },
+    {
+      "epoch": 0.05125815470643057,
+      "grad_norm": 0.44241392108033467,
+      "learning_rate": 8.540372670807453e-06,
+      "loss": 0.646,
+      "step": 55
+    },
+    {
+      "epoch": 0.05591798695246971,
+      "grad_norm": 0.3111820442035257,
+      "learning_rate": 9.316770186335403e-06,
+      "loss": 0.6088,
+      "step": 60
+    },
+    {
+      "epoch": 0.06057781919850885,
+      "grad_norm": 0.30251919432888047,
+      "learning_rate": 1.0093167701863353e-05,
+      "loss": 0.5932,
+      "step": 65
+    },
+    {
+      "epoch": 0.06523765144454799,
+      "grad_norm": 0.33324119756936554,
+      "learning_rate": 1.0869565217391305e-05,
+      "loss": 0.5892,
+      "step": 70
+    },
+    {
+      "epoch": 0.06989748369058714,
+      "grad_norm": 0.3063487771822568,
+      "learning_rate": 1.1645962732919255e-05,
+      "loss": 0.5713,
+      "step": 75
+    },
+    {
+      "epoch": 0.07455731593662628,
+      "grad_norm": 0.3071310049244509,
+      "learning_rate": 1.2422360248447205e-05,
+      "loss": 0.5618,
+      "step": 80
+    },
+    {
+      "epoch": 0.07921714818266543,
+      "grad_norm": 0.3358130788082856,
+      "learning_rate": 1.3198757763975155e-05,
+      "loss": 0.5524,
+      "step": 85
+    },
+    {
+      "epoch": 0.08387698042870456,
+      "grad_norm": 0.2925977780766404,
+      "learning_rate": 1.3975155279503105e-05,
+      "loss": 0.5628,
+      "step": 90
+    },
+    {
+      "epoch": 0.08853681267474371,
+      "grad_norm": 0.25884643717759576,
+      "learning_rate": 1.4751552795031057e-05,
+      "loss": 0.5442,
+      "step": 95
+    },
+    {
+      "epoch": 0.09319664492078285,
+      "grad_norm": 0.31707838864665394,
+      "learning_rate": 1.5527950310559007e-05,
+      "loss": 0.5296,
+      "step": 100
+    },
+    {
+      "epoch": 0.097856477166822,
+      "grad_norm": 0.3021647811110462,
+      "learning_rate": 1.630434782608696e-05,
+      "loss": 0.5243,
+      "step": 105
+    },
+    {
+      "epoch": 0.10251630941286113,
+      "grad_norm": 0.30442757517268515,
+      "learning_rate": 1.7080745341614907e-05,
+      "loss": 0.526,
+      "step": 110
+    },
+    {
+      "epoch": 0.10717614165890028,
+      "grad_norm": 0.3036499569102043,
+      "learning_rate": 1.785714285714286e-05,
+      "loss": 0.5231,
+      "step": 115
+    },
+    {
+      "epoch": 0.11183597390493942,
+      "grad_norm": 0.37387951427884036,
+      "learning_rate": 1.8633540372670807e-05,
+      "loss": 0.5098,
+      "step": 120
+    },
+    {
+      "epoch": 0.11649580615097857,
+      "grad_norm": 0.2871555813435635,
+      "learning_rate": 1.940993788819876e-05,
+      "loss": 0.5057,
+      "step": 125
+    },
+    {
+      "epoch": 0.1211556383970177,
+      "grad_norm": 0.3209503297683246,
+      "learning_rate": 2.0186335403726707e-05,
+      "loss": 0.5113,
+      "step": 130
+    },
+    {
+      "epoch": 0.12581547064305684,
+      "grad_norm": 0.41553906221283793,
+      "learning_rate": 2.096273291925466e-05,
+      "loss": 0.5105,
+      "step": 135
+    },
+    {
+      "epoch": 0.13047530288909598,
+      "grad_norm": 0.447991972215783,
+      "learning_rate": 2.173913043478261e-05,
+      "loss": 0.4951,
+      "step": 140
+    },
+    {
+      "epoch": 0.13513513513513514,
+      "grad_norm": 0.36952665164899706,
+      "learning_rate": 2.2515527950310562e-05,
+      "loss": 0.522,
+      "step": 145
+    },
+    {
+      "epoch": 0.13979496738117428,
+      "grad_norm": 0.43854865846501767,
+      "learning_rate": 2.329192546583851e-05,
+      "loss": 0.4958,
+      "step": 150
+    },
+    {
+      "epoch": 0.14445479962721341,
+      "grad_norm": 0.4551787765980115,
+      "learning_rate": 2.4068322981366462e-05,
+      "loss": 0.5041,
+      "step": 155
+    },
+    {
+      "epoch": 0.14911463187325255,
+      "grad_norm": 0.500295346094291,
+      "learning_rate": 2.484472049689441e-05,
+      "loss": 0.4847,
+      "step": 160
+    },
+    {
+      "epoch": 0.15377446411929171,
+      "grad_norm": 0.4867357618004153,
+      "learning_rate": 2.5621118012422362e-05,
+      "loss": 0.4961,
+      "step": 165
+    },
+    {
+      "epoch": 0.15843429636533085,
+      "grad_norm": 0.40984119146102166,
+      "learning_rate": 2.639751552795031e-05,
+      "loss": 0.4719,
+      "step": 170
+    },
+    {
+      "epoch": 0.16309412861137,
+      "grad_norm": 0.4604406630630476,
+      "learning_rate": 2.7173913043478262e-05,
+      "loss": 0.4852,
+      "step": 175
+    },
+    {
+      "epoch": 0.16775396085740912,
+      "grad_norm": 0.40948649685519667,
+      "learning_rate": 2.795031055900621e-05,
+      "loss": 0.478,
+      "step": 180
+    },
+    {
+      "epoch": 0.1724137931034483,
+      "grad_norm": 0.3653652404773955,
+      "learning_rate": 2.8726708074534165e-05,
+      "loss": 0.4809,
+      "step": 185
+    },
+    {
+      "epoch": 0.17707362534948742,
+      "grad_norm": 0.3788769696744858,
+      "learning_rate": 2.9503105590062114e-05,
+      "loss": 0.4907,
+      "step": 190
+    },
+    {
+      "epoch": 0.18173345759552656,
+      "grad_norm": 0.4455345320827965,
+      "learning_rate": 3.0279503105590062e-05,
+      "loss": 0.4802,
+      "step": 195
+    },
+    {
+      "epoch": 0.1863932898415657,
+      "grad_norm": 0.4872519077868042,
+      "learning_rate": 3.1055900621118014e-05,
+      "loss": 0.4818,
+      "step": 200
+    },
+    {
+      "epoch": 0.19105312208760486,
+      "grad_norm": 0.5176809127208468,
+      "learning_rate": 3.183229813664597e-05,
+      "loss": 0.4707,
+      "step": 205
+    },
+    {
+      "epoch": 0.195712954333644,
+      "grad_norm": 0.5721945537346641,
+      "learning_rate": 3.260869565217392e-05,
+      "loss": 0.4848,
+      "step": 210
+    },
+    {
+      "epoch": 0.20037278657968313,
+      "grad_norm": 0.3770305001602024,
+      "learning_rate": 3.3385093167701865e-05,
+      "loss": 0.4669,
+      "step": 215
+    },
+    {
+      "epoch": 0.20503261882572227,
+      "grad_norm": 0.4543945451638326,
+      "learning_rate": 3.4161490683229814e-05,
+      "loss": 0.4713,
+      "step": 220
+    },
+    {
+      "epoch": 0.2096924510717614,
+      "grad_norm": 0.49369340193910277,
+      "learning_rate": 3.493788819875777e-05,
+      "loss": 0.4534,
+      "step": 225
+    },
+    {
+      "epoch": 0.21435228331780057,
+      "grad_norm": 0.48464327291832504,
+      "learning_rate": 3.571428571428572e-05,
+      "loss": 0.4668,
+      "step": 230
+    },
+    {
+      "epoch": 0.2190121155638397,
+      "grad_norm": 0.4111365129541678,
+      "learning_rate": 3.6490683229813665e-05,
+      "loss": 0.4644,
+      "step": 235
+    },
+    {
+      "epoch": 0.22367194780987884,
+      "grad_norm": 0.6275047885479089,
+      "learning_rate": 3.7267080745341614e-05,
+      "loss": 0.4622,
+      "step": 240
+    },
+    {
+      "epoch": 0.22833178005591798,
+      "grad_norm": 0.556206325520564,
+      "learning_rate": 3.804347826086957e-05,
+      "loss": 0.4674,
+      "step": 245
+    },
+    {
+      "epoch": 0.23299161230195714,
+      "grad_norm": 0.4727137299853098,
+      "learning_rate": 3.881987577639752e-05,
+      "loss": 0.4592,
+      "step": 250
+    },
+    {
+      "epoch": 0.23765144454799628,
+      "grad_norm": 0.5175385392810632,
+      "learning_rate": 3.9596273291925465e-05,
+      "loss": 0.4679,
+      "step": 255
+    },
+    {
+      "epoch": 0.2423112767940354,
+      "grad_norm": 0.5522111161928154,
+      "learning_rate": 4.0372670807453414e-05,
+      "loss": 0.463,
+      "step": 260
+    },
+    {
+      "epoch": 0.24697110904007455,
+      "grad_norm": 0.533406707658831,
+      "learning_rate": 4.114906832298137e-05,
+      "loss": 0.4613,
+      "step": 265
+    },
+    {
+      "epoch": 0.2516309412861137,
+      "grad_norm": 0.5190269056190954,
+      "learning_rate": 4.192546583850932e-05,
+      "loss": 0.4596,
+      "step": 270
+    },
+    {
+      "epoch": 0.25629077353215285,
+      "grad_norm": 0.5124788795006116,
+      "learning_rate": 4.270186335403727e-05,
+      "loss": 0.4654,
+      "step": 275
+    },
+    {
+      "epoch": 0.26095060577819196,
+      "grad_norm": 0.6274404486819355,
+      "learning_rate": 4.347826086956522e-05,
+      "loss": 0.4576,
+      "step": 280
+    },
+    {
+      "epoch": 0.2656104380242311,
+      "grad_norm": 0.4751490435936714,
+      "learning_rate": 4.425465838509317e-05,
+      "loss": 0.4623,
+      "step": 285
+    },
+    {
+      "epoch": 0.2702702702702703,
+      "grad_norm": 0.7744692776296426,
+      "learning_rate": 4.5031055900621124e-05,
+      "loss": 0.4654,
+      "step": 290
+    },
+    {
+      "epoch": 0.2749301025163094,
+      "grad_norm": 0.7800040335094921,
+      "learning_rate": 4.580745341614907e-05,
+      "loss": 0.4682,
+      "step": 295
+    },
+    {
+      "epoch": 0.27958993476234856,
+      "grad_norm": 0.8281886923177356,
+      "learning_rate": 4.658385093167702e-05,
+      "loss": 0.4703,
+      "step": 300
+    },
+    {
+      "epoch": 0.2842497670083877,
+      "grad_norm": 0.5858908773046568,
+      "learning_rate": 4.736024844720497e-05,
+      "loss": 0.4578,
+      "step": 305
+    },
+    {
+      "epoch": 0.28890959925442683,
+      "grad_norm": 0.45985550866997615,
+      "learning_rate": 4.8136645962732924e-05,
+      "loss": 0.4497,
+      "step": 310
+    },
+    {
+      "epoch": 0.293569431500466,
+      "grad_norm": 0.5286612419877652,
+      "learning_rate": 4.891304347826087e-05,
+      "loss": 0.4575,
+      "step": 315
+    },
+    {
+      "epoch": 0.2982292637465051,
+      "grad_norm": 0.5552381000432096,
+      "learning_rate": 4.968944099378882e-05,
+      "loss": 0.4427,
+      "step": 320
+    },
+    {
+      "epoch": 0.30288909599254427,
+      "grad_norm": 0.3693124262013228,
+      "learning_rate": 4.994822229892993e-05,
+      "loss": 0.4405,
+      "step": 325
+    },
+    {
+      "epoch": 0.30754892823858343,
+      "grad_norm": 0.40179673300156293,
+      "learning_rate": 4.986192613047981e-05,
+      "loss": 0.4512,
+      "step": 330
+    },
+    {
+      "epoch": 0.31220876048462254,
+      "grad_norm": 0.6909331905128122,
+      "learning_rate": 4.977562996202969e-05,
+      "loss": 0.4518,
+      "step": 335
+    },
+    {
+      "epoch": 0.3168685927306617,
+      "grad_norm": 0.36848158859890456,
+      "learning_rate": 4.968933379357957e-05,
+      "loss": 0.4527,
+      "step": 340
+    },
+    {
+      "epoch": 0.32152842497670087,
+      "grad_norm": 0.43881556176124087,
+      "learning_rate": 4.9603037625129445e-05,
+      "loss": 0.4475,
+      "step": 345
+    },
+    {
+      "epoch": 0.32618825722274,
+      "grad_norm": 0.38991398427916507,
+      "learning_rate": 4.951674145667933e-05,
+      "loss": 0.461,
+      "step": 350
+    },
+    {
+      "epoch": 0.33084808946877914,
+      "grad_norm": 0.47212015510204275,
+      "learning_rate": 4.94304452882292e-05,
+      "loss": 0.4472,
+      "step": 355
+    },
+    {
+      "epoch": 0.33550792171481825,
+      "grad_norm": 0.44227904867586293,
+      "learning_rate": 4.934414911977908e-05,
+      "loss": 0.4448,
+      "step": 360
+    },
+    {
+      "epoch": 0.3401677539608574,
+      "grad_norm": 0.4466830759596917,
+      "learning_rate": 4.9257852951328965e-05,
+      "loss": 0.4412,
+      "step": 365
+    },
+    {
+      "epoch": 0.3448275862068966,
+      "grad_norm": 0.4019403755167571,
+      "learning_rate": 4.917155678287884e-05,
+      "loss": 0.4548,
+      "step": 370
+    },
+    {
+      "epoch": 0.3494874184529357,
+      "grad_norm": 0.46020164939301034,
+      "learning_rate": 4.908526061442872e-05,
+      "loss": 0.441,
+      "step": 375
+    },
+    {
+      "epoch": 0.35414725069897485,
+      "grad_norm": 0.41640820047800664,
+      "learning_rate": 4.89989644459786e-05,
+      "loss": 0.456,
+      "step": 380
+    },
+    {
+      "epoch": 0.35880708294501396,
+      "grad_norm": 0.43251075009831547,
+      "learning_rate": 4.891266827752848e-05,
+      "loss": 0.4575,
+      "step": 385
+    },
+    {
+      "epoch": 0.3634669151910531,
+      "grad_norm": 0.49587902382827737,
+      "learning_rate": 4.882637210907836e-05,
+      "loss": 0.4417,
+      "step": 390
+    },
+    {
+      "epoch": 0.3681267474370923,
+      "grad_norm": 0.45767354851720066,
+      "learning_rate": 4.874007594062824e-05,
+      "loss": 0.4481,
+      "step": 395
+    },
+    {
+      "epoch": 0.3727865796831314,
+      "grad_norm": 0.4109939412993998,
+      "learning_rate": 4.865377977217811e-05,
+      "loss": 0.4328,
+      "step": 400
+    },
+    {
+      "epoch": 0.37744641192917056,
+      "grad_norm": 0.4267479303093156,
+      "learning_rate": 4.8567483603728e-05,
+      "loss": 0.4512,
+      "step": 405
+    },
+    {
+      "epoch": 0.3821062441752097,
+      "grad_norm": 0.46022700390396754,
+      "learning_rate": 4.8481187435277875e-05,
+      "loss": 0.4473,
+      "step": 410
+    },
+    {
+      "epoch": 0.38676607642124883,
+      "grad_norm": 0.5818788628854474,
+      "learning_rate": 4.839489126682776e-05,
+      "loss": 0.4338,
+      "step": 415
+    },
+    {
+      "epoch": 0.391425908667288,
+      "grad_norm": 0.6348777601747888,
+      "learning_rate": 4.830859509837763e-05,
+      "loss": 0.4521,
+      "step": 420
+    },
+    {
+      "epoch": 0.3960857409133271,
+      "grad_norm": 0.3406273073070538,
+      "learning_rate": 4.822229892992751e-05,
+      "loss": 0.4495,
+      "step": 425
+    },
+    {
+      "epoch": 0.40074557315936626,
+      "grad_norm": 0.39490973933479806,
+      "learning_rate": 4.8136002761477395e-05,
+      "loss": 0.4364,
+      "step": 430
+    },
+    {
+      "epoch": 0.40540540540540543,
+      "grad_norm": 0.5621814721849863,
+      "learning_rate": 4.804970659302727e-05,
+      "loss": 0.4497,
+      "step": 435
+    },
+    {
+      "epoch": 0.41006523765144454,
+      "grad_norm": 0.5250828265047137,
+      "learning_rate": 4.796341042457715e-05,
+      "loss": 0.4438,
+      "step": 440
+    },
+    {
+      "epoch": 0.4147250698974837,
+      "grad_norm": 0.49605694640709724,
+      "learning_rate": 4.787711425612703e-05,
+      "loss": 0.4384,
+      "step": 445
+    },
+    {
+      "epoch": 0.4193849021435228,
+      "grad_norm": 0.49091569267750296,
+      "learning_rate": 4.779081808767691e-05,
+      "loss": 0.4292,
+      "step": 450
+    },
+    {
+      "epoch": 0.424044734389562,
+      "grad_norm": 0.5449853039218947,
+      "learning_rate": 4.770452191922679e-05,
+      "loss": 0.4359,
+      "step": 455
+    },
+    {
+      "epoch": 0.42870456663560114,
+      "grad_norm": 0.40595214751293,
+      "learning_rate": 4.761822575077667e-05,
+      "loss": 0.4327,
+      "step": 460
+    },
+    {
+      "epoch": 0.43336439888164024,
+      "grad_norm": 0.4959979164579244,
+      "learning_rate": 4.753192958232654e-05,
+      "loss": 0.4297,
+      "step": 465
+    },
+    {
+      "epoch": 0.4380242311276794,
+      "grad_norm": 0.42025573002718436,
+      "learning_rate": 4.744563341387643e-05,
+      "loss": 0.4426,
+      "step": 470
+    },
+    {
+      "epoch": 0.4426840633737186,
+      "grad_norm": 0.4082690281031086,
+      "learning_rate": 4.7359337245426306e-05,
+      "loss": 0.432,
+      "step": 475
+    },
+    {
+      "epoch": 0.4473438956197577,
+      "grad_norm": 0.46608619482702135,
+      "learning_rate": 4.7273041076976184e-05,
+      "loss": 0.4441,
+      "step": 480
+    },
+    {
+      "epoch": 0.45200372786579684,
+      "grad_norm": 0.41782415569185494,
+      "learning_rate": 4.718674490852606e-05,
+      "loss": 0.4367,
+      "step": 485
+    },
+    {
+      "epoch": 0.45666356011183595,
+      "grad_norm": 0.4159441336334524,
+      "learning_rate": 4.710044874007594e-05,
+      "loss": 0.431,
+      "step": 490
+    },
+    {
+      "epoch": 0.4613233923578751,
+      "grad_norm": 0.4267635887639151,
+      "learning_rate": 4.7014152571625826e-05,
+      "loss": 0.437,
+      "step": 495
+    },
+    {
+      "epoch": 0.4659832246039143,
+      "grad_norm": 0.536368733946403,
+      "learning_rate": 4.6927856403175704e-05,
+      "loss": 0.4398,
+      "step": 500
+    },
+    {
+      "epoch": 0.4706430568499534,
+      "grad_norm": 0.38365822324922083,
+      "learning_rate": 4.684156023472558e-05,
+      "loss": 0.4264,
+      "step": 505
+    },
+    {
+      "epoch": 0.47530288909599255,
+      "grad_norm": 0.49638181546901267,
+      "learning_rate": 4.675526406627546e-05,
+      "loss": 0.4349,
+      "step": 510
+    },
+    {
+      "epoch": 0.47996272134203166,
+      "grad_norm": 0.6039980202672868,
+      "learning_rate": 4.666896789782534e-05,
+      "loss": 0.4329,
+      "step": 515
+    },
+    {
+      "epoch": 0.4846225535880708,
+      "grad_norm": 0.531071639674577,
+      "learning_rate": 4.658267172937522e-05,
+      "loss": 0.4367,
+      "step": 520
+    },
+    {
+      "epoch": 0.48928238583411,
+      "grad_norm": 0.41927620753314093,
+      "learning_rate": 4.64963755609251e-05,
+      "loss": 0.4356,
+      "step": 525
+    },
+    {
+      "epoch": 0.4939422180801491,
+      "grad_norm": 0.49167935938181634,
+      "learning_rate": 4.641007939247497e-05,
+      "loss": 0.4324,
+      "step": 530
+    },
+    {
+      "epoch": 0.49860205032618826,
+      "grad_norm": 0.35194515788154196,
+      "learning_rate": 4.632378322402486e-05,
+      "loss": 0.4327,
+      "step": 535
+    },
+    {
+      "epoch": 0.5032618825722274,
+      "grad_norm": 0.3880164900857034,
+      "learning_rate": 4.6237487055574736e-05,
+      "loss": 0.4227,
+      "step": 540
+    },
+    {
+      "epoch": 0.5079217148182665,
+      "grad_norm": 0.4647026652109177,
+      "learning_rate": 4.6151190887124615e-05,
+      "loss": 0.4289,
+      "step": 545
+    },
+    {
+      "epoch": 0.5125815470643057,
+      "grad_norm": 0.4369224822745491,
+      "learning_rate": 4.606489471867449e-05,
+      "loss": 0.4292,
+      "step": 550
+    },
+    {
+      "epoch": 0.5172413793103449,
+      "grad_norm": 0.3516777481993619,
+      "learning_rate": 4.597859855022437e-05,
+      "loss": 0.4236,
+      "step": 555
+    },
+    {
+      "epoch": 0.5219012115563839,
+      "grad_norm": 0.4189980745198798,
+      "learning_rate": 4.589230238177425e-05,
+      "loss": 0.4442,
+      "step": 560
+    },
+    {
+      "epoch": 0.5265610438024231,
+      "grad_norm": 0.4546880509441404,
+      "learning_rate": 4.5806006213324134e-05,
+      "loss": 0.4219,
+      "step": 565
+    },
+    {
+      "epoch": 0.5312208760484622,
+      "grad_norm": 0.5415119247861921,
+      "learning_rate": 4.5719710044874006e-05,
+      "loss": 0.4218,
+      "step": 570
+    },
+    {
+      "epoch": 0.5358807082945014,
+      "grad_norm": 0.4163996631718338,
+      "learning_rate": 4.563341387642389e-05,
+      "loss": 0.428,
+      "step": 575
+    },
+    {
+      "epoch": 0.5405405405405406,
+      "grad_norm": 0.3619146303574906,
+      "learning_rate": 4.554711770797377e-05,
+      "loss": 0.428,
+      "step": 580
+    },
+    {
+      "epoch": 0.5452003727865797,
+      "grad_norm": 0.4528885219717091,
+      "learning_rate": 4.546082153952365e-05,
+      "loss": 0.4248,
+      "step": 585
+    },
+    {
+      "epoch": 0.5498602050326188,
+      "grad_norm": 0.42455500708425176,
+      "learning_rate": 4.5374525371073526e-05,
+      "loss": 0.4212,
+      "step": 590
+    },
+    {
+      "epoch": 0.554520037278658,
+      "grad_norm": 0.39895122252871523,
+      "learning_rate": 4.5288229202623404e-05,
+      "loss": 0.4233,
+      "step": 595
+    },
+    {
+      "epoch": 0.5591798695246971,
+      "grad_norm": 0.3990992719666481,
+      "learning_rate": 4.520193303417328e-05,
+      "loss": 0.4236,
+      "step": 600
+    },
+    {
+      "epoch": 0.5638397017707363,
+      "grad_norm": 0.3553585945932701,
+      "learning_rate": 4.511563686572317e-05,
+      "loss": 0.4223,
+      "step": 605
+    },
+    {
+      "epoch": 0.5684995340167754,
+      "grad_norm": 0.42046831338605023,
+      "learning_rate": 4.5029340697273045e-05,
+      "loss": 0.4334,
+      "step": 610
+    },
+    {
+      "epoch": 0.5731593662628145,
+      "grad_norm": 0.34391259032310467,
+      "learning_rate": 4.4943044528822923e-05,
+      "loss": 0.42,
+      "step": 615
+    },
+    {
+      "epoch": 0.5778191985088537,
+      "grad_norm": 0.4087272926784166,
+      "learning_rate": 4.48567483603728e-05,
+      "loss": 0.417,
+      "step": 620
+    },
+    {
+      "epoch": 0.5824790307548928,
+      "grad_norm": 0.3589614143705352,
+      "learning_rate": 4.477045219192268e-05,
+      "loss": 0.4387,
+      "step": 625
+    },
+    {
+      "epoch": 0.587138863000932,
+      "grad_norm": 0.3629824785143128,
+      "learning_rate": 4.4684156023472565e-05,
+      "loss": 0.4294,
+      "step": 630
+    },
+    {
+      "epoch": 0.5917986952469712,
+      "grad_norm": 0.47917310854378126,
+      "learning_rate": 4.4597859855022436e-05,
+      "loss": 0.4339,
+      "step": 635
+    },
+    {
+      "epoch": 0.5964585274930102,
+      "grad_norm": 0.4782662682845663,
+      "learning_rate": 4.4511563686572315e-05,
+      "loss": 0.4247,
+      "step": 640
+    },
+    {
+      "epoch": 0.6011183597390494,
+      "grad_norm": 0.471784972910924,
+      "learning_rate": 4.44252675181222e-05,
+      "loss": 0.4243,
+      "step": 645
+    },
+    {
+      "epoch": 0.6057781919850885,
+      "grad_norm": 0.42122505647880576,
+      "learning_rate": 4.433897134967208e-05,
+      "loss": 0.433,
+      "step": 650
+    },
+    {
+      "epoch": 0.6104380242311277,
+      "grad_norm": 0.4025673330141692,
+      "learning_rate": 4.4252675181221956e-05,
+      "loss": 0.4306,
+      "step": 655
+    },
+    {
+      "epoch": 0.6150978564771669,
+      "grad_norm": 0.33637741802274024,
+      "learning_rate": 4.4166379012771834e-05,
+      "loss": 0.4124,
+      "step": 660
+    },
+    {
+      "epoch": 0.6197576887232059,
+      "grad_norm": 0.46827398668452175,
+      "learning_rate": 4.408008284432171e-05,
+      "loss": 0.4132,
+      "step": 665
+    },
+    {
+      "epoch": 0.6244175209692451,
+      "grad_norm": 0.4372278275475368,
+      "learning_rate": 4.39937866758716e-05,
+      "loss": 0.4118,
+      "step": 670
+    },
+    {
+      "epoch": 0.6290773532152842,
+      "grad_norm": 0.41420214852536175,
+      "learning_rate": 4.3907490507421476e-05,
+      "loss": 0.4204,
+      "step": 675
+    },
+    {
+      "epoch": 0.6337371854613234,
+      "grad_norm": 0.4227796669483225,
+      "learning_rate": 4.382119433897135e-05,
+      "loss": 0.4233,
+      "step": 680
+    },
+    {
+      "epoch": 0.6383970177073626,
+      "grad_norm": 0.42862178932562206,
+      "learning_rate": 4.373489817052123e-05,
+      "loss": 0.4329,
+      "step": 685
+    },
+    {
+      "epoch": 0.6430568499534017,
+      "grad_norm": 0.4382533539684448,
+      "learning_rate": 4.364860200207111e-05,
+      "loss": 0.4151,
+      "step": 690
+    },
+    {
+      "epoch": 0.6477166821994408,
+      "grad_norm": 0.35948786609504135,
+      "learning_rate": 4.356230583362099e-05,
+      "loss": 0.4202,
+      "step": 695
+    },
+    {
+      "epoch": 0.65237651444548,
+      "grad_norm": 0.35194142984278265,
+      "learning_rate": 4.347600966517087e-05,
+      "loss": 0.4288,
+      "step": 700
+    },
+    {
+      "epoch": 0.6570363466915191,
+      "grad_norm": 0.4108441539595895,
+      "learning_rate": 4.3389713496720745e-05,
+      "loss": 0.4223,
+      "step": 705
+    },
+    {
+      "epoch": 0.6616961789375583,
+      "grad_norm": 0.48807209838383736,
+      "learning_rate": 4.330341732827063e-05,
+      "loss": 0.4229,
+      "step": 710
+    },
+    {
+      "epoch": 0.6663560111835974,
+      "grad_norm": 0.4177025446331477,
+      "learning_rate": 4.321712115982051e-05,
+      "loss": 0.4114,
+      "step": 715
+    },
+    {
+      "epoch": 0.6710158434296365,
+      "grad_norm": 0.3755176399576921,
+      "learning_rate": 4.3130824991370387e-05,
+      "loss": 0.4217,
+      "step": 720
+    },
+    {
+      "epoch": 0.6756756756756757,
+      "grad_norm": 0.36461380313474906,
+      "learning_rate": 4.3044528822920265e-05,
+      "loss": 0.4205,
+      "step": 725
+    },
+    {
+      "epoch": 0.6803355079217148,
+      "grad_norm": 0.4251420641738401,
+      "learning_rate": 4.295823265447014e-05,
+      "loss": 0.4153,
+      "step": 730
+    },
+    {
+      "epoch": 0.684995340167754,
+      "grad_norm": 0.4003093585866478,
+      "learning_rate": 4.287193648602002e-05,
+      "loss": 0.4193,
+      "step": 735
+    },
+    {
+      "epoch": 0.6896551724137931,
+      "grad_norm": 0.34876369977589905,
+      "learning_rate": 4.27856403175699e-05,
+      "loss": 0.4129,
+      "step": 740
+    },
+    {
+      "epoch": 0.6943150046598322,
+      "grad_norm": 0.3707569691409881,
+      "learning_rate": 4.269934414911978e-05,
+      "loss": 0.4087,
+      "step": 745
+    },
+    {
+      "epoch": 0.6989748369058714,
+      "grad_norm": 0.4316626941569482,
+      "learning_rate": 4.261304798066966e-05,
+      "loss": 0.4135,
+      "step": 750
+    },
+    {
+      "epoch": 0.7036346691519105,
+      "grad_norm": 0.39303618105943594,
+      "learning_rate": 4.252675181221954e-05,
+      "loss": 0.4159,
+      "step": 755
+    },
+    {
+      "epoch": 0.7082945013979497,
+      "grad_norm": 0.4267487540956833,
+      "learning_rate": 4.244045564376942e-05,
+      "loss": 0.419,
+      "step": 760
+    },
+    {
+      "epoch": 0.7129543336439889,
+      "grad_norm": 0.3664167792787496,
+      "learning_rate": 4.23541594753193e-05,
+      "loss": 0.403,
+      "step": 765
+    },
+    {
+      "epoch": 0.7176141658900279,
+      "grad_norm": 0.3577500899585447,
+      "learning_rate": 4.2267863306869176e-05,
+      "loss": 0.4122,
+      "step": 770
+    },
+    {
+      "epoch": 0.7222739981360671,
+      "grad_norm": 0.39507959012374977,
+      "learning_rate": 4.2181567138419054e-05,
+      "loss": 0.4249,
+      "step": 775
+    },
+    {
+      "epoch": 0.7269338303821062,
+      "grad_norm": 0.3688889563306457,
+      "learning_rate": 4.209527096996894e-05,
+      "loss": 0.4094,
+      "step": 780
+    },
+    {
+      "epoch": 0.7315936626281454,
+      "grad_norm": 0.3846331698540081,
+      "learning_rate": 4.200897480151881e-05,
+      "loss": 0.4167,
+      "step": 785
+    },
+    {
+      "epoch": 0.7362534948741846,
+      "grad_norm": 0.43990146629985694,
+      "learning_rate": 4.1922678633068695e-05,
+      "loss": 0.42,
+      "step": 790
+    },
+    {
+      "epoch": 0.7409133271202236,
+      "grad_norm": 0.4264573932403777,
+      "learning_rate": 4.1836382464618573e-05,
+      "loss": 0.4142,
+      "step": 795
+    },
+    {
+      "epoch": 0.7455731593662628,
+      "grad_norm": 0.43741863957364285,
+      "learning_rate": 4.175008629616845e-05,
+      "loss": 0.4124,
+      "step": 800
+    },
+    {
+      "epoch": 0.750232991612302,
+      "grad_norm": 0.38911299384649556,
+      "learning_rate": 4.166379012771833e-05,
+      "loss": 0.4122,
+      "step": 805
+    },
+    {
+      "epoch": 0.7548928238583411,
+      "grad_norm": 0.34798977131970454,
+      "learning_rate": 4.157749395926821e-05,
+      "loss": 0.4304,
+      "step": 810
+    },
+    {
+      "epoch": 0.7595526561043803,
+      "grad_norm": 0.360973732965553,
+      "learning_rate": 4.1491197790818086e-05,
+      "loss": 0.4048,
+      "step": 815
+    },
+    {
+      "epoch": 0.7642124883504194,
+      "grad_norm": 0.5342748114831669,
+      "learning_rate": 4.140490162236797e-05,
+      "loss": 0.4202,
+      "step": 820
+    },
+    {
+      "epoch": 0.7688723205964585,
+      "grad_norm": 0.4164036109796091,
+      "learning_rate": 4.131860545391785e-05,
+      "loss": 0.4124,
+      "step": 825
+    },
+    {
+      "epoch": 0.7735321528424977,
+      "grad_norm": 0.3661344541633312,
+      "learning_rate": 4.123230928546773e-05,
+      "loss": 0.4071,
+      "step": 830
+    },
+    {
+      "epoch": 0.7781919850885368,
+      "grad_norm": 0.3754811730264276,
+      "learning_rate": 4.1146013117017606e-05,
+      "loss": 0.4142,
+      "step": 835
+    },
+    {
+      "epoch": 0.782851817334576,
+      "grad_norm": 0.419200409043472,
+      "learning_rate": 4.1059716948567484e-05,
+      "loss": 0.4173,
+      "step": 840
+    },
+    {
+      "epoch": 0.7875116495806151,
+      "grad_norm": 0.43893714450759036,
+      "learning_rate": 4.097342078011737e-05,
+      "loss": 0.4253,
+      "step": 845
+    },
+    {
+      "epoch": 0.7921714818266542,
+      "grad_norm": 0.41322614598162755,
+      "learning_rate": 4.088712461166724e-05,
+      "loss": 0.4151,
+      "step": 850
+    },
+    {
+      "epoch": 0.7968313140726934,
+      "grad_norm": 0.5117391056019034,
+      "learning_rate": 4.080082844321712e-05,
+      "loss": 0.417,
+      "step": 855
+    },
+    {
+      "epoch": 0.8014911463187325,
+      "grad_norm": 0.39482750255468374,
+      "learning_rate": 4.0714532274767004e-05,
+      "loss": 0.4239,
+      "step": 860
+    },
+    {
+      "epoch": 0.8061509785647717,
+      "grad_norm": 0.35016978679527583,
+      "learning_rate": 4.062823610631688e-05,
+      "loss": 0.405,
+      "step": 865
+    },
+    {
+      "epoch": 0.8108108108108109,
+      "grad_norm": 0.32094903787924467,
+      "learning_rate": 4.054193993786676e-05,
+      "loss": 0.4194,
+      "step": 870
+    },
+    {
+      "epoch": 0.8154706430568499,
+      "grad_norm": 0.36889819669973173,
+      "learning_rate": 4.045564376941664e-05,
+      "loss": 0.4149,
+      "step": 875
+    },
+    {
+      "epoch": 0.8201304753028891,
+      "grad_norm": 0.3880605400422551,
+      "learning_rate": 4.036934760096652e-05,
+      "loss": 0.4059,
+      "step": 880
+    },
+    {
+      "epoch": 0.8247903075489282,
+      "grad_norm": 0.377773246265045,
+      "learning_rate": 4.02830514325164e-05,
+      "loss": 0.42,
+      "step": 885
+    },
+    {
+      "epoch": 0.8294501397949674,
+      "grad_norm": 0.38824868030619064,
+      "learning_rate": 4.019675526406628e-05,
+      "loss": 0.4067,
+      "step": 890
+    },
+    {
+      "epoch": 0.8341099720410066,
+      "grad_norm": 0.34547566635591975,
+      "learning_rate": 4.011045909561615e-05,
+      "loss": 0.4261,
+      "step": 895
+    },
+    {
+      "epoch": 0.8387698042870456,
+      "grad_norm": 0.3630734939739242,
+      "learning_rate": 4.0024162927166037e-05,
+      "loss": 0.409,
+      "step": 900
+    },
+    {
+      "epoch": 0.8434296365330848,
+      "grad_norm": 0.3215149282810052,
+      "learning_rate": 3.9937866758715915e-05,
+      "loss": 0.4105,
+      "step": 905
+    },
+    {
+      "epoch": 0.848089468779124,
+      "grad_norm": 0.5156466838457244,
+      "learning_rate": 3.98515705902658e-05,
+      "loss": 0.4167,
+      "step": 910
+    },
+    {
+      "epoch": 0.8527493010251631,
+      "grad_norm": 0.58719622036636,
+      "learning_rate": 3.976527442181567e-05,
+      "loss": 0.4094,
+      "step": 915
+    },
+    {
+      "epoch": 0.8574091332712023,
+      "grad_norm": 0.44663910672199086,
+      "learning_rate": 3.967897825336555e-05,
+      "loss": 0.4158,
+      "step": 920
+    },
+    {
+      "epoch": 0.8620689655172413,
+      "grad_norm": 0.3593752942361397,
+      "learning_rate": 3.9592682084915434e-05,
+      "loss": 0.4024,
+      "step": 925
+    },
+    {
+      "epoch": 0.8667287977632805,
+      "grad_norm": 0.5114184723138864,
+      "learning_rate": 3.950638591646531e-05,
+      "loss": 0.402,
+      "step": 930
+    },
+    {
+      "epoch": 0.8713886300093197,
+      "grad_norm": 0.3926412761807418,
+      "learning_rate": 3.942008974801519e-05,
+      "loss": 0.4139,
+      "step": 935
+    },
+    {
+      "epoch": 0.8760484622553588,
+      "grad_norm": 0.3111102224238461,
+      "learning_rate": 3.933379357956507e-05,
+      "loss": 0.4095,
+      "step": 940
+    },
+    {
+      "epoch": 0.880708294501398,
+      "grad_norm": 0.42810427582045707,
+      "learning_rate": 3.924749741111495e-05,
+      "loss": 0.4117,
+      "step": 945
+    },
+    {
+      "epoch": 0.8853681267474371,
+      "grad_norm": 0.48569011616159047,
+      "learning_rate": 3.916120124266483e-05,
+      "loss": 0.4195,
+      "step": 950
+    },
+    {
+      "epoch": 0.8900279589934762,
+      "grad_norm": 0.4190117006397043,
+      "learning_rate": 3.9074905074214704e-05,
+      "loss": 0.4184,
+      "step": 955
+    },
+    {
+      "epoch": 0.8946877912395154,
+      "grad_norm": 0.33914591782346504,
+      "learning_rate": 3.898860890576458e-05,
+      "loss": 0.4137,
+      "step": 960
+    },
+    {
+      "epoch": 0.8993476234855545,
+      "grad_norm": 0.43035587029435407,
+      "learning_rate": 3.890231273731447e-05,
+      "loss": 0.4053,
+      "step": 965
+    },
+    {
+      "epoch": 0.9040074557315937,
+      "grad_norm": 0.3101483588886324,
+      "learning_rate": 3.8816016568864345e-05,
+      "loss": 0.4031,
+      "step": 970
+    },
+    {
+      "epoch": 0.9086672879776329,
+      "grad_norm": 0.3322568146796224,
+      "learning_rate": 3.8729720400414224e-05,
+      "loss": 0.422,
+      "step": 975
+    },
+    {
+      "epoch": 0.9133271202236719,
+      "grad_norm": 0.3917813978291434,
+      "learning_rate": 3.86434242319641e-05,
+      "loss": 0.4091,
+      "step": 980
+    },
+    {
+      "epoch": 0.9179869524697111,
+      "grad_norm": 0.4742078695343531,
+      "learning_rate": 3.855712806351398e-05,
+      "loss": 0.4103,
+      "step": 985
+    },
+    {
+      "epoch": 0.9226467847157502,
+      "grad_norm": 0.45678911380863857,
+      "learning_rate": 3.8470831895063865e-05,
+      "loss": 0.4079,
+      "step": 990
+    },
+    {
+      "epoch": 0.9273066169617894,
+      "grad_norm": 0.4015797894562155,
+      "learning_rate": 3.838453572661374e-05,
+      "loss": 0.4036,
+      "step": 995
+    },
+    {
+      "epoch": 0.9319664492078286,
+      "grad_norm": 0.4393074039612925,
+      "learning_rate": 3.8298239558163615e-05,
+      "loss": 0.4066,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9366262814538676,
+      "grad_norm": 0.401959095434801,
+      "learning_rate": 3.82119433897135e-05,
+      "loss": 0.4113,
+      "step": 1005
+    },
+    {
+      "epoch": 0.9412861136999068,
+      "grad_norm": 0.38718845936575524,
+      "learning_rate": 3.812564722126338e-05,
+      "loss": 0.4042,
+      "step": 1010
+    },
+    {
+      "epoch": 0.9459459459459459,
+      "grad_norm": 0.40104704083420933,
+      "learning_rate": 3.8039351052813256e-05,
+      "loss": 0.4126,
+      "step": 1015
+    },
+    {
+      "epoch": 0.9506057781919851,
+      "grad_norm": 0.4415733445950018,
+      "learning_rate": 3.7953054884363134e-05,
+      "loss": 0.4063,
+      "step": 1020
+    },
+    {
+      "epoch": 0.9552656104380243,
+      "grad_norm": 0.37032743287501774,
+      "learning_rate": 3.786675871591301e-05,
+      "loss": 0.41,
+      "step": 1025
+    },
+    {
+      "epoch": 0.9599254426840633,
+      "grad_norm": 0.3024002245710326,
+      "learning_rate": 3.77804625474629e-05,
+      "loss": 0.4151,
+      "step": 1030
+    },
+    {
+      "epoch": 0.9645852749301025,
+      "grad_norm": 0.43466697925061876,
+      "learning_rate": 3.7694166379012776e-05,
+      "loss": 0.4074,
+      "step": 1035
+    },
+    {
+      "epoch": 0.9692451071761417,
+      "grad_norm": 0.4045497692524809,
+      "learning_rate": 3.7607870210562654e-05,
+      "loss": 0.4051,
+      "step": 1040
+    },
+    {
+      "epoch": 0.9739049394221808,
+      "grad_norm": 0.2870249098276766,
+      "learning_rate": 3.752157404211253e-05,
+      "loss": 0.3975,
+      "step": 1045
+    },
+    {
+      "epoch": 0.97856477166822,
+      "grad_norm": 0.44417765948538723,
+      "learning_rate": 3.743527787366241e-05,
+      "loss": 0.4273,
+      "step": 1050
+    },
+    {
+      "epoch": 0.983224603914259,
+      "grad_norm": 0.3866227301554364,
+      "learning_rate": 3.734898170521229e-05,
+      "loss": 0.4049,
+      "step": 1055
+    },
+    {
+      "epoch": 0.9878844361602982,
+      "grad_norm": 0.3834591341930073,
+      "learning_rate": 3.7262685536762174e-05,
+      "loss": 0.4041,
+      "step": 1060
+    },
+    {
+      "epoch": 0.9925442684063374,
+      "grad_norm": 0.3608189448212277,
+      "learning_rate": 3.7176389368312045e-05,
+      "loss": 0.4082,
+      "step": 1065
+    },
+    {
+      "epoch": 0.9972041006523765,
+      "grad_norm": 0.4677480050799797,
+      "learning_rate": 3.709009319986193e-05,
+      "loss": 0.4086,
+      "step": 1070
+    },
+    {
+      "epoch": 1.0018639328984156,
+      "grad_norm": 0.3925117221097306,
+      "learning_rate": 3.700379703141181e-05,
+      "loss": 0.3734,
+      "step": 1075
+    },
+    {
+      "epoch": 1.0065237651444547,
+      "grad_norm": 0.3242823046323612,
+      "learning_rate": 3.6917500862961687e-05,
+      "loss": 0.3492,
+      "step": 1080
+    },
+    {
+      "epoch": 1.011183597390494,
+      "grad_norm": 0.4138715616637518,
+      "learning_rate": 3.6831204694511565e-05,
+      "loss": 0.3557,
+      "step": 1085
+    },
+    {
+      "epoch": 1.015843429636533,
+      "grad_norm": 0.3816339832645457,
+      "learning_rate": 3.674490852606144e-05,
+      "loss": 0.3626,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0205032618825722,
+      "grad_norm": 0.3231428483280001,
+      "learning_rate": 3.665861235761132e-05,
+      "loss": 0.3487,
+      "step": 1095
+    },
+    {
+      "epoch": 1.0251630941286114,
+      "grad_norm": 0.4390336371531291,
+      "learning_rate": 3.6572316189161206e-05,
+      "loss": 0.3401,
+      "step": 1100
+    },
+    {
+      "epoch": 1.0298229263746506,
+      "grad_norm": 0.3739976173993305,
+      "learning_rate": 3.6486020020711085e-05,
+      "loss": 0.3468,
+      "step": 1105
+    },
+    {
+      "epoch": 1.0344827586206897,
+      "grad_norm": 0.30745527861808797,
+      "learning_rate": 3.639972385226096e-05,
+      "loss": 0.3473,
+      "step": 1110
+    },
+    {
+      "epoch": 1.0391425908667289,
+      "grad_norm": 0.311090717121424,
+      "learning_rate": 3.631342768381084e-05,
+      "loss": 0.3467,
+      "step": 1115
+    },
+    {
+      "epoch": 1.0438024231127678,
+      "grad_norm": 0.357597674408912,
+      "learning_rate": 3.622713151536072e-05,
+      "loss": 0.3464,
+      "step": 1120
+    },
+    {
+      "epoch": 1.048462255358807,
+      "grad_norm": 0.32526109191464125,
+      "learning_rate": 3.6140835346910604e-05,
+      "loss": 0.3505,
+      "step": 1125
+    },
+    {
+      "epoch": 1.0531220876048462,
+      "grad_norm": 0.3315794059992157,
+      "learning_rate": 3.6054539178460476e-05,
+      "loss": 0.3472,
+      "step": 1130
+    },
+    {
+      "epoch": 1.0577819198508853,
+      "grad_norm": 0.39159745949716,
+      "learning_rate": 3.5968243010010354e-05,
+      "loss": 0.3633,
+      "step": 1135
+    },
+    {
+      "epoch": 1.0624417520969245,
+      "grad_norm": 0.3268669622340021,
+      "learning_rate": 3.588194684156024e-05,
+      "loss": 0.3541,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0671015843429636,
+      "grad_norm": 0.3573697719413499,
+      "learning_rate": 3.579565067311012e-05,
+      "loss": 0.3454,
+      "step": 1145
+    },
+    {
+      "epoch": 1.0717614165890028,
+      "grad_norm": 0.3234870090563129,
+      "learning_rate": 3.5709354504659995e-05,
+      "loss": 0.342,
+      "step": 1150
+    },
+    {
+      "epoch": 1.076421248835042,
+      "grad_norm": 0.35361119617130937,
+      "learning_rate": 3.5623058336209874e-05,
+      "loss": 0.3531,
+      "step": 1155
+    },
+    {
+      "epoch": 1.0810810810810811,
+      "grad_norm": 0.31913782855515077,
+      "learning_rate": 3.553676216775975e-05,
+      "loss": 0.3443,
+      "step": 1160
+    },
+    {
+      "epoch": 1.0857409133271203,
+      "grad_norm": 0.29066190723337,
+      "learning_rate": 3.545046599930964e-05,
+      "loss": 0.3562,
+      "step": 1165
+    },
+    {
+      "epoch": 1.0904007455731595,
+      "grad_norm": 0.28057846055796526,
+      "learning_rate": 3.536416983085951e-05,
+      "loss": 0.3578,
+      "step": 1170
+    },
+    {
+      "epoch": 1.0950605778191984,
+      "grad_norm": 0.3528126877993132,
+      "learning_rate": 3.5277873662409386e-05,
+      "loss": 0.3444,
+      "step": 1175
+    },
+    {
+      "epoch": 1.0997204100652376,
+      "grad_norm": 0.32165919133814297,
+      "learning_rate": 3.519157749395927e-05,
+      "loss": 0.3504,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1043802423112767,
+      "grad_norm": 0.4088384463330471,
+      "learning_rate": 3.510528132550915e-05,
+      "loss": 0.3474,
+      "step": 1185
+    },
+    {
+      "epoch": 1.109040074557316,
+      "grad_norm": 0.3217473099595502,
+      "learning_rate": 3.501898515705903e-05,
+      "loss": 0.3495,
+      "step": 1190
+    },
+    {
+      "epoch": 1.113699906803355,
+      "grad_norm": 0.3333951141255346,
+      "learning_rate": 3.4932688988608906e-05,
+      "loss": 0.3608,
+      "step": 1195
+    },
+    {
+      "epoch": 1.1183597390493942,
+      "grad_norm": 0.36178211070864513,
+      "learning_rate": 3.4846392820158784e-05,
+      "loss": 0.3494,
+      "step": 1200
+    },
+    {
+      "epoch": 1.1230195712954334,
+      "grad_norm": 0.3931146552174344,
+      "learning_rate": 3.476009665170867e-05,
+      "loss": 0.3494,
+      "step": 1205
+    },
+    {
+      "epoch": 1.1276794035414726,
+      "grad_norm": 0.35642549099851184,
+      "learning_rate": 3.467380048325855e-05,
+      "loss": 0.3479,
+      "step": 1210
+    },
+    {
+      "epoch": 1.1323392357875117,
+      "grad_norm": 0.3113424243765187,
+      "learning_rate": 3.458750431480842e-05,
+      "loss": 0.3549,
+      "step": 1215
+    },
+    {
+      "epoch": 1.1369990680335509,
+      "grad_norm": 0.3671193715290013,
+      "learning_rate": 3.4501208146358304e-05,
+      "loss": 0.3537,
+      "step": 1220
+    },
+    {
+      "epoch": 1.14165890027959,
+      "grad_norm": 0.30756115076957946,
+      "learning_rate": 3.441491197790818e-05,
+      "loss": 0.3497,
+      "step": 1225
+    },
+    {
+      "epoch": 1.146318732525629,
+      "grad_norm": 0.2886729734663684,
+      "learning_rate": 3.432861580945806e-05,
+      "loss": 0.3512,
+      "step": 1230
+    },
+    {
+      "epoch": 1.1509785647716682,
+      "grad_norm": 0.3242211895005395,
+      "learning_rate": 3.424231964100794e-05,
+      "loss": 0.3502,
+      "step": 1235
+    },
+    {
+      "epoch": 1.1556383970177073,
+      "grad_norm": 0.38383015173739526,
+      "learning_rate": 3.415602347255782e-05,
+      "loss": 0.3489,
+      "step": 1240
+    },
+    {
+      "epoch": 1.1602982292637465,
+      "grad_norm": 0.28220156879742464,
+      "learning_rate": 3.40697273041077e-05,
+      "loss": 0.3559,
+      "step": 1245
+    },
+    {
+      "epoch": 1.1649580615097856,
+      "grad_norm": 0.41862621406850076,
+      "learning_rate": 3.398343113565758e-05,
+      "loss": 0.3501,
+      "step": 1250
+    },
+    {
+      "epoch": 1.1696178937558248,
+      "grad_norm": 0.2721035238920554,
+      "learning_rate": 3.389713496720746e-05,
+      "loss": 0.329,
+      "step": 1255
+    },
+    {
+      "epoch": 1.174277726001864,
+      "grad_norm": 0.2777723472168308,
+      "learning_rate": 3.381083879875734e-05,
+      "loss": 0.3487,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1789375582479031,
+      "grad_norm": 0.30260505363460816,
+      "learning_rate": 3.3724542630307215e-05,
+      "loss": 0.345,
+      "step": 1265
+    },
+    {
+      "epoch": 1.1835973904939423,
+      "grad_norm": 0.32084597957439803,
+      "learning_rate": 3.363824646185709e-05,
+      "loss": 0.354,
+      "step": 1270
+    },
+    {
+      "epoch": 1.1882572227399812,
+      "grad_norm": 0.2788324351707279,
+      "learning_rate": 3.355195029340698e-05,
+      "loss": 0.3481,
+      "step": 1275
+    },
+    {
+      "epoch": 1.1929170549860204,
+      "grad_norm": 0.32991275098127965,
+      "learning_rate": 3.346565412495685e-05,
+      "loss": 0.3589,
+      "step": 1280
+    },
+    {
+      "epoch": 1.1975768872320596,
+      "grad_norm": 0.3375689864632857,
+      "learning_rate": 3.3379357956506735e-05,
+      "loss": 0.3597,
+      "step": 1285
+    },
+    {
+      "epoch": 1.2022367194780987,
+      "grad_norm": 0.3367658315280557,
+      "learning_rate": 3.329306178805661e-05,
+      "loss": 0.3432,
+      "step": 1290
+    },
+    {
+      "epoch": 1.206896551724138,
+      "grad_norm": 0.34039045852928035,
+      "learning_rate": 3.320676561960649e-05,
+      "loss": 0.3481,
+      "step": 1295
+    },
+    {
+      "epoch": 1.211556383970177,
+      "grad_norm": 0.4410063394768305,
+      "learning_rate": 3.312046945115637e-05,
+      "loss": 0.3563,
+      "step": 1300
+    },
+    {
+      "epoch": 1.2162162162162162,
+      "grad_norm": 0.37770120291136555,
+      "learning_rate": 3.303417328270625e-05,
+      "loss": 0.3546,
+      "step": 1305
+    },
+    {
+      "epoch": 1.2208760484622554,
+      "grad_norm": 0.30686984772647424,
+      "learning_rate": 3.2947877114256126e-05,
+      "loss": 0.344,
+      "step": 1310
+    },
+    {
+      "epoch": 1.2255358807082946,
+      "grad_norm": 0.3153022389575515,
+      "learning_rate": 3.286158094580601e-05,
+      "loss": 0.364,
+      "step": 1315
+    },
+    {
+      "epoch": 1.2301957129543337,
+      "grad_norm": 0.274264578660015,
+      "learning_rate": 3.277528477735589e-05,
+      "loss": 0.3496,
+      "step": 1320
+    },
+    {
+      "epoch": 1.2348555452003729,
+      "grad_norm": 0.315982491275267,
+      "learning_rate": 3.268898860890577e-05,
+      "loss": 0.3569,
+      "step": 1325
+    },
+    {
+      "epoch": 1.2395153774464118,
+      "grad_norm": 0.33256551029664194,
+      "learning_rate": 3.2602692440455645e-05,
+      "loss": 0.3453,
+      "step": 1330
+    },
+    {
+      "epoch": 1.244175209692451,
+      "grad_norm": 0.34413275906180446,
+      "learning_rate": 3.2516396272005524e-05,
+      "loss": 0.3477,
+      "step": 1335
+    },
+    {
+      "epoch": 1.2488350419384902,
+      "grad_norm": 0.42922843597586463,
+      "learning_rate": 3.243010010355541e-05,
+      "loss": 0.3532,
+      "step": 1340
+    },
+    {
+      "epoch": 1.2534948741845293,
+      "grad_norm": 0.32129109214079155,
+      "learning_rate": 3.234380393510528e-05,
+      "loss": 0.3406,
+      "step": 1345
+    },
+    {
+      "epoch": 1.2581547064305685,
+      "grad_norm": 0.26270774966560256,
+      "learning_rate": 3.225750776665516e-05,
+      "loss": 0.348,
+      "step": 1350
+    },
+    {
+      "epoch": 1.2628145386766076,
+      "grad_norm": 0.315468672326717,
+      "learning_rate": 3.217121159820504e-05,
+      "loss": 0.3577,
+      "step": 1355
+    },
+    {
+      "epoch": 1.2674743709226468,
+      "grad_norm": 0.32475088370503385,
+      "learning_rate": 3.208491542975492e-05,
+      "loss": 0.3471,
+      "step": 1360
+    },
+    {
+      "epoch": 1.272134203168686,
+      "grad_norm": 0.3472189821685297,
+      "learning_rate": 3.19986192613048e-05,
+      "loss": 0.3524,
+      "step": 1365
+    },
+    {
+      "epoch": 1.2767940354147251,
+      "grad_norm": 0.35201175237308874,
+      "learning_rate": 3.191232309285468e-05,
+      "loss": 0.3515,
+      "step": 1370
+    },
+    {
+      "epoch": 1.281453867660764,
+      "grad_norm": 0.31322276160231,
+      "learning_rate": 3.1826026924404556e-05,
+      "loss": 0.3413,
+      "step": 1375
+    },
+    {
+      "epoch": 1.2861136999068035,
+      "grad_norm": 0.3125897715288331,
+      "learning_rate": 3.173973075595444e-05,
+      "loss": 0.3559,
+      "step": 1380
+    },
+    {
+      "epoch": 1.2907735321528424,
+      "grad_norm": 0.3500290640211496,
+      "learning_rate": 3.165343458750431e-05,
+      "loss": 0.3554,
+      "step": 1385
+    },
+    {
+      "epoch": 1.2954333643988816,
+      "grad_norm": 0.35924383933187914,
+      "learning_rate": 3.156713841905419e-05,
+      "loss": 0.3478,
+      "step": 1390
+    },
+    {
+      "epoch": 1.3000931966449207,
+      "grad_norm": 0.3240552623071767,
+      "learning_rate": 3.1480842250604076e-05,
+      "loss": 0.3522,
+      "step": 1395
+    },
+    {
+      "epoch": 1.30475302889096,
+      "grad_norm": 0.3776212492449141,
+      "learning_rate": 3.1394546082153954e-05,
+      "loss": 0.3463,
+      "step": 1400
+    },
+    {
+      "epoch": 1.309412861136999,
+      "grad_norm": 0.3723245498635331,
+      "learning_rate": 3.130824991370383e-05,
+      "loss": 0.3568,
+      "step": 1405
+    },
+    {
+      "epoch": 1.3140726933830382,
+      "grad_norm": 0.35075484625980147,
+      "learning_rate": 3.122195374525371e-05,
+      "loss": 0.3451,
+      "step": 1410
+    },
+    {
+      "epoch": 1.3187325256290774,
+      "grad_norm": 0.30197919303897663,
+      "learning_rate": 3.113565757680359e-05,
+      "loss": 0.3525,
+      "step": 1415
+    },
+    {
+      "epoch": 1.3233923578751166,
+      "grad_norm": 0.31435674602210045,
+      "learning_rate": 3.1049361408353474e-05,
+      "loss": 0.3398,
+      "step": 1420
+    },
+    {
+      "epoch": 1.3280521901211557,
+      "grad_norm": 0.26193370594807897,
+      "learning_rate": 3.096306523990335e-05,
+      "loss": 0.3477,
+      "step": 1425
+    },
+    {
+      "epoch": 1.3327120223671947,
+      "grad_norm": 0.27354032463967926,
+      "learning_rate": 3.0876769071453223e-05,
+      "loss": 0.3565,
+      "step": 1430
+    },
+    {
+      "epoch": 1.337371854613234,
+      "grad_norm": 0.31403549789712104,
+      "learning_rate": 3.079047290300311e-05,
+      "loss": 0.3545,
+      "step": 1435
+    },
+    {
+      "epoch": 1.342031686859273,
+      "grad_norm": 0.3164978494457341,
+      "learning_rate": 3.070417673455299e-05,
+      "loss": 0.3528,
+      "step": 1440
+    },
+    {
+      "epoch": 1.3466915191053122,
+      "grad_norm": 0.29866886965035916,
+      "learning_rate": 3.061788056610287e-05,
+      "loss": 0.3576,
+      "step": 1445
+    },
+    {
+      "epoch": 1.3513513513513513,
+      "grad_norm": 0.33112119030982146,
+      "learning_rate": 3.053158439765274e-05,
+      "loss": 0.3433,
+      "step": 1450
+    },
+    {
+      "epoch": 1.3560111835973905,
+      "grad_norm": 0.30484350511317343,
+      "learning_rate": 3.0445288229202625e-05,
+      "loss": 0.3509,
+      "step": 1455
+    },
+    {
+      "epoch": 1.3606710158434296,
+      "grad_norm": 0.30459690237321196,
+      "learning_rate": 3.0358992060752506e-05,
+      "loss": 0.357,
+      "step": 1460
+    },
+    {
+      "epoch": 1.3653308480894688,
+      "grad_norm": 0.32879838871626565,
+      "learning_rate": 3.027269589230238e-05,
+      "loss": 0.3509,
+      "step": 1465
+    },
+    {
+      "epoch": 1.369990680335508,
+      "grad_norm": 0.3249392520827432,
+      "learning_rate": 3.018639972385226e-05,
+      "loss": 0.3482,
+      "step": 1470
+    },
+    {
+      "epoch": 1.3746505125815471,
+      "grad_norm": 0.3474187101486429,
+      "learning_rate": 3.010010355540214e-05,
+      "loss": 0.3499,
+      "step": 1475
+    },
+    {
+      "epoch": 1.3793103448275863,
+      "grad_norm": 0.25523379854359524,
+      "learning_rate": 3.001380738695202e-05,
+      "loss": 0.3338,
+      "step": 1480
+    },
+    {
+      "epoch": 1.3839701770736252,
+      "grad_norm": 0.28033083753234334,
+      "learning_rate": 2.99275112185019e-05,
+      "loss": 0.3513,
+      "step": 1485
+    },
+    {
+      "epoch": 1.3886300093196646,
+      "grad_norm": 0.3288088359530075,
+      "learning_rate": 2.984121505005178e-05,
+      "loss": 0.3565,
+      "step": 1490
+    },
+    {
+      "epoch": 1.3932898415657036,
+      "grad_norm": 0.33858464934408805,
+      "learning_rate": 2.9754918881601657e-05,
+      "loss": 0.3647,
+      "step": 1495
+    },
+    {
+      "epoch": 1.3979496738117427,
+      "grad_norm": 0.3283654008215437,
+      "learning_rate": 2.966862271315154e-05,
+      "loss": 0.3606,
+      "step": 1500
+    },
+    {
+      "epoch": 1.402609506057782,
+      "grad_norm": 0.3700275695935878,
+      "learning_rate": 2.9582326544701417e-05,
+      "loss": 0.3642,
+      "step": 1505
+    },
+    {
+      "epoch": 1.407269338303821,
+      "grad_norm": 0.3275972102791232,
+      "learning_rate": 2.9496030376251292e-05,
+      "loss": 0.3497,
+      "step": 1510
+    },
+    {
+      "epoch": 1.4119291705498602,
+      "grad_norm": 0.2775244795834382,
+      "learning_rate": 2.9409734207801177e-05,
+      "loss": 0.3529,
+      "step": 1515
+    },
+    {
+      "epoch": 1.4165890027958994,
+      "grad_norm": 0.33772663347341314,
+      "learning_rate": 2.9323438039351052e-05,
+      "loss": 0.3526,
+      "step": 1520
+    },
+    {
+      "epoch": 1.4212488350419386,
+      "grad_norm": 0.31737217299385195,
+      "learning_rate": 2.9237141870900937e-05,
+      "loss": 0.3482,
+      "step": 1525
+    },
+    {
+      "epoch": 1.4259086672879777,
+      "grad_norm": 0.28261971890517357,
+      "learning_rate": 2.9150845702450812e-05,
+      "loss": 0.3473,
+      "step": 1530
+    },
+    {
+      "epoch": 1.4305684995340169,
+      "grad_norm": 0.29615924118648024,
+      "learning_rate": 2.906454953400069e-05,
+      "loss": 0.3522,
+      "step": 1535
+    },
+    {
+      "epoch": 1.4352283317800558,
+      "grad_norm": 0.3352703627712436,
+      "learning_rate": 2.897825336555057e-05,
+      "loss": 0.3484,
+      "step": 1540
+    },
+    {
+      "epoch": 1.439888164026095,
+      "grad_norm": 0.4373434216044478,
+      "learning_rate": 2.889195719710045e-05,
+      "loss": 0.3488,
+      "step": 1545
+    },
+    {
+      "epoch": 1.4445479962721341,
+      "grad_norm": 0.3924587730579784,
+      "learning_rate": 2.8805661028650328e-05,
+      "loss": 0.345,
+      "step": 1550
+    },
+    {
+      "epoch": 1.4492078285181733,
+      "grad_norm": 0.4091864022793969,
+      "learning_rate": 2.871936486020021e-05,
+      "loss": 0.3425,
+      "step": 1555
+    },
+    {
+      "epoch": 1.4538676607642125,
+      "grad_norm": 0.3735443689246875,
+      "learning_rate": 2.8633068691750088e-05,
+      "loss": 0.3449,
+      "step": 1560
+    },
+    {
+      "epoch": 1.4585274930102516,
+      "grad_norm": 0.37078512558076115,
+      "learning_rate": 2.854677252329997e-05,
+      "loss": 0.3475,
+      "step": 1565
+    },
+    {
+      "epoch": 1.4631873252562908,
+      "grad_norm": 0.34445437877669055,
+      "learning_rate": 2.8460476354849848e-05,
+      "loss": 0.3545,
+      "step": 1570
+    },
+    {
+      "epoch": 1.46784715750233,
+      "grad_norm": 0.40254864862904804,
+      "learning_rate": 2.8374180186399723e-05,
+      "loss": 0.3543,
+      "step": 1575
+    },
+    {
+      "epoch": 1.4725069897483691,
+      "grad_norm": 0.3464624270144264,
+      "learning_rate": 2.8287884017949608e-05,
+      "loss": 0.3555,
+      "step": 1580
+    },
+    {
+      "epoch": 1.477166821994408,
+      "grad_norm": 0.3406697399202642,
+      "learning_rate": 2.8201587849499482e-05,
+      "loss": 0.3486,
+      "step": 1585
+    },
+    {
+      "epoch": 1.4818266542404475,
+      "grad_norm": 0.34006016225440305,
+      "learning_rate": 2.811529168104936e-05,
+      "loss": 0.3587,
+      "step": 1590
+    },
+    {
+      "epoch": 1.4864864864864864,
+      "grad_norm": 0.31076893163516733,
+      "learning_rate": 2.8028995512599242e-05,
+      "loss": 0.3496,
+      "step": 1595
+    },
+    {
+      "epoch": 1.4911463187325256,
+      "grad_norm": 0.35622061929896803,
+      "learning_rate": 2.794269934414912e-05,
+      "loss": 0.351,
+      "step": 1600
+    },
+    {
+      "epoch": 1.4958061509785647,
+      "grad_norm": 0.28412573542460823,
+      "learning_rate": 2.7856403175699002e-05,
+      "loss": 0.3554,
+      "step": 1605
+    },
+    {
+      "epoch": 1.500465983224604,
+      "grad_norm": 0.34962920187865865,
+      "learning_rate": 2.777010700724888e-05,
+      "loss": 0.3504,
+      "step": 1610
+    },
+    {
+      "epoch": 1.505125815470643,
+      "grad_norm": 0.32148391556813743,
+      "learning_rate": 2.768381083879876e-05,
+      "loss": 0.3456,
+      "step": 1615
+    },
+    {
+      "epoch": 1.5097856477166822,
+      "grad_norm": 0.2979430610033821,
+      "learning_rate": 2.759751467034864e-05,
+      "loss": 0.349,
+      "step": 1620
+    },
+    {
+      "epoch": 1.5144454799627214,
+      "grad_norm": 0.274661962764332,
+      "learning_rate": 2.751121850189852e-05,
+      "loss": 0.3422,
+      "step": 1625
+    },
+    {
+      "epoch": 1.5191053122087603,
+      "grad_norm": 0.33331819001261015,
+      "learning_rate": 2.7424922333448393e-05,
+      "loss": 0.3358,
+      "step": 1630
+    },
+    {
+      "epoch": 1.5237651444547997,
+      "grad_norm": 0.37263713055207437,
+      "learning_rate": 2.7338626164998278e-05,
+      "loss": 0.343,
+      "step": 1635
+    },
+    {
+      "epoch": 1.5284249767008387,
+      "grad_norm": 0.2836229938139772,
+      "learning_rate": 2.7252329996548153e-05,
+      "loss": 0.3458,
+      "step": 1640
+    },
+    {
+      "epoch": 1.533084808946878,
+      "grad_norm": 0.28812507496364903,
+      "learning_rate": 2.7166033828098038e-05,
+      "loss": 0.361,
+      "step": 1645
+    },
+    {
+      "epoch": 1.537744641192917,
+      "grad_norm": 0.31374053879704117,
+      "learning_rate": 2.7079737659647913e-05,
+      "loss": 0.3501,
+      "step": 1650
+    },
+    {
+      "epoch": 1.5424044734389561,
+      "grad_norm": 0.3315497266948002,
+      "learning_rate": 2.699344149119779e-05,
+      "loss": 0.357,
+      "step": 1655
+    },
+    {
+      "epoch": 1.5470643056849953,
+      "grad_norm": 0.3422418207636451,
+      "learning_rate": 2.6907145322747673e-05,
+      "loss": 0.3512,
+      "step": 1660
+    },
+    {
+      "epoch": 1.5517241379310345,
+      "grad_norm": 0.3715397650391368,
+      "learning_rate": 2.682084915429755e-05,
+      "loss": 0.3421,
+      "step": 1665
+    },
+    {
+      "epoch": 1.5563839701770736,
+      "grad_norm": 0.34517149876929565,
+      "learning_rate": 2.673455298584743e-05,
+      "loss": 0.347,
+      "step": 1670
+    },
+    {
+      "epoch": 1.5610438024231128,
+      "grad_norm": 0.3273948636409803,
+      "learning_rate": 2.664825681739731e-05,
+      "loss": 0.3545,
+      "step": 1675
+    },
+    {
+      "epoch": 1.565703634669152,
+      "grad_norm": 0.2895984901541935,
+      "learning_rate": 2.6561960648947186e-05,
+      "loss": 0.3462,
+      "step": 1680
+    },
+    {
+      "epoch": 1.570363466915191,
+      "grad_norm": 0.2901125844043081,
+      "learning_rate": 2.647566448049707e-05,
+      "loss": 0.3421,
+      "step": 1685
+    },
+    {
+      "epoch": 1.5750232991612303,
+      "grad_norm": 0.2870864130794471,
+      "learning_rate": 2.6389368312046945e-05,
+      "loss": 0.3544,
+      "step": 1690
+    },
+    {
+      "epoch": 1.5796831314072692,
+      "grad_norm": 0.2651088709357085,
+      "learning_rate": 2.6303072143596824e-05,
+      "loss": 0.3585,
+      "step": 1695
+    },
+    {
+      "epoch": 1.5843429636533086,
+      "grad_norm": 0.32808480999934597,
+      "learning_rate": 2.6216775975146705e-05,
+      "loss": 0.3509,
+      "step": 1700
+    },
+    {
+      "epoch": 1.5890027958993476,
+      "grad_norm": 0.2988222621349023,
+      "learning_rate": 2.6130479806696584e-05,
+      "loss": 0.3484,
+      "step": 1705
+    },
+    {
+      "epoch": 1.5936626281453867,
+      "grad_norm": 0.36200483951270196,
+      "learning_rate": 2.6044183638246462e-05,
+      "loss": 0.3537,
+      "step": 1710
+    },
+    {
+      "epoch": 1.598322460391426,
+      "grad_norm": 0.24619168427061267,
+      "learning_rate": 2.5957887469796343e-05,
+      "loss": 0.3432,
+      "step": 1715
+    },
+    {
+      "epoch": 1.602982292637465,
+      "grad_norm": 0.27018050263181986,
+      "learning_rate": 2.587159130134622e-05,
+      "loss": 0.3476,
+      "step": 1720
+    },
+    {
+      "epoch": 1.6076421248835042,
+      "grad_norm": 0.2738158348907855,
+      "learning_rate": 2.5785295132896096e-05,
+      "loss": 0.3475,
+      "step": 1725
+    },
+    {
+      "epoch": 1.6123019571295434,
+      "grad_norm": 0.33689443334374064,
+      "learning_rate": 2.569899896444598e-05,
+      "loss": 0.3414,
+      "step": 1730
+    },
+    {
+      "epoch": 1.6169617893755825,
+      "grad_norm": 0.31348202595949626,
+      "learning_rate": 2.5612702795995856e-05,
+      "loss": 0.3532,
+      "step": 1735
+    },
+    {
+      "epoch": 1.6216216216216215,
+      "grad_norm": 0.2900821832234747,
+      "learning_rate": 2.552640662754574e-05,
+      "loss": 0.3466,
+      "step": 1740
+    },
+    {
+      "epoch": 1.6262814538676609,
+      "grad_norm": 0.2851244733459075,
+      "learning_rate": 2.5440110459095616e-05,
+      "loss": 0.3535,
+      "step": 1745
+    },
+    {
+      "epoch": 1.6309412861136998,
+      "grad_norm": 0.3693601184388067,
+      "learning_rate": 2.5353814290645494e-05,
+      "loss": 0.3522,
+      "step": 1750
+    },
+    {
+      "epoch": 1.6356011183597392,
+      "grad_norm": 0.29626890132059874,
+      "learning_rate": 2.5267518122195376e-05,
+      "loss": 0.3441,
+      "step": 1755
+    },
+    {
+      "epoch": 1.6402609506057781,
+      "grad_norm": 0.36710243494281314,
+      "learning_rate": 2.5181221953745254e-05,
+      "loss": 0.3526,
+      "step": 1760
+    },
+    {
+      "epoch": 1.6449207828518173,
+      "grad_norm": 0.2935136410260085,
+      "learning_rate": 2.5094925785295132e-05,
+      "loss": 0.3543,
+      "step": 1765
+    },
+    {
+      "epoch": 1.6495806150978565,
+      "grad_norm": 0.2843336784267725,
+      "learning_rate": 2.5008629616845014e-05,
+      "loss": 0.3474,
+      "step": 1770
+    },
+    {
+      "epoch": 1.6542404473438956,
+      "grad_norm": 0.3413623078154405,
+      "learning_rate": 2.4922333448394892e-05,
+      "loss": 0.34,
+      "step": 1775
+    },
+    {
+      "epoch": 1.6589002795899348,
+      "grad_norm": 0.3256075923701561,
+      "learning_rate": 2.483603727994477e-05,
+      "loss": 0.3497,
+      "step": 1780
+    },
+    {
+      "epoch": 1.6635601118359737,
+      "grad_norm": 0.2918964427060244,
+      "learning_rate": 2.4749741111494652e-05,
+      "loss": 0.3471,
+      "step": 1785
+    },
+    {
+      "epoch": 1.6682199440820131,
+      "grad_norm": 0.2850685569306027,
+      "learning_rate": 2.466344494304453e-05,
+      "loss": 0.3465,
+      "step": 1790
+    },
+    {
+      "epoch": 1.672879776328052,
+      "grad_norm": 0.2813413366207838,
+      "learning_rate": 2.457714877459441e-05,
+      "loss": 0.339,
+      "step": 1795
+    },
+    {
+      "epoch": 1.6775396085740915,
+      "grad_norm": 0.2874800136662209,
+      "learning_rate": 2.4490852606144287e-05,
+      "loss": 0.3391,
+      "step": 1800
+    },
+    {
+      "epoch": 1.6821994408201304,
+      "grad_norm": 0.29894669893728437,
+      "learning_rate": 2.440455643769417e-05,
+      "loss": 0.3422,
+      "step": 1805
+    },
+    {
+      "epoch": 1.6868592730661698,
+      "grad_norm": 0.27574525609872574,
+      "learning_rate": 2.4318260269244047e-05,
+      "loss": 0.3459,
+      "step": 1810
+    },
+    {
+      "epoch": 1.6915191053122087,
+      "grad_norm": 0.24779390120946074,
+      "learning_rate": 2.4231964100793925e-05,
+      "loss": 0.3335,
+      "step": 1815
+    },
+    {
+      "epoch": 1.696178937558248,
+      "grad_norm": 0.2708193368538403,
+      "learning_rate": 2.4145667932343803e-05,
+      "loss": 0.3447,
+      "step": 1820
+    },
+    {
+      "epoch": 1.700838769804287,
+      "grad_norm": 0.25621073372544567,
+      "learning_rate": 2.4059371763893685e-05,
+      "loss": 0.3495,
+      "step": 1825
+    },
+    {
+      "epoch": 1.7054986020503262,
+      "grad_norm": 0.2738238649435202,
+      "learning_rate": 2.3973075595443563e-05,
+      "loss": 0.3544,
+      "step": 1830
+    },
+    {
+      "epoch": 1.7101584342963654,
+      "grad_norm": 0.37045529725221055,
+      "learning_rate": 2.388677942699344e-05,
+      "loss": 0.3495,
+      "step": 1835
+    },
+    {
+      "epoch": 1.7148182665424043,
+      "grad_norm": 0.2872683227698006,
+      "learning_rate": 2.3800483258543323e-05,
+      "loss": 0.3456,
+      "step": 1840
+    },
+    {
+      "epoch": 1.7194780987884437,
+      "grad_norm": 0.2999306723313641,
+      "learning_rate": 2.37141870900932e-05,
+      "loss": 0.3502,
+      "step": 1845
+    },
+    {
+      "epoch": 1.7241379310344827,
+      "grad_norm": 0.25115358761380624,
+      "learning_rate": 2.3627890921643083e-05,
+      "loss": 0.3331,
+      "step": 1850
+    },
+    {
+      "epoch": 1.728797763280522,
+      "grad_norm": 0.30413899967026337,
+      "learning_rate": 2.3541594753192957e-05,
+      "loss": 0.3429,
+      "step": 1855
+    },
+    {
+      "epoch": 1.733457595526561,
+      "grad_norm": 0.30125880998725124,
+      "learning_rate": 2.345529858474284e-05,
+      "loss": 0.3435,
+      "step": 1860
+    },
+    {
+      "epoch": 1.7381174277726001,
+      "grad_norm": 0.32202456195609636,
+      "learning_rate": 2.3369002416292717e-05,
+      "loss": 0.3427,
+      "step": 1865
+    },
+    {
+      "epoch": 1.7427772600186393,
+      "grad_norm": 0.32197995562079285,
+      "learning_rate": 2.32827062478426e-05,
+      "loss": 0.3418,
+      "step": 1870
+    },
+    {
+      "epoch": 1.7474370922646785,
+      "grad_norm": 0.255236984530475,
+      "learning_rate": 2.3196410079392474e-05,
+      "loss": 0.3505,
+      "step": 1875
+    },
+    {
+      "epoch": 1.7520969245107176,
+      "grad_norm": 0.3214227473874937,
+      "learning_rate": 2.3110113910942355e-05,
+      "loss": 0.345,
+      "step": 1880
+    },
+    {
+      "epoch": 1.7567567567567568,
+      "grad_norm": 0.2902161243562964,
+      "learning_rate": 2.3023817742492234e-05,
+      "loss": 0.3503,
+      "step": 1885
+    },
+    {
+      "epoch": 1.761416589002796,
+      "grad_norm": 0.2798199356972451,
+      "learning_rate": 2.2937521574042115e-05,
+      "loss": 0.3481,
+      "step": 1890
+    },
+    {
+      "epoch": 1.766076421248835,
+      "grad_norm": 0.3205891408693067,
+      "learning_rate": 2.285122540559199e-05,
+      "loss": 0.3456,
+      "step": 1895
+    },
+    {
+      "epoch": 1.7707362534948743,
+      "grad_norm": 0.3460171767800496,
+      "learning_rate": 2.276492923714187e-05,
+      "loss": 0.3379,
+      "step": 1900
+    },
+    {
+      "epoch": 1.7753960857409132,
+      "grad_norm": 0.27153736535936507,
+      "learning_rate": 2.267863306869175e-05,
+      "loss": 0.3373,
+      "step": 1905
+    },
+    {
+      "epoch": 1.7800559179869526,
+      "grad_norm": 0.2522385351246555,
+      "learning_rate": 2.259233690024163e-05,
+      "loss": 0.3499,
+      "step": 1910
+    },
+    {
+      "epoch": 1.7847157502329916,
+      "grad_norm": 0.29837488535093165,
+      "learning_rate": 2.250604073179151e-05,
+      "loss": 0.3332,
+      "step": 1915
+    },
+    {
+      "epoch": 1.7893755824790307,
+      "grad_norm": 0.30736203767752707,
+      "learning_rate": 2.2419744563341388e-05,
+      "loss": 0.3491,
+      "step": 1920
+    },
+    {
+      "epoch": 1.7940354147250699,
+      "grad_norm": 0.26525661258403604,
+      "learning_rate": 2.233344839489127e-05,
+      "loss": 0.3459,
+      "step": 1925
+    },
+    {
+      "epoch": 1.798695246971109,
+      "grad_norm": 0.28662983654454316,
+      "learning_rate": 2.2247152226441148e-05,
+      "loss": 0.3423,
+      "step": 1930
+    },
+    {
+      "epoch": 1.8033550792171482,
+      "grad_norm": 0.3072372286013865,
+      "learning_rate": 2.2160856057991026e-05,
+      "loss": 0.3472,
+      "step": 1935
+    },
+    {
+      "epoch": 1.8080149114631874,
+      "grad_norm": 0.2686734450236742,
+      "learning_rate": 2.2074559889540904e-05,
+      "loss": 0.3351,
+      "step": 1940
+    },
+    {
+      "epoch": 1.8126747437092265,
+      "grad_norm": 0.27055764279116856,
+      "learning_rate": 2.1988263721090786e-05,
+      "loss": 0.3519,
+      "step": 1945
+    },
+    {
+      "epoch": 1.8173345759552655,
+      "grad_norm": 0.270801891724948,
+      "learning_rate": 2.1901967552640664e-05,
+      "loss": 0.3352,
+      "step": 1950
+    },
+    {
+      "epoch": 1.8219944082013049,
+      "grad_norm": 0.3376179200871101,
+      "learning_rate": 2.1815671384190542e-05,
+      "loss": 0.3433,
+      "step": 1955
+    },
+    {
+      "epoch": 1.8266542404473438,
+      "grad_norm": 0.3094291281432446,
+      "learning_rate": 2.172937521574042e-05,
+      "loss": 0.3468,
+      "step": 1960
+    },
+    {
+      "epoch": 1.8313140726933832,
+      "grad_norm": 0.30108696983896277,
+      "learning_rate": 2.1643079047290302e-05,
+      "loss": 0.3484,
+      "step": 1965
+    },
+    {
+      "epoch": 1.8359739049394221,
+      "grad_norm": 0.28808400729201145,
+      "learning_rate": 2.155678287884018e-05,
+      "loss": 0.3457,
+      "step": 1970
+    },
+    {
+      "epoch": 1.8406337371854613,
+      "grad_norm": 0.2626867135990665,
+      "learning_rate": 2.147048671039006e-05,
+      "loss": 0.3538,
+      "step": 1975
+    },
+    {
+      "epoch": 1.8452935694315005,
+      "grad_norm": 0.2545222699034913,
+      "learning_rate": 2.1384190541939937e-05,
+      "loss": 0.3441,
+      "step": 1980
+    },
+    {
+      "epoch": 1.8499534016775396,
+      "grad_norm": 0.2549762449152119,
+      "learning_rate": 2.129789437348982e-05,
+      "loss": 0.34,
+      "step": 1985
+    },
+    {
+      "epoch": 1.8546132339235788,
+      "grad_norm": 0.26994596327457904,
+      "learning_rate": 2.1211598205039697e-05,
+      "loss": 0.3588,
+      "step": 1990
+    },
+    {
+      "epoch": 1.8592730661696177,
+      "grad_norm": 0.3216618454834016,
+      "learning_rate": 2.1125302036589575e-05,
+      "loss": 0.3445,
+      "step": 1995
+    },
+    {
+      "epoch": 1.8639328984156571,
+      "grad_norm": 0.30002781053596494,
+      "learning_rate": 2.1039005868139457e-05,
+      "loss": 0.3434,
+      "step": 2000
+    },
+    {
+      "epoch": 1.868592730661696,
+      "grad_norm": 0.2624621146167971,
+      "learning_rate": 2.0952709699689335e-05,
+      "loss": 0.3371,
+      "step": 2005
+    },
+    {
+      "epoch": 1.8732525629077355,
+      "grad_norm": 0.2846552906265825,
+      "learning_rate": 2.0866413531239216e-05,
+      "loss": 0.3522,
+      "step": 2010
+    },
+    {
+      "epoch": 1.8779123951537744,
+      "grad_norm": 0.3035769653970306,
+      "learning_rate": 2.078011736278909e-05,
+      "loss": 0.3383,
+      "step": 2015
+    },
+    {
+      "epoch": 1.8825722273998136,
+      "grad_norm": 0.27140142490596453,
+      "learning_rate": 2.0693821194338973e-05,
+      "loss": 0.3333,
+      "step": 2020
+    },
+    {
+      "epoch": 1.8872320596458527,
+      "grad_norm": 0.3146195865656141,
+      "learning_rate": 2.060752502588885e-05,
+      "loss": 0.3392,
+      "step": 2025
+    },
+    {
+      "epoch": 1.8918918918918919,
+      "grad_norm": 0.24781220736275708,
+      "learning_rate": 2.0521228857438733e-05,
+      "loss": 0.3463,
+      "step": 2030
+    },
+    {
+      "epoch": 1.896551724137931,
+      "grad_norm": 0.2511766596522395,
+      "learning_rate": 2.0434932688988608e-05,
+      "loss": 0.3396,
+      "step": 2035
+    },
+    {
+      "epoch": 1.9012115563839702,
+      "grad_norm": 0.2443023930158746,
+      "learning_rate": 2.034863652053849e-05,
+      "loss": 0.3423,
+      "step": 2040
+    },
+    {
+      "epoch": 1.9058713886300094,
+      "grad_norm": 0.24918210888469716,
+      "learning_rate": 2.0262340352088367e-05,
+      "loss": 0.358,
+      "step": 2045
+    },
+    {
+      "epoch": 1.9105312208760483,
+      "grad_norm": 0.2837014945618899,
+      "learning_rate": 2.017604418363825e-05,
+      "loss": 0.3485,
+      "step": 2050
+    },
+    {
+      "epoch": 1.9151910531220877,
+      "grad_norm": 0.3127804701878722,
+      "learning_rate": 2.0089748015188127e-05,
+      "loss": 0.3429,
+      "step": 2055
+    },
+    {
+      "epoch": 1.9198508853681266,
+      "grad_norm": 0.2666255188065951,
+      "learning_rate": 2.0003451846738005e-05,
+      "loss": 0.3505,
+      "step": 2060
+    },
+    {
+      "epoch": 1.924510717614166,
+      "grad_norm": 0.29789654298155993,
+      "learning_rate": 1.9917155678287887e-05,
+      "loss": 0.34,
+      "step": 2065
+    },
+    {
+      "epoch": 1.929170549860205,
+      "grad_norm": 0.2642352382725658,
+      "learning_rate": 1.9830859509837765e-05,
+      "loss": 0.3475,
+      "step": 2070
+    },
+    {
+      "epoch": 1.9338303821062441,
+      "grad_norm": 0.2645416411826404,
+      "learning_rate": 1.9744563341387643e-05,
+      "loss": 0.3488,
+      "step": 2075
+    },
+    {
+      "epoch": 1.9384902143522833,
+      "grad_norm": 0.29553107175227916,
+      "learning_rate": 1.9658267172937522e-05,
+      "loss": 0.3439,
+      "step": 2080
+    },
+    {
+      "epoch": 1.9431500465983225,
+      "grad_norm": 0.3589091568313813,
+      "learning_rate": 1.9571971004487403e-05,
+      "loss": 0.3436,
+      "step": 2085
+    },
+    {
+      "epoch": 1.9478098788443616,
+      "grad_norm": 0.30654472343576,
+      "learning_rate": 1.948567483603728e-05,
+      "loss": 0.3495,
+      "step": 2090
+    },
+    {
+      "epoch": 1.9524697110904008,
+      "grad_norm": 0.3122541799556844,
+      "learning_rate": 1.939937866758716e-05,
+      "loss": 0.332,
+      "step": 2095
+    },
+    {
+      "epoch": 1.95712954333644,
+      "grad_norm": 0.34083389131210867,
+      "learning_rate": 1.9313082499137038e-05,
+      "loss": 0.3491,
+      "step": 2100
+    },
+    {
+      "epoch": 1.961789375582479,
+      "grad_norm": 0.3389679056885684,
+      "learning_rate": 1.922678633068692e-05,
+      "loss": 0.3435,
+      "step": 2105
+    },
+    {
+      "epoch": 1.9664492078285183,
+      "grad_norm": 0.284119291374539,
+      "learning_rate": 1.9140490162236798e-05,
+      "loss": 0.349,
+      "step": 2110
+    },
+    {
+      "epoch": 1.9711090400745572,
+      "grad_norm": 0.2751030635173434,
+      "learning_rate": 1.9054193993786676e-05,
+      "loss": 0.3562,
+      "step": 2115
+    },
+    {
+      "epoch": 1.9757688723205966,
+      "grad_norm": 0.2520969335383781,
+      "learning_rate": 1.8967897825336554e-05,
+      "loss": 0.3443,
+      "step": 2120
+    },
+    {
+      "epoch": 1.9804287045666356,
+      "grad_norm": 0.2416634545930341,
+      "learning_rate": 1.8881601656886436e-05,
+      "loss": 0.3417,
+      "step": 2125
+    },
+    {
+      "epoch": 1.9850885368126747,
+      "grad_norm": 0.28102065457654984,
+      "learning_rate": 1.8795305488436314e-05,
+      "loss": 0.3454,
+      "step": 2130
+    },
+    {
+      "epoch": 1.9897483690587139,
+      "grad_norm": 0.3225233365582484,
+      "learning_rate": 1.8709009319986192e-05,
+      "loss": 0.3338,
+      "step": 2135
+    },
+    {
+      "epoch": 1.994408201304753,
+      "grad_norm": 0.2766142574985382,
+      "learning_rate": 1.8622713151536074e-05,
+      "loss": 0.3409,
+      "step": 2140
+    },
+    {
+      "epoch": 1.9990680335507922,
+      "grad_norm": 0.24489555641581465,
+      "learning_rate": 1.8536416983085952e-05,
+      "loss": 0.3456,
+      "step": 2145
+    },
+    {
+      "epoch": 2.003727865796831,
+      "grad_norm": 0.26697517660954867,
+      "learning_rate": 1.8450120814635834e-05,
+      "loss": 0.2732,
+      "step": 2150
+    },
+    {
+      "epoch": 2.0083876980428705,
+      "grad_norm": 0.29713336900710496,
+      "learning_rate": 1.836382464618571e-05,
+      "loss": 0.2844,
+      "step": 2155
+    },
+    {
+      "epoch": 2.0130475302889095,
+      "grad_norm": 0.25425364630620323,
+      "learning_rate": 1.827752847773559e-05,
+      "loss": 0.2779,
+      "step": 2160
+    },
+    {
+      "epoch": 2.017707362534949,
+      "grad_norm": 0.2663192159735816,
+      "learning_rate": 1.819123230928547e-05,
+      "loss": 0.2803,
+      "step": 2165
+    },
+    {
+      "epoch": 2.022367194780988,
+      "grad_norm": 0.24826658297068152,
+      "learning_rate": 1.810493614083535e-05,
+      "loss": 0.2797,
+      "step": 2170
+    },
+    {
+      "epoch": 2.027027027027027,
+      "grad_norm": 0.23483010577743693,
+      "learning_rate": 1.8018639972385225e-05,
+      "loss": 0.2746,
+      "step": 2175
+    },
+    {
+      "epoch": 2.031686859273066,
+      "grad_norm": 0.2628640730877877,
+      "learning_rate": 1.7932343803935107e-05,
+      "loss": 0.2793,
+      "step": 2180
+    },
+    {
+      "epoch": 2.0363466915191055,
+      "grad_norm": 0.23805358955056016,
+      "learning_rate": 1.7846047635484985e-05,
+      "loss": 0.2758,
+      "step": 2185
+    },
+    {
+      "epoch": 2.0410065237651445,
+      "grad_norm": 0.24733983972944648,
+      "learning_rate": 1.7759751467034866e-05,
+      "loss": 0.2757,
+      "step": 2190
+    },
+    {
+      "epoch": 2.0456663560111834,
+      "grad_norm": 0.3055543356817937,
+      "learning_rate": 1.767345529858474e-05,
+      "loss": 0.2856,
+      "step": 2195
+    },
+    {
+      "epoch": 2.050326188257223,
+      "grad_norm": 0.2337073038384121,
+      "learning_rate": 1.7587159130134623e-05,
+      "loss": 0.2855,
+      "step": 2200
+    },
+    {
+      "epoch": 2.0549860205032617,
+      "grad_norm": 0.23937107448773656,
+      "learning_rate": 1.75008629616845e-05,
+      "loss": 0.2732,
+      "step": 2205
+    },
+    {
+      "epoch": 2.059645852749301,
+      "grad_norm": 0.2226874223201284,
+      "learning_rate": 1.7414566793234383e-05,
+      "loss": 0.2756,
+      "step": 2210
+    },
+    {
+      "epoch": 2.06430568499534,
+      "grad_norm": 0.2380895456546044,
+      "learning_rate": 1.732827062478426e-05,
+      "loss": 0.2777,
+      "step": 2215
+    },
+    {
+      "epoch": 2.0689655172413794,
+      "grad_norm": 0.2644818134374498,
+      "learning_rate": 1.724197445633414e-05,
+      "loss": 0.2726,
+      "step": 2220
+    },
+    {
+      "epoch": 2.0736253494874184,
+      "grad_norm": 0.2426869688707367,
+      "learning_rate": 1.715567828788402e-05,
+      "loss": 0.2708,
+      "step": 2225
+    },
+    {
+      "epoch": 2.0782851817334578,
+      "grad_norm": 0.2391482764535829,
+      "learning_rate": 1.70693821194339e-05,
+      "loss": 0.2711,
+      "step": 2230
+    },
+    {
+      "epoch": 2.0829450139794967,
+      "grad_norm": 0.210165128542321,
+      "learning_rate": 1.6983085950983777e-05,
+      "loss": 0.2714,
+      "step": 2235
+    },
+    {
+      "epoch": 2.0876048462255357,
+      "grad_norm": 0.22837868460684904,
+      "learning_rate": 1.6896789782533655e-05,
+      "loss": 0.2849,
+      "step": 2240
+    },
+    {
+      "epoch": 2.092264678471575,
+      "grad_norm": 0.23559601311506279,
+      "learning_rate": 1.6810493614083537e-05,
+      "loss": 0.277,
+      "step": 2245
+    },
+    {
+      "epoch": 2.096924510717614,
+      "grad_norm": 0.2662351451201318,
+      "learning_rate": 1.6724197445633415e-05,
+      "loss": 0.28,
+      "step": 2250
+    },
+    {
+      "epoch": 2.1015843429636534,
+      "grad_norm": 0.2738683927142175,
+      "learning_rate": 1.6637901277183294e-05,
+      "loss": 0.2835,
+      "step": 2255
+    },
+    {
+      "epoch": 2.1062441752096923,
+      "grad_norm": 0.22041016833826543,
+      "learning_rate": 1.6551605108733172e-05,
+      "loss": 0.2772,
+      "step": 2260
+    },
+    {
+      "epoch": 2.1109040074557317,
+      "grad_norm": 0.2497373742051042,
+      "learning_rate": 1.6465308940283053e-05,
+      "loss": 0.2792,
+      "step": 2265
+    },
+    {
+      "epoch": 2.1155638397017706,
+      "grad_norm": 0.2390321080094361,
+      "learning_rate": 1.637901277183293e-05,
+      "loss": 0.2832,
+      "step": 2270
+    },
+    {
+      "epoch": 2.12022367194781,
+      "grad_norm": 0.209579439472578,
+      "learning_rate": 1.629271660338281e-05,
+      "loss": 0.2758,
+      "step": 2275
+    },
+    {
+      "epoch": 2.124883504193849,
+      "grad_norm": 0.28525949537957784,
+      "learning_rate": 1.620642043493269e-05,
+      "loss": 0.2813,
+      "step": 2280
+    },
+    {
+      "epoch": 2.1295433364398884,
+      "grad_norm": 0.236227199647008,
+      "learning_rate": 1.612012426648257e-05,
+      "loss": 0.2743,
+      "step": 2285
+    },
+    {
+      "epoch": 2.1342031686859273,
+      "grad_norm": 0.22821560445761854,
+      "learning_rate": 1.6033828098032448e-05,
+      "loss": 0.2762,
+      "step": 2290
+    },
+    {
+      "epoch": 2.1388630009319662,
+      "grad_norm": 0.24440972075031359,
+      "learning_rate": 1.5947531929582326e-05,
+      "loss": 0.2799,
+      "step": 2295
+    },
+    {
+      "epoch": 2.1435228331780056,
+      "grad_norm": 0.2581682742335318,
+      "learning_rate": 1.5861235761132208e-05,
+      "loss": 0.2806,
+      "step": 2300
+    },
+    {
+      "epoch": 2.1481826654240446,
+      "grad_norm": 0.2540521626388728,
+      "learning_rate": 1.5774939592682086e-05,
+      "loss": 0.273,
+      "step": 2305
+    },
+    {
+      "epoch": 2.152842497670084,
+      "grad_norm": 0.2515323075058934,
+      "learning_rate": 1.5688643424231964e-05,
+      "loss": 0.2784,
+      "step": 2310
+    },
+    {
+      "epoch": 2.157502329916123,
+      "grad_norm": 0.2400499782400618,
+      "learning_rate": 1.5602347255781842e-05,
+      "loss": 0.2793,
+      "step": 2315
+    },
+    {
+      "epoch": 2.1621621621621623,
+      "grad_norm": 0.22998127399543042,
+      "learning_rate": 1.5516051087331724e-05,
+      "loss": 0.2716,
+      "step": 2320
+    },
+    {
+      "epoch": 2.1668219944082012,
+      "grad_norm": 0.25446021197505714,
+      "learning_rate": 1.5429754918881602e-05,
+      "loss": 0.2829,
+      "step": 2325
+    },
+    {
+      "epoch": 2.1714818266542406,
+      "grad_norm": 0.24146900518073713,
+      "learning_rate": 1.534345875043148e-05,
+      "loss": 0.2761,
+      "step": 2330
+    },
+    {
+      "epoch": 2.1761416589002796,
+      "grad_norm": 0.21471437729750617,
+      "learning_rate": 1.525716258198136e-05,
+      "loss": 0.2748,
+      "step": 2335
+    },
+    {
+      "epoch": 2.180801491146319,
+      "grad_norm": 0.274661106980947,
+      "learning_rate": 1.517086641353124e-05,
+      "loss": 0.2749,
+      "step": 2340
+    },
+    {
+      "epoch": 2.185461323392358,
+      "grad_norm": 0.23572577764012273,
+      "learning_rate": 1.508457024508112e-05,
+      "loss": 0.2868,
+      "step": 2345
+    },
+    {
+      "epoch": 2.190121155638397,
+      "grad_norm": 0.23986113910915802,
+      "learning_rate": 1.4998274076630997e-05,
+      "loss": 0.2672,
+      "step": 2350
+    },
+    {
+      "epoch": 2.194780987884436,
+      "grad_norm": 0.22387516179145728,
+      "learning_rate": 1.4911977908180877e-05,
+      "loss": 0.2834,
+      "step": 2355
+    },
+    {
+      "epoch": 2.199440820130475,
+      "grad_norm": 0.22489661072774603,
+      "learning_rate": 1.4825681739730757e-05,
+      "loss": 0.2819,
+      "step": 2360
+    },
+    {
+      "epoch": 2.2041006523765145,
+      "grad_norm": 0.23676616542943682,
+      "learning_rate": 1.4739385571280637e-05,
+      "loss": 0.2735,
+      "step": 2365
+    },
+    {
+      "epoch": 2.2087604846225535,
+      "grad_norm": 0.22185340303804596,
+      "learning_rate": 1.4653089402830513e-05,
+      "loss": 0.2667,
+      "step": 2370
+    },
+    {
+      "epoch": 2.213420316868593,
+      "grad_norm": 0.22897041963602588,
+      "learning_rate": 1.4566793234380393e-05,
+      "loss": 0.2746,
+      "step": 2375
+    },
+    {
+      "epoch": 2.218080149114632,
+      "grad_norm": 0.20996348498449105,
+      "learning_rate": 1.4480497065930273e-05,
+      "loss": 0.2811,
+      "step": 2380
+    },
+    {
+      "epoch": 2.222739981360671,
+      "grad_norm": 0.21769046750185295,
+      "learning_rate": 1.4394200897480153e-05,
+      "loss": 0.2693,
+      "step": 2385
+    },
+    {
+      "epoch": 2.22739981360671,
+      "grad_norm": 0.22624807640590616,
+      "learning_rate": 1.4307904729030031e-05,
+      "loss": 0.2771,
+      "step": 2390
+    },
+    {
+      "epoch": 2.2320596458527495,
+      "grad_norm": 0.2671892007641654,
+      "learning_rate": 1.4221608560579911e-05,
+      "loss": 0.2799,
+      "step": 2395
+    },
+    {
+      "epoch": 2.2367194780987885,
+      "grad_norm": 0.23279040425941586,
+      "learning_rate": 1.4135312392129791e-05,
+      "loss": 0.276,
+      "step": 2400
+    },
+    {
+      "epoch": 2.2413793103448274,
+      "grad_norm": 0.20911596275639371,
+      "learning_rate": 1.404901622367967e-05,
+      "loss": 0.2643,
+      "step": 2405
+    },
+    {
+      "epoch": 2.246039142590867,
+      "grad_norm": 0.28113800038856673,
+      "learning_rate": 1.3962720055229547e-05,
+      "loss": 0.2836,
+      "step": 2410
+    },
+    {
+      "epoch": 2.2506989748369057,
+      "grad_norm": 0.24172264177329242,
+      "learning_rate": 1.3876423886779427e-05,
+      "loss": 0.2788,
+      "step": 2415
+    },
+    {
+      "epoch": 2.255358807082945,
+      "grad_norm": 0.21385121081517755,
+      "learning_rate": 1.3790127718329307e-05,
+      "loss": 0.2768,
+      "step": 2420
+    },
+    {
+      "epoch": 2.260018639328984,
+      "grad_norm": 0.24514455785444444,
+      "learning_rate": 1.3703831549879187e-05,
+      "loss": 0.2795,
+      "step": 2425
+    },
+    {
+      "epoch": 2.2646784715750234,
+      "grad_norm": 0.2493272686035274,
+      "learning_rate": 1.3617535381429064e-05,
+      "loss": 0.275,
+      "step": 2430
+    },
+    {
+      "epoch": 2.2693383038210624,
+      "grad_norm": 0.2229154462153395,
+      "learning_rate": 1.3531239212978944e-05,
+      "loss": 0.2764,
+      "step": 2435
+    },
+    {
+      "epoch": 2.2739981360671018,
+      "grad_norm": 0.24328259700078486,
+      "learning_rate": 1.3444943044528824e-05,
+      "loss": 0.2751,
+      "step": 2440
+    },
+    {
+      "epoch": 2.2786579683131407,
+      "grad_norm": 0.22808506603096315,
+      "learning_rate": 1.3358646876078703e-05,
+      "loss": 0.2836,
+      "step": 2445
+    },
+    {
+      "epoch": 2.28331780055918,
+      "grad_norm": 0.2611178749618171,
+      "learning_rate": 1.327235070762858e-05,
+      "loss": 0.2748,
+      "step": 2450
+    },
+    {
+      "epoch": 2.287977632805219,
+      "grad_norm": 0.22865462655212188,
+      "learning_rate": 1.318605453917846e-05,
+      "loss": 0.2795,
+      "step": 2455
+    },
+    {
+      "epoch": 2.292637465051258,
+      "grad_norm": 0.25090666476157925,
+      "learning_rate": 1.309975837072834e-05,
+      "loss": 0.2787,
+      "step": 2460
+    },
+    {
+      "epoch": 2.2972972972972974,
+      "grad_norm": 0.24278475703063251,
+      "learning_rate": 1.301346220227822e-05,
+      "loss": 0.2745,
+      "step": 2465
+    },
+    {
+      "epoch": 2.3019571295433363,
+      "grad_norm": 0.26270405891658377,
+      "learning_rate": 1.2927166033828098e-05,
+      "loss": 0.2764,
+      "step": 2470
+    },
+    {
+      "epoch": 2.3066169617893757,
+      "grad_norm": 0.26238764461873265,
+      "learning_rate": 1.2840869865377978e-05,
+      "loss": 0.2789,
+      "step": 2475
+    },
+    {
+      "epoch": 2.3112767940354146,
+      "grad_norm": 0.2326308848932903,
+      "learning_rate": 1.2754573696927858e-05,
+      "loss": 0.2708,
+      "step": 2480
+    },
+    {
+      "epoch": 2.315936626281454,
+      "grad_norm": 0.2306547825900092,
+      "learning_rate": 1.2668277528477738e-05,
+      "loss": 0.2799,
+      "step": 2485
+    },
+    {
+      "epoch": 2.320596458527493,
+      "grad_norm": 0.23062836705342,
+      "learning_rate": 1.2581981360027614e-05,
+      "loss": 0.2733,
+      "step": 2490
+    },
+    {
+      "epoch": 2.325256290773532,
+      "grad_norm": 0.26803762016431265,
+      "learning_rate": 1.2495685191577494e-05,
+      "loss": 0.2873,
+      "step": 2495
+    },
+    {
+      "epoch": 2.3299161230195713,
+      "grad_norm": 0.2426184142905924,
+      "learning_rate": 1.2409389023127374e-05,
+      "loss": 0.2746,
+      "step": 2500
+    },
+    {
+      "epoch": 2.3345759552656107,
+      "grad_norm": 0.2416943056672211,
+      "learning_rate": 1.2323092854677252e-05,
+      "loss": 0.2756,
+      "step": 2505
+    },
+    {
+      "epoch": 2.3392357875116496,
+      "grad_norm": 0.22953759712278493,
+      "learning_rate": 1.2236796686227132e-05,
+      "loss": 0.2843,
+      "step": 2510
+    },
+    {
+      "epoch": 2.3438956197576886,
+      "grad_norm": 0.21385164880927332,
+      "learning_rate": 1.215050051777701e-05,
+      "loss": 0.2808,
+      "step": 2515
+    },
+    {
+      "epoch": 2.348555452003728,
+      "grad_norm": 0.20765216611168574,
+      "learning_rate": 1.206420434932689e-05,
+      "loss": 0.2806,
+      "step": 2520
+    },
+    {
+      "epoch": 2.353215284249767,
+      "grad_norm": 0.22612549075955793,
+      "learning_rate": 1.1977908180876769e-05,
+      "loss": 0.2796,
+      "step": 2525
+    },
+    {
+      "epoch": 2.3578751164958063,
+      "grad_norm": 0.24251455141533623,
+      "learning_rate": 1.1891612012426649e-05,
+      "loss": 0.2828,
+      "step": 2530
+    },
+    {
+      "epoch": 2.362534948741845,
+      "grad_norm": 0.2268119367796483,
+      "learning_rate": 1.1805315843976528e-05,
+      "loss": 0.282,
+      "step": 2535
+    },
+    {
+      "epoch": 2.3671947809878846,
+      "grad_norm": 0.2167177550320992,
+      "learning_rate": 1.1719019675526408e-05,
+      "loss": 0.2684,
+      "step": 2540
+    },
+    {
+      "epoch": 2.3718546132339235,
+      "grad_norm": 0.21378424339904822,
+      "learning_rate": 1.1632723507076287e-05,
+      "loss": 0.2816,
+      "step": 2545
+    },
+    {
+      "epoch": 2.3765144454799625,
+      "grad_norm": 0.27089608172872354,
+      "learning_rate": 1.1546427338626167e-05,
+      "loss": 0.2716,
+      "step": 2550
+    },
+    {
+      "epoch": 2.381174277726002,
+      "grad_norm": 0.2216261000001064,
+      "learning_rate": 1.1460131170176045e-05,
+      "loss": 0.2713,
+      "step": 2555
+    },
+    {
+      "epoch": 2.385834109972041,
+      "grad_norm": 0.22470727876994512,
+      "learning_rate": 1.1373835001725925e-05,
+      "loss": 0.275,
+      "step": 2560
+    },
+    {
+      "epoch": 2.39049394221808,
+      "grad_norm": 0.2396455051127378,
+      "learning_rate": 1.1287538833275803e-05,
+      "loss": 0.2874,
+      "step": 2565
+    },
+    {
+      "epoch": 2.395153774464119,
+      "grad_norm": 0.2450640166805578,
+      "learning_rate": 1.1201242664825683e-05,
+      "loss": 0.2801,
+      "step": 2570
+    },
+    {
+      "epoch": 2.3998136067101585,
+      "grad_norm": 0.20426581346179554,
+      "learning_rate": 1.1114946496375561e-05,
+      "loss": 0.2743,
+      "step": 2575
+    },
+    {
+      "epoch": 2.4044734389561975,
+      "grad_norm": 0.19422133756032336,
+      "learning_rate": 1.1028650327925441e-05,
+      "loss": 0.2794,
+      "step": 2580
+    },
+    {
+      "epoch": 2.409133271202237,
+      "grad_norm": 0.2213276049565462,
+      "learning_rate": 1.094235415947532e-05,
+      "loss": 0.2913,
+      "step": 2585
+    },
+    {
+      "epoch": 2.413793103448276,
+      "grad_norm": 0.23189143153792158,
+      "learning_rate": 1.0856057991025199e-05,
+      "loss": 0.2723,
+      "step": 2590
+    },
+    {
+      "epoch": 2.418452935694315,
+      "grad_norm": 0.2187547564767661,
+      "learning_rate": 1.0769761822575077e-05,
+      "loss": 0.2779,
+      "step": 2595
+    },
+    {
+      "epoch": 2.423112767940354,
+      "grad_norm": 0.22765146036127656,
+      "learning_rate": 1.0683465654124957e-05,
+      "loss": 0.2764,
+      "step": 2600
+    },
+    {
+      "epoch": 2.427772600186393,
+      "grad_norm": 0.20506008234988998,
+      "learning_rate": 1.0597169485674835e-05,
+      "loss": 0.2809,
+      "step": 2605
+    },
+    {
+      "epoch": 2.4324324324324325,
+      "grad_norm": 0.22311257370815907,
+      "learning_rate": 1.0510873317224715e-05,
+      "loss": 0.2756,
+      "step": 2610
+    },
+    {
+      "epoch": 2.4370922646784714,
+      "grad_norm": 0.23071608639070978,
+      "learning_rate": 1.0424577148774595e-05,
+      "loss": 0.2726,
+      "step": 2615
+    },
+    {
+      "epoch": 2.441752096924511,
+      "grad_norm": 0.2554285691099882,
+      "learning_rate": 1.0338280980324475e-05,
+      "loss": 0.2778,
+      "step": 2620
+    },
+    {
+      "epoch": 2.4464119291705497,
+      "grad_norm": 0.2724645083211829,
+      "learning_rate": 1.0251984811874353e-05,
+      "loss": 0.2745,
+      "step": 2625
+    },
+    {
+      "epoch": 2.451071761416589,
+      "grad_norm": 0.19701756362723155,
+      "learning_rate": 1.0165688643424233e-05,
+      "loss": 0.272,
+      "step": 2630
+    },
+    {
+      "epoch": 2.455731593662628,
+      "grad_norm": 0.23117211812989907,
+      "learning_rate": 1.0079392474974112e-05,
+      "loss": 0.2705,
+      "step": 2635
+    },
+    {
+      "epoch": 2.4603914259086674,
+      "grad_norm": 0.20606293533507136,
+      "learning_rate": 9.993096306523992e-06,
+      "loss": 0.2724,
+      "step": 2640
+    },
+    {
+      "epoch": 2.4650512581547064,
+      "grad_norm": 0.20172978733840546,
+      "learning_rate": 9.90680013807387e-06,
+      "loss": 0.2674,
+      "step": 2645
+    },
+    {
+      "epoch": 2.4697110904007458,
+      "grad_norm": 0.24421681228009895,
+      "learning_rate": 9.82050396962375e-06,
+      "loss": 0.2816,
+      "step": 2650
+    },
+    {
+      "epoch": 2.4743709226467847,
+      "grad_norm": 0.22242379953820635,
+      "learning_rate": 9.734207801173628e-06,
+      "loss": 0.2764,
+      "step": 2655
+    },
+    {
+      "epoch": 2.4790307548928237,
+      "grad_norm": 0.2039986747959715,
+      "learning_rate": 9.647911632723508e-06,
+      "loss": 0.2715,
+      "step": 2660
+    },
+    {
+      "epoch": 2.483690587138863,
+      "grad_norm": 0.21712699509899772,
+      "learning_rate": 9.561615464273386e-06,
+      "loss": 0.2774,
+      "step": 2665
+    },
+    {
+      "epoch": 2.488350419384902,
+      "grad_norm": 0.22270653929899237,
+      "learning_rate": 9.475319295823266e-06,
+      "loss": 0.277,
+      "step": 2670
+    },
+    {
+      "epoch": 2.4930102516309414,
+      "grad_norm": 0.2186039907677849,
+      "learning_rate": 9.389023127373144e-06,
+      "loss": 0.2814,
+      "step": 2675
+    },
+    {
+      "epoch": 2.4976700838769803,
+      "grad_norm": 0.22143379415501022,
+      "learning_rate": 9.302726958923024e-06,
+      "loss": 0.2773,
+      "step": 2680
+    },
+    {
+      "epoch": 2.5023299161230197,
+      "grad_norm": 0.21468489845047237,
+      "learning_rate": 9.216430790472904e-06,
+      "loss": 0.2747,
+      "step": 2685
+    },
+    {
+      "epoch": 2.5069897483690586,
+      "grad_norm": 0.19241778534453122,
+      "learning_rate": 9.130134622022784e-06,
+      "loss": 0.2856,
+      "step": 2690
+    },
+    {
+      "epoch": 2.511649580615098,
+      "grad_norm": 0.20393496267357739,
+      "learning_rate": 9.043838453572662e-06,
+      "loss": 0.2744,
+      "step": 2695
+    },
+    {
+      "epoch": 2.516309412861137,
+      "grad_norm": 0.20389587633359454,
+      "learning_rate": 8.957542285122542e-06,
+      "loss": 0.2725,
+      "step": 2700
+    },
+    {
+      "epoch": 2.5209692451071763,
+      "grad_norm": 0.2298880584116301,
+      "learning_rate": 8.87124611667242e-06,
+      "loss": 0.2759,
+      "step": 2705
+    },
+    {
+      "epoch": 2.5256290773532153,
+      "grad_norm": 0.21973095650040791,
+      "learning_rate": 8.7849499482223e-06,
+      "loss": 0.2782,
+      "step": 2710
+    },
+    {
+      "epoch": 2.5302889095992542,
+      "grad_norm": 0.2026001161863883,
+      "learning_rate": 8.698653779772179e-06,
+      "loss": 0.2749,
+      "step": 2715
+    },
+    {
+      "epoch": 2.5349487418452936,
+      "grad_norm": 0.21961815538711066,
+      "learning_rate": 8.612357611322058e-06,
+      "loss": 0.2829,
+      "step": 2720
+    },
+    {
+      "epoch": 2.5396085740913326,
+      "grad_norm": 0.22671573479027304,
+      "learning_rate": 8.526061442871937e-06,
+      "loss": 0.2662,
+      "step": 2725
+    },
+    {
+      "epoch": 2.544268406337372,
+      "grad_norm": 0.20534062064048056,
+      "learning_rate": 8.439765274421817e-06,
+      "loss": 0.2735,
+      "step": 2730
+    },
+    {
+      "epoch": 2.548928238583411,
+      "grad_norm": 0.2210584735777573,
+      "learning_rate": 8.353469105971695e-06,
+      "loss": 0.2754,
+      "step": 2735
+    },
+    {
+      "epoch": 2.5535880708294503,
+      "grad_norm": 0.24283622848184136,
+      "learning_rate": 8.267172937521575e-06,
+      "loss": 0.2712,
+      "step": 2740
+    },
+    {
+      "epoch": 2.558247903075489,
+      "grad_norm": 0.22087685642352617,
+      "learning_rate": 8.180876769071453e-06,
+      "loss": 0.275,
+      "step": 2745
+    },
+    {
+      "epoch": 2.562907735321528,
+      "grad_norm": 0.20106499557719176,
+      "learning_rate": 8.094580600621333e-06,
+      "loss": 0.2751,
+      "step": 2750
+    },
+    {
+      "epoch": 2.5675675675675675,
+      "grad_norm": 0.2388933488779767,
+      "learning_rate": 8.008284432171211e-06,
+      "loss": 0.2759,
+      "step": 2755
+    },
+    {
+      "epoch": 2.572227399813607,
+      "grad_norm": 0.25158472015358696,
+      "learning_rate": 7.921988263721091e-06,
+      "loss": 0.2808,
+      "step": 2760
+    },
+    {
+      "epoch": 2.576887232059646,
+      "grad_norm": 0.23558810936058092,
+      "learning_rate": 7.835692095270971e-06,
+      "loss": 0.2819,
+      "step": 2765
+    },
+    {
+      "epoch": 2.581547064305685,
+      "grad_norm": 0.19968560251718045,
+      "learning_rate": 7.749395926820851e-06,
+      "loss": 0.2692,
+      "step": 2770
+    },
+    {
+      "epoch": 2.586206896551724,
+      "grad_norm": 0.22080088082989097,
+      "learning_rate": 7.663099758370729e-06,
+      "loss": 0.2731,
+      "step": 2775
+    },
+    {
+      "epoch": 2.590866728797763,
+      "grad_norm": 0.22343341214048187,
+      "learning_rate": 7.576803589920608e-06,
+      "loss": 0.2816,
+      "step": 2780
+    },
+    {
+      "epoch": 2.5955265610438025,
+      "grad_norm": 0.2007254819071555,
+      "learning_rate": 7.490507421470487e-06,
+      "loss": 0.2751,
+      "step": 2785
+    },
+    {
+      "epoch": 2.6001863932898415,
+      "grad_norm": 0.19397542938514054,
+      "learning_rate": 7.4042112530203655e-06,
+      "loss": 0.2804,
+      "step": 2790
+    },
+    {
+      "epoch": 2.604846225535881,
+      "grad_norm": 0.21683577031938794,
+      "learning_rate": 7.317915084570245e-06,
+      "loss": 0.2808,
+      "step": 2795
+    },
+    {
+      "epoch": 2.60950605778192,
+      "grad_norm": 0.20491594550414058,
+      "learning_rate": 7.231618916120124e-06,
+      "loss": 0.2723,
+      "step": 2800
+    },
+    {
+      "epoch": 2.6141658900279587,
+      "grad_norm": 0.1955876556158969,
+      "learning_rate": 7.1453227476700035e-06,
+      "loss": 0.2811,
+      "step": 2805
+    },
+    {
+      "epoch": 2.618825722273998,
+      "grad_norm": 0.21753242331383046,
+      "learning_rate": 7.059026579219883e-06,
+      "loss": 0.2729,
+      "step": 2810
+    },
+    {
+      "epoch": 2.6234855545200375,
+      "grad_norm": 0.20145126986430875,
+      "learning_rate": 6.9727304107697625e-06,
+      "loss": 0.2768,
+      "step": 2815
+    },
+    {
+      "epoch": 2.6281453867660765,
+      "grad_norm": 0.22414952828658224,
+      "learning_rate": 6.886434242319641e-06,
+      "loss": 0.2844,
+      "step": 2820
+    },
+    {
+      "epoch": 2.6328052190121154,
+      "grad_norm": 0.22444611978602524,
+      "learning_rate": 6.800138073869521e-06,
+      "loss": 0.2814,
+      "step": 2825
+    },
+    {
+      "epoch": 2.637465051258155,
+      "grad_norm": 0.2067453059387536,
+      "learning_rate": 6.713841905419399e-06,
+      "loss": 0.276,
+      "step": 2830
+    },
+    {
+      "epoch": 2.6421248835041937,
+      "grad_norm": 0.23782780990583297,
+      "learning_rate": 6.627545736969279e-06,
+      "loss": 0.281,
+      "step": 2835
+    },
+    {
+      "epoch": 2.646784715750233,
+      "grad_norm": 0.1977330966266734,
+      "learning_rate": 6.541249568519157e-06,
+      "loss": 0.2764,
+      "step": 2840
+    },
+    {
+      "epoch": 2.651444547996272,
+      "grad_norm": 0.20068601677851078,
+      "learning_rate": 6.454953400069037e-06,
+      "loss": 0.2718,
+      "step": 2845
+    },
+    {
+      "epoch": 2.6561043802423114,
+      "grad_norm": 0.21584080255074678,
+      "learning_rate": 6.368657231618916e-06,
+      "loss": 0.2808,
+      "step": 2850
+    },
+    {
+      "epoch": 2.6607642124883504,
+      "grad_norm": 0.21902789632328576,
+      "learning_rate": 6.282361063168796e-06,
+      "loss": 0.2749,
+      "step": 2855
+    },
+    {
+      "epoch": 2.6654240447343893,
+      "grad_norm": 0.2203236216621516,
+      "learning_rate": 6.196064894718675e-06,
+      "loss": 0.2828,
+      "step": 2860
+    },
+    {
+      "epoch": 2.6700838769804287,
+      "grad_norm": 0.20819846449081367,
+      "learning_rate": 6.109768726268554e-06,
+      "loss": 0.2764,
+      "step": 2865
+    },
+    {
+      "epoch": 2.674743709226468,
+      "grad_norm": 0.20104850465228524,
+      "learning_rate": 6.023472557818433e-06,
+      "loss": 0.2754,
+      "step": 2870
+    },
+    {
+      "epoch": 2.679403541472507,
+      "grad_norm": 0.22211394517376387,
+      "learning_rate": 5.937176389368312e-06,
+      "loss": 0.2863,
+      "step": 2875
+    },
+    {
+      "epoch": 2.684063373718546,
+      "grad_norm": 0.1933728152830278,
+      "learning_rate": 5.850880220918191e-06,
+      "loss": 0.2814,
+      "step": 2880
+    },
+    {
+      "epoch": 2.6887232059645854,
+      "grad_norm": 0.21703291309436695,
+      "learning_rate": 5.76458405246807e-06,
+      "loss": 0.2749,
+      "step": 2885
+    },
+    {
+      "epoch": 2.6933830382106243,
+      "grad_norm": 0.21200419782946417,
+      "learning_rate": 5.67828788401795e-06,
+      "loss": 0.2801,
+      "step": 2890
+    },
+    {
+      "epoch": 2.6980428704566637,
+      "grad_norm": 0.21571072517568418,
+      "learning_rate": 5.591991715567829e-06,
+      "loss": 0.2762,
+      "step": 2895
+    },
+    {
+      "epoch": 2.7027027027027026,
+      "grad_norm": 0.2354145464181523,
+      "learning_rate": 5.5056955471177085e-06,
+      "loss": 0.2715,
+      "step": 2900
+    },
+    {
+      "epoch": 2.707362534948742,
+      "grad_norm": 0.23165631413429835,
+      "learning_rate": 5.4193993786675876e-06,
+      "loss": 0.2752,
+      "step": 2905
+    },
+    {
+      "epoch": 2.712022367194781,
+      "grad_norm": 0.21226409476551816,
+      "learning_rate": 5.333103210217467e-06,
+      "loss": 0.2826,
+      "step": 2910
+    },
+    {
+      "epoch": 2.71668219944082,
+      "grad_norm": 0.21425923226801386,
+      "learning_rate": 5.246807041767346e-06,
+      "loss": 0.2723,
+      "step": 2915
+    },
+    {
+      "epoch": 2.7213420316868593,
+      "grad_norm": 0.19799150060948206,
+      "learning_rate": 5.160510873317225e-06,
+      "loss": 0.2709,
+      "step": 2920
+    },
+    {
+      "epoch": 2.7260018639328987,
+      "grad_norm": 0.2020341569214513,
+      "learning_rate": 5.074214704867105e-06,
+      "loss": 0.2734,
+      "step": 2925
+    },
+    {
+      "epoch": 2.7306616961789376,
+      "grad_norm": 0.21731764039831938,
+      "learning_rate": 4.987918536416984e-06,
+      "loss": 0.2831,
+      "step": 2930
+    },
+    {
+      "epoch": 2.7353215284249766,
+      "grad_norm": 0.1929509014617383,
+      "learning_rate": 4.901622367966863e-06,
+      "loss": 0.2736,
+      "step": 2935
+    },
+    {
+      "epoch": 2.739981360671016,
+      "grad_norm": 0.21804886386289127,
+      "learning_rate": 4.815326199516742e-06,
+      "loss": 0.277,
+      "step": 2940
+    },
+    {
+      "epoch": 2.744641192917055,
+      "grad_norm": 0.21374173524781143,
+      "learning_rate": 4.729030031066621e-06,
+      "loss": 0.2763,
+      "step": 2945
+    },
+    {
+      "epoch": 2.7493010251630943,
+      "grad_norm": 0.2125706232740658,
+      "learning_rate": 4.6427338626165e-06,
+      "loss": 0.2801,
+      "step": 2950
+    },
+    {
+      "epoch": 2.753960857409133,
+      "grad_norm": 0.21514857161669884,
+      "learning_rate": 4.556437694166379e-06,
+      "loss": 0.2699,
+      "step": 2955
+    },
+    {
+      "epoch": 2.7586206896551726,
+      "grad_norm": 0.19358252363225883,
+      "learning_rate": 4.470141525716258e-06,
+      "loss": 0.2722,
+      "step": 2960
+    },
+    {
+      "epoch": 2.7632805219012115,
+      "grad_norm": 0.19587310589362347,
+      "learning_rate": 4.383845357266138e-06,
+      "loss": 0.277,
+      "step": 2965
+    },
+    {
+      "epoch": 2.7679403541472505,
+      "grad_norm": 0.21357209274379946,
+      "learning_rate": 4.297549188816017e-06,
+      "loss": 0.2775,
+      "step": 2970
+    },
+    {
+      "epoch": 2.77260018639329,
+      "grad_norm": 0.1964853253114847,
+      "learning_rate": 4.211253020365896e-06,
+      "loss": 0.2756,
+      "step": 2975
+    },
+    {
+      "epoch": 2.7772600186393293,
+      "grad_norm": 0.2052227692129964,
+      "learning_rate": 4.124956851915775e-06,
+      "loss": 0.2735,
+      "step": 2980
+    },
+    {
+      "epoch": 2.781919850885368,
+      "grad_norm": 0.19801125225690663,
+      "learning_rate": 4.0386606834656544e-06,
+      "loss": 0.2807,
+      "step": 2985
+    },
+    {
+      "epoch": 2.786579683131407,
+      "grad_norm": 0.19564352690896755,
+      "learning_rate": 3.9523645150155335e-06,
+      "loss": 0.2757,
+      "step": 2990
+    },
+    {
+      "epoch": 2.7912395153774465,
+      "grad_norm": 0.20301730710443705,
+      "learning_rate": 3.8660683465654126e-06,
+      "loss": 0.2713,
+      "step": 2995
+    },
+    {
+      "epoch": 2.7958993476234855,
+      "grad_norm": 0.1917733674204406,
+      "learning_rate": 3.779772178115292e-06,
+      "loss": 0.2723,
+      "step": 3000
+    },
+    {
+      "epoch": 2.800559179869525,
+      "grad_norm": 0.20584113686312208,
+      "learning_rate": 3.693476009665171e-06,
+      "loss": 0.2827,
+      "step": 3005
+    },
+    {
+      "epoch": 2.805219012115564,
+      "grad_norm": 0.20069010472200335,
+      "learning_rate": 3.6071798412150506e-06,
+      "loss": 0.2859,
+      "step": 3010
+    },
+    {
+      "epoch": 2.809878844361603,
+      "grad_norm": 0.2155639428033643,
+      "learning_rate": 3.5208836727649297e-06,
+      "loss": 0.2726,
+      "step": 3015
+    },
+    {
+      "epoch": 2.814538676607642,
+      "grad_norm": 0.2063711645297585,
+      "learning_rate": 3.434587504314809e-06,
+      "loss": 0.2735,
+      "step": 3020
+    },
+    {
+      "epoch": 2.819198508853681,
+      "grad_norm": 0.19346922401332795,
+      "learning_rate": 3.348291335864688e-06,
+      "loss": 0.2816,
+      "step": 3025
+    },
+    {
+      "epoch": 2.8238583410997204,
+      "grad_norm": 0.20019645072426784,
+      "learning_rate": 3.2619951674145674e-06,
+      "loss": 0.2892,
+      "step": 3030
+    },
+    {
+      "epoch": 2.8285181733457594,
+      "grad_norm": 0.2002316070918464,
+      "learning_rate": 3.1756989989644464e-06,
+      "loss": 0.2797,
+      "step": 3035
+    },
+    {
+      "epoch": 2.8331780055917988,
+      "grad_norm": 0.1940197752949764,
+      "learning_rate": 3.089402830514325e-06,
+      "loss": 0.2751,
+      "step": 3040
+    },
+    {
+      "epoch": 2.8378378378378377,
+      "grad_norm": 0.19816493543307628,
+      "learning_rate": 3.0031066620642046e-06,
+      "loss": 0.2747,
+      "step": 3045
+    },
+    {
+      "epoch": 2.842497670083877,
+      "grad_norm": 0.2052909545356945,
+      "learning_rate": 2.9168104936140837e-06,
+      "loss": 0.2777,
+      "step": 3050
+    },
+    {
+      "epoch": 2.847157502329916,
+      "grad_norm": 0.20260105000590387,
+      "learning_rate": 2.8305143251639627e-06,
+      "loss": 0.2741,
+      "step": 3055
+    },
+    {
+      "epoch": 2.8518173345759554,
+      "grad_norm": 0.1928915769100951,
+      "learning_rate": 2.7442181567138422e-06,
+      "loss": 0.2721,
+      "step": 3060
+    },
+    {
+      "epoch": 2.8564771668219944,
+      "grad_norm": 0.181512581519191,
+      "learning_rate": 2.6579219882637213e-06,
+      "loss": 0.2721,
+      "step": 3065
+    },
+    {
+      "epoch": 2.8611369990680338,
+      "grad_norm": 0.19194747429243134,
+      "learning_rate": 2.5716258198136004e-06,
+      "loss": 0.2719,
+      "step": 3070
+    },
+    {
+      "epoch": 2.8657968313140727,
+      "grad_norm": 0.19008254008126735,
+      "learning_rate": 2.4853296513634795e-06,
+      "loss": 0.2754,
+      "step": 3075
+    },
+    {
+      "epoch": 2.8704566635601116,
+      "grad_norm": 0.20105614451353776,
+      "learning_rate": 2.399033482913359e-06,
+      "loss": 0.2813,
+      "step": 3080
+    },
+    {
+      "epoch": 2.875116495806151,
+      "grad_norm": 0.1881083592669046,
+      "learning_rate": 2.312737314463238e-06,
+      "loss": 0.271,
+      "step": 3085
+    },
+    {
+      "epoch": 2.87977632805219,
+      "grad_norm": 0.19947871205314419,
+      "learning_rate": 2.226441146013117e-06,
+      "loss": 0.271,
+      "step": 3090
+    },
+    {
+      "epoch": 2.8844361602982294,
+      "grad_norm": 0.17963512329176656,
+      "learning_rate": 2.140144977562996e-06,
+      "loss": 0.2746,
+      "step": 3095
+    },
+    {
+      "epoch": 2.8890959925442683,
+      "grad_norm": 0.1967806751007597,
+      "learning_rate": 2.0538488091128757e-06,
+      "loss": 0.2721,
+      "step": 3100
+    },
+    {
+      "epoch": 2.8937558247903077,
+      "grad_norm": 0.18863156329787908,
+      "learning_rate": 1.9675526406627547e-06,
+      "loss": 0.2691,
+      "step": 3105
+    },
+    {
+      "epoch": 2.8984156570363466,
+      "grad_norm": 0.196717106793742,
+      "learning_rate": 1.8812564722126338e-06,
+      "loss": 0.2803,
+      "step": 3110
+    },
+    {
+      "epoch": 2.9030754892823856,
+      "grad_norm": 0.18397307772060773,
+      "learning_rate": 1.794960303762513e-06,
+      "loss": 0.2674,
+      "step": 3115
+    },
+    {
+      "epoch": 2.907735321528425,
+      "grad_norm": 0.19235790038860878,
+      "learning_rate": 1.7086641353123924e-06,
+      "loss": 0.278,
+      "step": 3120
+    },
+    {
+      "epoch": 2.9123951537744643,
+      "grad_norm": 0.18631332529515698,
+      "learning_rate": 1.6223679668622715e-06,
+      "loss": 0.27,
+      "step": 3125
+    },
+    {
+      "epoch": 2.9170549860205033,
+      "grad_norm": 0.20134923103840421,
+      "learning_rate": 1.5360717984121505e-06,
+      "loss": 0.2692,
+      "step": 3130
+    },
+    {
+      "epoch": 2.9217148182665422,
+      "grad_norm": 0.19154288558365484,
+      "learning_rate": 1.4497756299620296e-06,
+      "loss": 0.2728,
+      "step": 3135
+    },
+    {
+      "epoch": 2.9263746505125816,
+      "grad_norm": 0.2023631652179777,
+      "learning_rate": 1.363479461511909e-06,
+      "loss": 0.2698,
+      "step": 3140
+    },
+    {
+      "epoch": 2.9310344827586206,
+      "grad_norm": 0.179577664887894,
+      "learning_rate": 1.277183293061788e-06,
+      "loss": 0.2734,
+      "step": 3145
+    },
+    {
+      "epoch": 2.93569431500466,
+      "grad_norm": 0.18851689622671344,
+      "learning_rate": 1.1908871246116673e-06,
+      "loss": 0.2755,
+      "step": 3150
+    },
+    {
+      "epoch": 2.940354147250699,
+      "grad_norm": 0.18643521633985904,
+      "learning_rate": 1.1045909561615463e-06,
+      "loss": 0.2749,
+      "step": 3155
+    },
+    {
+      "epoch": 2.9450139794967383,
+      "grad_norm": 0.19111738910054296,
+      "learning_rate": 1.0182947877114256e-06,
+      "loss": 0.277,
+      "step": 3160
+    },
+    {
+      "epoch": 2.949673811742777,
+      "grad_norm": 0.18254811169542898,
+      "learning_rate": 9.319986192613048e-07,
+      "loss": 0.2701,
+      "step": 3165
+    },
+    {
+      "epoch": 2.954333643988816,
+      "grad_norm": 0.19479015937246086,
+      "learning_rate": 8.45702450811184e-07,
+      "loss": 0.2729,
+      "step": 3170
+    },
+    {
+      "epoch": 2.9589934762348555,
+      "grad_norm": 0.18494378963619557,
+      "learning_rate": 7.594062823610632e-07,
+      "loss": 0.2762,
+      "step": 3175
+    },
+    {
+      "epoch": 2.963653308480895,
+      "grad_norm": 0.19622969519521213,
+      "learning_rate": 6.731101139109423e-07,
+      "loss": 0.2795,
+      "step": 3180
+    },
+    {
+      "epoch": 2.968313140726934,
+      "grad_norm": 0.18692020486752917,
+      "learning_rate": 5.868139454608215e-07,
+      "loss": 0.2737,
+      "step": 3185
+    },
+    {
+      "epoch": 2.972972972972973,
+      "grad_norm": 0.17786403177932492,
+      "learning_rate": 5.005177770107007e-07,
+      "loss": 0.2731,
+      "step": 3190
+    },
+    {
+      "epoch": 2.977632805219012,
+      "grad_norm": 0.18946790797590035,
+      "learning_rate": 4.142216085605799e-07,
+      "loss": 0.2732,
+      "step": 3195
+    },
+    {
+      "epoch": 2.982292637465051,
+      "grad_norm": 0.18523285794263772,
+      "learning_rate": 3.279254401104591e-07,
+      "loss": 0.2805,
+      "step": 3200
+    },
+    {
+      "epoch": 2.9869524697110905,
+      "grad_norm": 0.19300091016125856,
+      "learning_rate": 2.416292716603383e-07,
+      "loss": 0.2744,
+      "step": 3205
+    },
+    {
+      "epoch": 2.9916123019571295,
+      "grad_norm": 0.1940562662328004,
+      "learning_rate": 1.5533310321021747e-07,
+      "loss": 0.2779,
+      "step": 3210
+    },
+    {
+      "epoch": 2.996272134203169,
+      "grad_norm": 0.18987250111549814,
+      "learning_rate": 6.903693476009665e-08,
+      "loss": 0.2761,
+      "step": 3215
+    },
+    {
+      "epoch": 3.0,
+      "step": 3219,
+      "total_flos": 2.7558554468194714e+18,
+      "train_loss": 0.09220795636593314,
+      "train_runtime": 28393.5776,
+      "train_samples_per_second": 1.814,
+      "train_steps_per_second": 0.113
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3219,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.7558554468194714e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}