diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,4867 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 3450,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0,
+      "grad_norm": 29.54340489747315,
+      "learning_rate": 2.8571428571428575e-07,
+      "loss": 1.2261,
+      "step": 1
+    },
+    {
+      "epoch": 0.0,
+      "grad_norm": 22.808761745462473,
+      "learning_rate": 1.4285714285714286e-06,
+      "loss": 1.2154,
+      "step": 5
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 9.8799538809059,
+      "learning_rate": 2.8571428571428573e-06,
+      "loss": 0.9962,
+      "step": 10
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 4.989439314642267,
+      "learning_rate": 4.2857142857142855e-06,
+      "loss": 0.919,
+      "step": 15
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 4.393370598144475,
+      "learning_rate": 5.7142857142857145e-06,
+      "loss": 0.8978,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 3.592829916520255,
+      "learning_rate": 7.1428571428571436e-06,
+      "loss": 0.8589,
+      "step": 25
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.2538364586400332,
+      "learning_rate": 8.571428571428571e-06,
+      "loss": 0.8447,
+      "step": 30
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 2.9999553359801014,
+      "learning_rate": 1e-05,
+      "loss": 0.8595,
+      "step": 35
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 3.262257995183099,
+      "learning_rate": 9.99994710707541e-06,
+      "loss": 0.8109,
+      "step": 40
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.891385395889321,
+      "learning_rate": 9.999788429420697e-06,
+      "loss": 0.8061,
+      "step": 45
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 3.0405298806513694,
+      "learning_rate": 9.999523970393038e-06,
+      "loss": 0.8585,
+      "step": 50
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 2.966926069119011,
+      "learning_rate": 9.999153735587632e-06,
+      "loss": 0.8167,
+      "step": 55
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 3.023657776322375,
+      "learning_rate": 9.998677732837604e-06,
+      "loss": 0.8082,
+      "step": 60
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.8642622008411758,
+      "learning_rate": 9.99809597221382e-06,
+      "loss": 0.7945,
+      "step": 65
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 2.755547941464214,
+      "learning_rate": 9.997408466024692e-06,
+      "loss": 0.7914,
+      "step": 70
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.5705353797466217,
+      "learning_rate": 9.996615228815906e-06,
+      "loss": 0.7822,
+      "step": 75
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.9040071983966396,
+      "learning_rate": 9.995716277370114e-06,
+      "loss": 0.7772,
+      "step": 80
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 2.8339684851394584,
+      "learning_rate": 9.994711630706585e-06,
+      "loss": 0.7888,
+      "step": 85
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.6456345105507477,
+      "learning_rate": 9.9936013100808e-06,
+      "loss": 0.7683,
+      "step": 90
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 2.8161218216336477,
+      "learning_rate": 9.992385338984e-06,
+      "loss": 0.7673,
+      "step": 95
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 3.3076899038616623,
+      "learning_rate": 9.991063743142693e-06,
+      "loss": 0.7758,
+      "step": 100
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 2.5788476713979374,
+      "learning_rate": 9.989636550518105e-06,
+      "loss": 0.7825,
+      "step": 105
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.725324467749148,
+      "learning_rate": 9.988103791305594e-06,
+      "loss": 0.7791,
+      "step": 110
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.699154929977825,
+      "learning_rate": 9.986465497934008e-06,
+      "loss": 0.7696,
+      "step": 115
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 2.6184415520253643,
+      "learning_rate": 9.984721705064994e-06,
+      "loss": 0.7772,
+      "step": 120
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.6932563939329257,
+      "learning_rate": 9.98287244959228e-06,
+      "loss": 0.7758,
+      "step": 125
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 2.558292453219913,
+      "learning_rate": 9.980917770640873e-06,
+      "loss": 0.7488,
+      "step": 130
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.8553332314156643,
+      "learning_rate": 9.97885770956625e-06,
+      "loss": 0.7602,
+      "step": 135
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 2.7454816786831597,
+      "learning_rate": 9.976692309953472e-06,
+      "loss": 0.7719,
+      "step": 140
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.6722049507067918,
+      "learning_rate": 9.974421617616267e-06,
+      "loss": 0.7811,
+      "step": 145
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.768310513464075,
+      "learning_rate": 9.97204568059606e-06,
+      "loss": 0.7793,
+      "step": 150
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 2.770637299540184,
+      "learning_rate": 9.969564549160952e-06,
+      "loss": 0.7469,
+      "step": 155
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.7539240146037227,
+      "learning_rate": 9.96697827580466e-06,
+      "loss": 0.7773,
+      "step": 160
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 2.7200265410438296,
+      "learning_rate": 9.964286915245414e-06,
+      "loss": 0.7615,
+      "step": 165
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.66070428599963,
+      "learning_rate": 9.961490524424781e-06,
+      "loss": 0.7626,
+      "step": 170
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 2.7255648035063764,
+      "learning_rate": 9.958589162506481e-06,
+      "loss": 0.7359,
+      "step": 175
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.786041603773328,
+      "learning_rate": 9.955582890875118e-06,
+      "loss": 0.7739,
+      "step": 180
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 2.5795832004050374,
+      "learning_rate": 9.952471773134893e-06,
+      "loss": 0.7696,
+      "step": 185
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.627315901008289,
+      "learning_rate": 9.949255875108252e-06,
+      "loss": 0.7755,
+      "step": 190
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.5724268911345116,
+      "learning_rate": 9.945935264834495e-06,
+      "loss": 0.754,
+      "step": 195
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 2.7623409960370666,
+      "learning_rate": 9.942510012568338e-06,
+      "loss": 0.7532,
+      "step": 200
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.6345064087955845,
+      "learning_rate": 9.938980190778426e-06,
+      "loss": 0.7516,
+      "step": 205
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 2.7160427754334364,
+      "learning_rate": 9.935345874145797e-06,
+      "loss": 0.768,
+      "step": 210
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.5653032187673137,
+      "learning_rate": 9.931607139562304e-06,
+      "loss": 0.7407,
+      "step": 215
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 2.616121275524974,
+      "learning_rate": 9.927764066128992e-06,
+      "loss": 0.7691,
+      "step": 220
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.570280959436263,
+      "learning_rate": 9.923816735154417e-06,
+      "loss": 0.7281,
+      "step": 225
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.472353677171501,
+      "learning_rate": 9.919765230152932e-06,
+      "loss": 0.7485,
+      "step": 230
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 2.539746453371004,
+      "learning_rate": 9.915609636842914e-06,
+      "loss": 0.7473,
+      "step": 235
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.5177181887377422,
+      "learning_rate": 9.911350043144958e-06,
+      "loss": 0.7426,
+      "step": 240
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 2.4720930131495407,
+      "learning_rate": 9.906986539180012e-06,
+      "loss": 0.7442,
+      "step": 245
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.5270618999743983,
+      "learning_rate": 9.90251921726747e-06,
+      "loss": 0.7534,
+      "step": 250
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 2.4644331518869698,
+      "learning_rate": 9.89794817192322e-06,
+      "loss": 0.7433,
+      "step": 255
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.7138902839835968,
+      "learning_rate": 9.893273499857642e-06,
+      "loss": 0.7367,
+      "step": 260
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.6044202431283145,
+      "learning_rate": 9.888495299973574e-06,
+      "loss": 0.7492,
+      "step": 265
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 2.56968357976584,
+      "learning_rate": 9.883613673364197e-06,
+      "loss": 0.7358,
+      "step": 270
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.665696800098671,
+      "learning_rate": 9.878628723310914e-06,
+      "loss": 0.7619,
+      "step": 275
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 2.953015088180945,
+      "learning_rate": 9.873540555281162e-06,
+      "loss": 0.7686,
+      "step": 280
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.50953540021068,
+      "learning_rate": 9.868349276926174e-06,
+      "loss": 0.75,
+      "step": 285
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 2.3798185960845113,
+      "learning_rate": 9.863054998078711e-06,
+      "loss": 0.7116,
+      "step": 290
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.640089489730164,
+      "learning_rate": 9.857657830750727e-06,
+      "loss": 0.7442,
+      "step": 295
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 2.7095039564753542,
+      "learning_rate": 9.85215788913101e-06,
+      "loss": 0.752,
+      "step": 300
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.513429446875121,
+      "learning_rate": 9.846555289582757e-06,
+      "loss": 0.7149,
+      "step": 305
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.4773281491463934,
+      "learning_rate": 9.840850150641117e-06,
+      "loss": 0.7586,
+      "step": 310
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 2.5694933133678375,
+      "learning_rate": 9.835042593010688e-06,
+      "loss": 0.722,
+      "step": 315
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.4529693819685936,
+      "learning_rate": 9.82913273956295e-06,
+      "loss": 0.7158,
+      "step": 320
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 2.476730985919433,
+      "learning_rate": 9.823120715333677e-06,
+      "loss": 0.7209,
+      "step": 325
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.557478344233571,
+      "learning_rate": 9.817006647520285e-06,
+      "loss": 0.7504,
+      "step": 330
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 2.5248657390840425,
+      "learning_rate": 9.810790665479147e-06,
+      "loss": 0.7273,
+      "step": 335
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.550311193358409,
+      "learning_rate": 9.80447290072285e-06,
+      "loss": 0.7369,
+      "step": 340
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.6451207143889386,
+      "learning_rate": 9.798053486917417e-06,
+      "loss": 0.716,
+      "step": 345
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 2.4625230913036917,
+      "learning_rate": 9.791532559879475e-06,
+      "loss": 0.7258,
+      "step": 350
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.8082861667365204,
+      "learning_rate": 9.784910257573384e-06,
+      "loss": 0.7387,
+      "step": 355
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 2.3838498876394754,
+      "learning_rate": 9.77818672010832e-06,
+      "loss": 0.7374,
+      "step": 360
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.5237932886610985,
+      "learning_rate": 9.771362089735308e-06,
+      "loss": 0.7139,
+      "step": 365
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 2.5238938887122497,
+      "learning_rate": 9.764436510844211e-06,
+      "loss": 0.7311,
+      "step": 370
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.3706005076627092,
+      "learning_rate": 9.757410129960677e-06,
+      "loss": 0.7266,
+      "step": 375
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.4442632943571554,
+      "learning_rate": 9.750283095743038e-06,
+      "loss": 0.7342,
+      "step": 380
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 2.3138128605282935,
+      "learning_rate": 9.743055558979172e-06,
+      "loss": 0.7397,
+      "step": 385
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.2847654770128494,
+      "learning_rate": 9.735727672583298e-06,
+      "loss": 0.7399,
+      "step": 390
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 2.5373133223306623,
+      "learning_rate": 9.728299591592754e-06,
+      "loss": 0.7286,
+      "step": 395
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.452038199405589,
+      "learning_rate": 9.720771473164711e-06,
+      "loss": 0.7236,
+      "step": 400
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 2.5337742582348364,
+      "learning_rate": 9.713143476572853e-06,
+      "loss": 0.7216,
+      "step": 405
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.4617731404164105,
+      "learning_rate": 9.705415763203992e-06,
+      "loss": 0.7099,
+      "step": 410
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 2.4416863074195403,
+      "learning_rate": 9.697588496554679e-06,
+      "loss": 0.6988,
+      "step": 415
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.3926416774659636,
+      "learning_rate": 9.68966184222772e-06,
+      "loss": 0.7146,
+      "step": 420
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.4534759050847894,
+      "learning_rate": 9.681635967928687e-06,
+      "loss": 0.7002,
+      "step": 425
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 2.387101177885321,
+      "learning_rate": 9.673511043462367e-06,
+      "loss": 0.6882,
+      "step": 430
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.351201127578531,
+      "learning_rate": 9.665287240729166e-06,
+      "loss": 0.7016,
+      "step": 435
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 2.4028637732485207,
+      "learning_rate": 9.656964733721476e-06,
+      "loss": 0.7087,
+      "step": 440
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.5222610742079596,
+      "learning_rate": 9.648543698519993e-06,
+      "loss": 0.7098,
+      "step": 445
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 2.456149366911768,
+      "learning_rate": 9.640024313289982e-06,
+      "loss": 0.7146,
+      "step": 450
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.5926596049197155,
+      "learning_rate": 9.63140675827753e-06,
+      "loss": 0.7165,
+      "step": 455
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.4065288074871436,
+      "learning_rate": 9.62269121580571e-06,
+      "loss": 0.7242,
+      "step": 460
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 2.4153475258252635,
+      "learning_rate": 9.613877870270735e-06,
+      "loss": 0.7117,
+      "step": 465
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.456326505795922,
+      "learning_rate": 9.604966908138052e-06,
+      "loss": 0.728,
+      "step": 470
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 2.460254090649474,
+      "learning_rate": 9.5959585179384e-06,
+      "loss": 0.6941,
+      "step": 475
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 2.52985016903139,
+      "learning_rate": 9.586852890263822e-06,
+      "loss": 0.7242,
+      "step": 480
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 2.34433015717048,
+      "learning_rate": 9.577650217763627e-06,
+      "loss": 0.6942,
+      "step": 485
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.578889043173228,
+      "learning_rate": 9.568350695140322e-06,
+      "loss": 0.7108,
+      "step": 490
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.5642361656083468,
+      "learning_rate": 9.558954519145487e-06,
+      "loss": 0.7283,
+      "step": 495
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 2.4744562180788723,
+      "learning_rate": 9.549461888575611e-06,
+      "loss": 0.7039,
+      "step": 500
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.4916298622893134,
+      "learning_rate": 9.539873004267892e-06,
+      "loss": 0.7058,
+      "step": 505
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 2.462034950187244,
+      "learning_rate": 9.530188069095985e-06,
+      "loss": 0.7157,
+      "step": 510
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.449172857073642,
+      "learning_rate": 9.520407287965707e-06,
+      "loss": 0.722,
+      "step": 515
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 2.3756831201714084,
+      "learning_rate": 9.510530867810706e-06,
+      "loss": 0.7237,
+      "step": 520
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.3708673455593887,
+      "learning_rate": 9.500559017588081e-06,
+      "loss": 0.6945,
+      "step": 525
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 2.4107569191063734,
+      "learning_rate": 9.490491948273961e-06,
+      "loss": 0.6866,
+      "step": 530
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.415351076693774,
+      "learning_rate": 9.48032987285904e-06,
+      "loss": 0.703,
+      "step": 535
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.390041383486964,
+      "learning_rate": 9.470073006344074e-06,
+      "loss": 0.6959,
+      "step": 540
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 2.4170577155298303,
+      "learning_rate": 9.45972156573533e-06,
+      "loss": 0.7158,
+      "step": 545
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.357444455763281,
+      "learning_rate": 9.449275770039995e-06,
+      "loss": 0.7149,
+      "step": 550
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 2.5525262377117293,
+      "learning_rate": 9.438735840261541e-06,
+      "loss": 0.7186,
+      "step": 555
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.2844487560246947,
+      "learning_rate": 9.428101999395057e-06,
+      "loss": 0.6957,
+      "step": 560
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 2.5637917700196384,
+      "learning_rate": 9.417374472422514e-06,
+      "loss": 0.7119,
+      "step": 565
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.401316100648321,
+      "learning_rate": 9.406553486308028e-06,
+      "loss": 0.6846,
+      "step": 570
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.394178128316637,
+      "learning_rate": 9.395639269993035e-06,
+      "loss": 0.6883,
+      "step": 575
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 2.4032333085550035,
+      "learning_rate": 9.384632054391468e-06,
+      "loss": 0.7242,
+      "step": 580
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.412056956339954,
+      "learning_rate": 9.373532072384852e-06,
+      "loss": 0.7175,
+      "step": 585
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 2.5298377835998775,
+      "learning_rate": 9.362339558817395e-06,
+      "loss": 0.7131,
+      "step": 590
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.5730596244972954,
+      "learning_rate": 9.351054750491005e-06,
+      "loss": 0.7021,
+      "step": 595
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 2.342248771961136,
+      "learning_rate": 9.339677886160293e-06,
+      "loss": 0.7213,
+      "step": 600
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.3406275850155533,
+      "learning_rate": 9.328209206527503e-06,
+      "loss": 0.7072,
+      "step": 605
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.4480685504384345,
+      "learning_rate": 9.316648954237441e-06,
+      "loss": 0.6935,
+      "step": 610
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 2.3571630508729635,
+      "learning_rate": 9.304997373872332e-06,
+      "loss": 0.6822,
+      "step": 615
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.411752568938528,
+      "learning_rate": 9.293254711946634e-06,
+      "loss": 0.7108,
+      "step": 620
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 2.3189554578732,
+      "learning_rate": 9.281421216901844e-06,
+      "loss": 0.7005,
+      "step": 625
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.300017975189379,
+      "learning_rate": 9.269497139101224e-06,
+      "loss": 0.7102,
+      "step": 630
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 2.404149990858198,
+      "learning_rate": 9.257482730824516e-06,
+      "loss": 0.703,
+      "step": 635
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.3029320596032377,
+      "learning_rate": 9.245378246262593e-06,
+      "loss": 0.6684,
+      "step": 640
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 2.5032720918650844,
+      "learning_rate": 9.233183941512093e-06,
+      "loss": 0.7059,
+      "step": 645
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.238997245234633,
+      "learning_rate": 9.220900074569994e-06,
+      "loss": 0.6768,
+      "step": 650
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.2703438372445848,
+      "learning_rate": 9.208526905328151e-06,
+      "loss": 0.678,
+      "step": 655
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 2.4396238742216556,
+      "learning_rate": 9.19606469556781e-06,
+      "loss": 0.6916,
+      "step": 660
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 2.4993574577483377,
+      "learning_rate": 9.183513708954058e-06,
+      "loss": 0.68,
+      "step": 665
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 2.3486701594923316,
+      "learning_rate": 9.170874211030252e-06,
+      "loss": 0.6775,
+      "step": 670
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.3102513888146206,
+      "learning_rate": 9.158146469212394e-06,
+      "loss": 0.6987,
+      "step": 675
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.294015980752727,
+      "learning_rate": 9.145330752783482e-06,
+      "loss": 0.7246,
+      "step": 680
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.2677057179765616,
+      "learning_rate": 9.132427332887804e-06,
+      "loss": 0.6748,
+      "step": 685
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.347417382651582,
+      "learning_rate": 9.119436482525205e-06,
+      "loss": 0.7132,
+      "step": 690
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 2.3961161008418483,
+      "learning_rate": 9.106358476545313e-06,
+      "loss": 0.6917,
+      "step": 695
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 2.5050859884311696,
+      "learning_rate": 9.093193591641723e-06,
+      "loss": 0.6794,
+      "step": 700
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 2.404050787157897,
+      "learning_rate": 9.079942106346138e-06,
+      "loss": 0.6964,
+      "step": 705
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.6207202710260895,
+      "learning_rate": 9.066604301022485e-06,
+      "loss": 0.6901,
+      "step": 710
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 2.2800088504281826,
+      "learning_rate": 9.053180457860978e-06,
+      "loss": 0.6977,
+      "step": 715
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 2.2856799216567762,
+      "learning_rate": 9.039670860872145e-06,
+      "loss": 0.6922,
+      "step": 720
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 2.4110081835843444,
+      "learning_rate": 9.026075795880822e-06,
+      "loss": 0.7091,
+      "step": 725
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 2.251437710188116,
+      "learning_rate": 9.01239555052011e-06,
+      "loss": 0.6915,
+      "step": 730
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 2.3032036266738047,
+      "learning_rate": 8.998630414225284e-06,
+      "loss": 0.6904,
+      "step": 735
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 2.2499145731750145,
+      "learning_rate": 8.98478067822767e-06,
+      "loss": 0.6847,
+      "step": 740
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.2911334845642406,
+      "learning_rate": 8.970846635548483e-06,
+      "loss": 0.686,
+      "step": 745
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 2.2920826486778347,
+      "learning_rate": 8.956828580992633e-06,
+      "loss": 0.6692,
+      "step": 750
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 2.312834518329523,
+      "learning_rate": 8.942726811142478e-06,
+      "loss": 0.7089,
+      "step": 755
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 2.2535035010232662,
+      "learning_rate": 8.928541624351563e-06,
+      "loss": 0.701,
+      "step": 760
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.243816611260628,
+      "learning_rate": 8.914273320738288e-06,
+      "loss": 0.6556,
+      "step": 765
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.3883240307823557,
+      "learning_rate": 8.899922202179581e-06,
+      "loss": 0.6784,
+      "step": 770
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 2.2760434140090466,
+      "learning_rate": 8.88548857230449e-06,
+      "loss": 0.6978,
+      "step": 775
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 2.277248154597817,
+      "learning_rate": 8.870972736487775e-06,
+      "loss": 0.6947,
+      "step": 780
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 2.2442652725243772,
+      "learning_rate": 8.856375001843442e-06,
+      "loss": 0.6729,
+      "step": 785
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.2370317874182617,
+      "learning_rate": 8.84169567721824e-06,
+      "loss": 0.6933,
+      "step": 790
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 2.2607128250963595,
+      "learning_rate": 8.826935073185135e-06,
+      "loss": 0.6713,
+      "step": 795
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.3744496402305426,
+      "learning_rate": 8.812093502036732e-06,
+      "loss": 0.6818,
+      "step": 800
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.383599476158656,
+      "learning_rate": 8.797171277778672e-06,
+      "loss": 0.6923,
+      "step": 805
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 2.225223944515689,
+      "learning_rate": 8.782168716122988e-06,
+      "loss": 0.6863,
+      "step": 810
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 2.336363416075415,
+      "learning_rate": 8.767086134481427e-06,
+      "loss": 0.6769,
+      "step": 815
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 2.3083758969360098,
+      "learning_rate": 8.751923851958728e-06,
+      "loss": 0.6579,
+      "step": 820
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 2.1851157321909507,
+      "learning_rate": 8.73668218934588e-06,
+      "loss": 0.6782,
+      "step": 825
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 2.31411684509431,
+      "learning_rate": 8.721361469113324e-06,
+      "loss": 0.709,
+      "step": 830
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 2.1474931353306626,
+      "learning_rate": 8.705962015404143e-06,
+      "loss": 0.6859,
+      "step": 835
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 2.187437489420089,
+      "learning_rate": 8.690484154027192e-06,
+      "loss": 0.6922,
+      "step": 840
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 2.8268613822761366,
+      "learning_rate": 8.674928212450216e-06,
+      "loss": 0.7038,
+      "step": 845
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 2.324162285544936,
+      "learning_rate": 8.659294519792909e-06,
+      "loss": 0.6731,
+      "step": 850
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 2.282125991178541,
+      "learning_rate": 8.643583406819965e-06,
+      "loss": 0.6586,
+      "step": 855
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 2.1777712326813927,
+      "learning_rate": 8.627795205934069e-06,
+      "loss": 0.6742,
+      "step": 860
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 2.148591637467636,
+      "learning_rate": 8.611930251168867e-06,
+      "loss": 0.6928,
+      "step": 865
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 2.242091930760005,
+      "learning_rate": 8.595988878181904e-06,
+      "loss": 0.6632,
+      "step": 870
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 2.1450868723403196,
+      "learning_rate": 8.579971424247514e-06,
+      "loss": 0.6625,
+      "step": 875
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.3648432919704163,
+      "learning_rate": 8.563878228249686e-06,
+      "loss": 0.6519,
+      "step": 880
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.3029446616572975,
+      "learning_rate": 8.547709630674909e-06,
+      "loss": 0.6667,
+      "step": 885
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 2.2718026788730885,
+      "learning_rate": 8.531465973604946e-06,
+      "loss": 0.6711,
+      "step": 890
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 2.176847727890739,
+      "learning_rate": 8.515147600709604e-06,
+      "loss": 0.6911,
+      "step": 895
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 2.159887584606336,
+      "learning_rate": 8.498754857239472e-06,
+      "loss": 0.672,
+      "step": 900
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 2.1436273681591804,
+      "learning_rate": 8.482288090018608e-06,
+      "loss": 0.6837,
+      "step": 905
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 2.320916872081505,
+      "learning_rate": 8.465747647437205e-06,
+      "loss": 0.6873,
+      "step": 910
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.2640460015839854,
+      "learning_rate": 8.449133879444211e-06,
+      "loss": 0.6687,
+      "step": 915
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.191546885776604,
+      "learning_rate": 8.43244713753994e-06,
+      "loss": 0.6687,
+      "step": 920
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 2.286636883588422,
+      "learning_rate": 8.415687774768625e-06,
+      "loss": 0.6827,
+      "step": 925
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.119469288532559,
+      "learning_rate": 8.398856145710953e-06,
+      "loss": 0.6464,
+      "step": 930
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 2.2317587056196433,
+      "learning_rate": 8.381952606476552e-06,
+      "loss": 0.6769,
+      "step": 935
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 2.218334601694794,
+      "learning_rate": 8.364977514696474e-06,
+      "loss": 0.6547,
+      "step": 940
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 2.211113896309907,
+      "learning_rate": 8.347931229515625e-06,
+      "loss": 0.6708,
+      "step": 945
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 2.2852199925944094,
+      "learning_rate": 8.330814111585149e-06,
+      "loss": 0.6634,
+      "step": 950
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 2.175125532278546,
+      "learning_rate": 8.31362652305482e-06,
+      "loss": 0.6636,
+      "step": 955
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 2.194239117414355,
+      "learning_rate": 8.296368827565365e-06,
+      "loss": 0.6745,
+      "step": 960
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 2.288161571096483,
+      "learning_rate": 8.279041390240781e-06,
+      "loss": 0.6753,
+      "step": 965
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 2.337786013162235,
+      "learning_rate": 8.261644577680603e-06,
+      "loss": 0.6637,
+      "step": 970
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.232975975441481,
+      "learning_rate": 8.244178757952149e-06,
+      "loss": 0.6751,
+      "step": 975
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 2.1659815604405073,
+      "learning_rate": 8.22664430058273e-06,
+      "loss": 0.6695,
+      "step": 980
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.1750265665614914,
+      "learning_rate": 8.209041576551842e-06,
+      "loss": 0.6807,
+      "step": 985
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 2.117500758919553,
+      "learning_rate": 8.191370958283305e-06,
+      "loss": 0.6738,
+      "step": 990
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 2.186272470050956,
+      "learning_rate": 8.17363281963739e-06,
+      "loss": 0.6428,
+      "step": 995
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 2.3398838498470407,
+      "learning_rate": 8.155827535902912e-06,
+      "loss": 0.6865,
+      "step": 1000
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 2.1027998540986785,
+      "learning_rate": 8.137955483789279e-06,
+      "loss": 0.6429,
+      "step": 1005
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.1675023981221475,
+      "learning_rate": 8.120017041418539e-06,
+      "loss": 0.6612,
+      "step": 1010
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 2.3164494379167606,
+      "learning_rate": 8.102012588317356e-06,
+      "loss": 0.6623,
+      "step": 1015
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 2.0641139366003545,
+      "learning_rate": 8.083942505409009e-06,
+      "loss": 0.6545,
+      "step": 1020
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 2.221849548889518,
+      "learning_rate": 8.065807175005307e-06,
+      "loss": 0.6749,
+      "step": 1025
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.104874622991842,
+      "learning_rate": 8.047606980798517e-06,
+      "loss": 0.6701,
+      "step": 1030
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.191896965558964,
+      "learning_rate": 8.029342307853238e-06,
+      "loss": 0.6732,
+      "step": 1035
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 2.110125648682477,
+      "learning_rate": 8.011013542598258e-06,
+      "loss": 0.6753,
+      "step": 1040
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.052875709500674,
+      "learning_rate": 7.992621072818377e-06,
+      "loss": 0.6505,
+      "step": 1045
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 2.226161904353228,
+      "learning_rate": 7.9741652876462e-06,
+      "loss": 0.6687,
+      "step": 1050
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 2.055919642942431,
+      "learning_rate": 7.95564657755391e-06,
+      "loss": 0.6422,
+      "step": 1055
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 2.2138333531678027,
+      "learning_rate": 7.937065334345002e-06,
+      "loss": 0.6648,
+      "step": 1060
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.106138856322583,
+      "learning_rate": 7.918421951145993e-06,
+      "loss": 0.676,
+      "step": 1065
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.217741214632785,
+      "learning_rate": 7.899716822398107e-06,
+      "loss": 0.6608,
+      "step": 1070
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 2.1229141865243735,
+      "learning_rate": 7.880950343848933e-06,
+      "loss": 0.6803,
+      "step": 1075
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.1730705138012483,
+      "learning_rate": 7.862122912544043e-06,
+      "loss": 0.638,
+      "step": 1080
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 2.1759981588146267,
+      "learning_rate": 7.843234926818595e-06,
+      "loss": 0.6573,
+      "step": 1085
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.246051922867301,
+      "learning_rate": 7.824286786288919e-06,
+      "loss": 0.6498,
+      "step": 1090
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 2.221038445486671,
+      "learning_rate": 7.805278891844036e-06,
+      "loss": 0.6625,
+      "step": 1095
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 2.1160883449068635,
+      "learning_rate": 7.786211645637197e-06,
+      "loss": 0.6706,
+      "step": 1100
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 2.5117047903095187,
+      "learning_rate": 7.76708545107737e-06,
+      "loss": 0.6483,
+      "step": 1105
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.199168846682398,
+      "learning_rate": 7.747900712820705e-06,
+      "loss": 0.6527,
+      "step": 1110
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.2087885312824143,
+      "learning_rate": 7.72865783676196e-06,
+      "loss": 0.6905,
+      "step": 1115
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 2.2279626457852717,
+      "learning_rate": 7.709357230025937e-06,
+      "loss": 0.6544,
+      "step": 1120
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 2.189165894933891,
+      "learning_rate": 7.689999300958853e-06,
+      "loss": 0.6611,
+      "step": 1125
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 2.1316909692662027,
+      "learning_rate": 7.670584459119695e-06,
+      "loss": 0.6618,
+      "step": 1130
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.147920589091614,
+      "learning_rate": 7.651113115271573e-06,
+      "loss": 0.656,
+      "step": 1135
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 2.146250338167946,
+      "learning_rate": 7.631585681373014e-06,
+      "loss": 0.6638,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.119468285060325,
+      "learning_rate": 7.612002570569254e-06,
+      "loss": 0.6513,
+      "step": 1145
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.198770694512473,
+      "learning_rate": 7.592364197183495e-06,
+      "loss": 0.6586,
+      "step": 1150
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 2.400108030261852,
+      "learning_rate": 7.572670976708137e-06,
+      "loss": 0.4731,
+      "step": 1155
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 2.364746913129791,
+      "learning_rate": 7.552923325795991e-06,
+      "loss": 0.4555,
+      "step": 1160
+    },
+    {
+      "epoch": 1.01,
+      "grad_norm": 2.4055012226169095,
+      "learning_rate": 7.5331216622514595e-06,
+      "loss": 0.437,
+      "step": 1165
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 2.313279146739233,
+      "learning_rate": 7.513266405021704e-06,
+      "loss": 0.4398,
+      "step": 1170
+    },
+    {
+      "epoch": 1.02,
+      "grad_norm": 2.3317274950511107,
+      "learning_rate": 7.4933579741877715e-06,
+      "loss": 0.4319,
+      "step": 1175
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 2.296967740213265,
+      "learning_rate": 7.473396790955715e-06,
+      "loss": 0.4581,
+      "step": 1180
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 2.3201979906079515,
+      "learning_rate": 7.4533832776476785e-06,
+      "loss": 0.446,
+      "step": 1185
+    },
+    {
+      "epoch": 1.03,
+      "grad_norm": 2.4563492503930076,
+      "learning_rate": 7.433317857692963e-06,
+      "loss": 0.4573,
+      "step": 1190
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 2.2940746118411637,
+      "learning_rate": 7.413200955619066e-06,
+      "loss": 0.4619,
+      "step": 1195
+    },
+    {
+      "epoch": 1.04,
+      "grad_norm": 2.398133267433227,
+      "learning_rate": 7.3930329970426984e-06,
+      "loss": 0.4575,
+      "step": 1200
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 2.273563546691222,
+      "learning_rate": 7.372814408660789e-06,
+      "loss": 0.4588,
+      "step": 1205
+    },
+    {
+      "epoch": 1.05,
+      "grad_norm": 2.267209839029385,
+      "learning_rate": 7.352545618241444e-06,
+      "loss": 0.4441,
+      "step": 1210
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 2.323736959536939,
+      "learning_rate": 7.332227054614904e-06,
+      "loss": 0.4508,
+      "step": 1215
+    },
+    {
+      "epoch": 1.06,
+      "grad_norm": 2.3838571633248002,
+      "learning_rate": 7.311859147664473e-06,
+      "loss": 0.4531,
+      "step": 1220
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 2.182188209979992,
+      "learning_rate": 7.291442328317414e-06,
+      "loss": 0.4505,
+      "step": 1225
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 2.3442142988685966,
+      "learning_rate": 7.270977028535846e-06,
+      "loss": 0.4412,
+      "step": 1230
+    },
+    {
+      "epoch": 1.07,
+      "grad_norm": 2.323598372799621,
+      "learning_rate": 7.250463681307589e-06,
+      "loss": 0.4439,
+      "step": 1235
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 2.2106474129172375,
+      "learning_rate": 7.229902720637014e-06,
+      "loss": 0.4599,
+      "step": 1240
+    },
+    {
+      "epoch": 1.08,
+      "grad_norm": 2.33819175232935,
+      "learning_rate": 7.2092945815358605e-06,
+      "loss": 0.4667,
+      "step": 1245
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 2.3341823848207914,
+      "learning_rate": 7.1886397000140265e-06,
+      "loss": 0.449,
+      "step": 1250
+    },
+    {
+      "epoch": 1.09,
+      "grad_norm": 2.277246806756992,
+      "learning_rate": 7.167938513070345e-06,
+      "loss": 0.4299,
+      "step": 1255
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 2.3424708082894337,
+      "learning_rate": 7.147191458683349e-06,
+      "loss": 0.4491,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 2.2855389092781038,
+      "learning_rate": 7.126398975801989e-06,
+      "loss": 0.4319,
+      "step": 1265
+    },
+    {
+      "epoch": 1.1,
+      "grad_norm": 2.2355767437622793,
+      "learning_rate": 7.105561504336357e-06,
+      "loss": 0.4708,
+      "step": 1270
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 2.334913640645696,
+      "learning_rate": 7.084679485148376e-06,
+      "loss": 0.4487,
+      "step": 1275
+    },
+    {
+      "epoch": 1.11,
+      "grad_norm": 2.2651259873576133,
+      "learning_rate": 7.063753360042471e-06,
+      "loss": 0.4505,
+      "step": 1280
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 2.2132949455185242,
+      "learning_rate": 7.042783571756229e-06,
+      "loss": 0.4479,
+      "step": 1285
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 2.27387092173793,
+      "learning_rate": 7.021770563951018e-06,
+      "loss": 0.4462,
+      "step": 1290
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 2.1617502787818226,
+      "learning_rate": 7.0007147812026136e-06,
+      "loss": 0.455,
+      "step": 1295
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 2.1953103025627154,
+      "learning_rate": 6.979616668991791e-06,
+      "loss": 0.4657,
+      "step": 1300
+    },
+    {
+      "epoch": 1.13,
+      "grad_norm": 2.2881259063861505,
+      "learning_rate": 6.958476673694888e-06,
+      "loss": 0.4477,
+      "step": 1305
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 2.2281844588924327,
+      "learning_rate": 6.93729524257438e-06,
+      "loss": 0.457,
+      "step": 1310
+    },
+    {
+      "epoch": 1.14,
+      "grad_norm": 2.1988173912911098,
+      "learning_rate": 6.9160728237694e-06,
+      "loss": 0.4474,
+      "step": 1315
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 2.3678698134060094,
+      "learning_rate": 6.89480986628627e-06,
+      "loss": 0.4537,
+      "step": 1320
+    },
+    {
+      "epoch": 1.15,
+      "grad_norm": 2.2484647200139642,
+      "learning_rate": 6.873506819988986e-06,
+      "loss": 0.4598,
+      "step": 1325
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 2.3670299685805625,
+      "learning_rate": 6.852164135589725e-06,
+      "loss": 0.4477,
+      "step": 1330
+    },
+    {
+      "epoch": 1.16,
+      "grad_norm": 2.2380787201135797,
+      "learning_rate": 6.830782264639281e-06,
+      "loss": 0.4344,
+      "step": 1335
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 2.3442882798353297,
+      "learning_rate": 6.809361659517528e-06,
+      "loss": 0.4591,
+      "step": 1340
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 2.328235841696171,
+      "learning_rate": 6.78790277342385e-06,
+      "loss": 0.4541,
+      "step": 1345
+    },
+    {
+      "epoch": 1.17,
+      "grad_norm": 2.2793368086904886,
+      "learning_rate": 6.766406060367544e-06,
+      "loss": 0.4372,
+      "step": 1350
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 2.289606269955421,
+      "learning_rate": 6.744871975158216e-06,
+      "loss": 0.465,
+      "step": 1355
+    },
+    {
+      "epoch": 1.18,
+      "grad_norm": 2.147843921505496,
+      "learning_rate": 6.723300973396167e-06,
+      "loss": 0.4443,
+      "step": 1360
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 2.43326337934887,
+      "learning_rate": 6.701693511462744e-06,
+      "loss": 0.4763,
+      "step": 1365
+    },
+    {
+      "epoch": 1.19,
+      "grad_norm": 2.462997342629488,
+      "learning_rate": 6.680050046510689e-06,
+      "loss": 0.4711,
+      "step": 1370
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 2.312063338092545,
+      "learning_rate": 6.658371036454464e-06,
+      "loss": 0.4577,
+      "step": 1375
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 2.480660839404198,
+      "learning_rate": 6.636656939960569e-06,
+      "loss": 0.457,
+      "step": 1380
+    },
+    {
+      "epoch": 1.2,
+      "grad_norm": 2.2605004693409083,
+      "learning_rate": 6.614908216437832e-06,
+      "loss": 0.4674,
+      "step": 1385
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 2.254053949935403,
+      "learning_rate": 6.59312532602769e-06,
+      "loss": 0.4458,
+      "step": 1390
+    },
+    {
+      "epoch": 1.21,
+      "grad_norm": 2.425633478739145,
+      "learning_rate": 6.57130872959445e-06,
+      "loss": 0.4484,
+      "step": 1395
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 2.3799822205570726,
+      "learning_rate": 6.549458888715555e-06,
+      "loss": 0.4355,
+      "step": 1400
+    },
+    {
+      "epoch": 1.22,
+      "grad_norm": 2.2426442008904037,
+      "learning_rate": 6.527576265671796e-06,
+      "loss": 0.4532,
+      "step": 1405
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 2.267315783289316,
+      "learning_rate": 6.505661323437544e-06,
+      "loss": 0.4507,
+      "step": 1410
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 2.283183569643785,
+      "learning_rate": 6.483714525670956e-06,
+      "loss": 0.4509,
+      "step": 1415
+    },
+    {
+      "epoch": 1.23,
+      "grad_norm": 2.3514735292836537,
+      "learning_rate": 6.4617363367041605e-06,
+      "loss": 0.4508,
+      "step": 1420
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 2.397931859781916,
+      "learning_rate": 6.439727221533431e-06,
+      "loss": 0.4545,
+      "step": 1425
+    },
+    {
+      "epoch": 1.24,
+      "grad_norm": 2.2919163085343475,
+      "learning_rate": 6.417687645809358e-06,
+      "loss": 0.4445,
+      "step": 1430
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 2.4071810549228827,
+      "learning_rate": 6.395618075826987e-06,
+      "loss": 0.4591,
+      "step": 1435
+    },
+    {
+      "epoch": 1.25,
+      "grad_norm": 2.209407168401749,
+      "learning_rate": 6.373518978515958e-06,
+      "loss": 0.4391,
+      "step": 1440
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 2.356872933625598,
+      "learning_rate": 6.351390821430626e-06,
+      "loss": 0.4418,
+      "step": 1445
+    },
+    {
+      "epoch": 1.26,
+      "grad_norm": 2.3133423881622166,
+      "learning_rate": 6.329234072740169e-06,
+      "loss": 0.4578,
+      "step": 1450
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 2.3665337891549165,
+      "learning_rate": 6.3070492012186836e-06,
+      "loss": 0.4486,
+      "step": 1455
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 2.3090803550297236,
+      "learning_rate": 6.284836676235262e-06,
+      "loss": 0.4391,
+      "step": 1460
+    },
+    {
+      "epoch": 1.27,
+      "grad_norm": 2.4024477075697863,
+      "learning_rate": 6.262596967744069e-06,
+      "loss": 0.4403,
+      "step": 1465
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 2.170215764608795,
+      "learning_rate": 6.240330546274394e-06,
+      "loss": 0.4419,
+      "step": 1470
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 2.2218725089875293,
+      "learning_rate": 6.218037882920698e-06,
+      "loss": 0.4443,
+      "step": 1475
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 2.212289537711609,
+      "learning_rate": 6.195719449332645e-06,
+      "loss": 0.4337,
+      "step": 1480
+    },
+    {
+      "epoch": 1.29,
+      "grad_norm": 2.2825227332103335,
+      "learning_rate": 6.173375717705124e-06,
+      "loss": 0.4491,
+      "step": 1485
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 2.209958254432096,
+      "learning_rate": 6.151007160768265e-06,
+      "loss": 0.4757,
+      "step": 1490
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 2.303785841368444,
+      "learning_rate": 6.128614251777417e-06,
+      "loss": 0.439,
+      "step": 1495
+    },
+    {
+      "epoch": 1.3,
+      "grad_norm": 2.3058301906040084,
+      "learning_rate": 6.106197464503168e-06,
+      "loss": 0.4589,
+      "step": 1500
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 2.286045686932392,
+      "learning_rate": 6.083757273221288e-06,
+      "loss": 0.4455,
+      "step": 1505
+    },
+    {
+      "epoch": 1.31,
+      "grad_norm": 2.2213791680934203,
+      "learning_rate": 6.061294152702717e-06,
+      "loss": 0.4264,
+      "step": 1510
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 2.3127729571587112,
+      "learning_rate": 6.03880857820351e-06,
+      "loss": 0.4397,
+      "step": 1515
+    },
+    {
+      "epoch": 1.32,
+      "grad_norm": 2.4109016633221443,
+      "learning_rate": 6.016301025454787e-06,
+      "loss": 0.4369,
+      "step": 1520
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 2.3588172620745094,
+      "learning_rate": 5.993771970652661e-06,
+      "loss": 0.4422,
+      "step": 1525
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 2.3352138929079795,
+      "learning_rate": 5.971221890448175e-06,
+      "loss": 0.459,
+      "step": 1530
+    },
+    {
+      "epoch": 1.33,
+      "grad_norm": 2.3469890107365017,
+      "learning_rate": 5.948651261937203e-06,
+      "loss": 0.4445,
+      "step": 1535
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 2.2763593608974957,
+      "learning_rate": 5.926060562650365e-06,
+      "loss": 0.4534,
+      "step": 1540
+    },
+    {
+      "epoch": 1.34,
+      "grad_norm": 2.3051490949637183,
+      "learning_rate": 5.903450270542925e-06,
+      "loss": 0.4731,
+      "step": 1545
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 2.470815165119516,
+      "learning_rate": 5.880820863984672e-06,
+      "loss": 0.4687,
+      "step": 1550
+    },
+    {
+      "epoch": 1.35,
+      "grad_norm": 2.2612634510823266,
+      "learning_rate": 5.858172821749804e-06,
+      "loss": 0.4493,
+      "step": 1555
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 2.275063055842053,
+      "learning_rate": 5.835506623006798e-06,
+      "loss": 0.4537,
+      "step": 1560
+    },
+    {
+      "epoch": 1.36,
+      "grad_norm": 2.1888990341084384,
+      "learning_rate": 5.8128227473082676e-06,
+      "loss": 0.441,
+      "step": 1565
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 2.409270397966297,
+      "learning_rate": 5.790121674580825e-06,
+      "loss": 0.4557,
+      "step": 1570
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 2.321494986142115,
+      "learning_rate": 5.7674038851149225e-06,
+      "loss": 0.4415,
+      "step": 1575
+    },
+    {
+      "epoch": 1.37,
+      "grad_norm": 2.246796045328034,
+      "learning_rate": 5.744669859554689e-06,
+      "loss": 0.437,
+      "step": 1580
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 2.3188020814924726,
+      "learning_rate": 5.721920078887764e-06,
+      "loss": 0.4456,
+      "step": 1585
+    },
+    {
+      "epoch": 1.38,
+      "grad_norm": 2.2417028323394823,
+      "learning_rate": 5.699155024435123e-06,
+      "loss": 0.4483,
+      "step": 1590
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 2.2662638123642176,
+      "learning_rate": 5.676375177840886e-06,
+      "loss": 0.4454,
+      "step": 1595
+    },
+    {
+      "epoch": 1.39,
+      "grad_norm": 2.371570368123361,
+      "learning_rate": 5.653581021062139e-06,
+      "loss": 0.4478,
+      "step": 1600
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 2.2716775566076364,
+      "learning_rate": 5.630773036358727e-06,
+      "loss": 0.4473,
+      "step": 1605
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 2.354064671717741,
+      "learning_rate": 5.607951706283056e-06,
+      "loss": 0.4464,
+      "step": 1610
+    },
+    {
+      "epoch": 1.4,
+      "grad_norm": 2.2083958223815707,
+      "learning_rate": 5.585117513669883e-06,
+      "loss": 0.4488,
+      "step": 1615
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 2.248597914210575,
+      "learning_rate": 5.562270941626099e-06,
+      "loss": 0.4398,
+      "step": 1620
+    },
+    {
+      "epoch": 1.41,
+      "grad_norm": 2.682659482628619,
+      "learning_rate": 5.539412473520508e-06,
+      "loss": 0.4599,
+      "step": 1625
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 2.2835988156541123,
+      "learning_rate": 5.516542592973604e-06,
+      "loss": 0.4453,
+      "step": 1630
+    },
+    {
+      "epoch": 1.42,
+      "grad_norm": 2.3120069870482833,
+      "learning_rate": 5.493661783847331e-06,
+      "loss": 0.4433,
+      "step": 1635
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.2910069598145486,
+      "learning_rate": 5.470770530234856e-06,
+      "loss": 0.4442,
+      "step": 1640
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.3021256006635005,
+      "learning_rate": 5.447869316450318e-06,
+      "loss": 0.4561,
+      "step": 1645
+    },
+    {
+      "epoch": 1.43,
+      "grad_norm": 2.3209867511334923,
+      "learning_rate": 5.424958627018587e-06,
+      "loss": 0.4528,
+      "step": 1650
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 2.261233409701518,
+      "learning_rate": 5.402038946665011e-06,
+      "loss": 0.4312,
+      "step": 1655
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 2.231273732412954,
+      "learning_rate": 5.3791107603051605e-06,
+      "loss": 0.4332,
+      "step": 1660
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 2.230611232382257,
+      "learning_rate": 5.356174553034566e-06,
+      "loss": 0.433,
+      "step": 1665
+    },
+    {
+      "epoch": 1.45,
+      "grad_norm": 2.275712471392273,
+      "learning_rate": 5.3332308101184616e-06,
+      "loss": 0.4523,
+      "step": 1670
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 2.28631570449613,
+      "learning_rate": 5.310280016981513e-06,
+      "loss": 0.4264,
+      "step": 1675
+    },
+    {
+      "epoch": 1.46,
+      "grad_norm": 2.2025636683152223,
+      "learning_rate": 5.287322659197548e-06,
+      "loss": 0.4336,
+      "step": 1680
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 2.3332300388398233,
+      "learning_rate": 5.264359222479284e-06,
+      "loss": 0.432,
+      "step": 1685
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 2.245112797694023,
+      "learning_rate": 5.2413901926680535e-06,
+      "loss": 0.4504,
+      "step": 1690
+    },
+    {
+      "epoch": 1.47,
+      "grad_norm": 2.347054603456251,
+      "learning_rate": 5.218416055723517e-06,
+      "loss": 0.4462,
+      "step": 1695
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 2.3254478674679087,
+      "learning_rate": 5.195437297713397e-06,
+      "loss": 0.4558,
+      "step": 1700
+    },
+    {
+      "epoch": 1.48,
+      "grad_norm": 2.319391634753776,
+      "learning_rate": 5.172454404803176e-06,
+      "loss": 0.4414,
+      "step": 1705
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 2.314982725356784,
+      "learning_rate": 5.149467863245824e-06,
+      "loss": 0.4596,
+      "step": 1710
+    },
+    {
+      "epoch": 1.49,
+      "grad_norm": 2.1793462624109834,
+      "learning_rate": 5.126478159371503e-06,
+      "loss": 0.4473,
+      "step": 1715
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 2.2896649530853166,
+      "learning_rate": 5.103485779577285e-06,
+      "loss": 0.4351,
+      "step": 1720
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 2.1355102943767097,
+      "learning_rate": 5.0804912103168504e-06,
+      "loss": 0.4307,
+      "step": 1725
+    },
+    {
+      "epoch": 1.5,
+      "grad_norm": 2.1597304988418915,
+      "learning_rate": 5.057494938090212e-06,
+      "loss": 0.4391,
+      "step": 1730
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 2.277000849546902,
+      "learning_rate": 5.034497449433402e-06,
+      "loss": 0.4448,
+      "step": 1735
+    },
+    {
+      "epoch": 1.51,
+      "grad_norm": 2.2763155206540917,
+      "learning_rate": 5.011499230908195e-06,
+      "loss": 0.4544,
+      "step": 1740
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 2.2298952898945976,
+      "learning_rate": 4.988500769091808e-06,
+      "loss": 0.4538,
+      "step": 1745
+    },
+    {
+      "epoch": 1.52,
+      "grad_norm": 2.308312147908615,
+      "learning_rate": 4.9655025505666e-06,
+      "loss": 0.4553,
+      "step": 1750
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 2.279721233089429,
+      "learning_rate": 4.94250506190979e-06,
+      "loss": 0.4435,
+      "step": 1755
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 2.2783363369895246,
+      "learning_rate": 4.9195087896831495e-06,
+      "loss": 0.4518,
+      "step": 1760
+    },
+    {
+      "epoch": 1.53,
+      "grad_norm": 2.265802433578123,
+      "learning_rate": 4.896514220422718e-06,
+      "loss": 0.4438,
+      "step": 1765
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 2.256465497081325,
+      "learning_rate": 4.873521840628498e-06,
+      "loss": 0.4583,
+      "step": 1770
+    },
+    {
+      "epoch": 1.54,
+      "grad_norm": 2.194603255396972,
+      "learning_rate": 4.850532136754178e-06,
+      "loss": 0.4378,
+      "step": 1775
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 2.2121127647515393,
+      "learning_rate": 4.827545595196825e-06,
+      "loss": 0.4593,
+      "step": 1780
+    },
+    {
+      "epoch": 1.55,
+      "grad_norm": 2.2787788765852577,
+      "learning_rate": 4.804562702286606e-06,
+      "loss": 0.4575,
+      "step": 1785
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 2.3657941610837496,
+      "learning_rate": 4.7815839442764846e-06,
+      "loss": 0.4315,
+      "step": 1790
+    },
+    {
+      "epoch": 1.56,
+      "grad_norm": 2.394715564020185,
+      "learning_rate": 4.758609807331948e-06,
+      "loss": 0.4531,
+      "step": 1795
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.2870566209131553,
+      "learning_rate": 4.735640777520716e-06,
+      "loss": 0.4346,
+      "step": 1800
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.1772760958721955,
+      "learning_rate": 4.712677340802454e-06,
+      "loss": 0.424,
+      "step": 1805
+    },
+    {
+      "epoch": 1.57,
+      "grad_norm": 2.2247434098325938,
+      "learning_rate": 4.6897199830184885e-06,
+      "loss": 0.4095,
+      "step": 1810
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 2.290681584181675,
+      "learning_rate": 4.66676918988154e-06,
+      "loss": 0.4357,
+      "step": 1815
+    },
+    {
+      "epoch": 1.58,
+      "grad_norm": 2.2388519615862252,
+      "learning_rate": 4.643825446965435e-06,
+      "loss": 0.4301,
+      "step": 1820
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 2.202537044622834,
+      "learning_rate": 4.620889239694842e-06,
+      "loss": 0.4566,
+      "step": 1825
+    },
+    {
+      "epoch": 1.59,
+      "grad_norm": 2.3671432882023424,
+      "learning_rate": 4.5979610533349904e-06,
+      "loss": 0.4361,
+      "step": 1830
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 2.2139507899755095,
+      "learning_rate": 4.575041372981414e-06,
+      "loss": 0.454,
+      "step": 1835
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 2.33465335246055,
+      "learning_rate": 4.5521306835496825e-06,
+      "loss": 0.4328,
+      "step": 1840
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 2.1273690709843462,
+      "learning_rate": 4.5292294697651476e-06,
+      "loss": 0.4381,
+      "step": 1845
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 2.3252935736092106,
+      "learning_rate": 4.50633821615267e-06,
+      "loss": 0.4612,
+      "step": 1850
+    },
+    {
+      "epoch": 1.61,
+      "grad_norm": 2.3008242282079108,
+      "learning_rate": 4.483457407026398e-06,
+      "loss": 0.4305,
+      "step": 1855
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 2.174985689846034,
+      "learning_rate": 4.460587526479493e-06,
+      "loss": 0.4544,
+      "step": 1860
+    },
+    {
+      "epoch": 1.62,
+      "grad_norm": 2.3490692073745856,
+      "learning_rate": 4.437729058373903e-06,
+      "loss": 0.4354,
+      "step": 1865
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 2.242430931619391,
+      "learning_rate": 4.4148824863301185e-06,
+      "loss": 0.4253,
+      "step": 1870
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 2.264671767323145,
+      "learning_rate": 4.392048293716945e-06,
+      "loss": 0.4247,
+      "step": 1875
+    },
+    {
+      "epoch": 1.63,
+      "grad_norm": 2.2937868058141415,
+      "learning_rate": 4.369226963641274e-06,
+      "loss": 0.4649,
+      "step": 1880
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 2.2699213674438714,
+      "learning_rate": 4.346418978937863e-06,
+      "loss": 0.4414,
+      "step": 1885
+    },
+    {
+      "epoch": 1.64,
+      "grad_norm": 2.3538952460044364,
+      "learning_rate": 4.323624822159116e-06,
+      "loss": 0.4331,
+      "step": 1890
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 2.2783366908821447,
+      "learning_rate": 4.300844975564878e-06,
+      "loss": 0.4215,
+      "step": 1895
+    },
+    {
+      "epoch": 1.65,
+      "grad_norm": 2.2025691509337872,
+      "learning_rate": 4.278079921112236e-06,
+      "loss": 0.4361,
+      "step": 1900
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 2.2754937791696697,
+      "learning_rate": 4.2553301404453125e-06,
+      "loss": 0.4454,
+      "step": 1905
+    },
+    {
+      "epoch": 1.66,
+      "grad_norm": 2.494049937308044,
+      "learning_rate": 4.232596114885078e-06,
+      "loss": 0.4428,
+      "step": 1910
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 2.324068297169444,
+      "learning_rate": 4.209878325419176e-06,
+      "loss": 0.4555,
+      "step": 1915
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 2.2124148165274407,
+      "learning_rate": 4.187177252691734e-06,
+      "loss": 0.4207,
+      "step": 1920
+    },
+    {
+      "epoch": 1.67,
+      "grad_norm": 2.330370938836315,
+      "learning_rate": 4.164493376993205e-06,
+      "loss": 0.4433,
+      "step": 1925
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 2.280092234160569,
+      "learning_rate": 4.1418271782501974e-06,
+      "loss": 0.4301,
+      "step": 1930
+    },
+    {
+      "epoch": 1.68,
+      "grad_norm": 2.2780438553394604,
+      "learning_rate": 4.119179136015329e-06,
+      "loss": 0.4585,
+      "step": 1935
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 2.2902061321424836,
+      "learning_rate": 4.096549729457074e-06,
+      "loss": 0.4543,
+      "step": 1940
+    },
+    {
+      "epoch": 1.69,
+      "grad_norm": 2.247668306299449,
+      "learning_rate": 4.0739394373496364e-06,
+      "loss": 0.4374,
+      "step": 1945
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 2.210281346137928,
+      "learning_rate": 4.051348738062798e-06,
+      "loss": 0.4336,
+      "step": 1950
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 2.2955680525797395,
+      "learning_rate": 4.028778109551826e-06,
+      "loss": 0.4306,
+      "step": 1955
+    },
+    {
+      "epoch": 1.7,
+      "grad_norm": 2.2066314731577976,
+      "learning_rate": 4.006228029347339e-06,
+      "loss": 0.4398,
+      "step": 1960
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 2.144292696403874,
+      "learning_rate": 3.983698974545216e-06,
+      "loss": 0.4209,
+      "step": 1965
+    },
+    {
+      "epoch": 1.71,
+      "grad_norm": 2.4140585929310223,
+      "learning_rate": 3.961191421796492e-06,
+      "loss": 0.4414,
+      "step": 1970
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 2.259932100532896,
+      "learning_rate": 3.938705847297285e-06,
+      "loss": 0.4442,
+      "step": 1975
+    },
+    {
+      "epoch": 1.72,
+      "grad_norm": 2.2955992868486095,
+      "learning_rate": 3.916242726778712e-06,
+      "loss": 0.4388,
+      "step": 1980
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 2.3276261180040896,
+      "learning_rate": 3.893802535496834e-06,
+      "loss": 0.4352,
+      "step": 1985
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 2.2603623578249734,
+      "learning_rate": 3.871385748222584e-06,
+      "loss": 0.4224,
+      "step": 1990
+    },
+    {
+      "epoch": 1.73,
+      "grad_norm": 2.193353286563046,
+      "learning_rate": 3.848992839231738e-06,
+      "loss": 0.4326,
+      "step": 1995
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 2.288992981491336,
+      "learning_rate": 3.826624282294876e-06,
+      "loss": 0.4291,
+      "step": 2000
+    },
+    {
+      "epoch": 1.74,
+      "grad_norm": 2.224041598232687,
+      "learning_rate": 3.804280550667357e-06,
+      "loss": 0.4165,
+      "step": 2005
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 2.2588806957365475,
+      "learning_rate": 3.781962117079304e-06,
+      "loss": 0.4401,
+      "step": 2010
+    },
+    {
+      "epoch": 1.75,
+      "grad_norm": 2.2237942651038254,
+      "learning_rate": 3.759669453725607e-06,
+      "loss": 0.4368,
+      "step": 2015
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 2.340413330784164,
+      "learning_rate": 3.737403032255932e-06,
+      "loss": 0.433,
+      "step": 2020
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 2.3588321620415402,
+      "learning_rate": 3.71516332376474e-06,
+      "loss": 0.4318,
+      "step": 2025
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 3.8365706672083375,
+      "learning_rate": 3.6929507987813185e-06,
+      "loss": 0.4398,
+      "step": 2030
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 2.2030009620446194,
+      "learning_rate": 3.670765927259832e-06,
+      "loss": 0.4376,
+      "step": 2035
+    },
+    {
+      "epoch": 1.77,
+      "grad_norm": 2.1843672172551463,
+      "learning_rate": 3.6486091785693744e-06,
+      "loss": 0.4202,
+      "step": 2040
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 2.226700778755956,
+      "learning_rate": 3.626481021484045e-06,
+      "loss": 0.4115,
+      "step": 2045
+    },
+    {
+      "epoch": 1.78,
+      "grad_norm": 2.3004579747757705,
+      "learning_rate": 3.6043819241730148e-06,
+      "loss": 0.4331,
+      "step": 2050
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 2.213263439511049,
+      "learning_rate": 3.582312354190643e-06,
+      "loss": 0.4231,
+      "step": 2055
+    },
+    {
+      "epoch": 1.79,
+      "grad_norm": 2.2802111043844926,
+      "learning_rate": 3.5602727784665692e-06,
+      "loss": 0.4187,
+      "step": 2060
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 2.3209983286675158,
+      "learning_rate": 3.538263663295841e-06,
+      "loss": 0.4433,
+      "step": 2065
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 2.211102964823549,
+      "learning_rate": 3.516285474329045e-06,
+      "loss": 0.4329,
+      "step": 2070
+    },
+    {
+      "epoch": 1.8,
+      "grad_norm": 2.2154473812994255,
+      "learning_rate": 3.4943386765624564e-06,
+      "loss": 0.4464,
+      "step": 2075
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 2.379437949936965,
+      "learning_rate": 3.472423734328205e-06,
+      "loss": 0.4494,
+      "step": 2080
+    },
+    {
+      "epoch": 1.81,
+      "grad_norm": 2.2412367250685374,
+      "learning_rate": 3.450541111284447e-06,
+      "loss": 0.444,
+      "step": 2085
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 2.2888382910956655,
+      "learning_rate": 3.4286912704055507e-06,
+      "loss": 0.4525,
+      "step": 2090
+    },
+    {
+      "epoch": 1.82,
+      "grad_norm": 2.270694169670531,
+      "learning_rate": 3.4068746739723124e-06,
+      "loss": 0.4465,
+      "step": 2095
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 2.403612322315147,
+      "learning_rate": 3.385091783562168e-06,
+      "loss": 0.4423,
+      "step": 2100
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 2.3531894071091233,
+      "learning_rate": 3.363343060039431e-06,
+      "loss": 0.4234,
+      "step": 2105
+    },
+    {
+      "epoch": 1.83,
+      "grad_norm": 2.2914909080973622,
+      "learning_rate": 3.3416289635455367e-06,
+      "loss": 0.4485,
+      "step": 2110
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 2.265368076663664,
+      "learning_rate": 3.319949953489313e-06,
+      "loss": 0.4503,
+      "step": 2115
+    },
+    {
+      "epoch": 1.84,
+      "grad_norm": 2.1959246230192666,
+      "learning_rate": 3.2983064885372574e-06,
+      "loss": 0.4174,
+      "step": 2120
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 2.2566063733367145,
+      "learning_rate": 3.2766990266038358e-06,
+      "loss": 0.4202,
+      "step": 2125
+    },
+    {
+      "epoch": 1.85,
+      "grad_norm": 2.421453614794928,
+      "learning_rate": 3.255128024841786e-06,
+      "loss": 0.4445,
+      "step": 2130
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 2.2481623701541027,
+      "learning_rate": 3.233593939632458e-06,
+      "loss": 0.4356,
+      "step": 2135
+    },
+    {
+      "epoch": 1.86,
+      "grad_norm": 2.241280678645406,
+      "learning_rate": 3.21209722657615e-06,
+      "loss": 0.4285,
+      "step": 2140
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 2.322803716341616,
+      "learning_rate": 3.1906383404824735e-06,
+      "loss": 0.4439,
+      "step": 2145
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 2.322285084946949,
+      "learning_rate": 3.169217735360721e-06,
+      "loss": 0.428,
+      "step": 2150
+    },
+    {
+      "epoch": 1.87,
+      "grad_norm": 2.2276484946529838,
+      "learning_rate": 3.1478358644102763e-06,
+      "loss": 0.4263,
+      "step": 2155
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 2.303959342499357,
+      "learning_rate": 3.1264931800110143e-06,
+      "loss": 0.435,
+      "step": 2160
+    },
+    {
+      "epoch": 1.88,
+      "grad_norm": 2.3968857618720576,
+      "learning_rate": 3.1051901337137337e-06,
+      "loss": 0.4344,
+      "step": 2165
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 2.2864487346327955,
+      "learning_rate": 3.083927176230601e-06,
+      "loss": 0.4146,
+      "step": 2170
+    },
+    {
+      "epoch": 1.89,
+      "grad_norm": 2.2707616791627934,
+      "learning_rate": 3.062704757425622e-06,
+      "loss": 0.4255,
+      "step": 2175
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 2.207654035345,
+      "learning_rate": 3.041523326305112e-06,
+      "loss": 0.436,
+      "step": 2180
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 2.1599866413154234,
+      "learning_rate": 3.020383331008212e-06,
+      "loss": 0.4135,
+      "step": 2185
+    },
+    {
+      "epoch": 1.9,
+      "grad_norm": 2.2816200841515735,
+      "learning_rate": 2.9992852187973877e-06,
+      "loss": 0.4195,
+      "step": 2190
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 2.229660770796668,
+      "learning_rate": 2.978229436048983e-06,
+      "loss": 0.4182,
+      "step": 2195
+    },
+    {
+      "epoch": 1.91,
+      "grad_norm": 2.270385764605724,
+      "learning_rate": 2.957216428243772e-06,
+      "loss": 0.4293,
+      "step": 2200
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 2.212249171865807,
+      "learning_rate": 2.9362466399575295e-06,
+      "loss": 0.4269,
+      "step": 2205
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 2.3022153629135467,
+      "learning_rate": 2.915320514851627e-06,
+      "loss": 0.4252,
+      "step": 2210
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 2.1798650360818383,
+      "learning_rate": 2.894438495663644e-06,
+      "loss": 0.4186,
+      "step": 2215
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 2.266858831074868,
+      "learning_rate": 2.8736010241980115e-06,
+      "loss": 0.4366,
+      "step": 2220
+    },
+    {
+      "epoch": 1.93,
+      "grad_norm": 2.1667781306193334,
+      "learning_rate": 2.8528085413166527e-06,
+      "loss": 0.4136,
+      "step": 2225
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 2.2017355573681985,
+      "learning_rate": 2.8320614869296566e-06,
+      "loss": 0.4342,
+      "step": 2230
+    },
+    {
+      "epoch": 1.94,
+      "grad_norm": 2.2079876034896455,
+      "learning_rate": 2.8113602999859764e-06,
+      "loss": 0.4345,
+      "step": 2235
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 2.202004834836398,
+      "learning_rate": 2.7907054184641412e-06,
+      "loss": 0.4152,
+      "step": 2240
+    },
+    {
+      "epoch": 1.95,
+      "grad_norm": 2.16840408850076,
+      "learning_rate": 2.7700972793629866e-06,
+      "loss": 0.4222,
+      "step": 2245
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 2.223578396387233,
+      "learning_rate": 2.7495363186924125e-06,
+      "loss": 0.4349,
+      "step": 2250
+    },
+    {
+      "epoch": 1.96,
+      "grad_norm": 2.261505217210584,
+      "learning_rate": 2.7290229714641546e-06,
+      "loss": 0.4334,
+      "step": 2255
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 2.2233867732051404,
+      "learning_rate": 2.708557671682586e-06,
+      "loss": 0.4338,
+      "step": 2260
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 2.283982826150082,
+      "learning_rate": 2.68814085233553e-06,
+      "loss": 0.4255,
+      "step": 2265
+    },
+    {
+      "epoch": 1.97,
+      "grad_norm": 2.237726467608789,
+      "learning_rate": 2.6677729453850964e-06,
+      "loss": 0.4211,
+      "step": 2270
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 2.331024809245345,
+      "learning_rate": 2.6474543817585575e-06,
+      "loss": 0.4407,
+      "step": 2275
+    },
+    {
+      "epoch": 1.98,
+      "grad_norm": 2.2236864768079294,
+      "learning_rate": 2.627185591339212e-06,
+      "loss": 0.4317,
+      "step": 2280
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 2.2713302070160997,
+      "learning_rate": 2.6069670029573036e-06,
+      "loss": 0.4206,
+      "step": 2285
+    },
+    {
+      "epoch": 1.99,
+      "grad_norm": 2.2901608785310743,
+      "learning_rate": 2.5867990443809373e-06,
+      "loss": 0.4385,
+      "step": 2290
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 2.2797959852072096,
+      "learning_rate": 2.5666821423070386e-06,
+      "loss": 0.4167,
+      "step": 2295
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 2.25422106532395,
+      "learning_rate": 2.546616722352321e-06,
+      "loss": 0.4267,
+      "step": 2300
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 3.078936861391817,
+      "learning_rate": 2.526603209044286e-06,
+      "loss": 0.2427,
+      "step": 2305
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 3.0338752203541777,
+      "learning_rate": 2.50664202581223e-06,
+      "loss": 0.2334,
+      "step": 2310
+    },
+    {
+      "epoch": 2.01,
+      "grad_norm": 2.6674179643190516,
+      "learning_rate": 2.486733594978298e-06,
+      "loss": 0.2044,
+      "step": 2315
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 2.5584075286035772,
+      "learning_rate": 2.466878337748541e-06,
+      "loss": 0.2142,
+      "step": 2320
+    },
+    {
+      "epoch": 2.02,
+      "grad_norm": 2.4273031380880057,
+      "learning_rate": 2.447076674204011e-06,
+      "loss": 0.2183,
+      "step": 2325
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.281976180794783,
+      "learning_rate": 2.427329023291864e-06,
+      "loss": 0.2129,
+      "step": 2330
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.3648221730843395,
+      "learning_rate": 2.407635802816506e-06,
+      "loss": 0.2317,
+      "step": 2335
+    },
+    {
+      "epoch": 2.03,
+      "grad_norm": 2.3911736141745537,
+      "learning_rate": 2.387997429430746e-06,
+      "loss": 0.2245,
+      "step": 2340
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 2.595042568929118,
+      "learning_rate": 2.3684143186269887e-06,
+      "loss": 0.2236,
+      "step": 2345
+    },
+    {
+      "epoch": 2.04,
+      "grad_norm": 2.4778116434984025,
+      "learning_rate": 2.3488868847284296e-06,
+      "loss": 0.2211,
+      "step": 2350
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 2.4910327415757756,
+      "learning_rate": 2.329415540880307e-06,
+      "loss": 0.2211,
+      "step": 2355
+    },
+    {
+      "epoch": 2.05,
+      "grad_norm": 2.4290998048021977,
+      "learning_rate": 2.3100006990411476e-06,
+      "loss": 0.2183,
+      "step": 2360
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 2.356104409371097,
+      "learning_rate": 2.290642769974063e-06,
+      "loss": 0.2151,
+      "step": 2365
+    },
+    {
+      "epoch": 2.06,
+      "grad_norm": 2.520455659230371,
+      "learning_rate": 2.271342163238041e-06,
+      "loss": 0.2299,
+      "step": 2370
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 2.33654416831311,
+      "learning_rate": 2.252099287179298e-06,
+      "loss": 0.2137,
+      "step": 2375
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 2.565006159671707,
+      "learning_rate": 2.2329145489226307e-06,
+      "loss": 0.2279,
+      "step": 2380
+    },
+    {
+      "epoch": 2.07,
+      "grad_norm": 2.5151919660325968,
+      "learning_rate": 2.2137883543628047e-06,
+      "loss": 0.2185,
+      "step": 2385
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 2.4666107209087187,
+      "learning_rate": 2.1947211081559666e-06,
+      "loss": 0.2073,
+      "step": 2390
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 2.4483308704122195,
+      "learning_rate": 2.1757132137110826e-06,
+      "loss": 0.2184,
+      "step": 2395
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 2.4022068639942367,
+      "learning_rate": 2.1567650731814045e-06,
+      "loss": 0.2216,
+      "step": 2400
+    },
+    {
+      "epoch": 2.09,
+      "grad_norm": 2.4475371432336663,
+      "learning_rate": 2.1378770874559607e-06,
+      "loss": 0.2049,
+      "step": 2405
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 2.5467512300119015,
+      "learning_rate": 2.1190496561510693e-06,
+      "loss": 0.2056,
+      "step": 2410
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 2.3365808578736518,
+      "learning_rate": 2.100283177601892e-06,
+      "loss": 0.2051,
+      "step": 2415
+    },
+    {
+      "epoch": 2.1,
+      "grad_norm": 2.3441486856497225,
+      "learning_rate": 2.081578048854007e-06,
+      "loss": 0.2138,
+      "step": 2420
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 2.635268753849708,
+      "learning_rate": 2.0629346656549996e-06,
+      "loss": 0.2185,
+      "step": 2425
+    },
+    {
+      "epoch": 2.11,
+      "grad_norm": 2.4875959650107458,
+      "learning_rate": 2.044353422446091e-06,
+      "loss": 0.2205,
+      "step": 2430
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 2.453382098035225,
+      "learning_rate": 2.0258347123538013e-06,
+      "loss": 0.2194,
+      "step": 2435
+    },
+    {
+      "epoch": 2.12,
+      "grad_norm": 2.5535540646032593,
+      "learning_rate": 2.0073789271816246e-06,
+      "loss": 0.225,
+      "step": 2440
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 2.3990094762895153,
+      "learning_rate": 1.9889864574017433e-06,
+      "loss": 0.2212,
+      "step": 2445
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 2.572526283977743,
+      "learning_rate": 1.9706576921467627e-06,
+      "loss": 0.2189,
+      "step": 2450
+    },
+    {
+      "epoch": 2.13,
+      "grad_norm": 2.4878028844723907,
+      "learning_rate": 1.952393019201484e-06,
+      "loss": 0.2266,
+      "step": 2455
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 2.3992447329696107,
+      "learning_rate": 1.934192824994694e-06,
+      "loss": 0.2011,
+      "step": 2460
+    },
+    {
+      "epoch": 2.14,
+      "grad_norm": 3.230696063743111,
+      "learning_rate": 1.916057494590994e-06,
+      "loss": 0.2138,
+      "step": 2465
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 2.4556476782365957,
+      "learning_rate": 1.8979874116826436e-06,
+      "loss": 0.2128,
+      "step": 2470
+    },
+    {
+      "epoch": 2.15,
+      "grad_norm": 2.5017117854146984,
+      "learning_rate": 1.8799829585814627e-06,
+      "loss": 0.2127,
+      "step": 2475
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 2.3513868209738744,
+      "learning_rate": 1.8620445162107204e-06,
+      "loss": 0.2047,
+      "step": 2480
+    },
+    {
+      "epoch": 2.16,
+      "grad_norm": 2.3299203264727852,
+      "learning_rate": 1.8441724640970904e-06,
+      "loss": 0.2101,
+      "step": 2485
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 2.4798119847923235,
+      "learning_rate": 1.826367180362612e-06,
+      "loss": 0.2171,
+      "step": 2490
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 2.538341375619424,
+      "learning_rate": 1.808629041716698e-06,
+      "loss": 0.2213,
+      "step": 2495
+    },
+    {
+      "epoch": 2.17,
+      "grad_norm": 2.472603273914458,
+      "learning_rate": 1.7909584234481591e-06,
+      "loss": 0.2141,
+      "step": 2500
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 2.4757251006111747,
+      "learning_rate": 1.7733556994172719e-06,
+      "loss": 0.2076,
+      "step": 2505
+    },
+    {
+      "epoch": 2.18,
+      "grad_norm": 2.4403116297970606,
+      "learning_rate": 1.7558212420478533e-06,
+      "loss": 0.2108,
+      "step": 2510
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 2.3937822033927123,
+      "learning_rate": 1.7383554223193977e-06,
+      "loss": 0.2183,
+      "step": 2515
+    },
+    {
+      "epoch": 2.19,
+      "grad_norm": 2.4812019825904854,
+      "learning_rate": 1.720958609759219e-06,
+      "loss": 0.2036,
+      "step": 2520
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 2.815408112564539,
+      "learning_rate": 1.703631172434636e-06,
+      "loss": 0.2217,
+      "step": 2525
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 2.561348994388801,
+      "learning_rate": 1.6863734769451822e-06,
+      "loss": 0.2194,
+      "step": 2530
+    },
+    {
+      "epoch": 2.2,
+      "grad_norm": 2.3832304794284993,
+      "learning_rate": 1.6691858884148527e-06,
+      "loss": 0.2063,
+      "step": 2535
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 2.4670748321250917,
+      "learning_rate": 1.6520687704843762e-06,
+      "loss": 0.213,
+      "step": 2540
+    },
+    {
+      "epoch": 2.21,
+      "grad_norm": 2.509183464514469,
+      "learning_rate": 1.6350224853035268e-06,
+      "loss": 0.2193,
+      "step": 2545
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 2.4993326189673923,
+      "learning_rate": 1.6180473935234508e-06,
+      "loss": 0.2091,
+      "step": 2550
+    },
+    {
+      "epoch": 2.22,
+      "grad_norm": 2.4702996837153424,
+      "learning_rate": 1.6011438542890484e-06,
+      "loss": 0.2254,
+      "step": 2555
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 2.557538672084199,
+      "learning_rate": 1.584312225231373e-06,
+      "loss": 0.2146,
+      "step": 2560
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 2.5982075216250307,
+      "learning_rate": 1.56755286246006e-06,
+      "loss": 0.2216,
+      "step": 2565
+    },
+    {
+      "epoch": 2.23,
+      "grad_norm": 2.4959152897024275,
+      "learning_rate": 1.5508661205557902e-06,
+      "loss": 0.2067,
+      "step": 2570
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 2.4985893772429204,
+      "learning_rate": 1.5342523525627973e-06,
+      "loss": 0.2214,
+      "step": 2575
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 2.409035587015272,
+      "learning_rate": 1.5177119099813925e-06,
+      "loss": 0.2147,
+      "step": 2580
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 2.5968730449121766,
+      "learning_rate": 1.5012451427605295e-06,
+      "loss": 0.2184,
+      "step": 2585
+    },
+    {
+      "epoch": 2.25,
+      "grad_norm": 2.5561487275537393,
+      "learning_rate": 1.484852399290398e-06,
+      "loss": 0.2052,
+      "step": 2590
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 2.632110553434636,
+      "learning_rate": 1.468534026395056e-06,
+      "loss": 0.2159,
+      "step": 2595
+    },
+    {
+      "epoch": 2.26,
+      "grad_norm": 2.6369225582838727,
+      "learning_rate": 1.4522903693250906e-06,
+      "loss": 0.2359,
+      "step": 2600
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 2.4423086751769447,
+      "learning_rate": 1.4361217717503145e-06,
+      "loss": 0.2131,
+      "step": 2605
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 2.468483025647043,
+      "learning_rate": 1.4200285757524896e-06,
+      "loss": 0.2159,
+      "step": 2610
+    },
+    {
+      "epoch": 2.27,
+      "grad_norm": 2.4975935128764726,
+      "learning_rate": 1.4040111218180968e-06,
+      "loss": 0.2034,
+      "step": 2615
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 2.5909634014951712,
+      "learning_rate": 1.388069748831133e-06,
+      "loss": 0.2242,
+      "step": 2620
+    },
+    {
+      "epoch": 2.28,
+      "grad_norm": 2.5919849822402985,
+      "learning_rate": 1.3722047940659328e-06,
+      "loss": 0.2194,
+      "step": 2625
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 2.455607537878878,
+      "learning_rate": 1.356416593180036e-06,
+      "loss": 0.2227,
+      "step": 2630
+    },
+    {
+      "epoch": 2.29,
+      "grad_norm": 2.5525735792403075,
+      "learning_rate": 1.3407054802070923e-06,
+      "loss": 0.2177,
+      "step": 2635
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.5763191988421417,
+      "learning_rate": 1.3250717875497865e-06,
+      "loss": 0.2033,
+      "step": 2640
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.5268479767549334,
+      "learning_rate": 1.3095158459728092e-06,
+      "loss": 0.2055,
+      "step": 2645
+    },
+    {
+      "epoch": 2.3,
+      "grad_norm": 2.5486803652093633,
+      "learning_rate": 1.294037984595859e-06,
+      "loss": 0.2083,
+      "step": 2650
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 2.566098393683622,
+      "learning_rate": 1.2786385308866772e-06,
+      "loss": 0.2154,
+      "step": 2655
+    },
+    {
+      "epoch": 2.31,
+      "grad_norm": 3.188245118367528,
+      "learning_rate": 1.2633178106541217e-06,
+      "loss": 0.2129,
+      "step": 2660
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 2.5401452854445794,
+      "learning_rate": 1.248076148041274e-06,
+      "loss": 0.204,
+      "step": 2665
+    },
+    {
+      "epoch": 2.32,
+      "grad_norm": 2.4149301452746132,
+      "learning_rate": 1.2329138655185736e-06,
+      "loss": 0.2119,
+      "step": 2670
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 2.4601647903220676,
+      "learning_rate": 1.2178312838770119e-06,
+      "loss": 0.212,
+      "step": 2675
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 2.503416524017546,
+      "learning_rate": 1.2028287222213286e-06,
+      "loss": 0.2095,
+      "step": 2680
+    },
+    {
+      "epoch": 2.33,
+      "grad_norm": 2.514672386421696,
+      "learning_rate": 1.1879064979632705e-06,
+      "loss": 0.209,
+      "step": 2685
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 2.589514084936205,
+      "learning_rate": 1.1730649268148663e-06,
+      "loss": 0.2134,
+      "step": 2690
+    },
+    {
+      "epoch": 2.34,
+      "grad_norm": 2.3972054242854246,
+      "learning_rate": 1.1583043227817608e-06,
+      "loss": 0.1996,
+      "step": 2695
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 2.9915330230427144,
+      "learning_rate": 1.1436249981565577e-06,
+      "loss": 0.2097,
+      "step": 2700
+    },
+    {
+      "epoch": 2.35,
+      "grad_norm": 2.4095232184813056,
+      "learning_rate": 1.1290272635122256e-06,
+      "loss": 0.2089,
+      "step": 2705
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 2.40401180034597,
+      "learning_rate": 1.114511427695512e-06,
+      "loss": 0.2043,
+      "step": 2710
+    },
+    {
+      "epoch": 2.36,
+      "grad_norm": 2.5990161443829143,
+      "learning_rate": 1.1000777978204214e-06,
+      "loss": 0.2138,
+      "step": 2715
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 2.412272176972184,
+      "learning_rate": 1.0857266792617122e-06,
+      "loss": 0.2129,
+      "step": 2720
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 2.490425041886623,
+      "learning_rate": 1.0714583756484382e-06,
+      "loss": 0.2081,
+      "step": 2725
+    },
+    {
+      "epoch": 2.37,
+      "grad_norm": 2.545431926941494,
+      "learning_rate": 1.057273188857521e-06,
+      "loss": 0.1969,
+      "step": 2730
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 2.5491354697389994,
+      "learning_rate": 1.0431714190073673e-06,
+      "loss": 0.2099,
+      "step": 2735
+    },
+    {
+      "epoch": 2.38,
+      "grad_norm": 2.535722843097943,
+      "learning_rate": 1.0291533644515167e-06,
+      "loss": 0.2188,
+      "step": 2740
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 2.4412268635568823,
+      "learning_rate": 1.0152193217723316e-06,
+      "loss": 0.2066,
+      "step": 2745
+    },
+    {
+      "epoch": 2.39,
+      "grad_norm": 2.4284105675604466,
+      "learning_rate": 1.0013695857747175e-06,
+      "loss": 0.2046,
+      "step": 2750
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 2.5647064944739415,
+      "learning_rate": 9.876044494798897e-07,
+      "loss": 0.2028,
+      "step": 2755
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 2.4550051723816115,
+      "learning_rate": 9.739242041191782e-07,
+      "loss": 0.2041,
+      "step": 2760
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 2.500586072069052,
+      "learning_rate": 9.60329139127857e-07,
+      "loss": 0.2034,
+      "step": 2765
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 2.486030234412699,
+      "learning_rate": 9.46819542139023e-07,
+      "loss": 0.2105,
+      "step": 2770
+    },
+    {
+      "epoch": 2.41,
+      "grad_norm": 2.47132818523371,
+      "learning_rate": 9.333956989775151e-07,
+      "loss": 0.2062,
+      "step": 2775
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 2.5162354138555383,
+      "learning_rate": 9.200578936538629e-07,
+      "loss": 0.2042,
+      "step": 2780
+    },
+    {
+      "epoch": 2.42,
+      "grad_norm": 2.6450756972218112,
+      "learning_rate": 9.06806408358279e-07,
+      "loss": 0.2068,
+      "step": 2785
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 2.51599107747933,
+      "learning_rate": 8.93641523454688e-07,
+      "loss": 0.2167,
+      "step": 2790
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 2.452817624703225,
+      "learning_rate": 8.805635174747962e-07,
+      "loss": 0.2112,
+      "step": 2795
+    },
+    {
+      "epoch": 2.43,
+      "grad_norm": 2.499623368832961,
+      "learning_rate": 8.67572667112197e-07,
+      "loss": 0.2088,
+      "step": 2800
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 2.4658751702963886,
+      "learning_rate": 8.546692472165197e-07,
+      "loss": 0.206,
+      "step": 2805
+    },
+    {
+      "epoch": 2.44,
+      "grad_norm": 2.611752158683941,
+      "learning_rate": 8.418535307876058e-07,
+      "loss": 0.2188,
+      "step": 2810
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 2.604795643200606,
+      "learning_rate": 8.291257889697485e-07,
+      "loss": 0.2034,
+      "step": 2815
+    },
+    {
+      "epoch": 2.45,
+      "grad_norm": 2.522685491865494,
+      "learning_rate": 8.16486291045942e-07,
+      "loss": 0.2092,
+      "step": 2820
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 2.674101175782956,
+      "learning_rate": 8.039353044321918e-07,
+      "loss": 0.2073,
+      "step": 2825
+    },
+    {
+      "epoch": 2.46,
+      "grad_norm": 2.580255680760457,
+      "learning_rate": 7.914730946718507e-07,
+      "loss": 0.2169,
+      "step": 2830
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 2.5015240037459154,
+      "learning_rate": 7.790999254300081e-07,
+      "loss": 0.2041,
+      "step": 2835
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 2.4392301749269687,
+      "learning_rate": 7.66816058487907e-07,
+      "loss": 0.2005,
+      "step": 2840
+    },
+    {
+      "epoch": 2.47,
+      "grad_norm": 2.3498058173364234,
+      "learning_rate": 7.546217537374073e-07,
+      "loss": 0.2026,
+      "step": 2845
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 2.4522810673328204,
+      "learning_rate": 7.425172691754851e-07,
+      "loss": 0.2123,
+      "step": 2850
+    },
+    {
+      "epoch": 2.48,
+      "grad_norm": 2.4816500907084356,
+      "learning_rate": 7.305028608987763e-07,
+      "loss": 0.2218,
+      "step": 2855
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 2.4611236830391565,
+      "learning_rate": 7.185787830981572e-07,
+      "loss": 0.2082,
+      "step": 2860
+    },
+    {
+      "epoch": 2.49,
+      "grad_norm": 2.6003284220480687,
+      "learning_rate": 7.067452880533665e-07,
+      "loss": 0.2076,
+      "step": 2865
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 2.4885849533798625,
+      "learning_rate": 6.950026261276699e-07,
+      "loss": 0.2173,
+      "step": 2870
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 2.5433749317969188,
+      "learning_rate": 6.833510457625586e-07,
+      "loss": 0.2183,
+      "step": 2875
+    },
+    {
+      "epoch": 2.5,
+      "grad_norm": 2.618342149652622,
+      "learning_rate": 6.717907934724982e-07,
+      "loss": 0.2072,
+      "step": 2880
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 2.470474590569744,
+      "learning_rate": 6.603221138397104e-07,
+      "loss": 0.2107,
+      "step": 2885
+    },
+    {
+      "epoch": 2.51,
+      "grad_norm": 2.56294794338806,
+      "learning_rate": 6.489452495089959e-07,
+      "loss": 0.2141,
+      "step": 2890
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 2.5169368315503546,
+      "learning_rate": 6.376604411826071e-07,
+      "loss": 0.2062,
+      "step": 2895
+    },
+    {
+      "epoch": 2.52,
+      "grad_norm": 2.57841805322415,
+      "learning_rate": 6.264679276151486e-07,
+      "loss": 0.2067,
+      "step": 2900
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 2.341138644238091,
+      "learning_rate": 6.153679456085344e-07,
+      "loss": 0.2003,
+      "step": 2905
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 2.5007286495912626,
+      "learning_rate": 6.043607300069654e-07,
+      "loss": 0.2106,
+      "step": 2910
+    },
+    {
+      "epoch": 2.53,
+      "grad_norm": 2.455825720793518,
+      "learning_rate": 5.934465136919737e-07,
+      "loss": 0.2083,
+      "step": 2915
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 2.5627510227859283,
+      "learning_rate": 5.826255275774861e-07,
+      "loss": 0.2049,
+      "step": 2920
+    },
+    {
+      "epoch": 2.54,
+      "grad_norm": 2.5412171669600037,
+      "learning_rate": 5.718980006049446e-07,
+      "loss": 0.2129,
+      "step": 2925
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 2.624753714196973,
+      "learning_rate": 5.612641597384588e-07,
+      "loss": 0.2017,
+      "step": 2930
+    },
+    {
+      "epoch": 2.55,
+      "grad_norm": 2.380115969904747,
+      "learning_rate": 5.507242299600063e-07,
+      "loss": 0.1942,
+      "step": 2935
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 2.3667094450521615,
+      "learning_rate": 5.402784342646711e-07,
+      "loss": 0.2042,
+      "step": 2940
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 2.4494439523093385,
+      "learning_rate": 5.299269936559276e-07,
+      "loss": 0.2082,
+      "step": 2945
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 2.51623944394463,
+      "learning_rate": 5.196701271409616e-07,
+      "loss": 0.2077,
+      "step": 2950
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 2.455262291358995,
+      "learning_rate": 5.095080517260398e-07,
+      "loss": 0.2076,
+      "step": 2955
+    },
+    {
+      "epoch": 2.57,
+      "grad_norm": 2.402969127726828,
+      "learning_rate": 4.994409824119189e-07,
+      "loss": 0.2104,
+      "step": 2960
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 2.5411078463886976,
+      "learning_rate": 4.894691321892947e-07,
+      "loss": 0.2036,
+      "step": 2965
+    },
+    {
+      "epoch": 2.58,
+      "grad_norm": 2.5610023367407013,
+      "learning_rate": 4.795927120342941e-07,
+      "loss": 0.2156,
+      "step": 2970
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 2.3831696328366587,
+      "learning_rate": 4.698119309040161e-07,
+      "loss": 0.2198,
+      "step": 2975
+    },
+    {
+      "epoch": 2.59,
+      "grad_norm": 2.451514189548151,
+      "learning_rate": 4.601269957321092e-07,
+      "loss": 0.2131,
+      "step": 2980
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 2.488054419848398,
+      "learning_rate": 4.5053811142439065e-07,
+      "loss": 0.2004,
+      "step": 2985
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 2.5153011259562548,
+      "learning_rate": 4.410454808545145e-07,
+      "loss": 0.2087,
+      "step": 2990
+    },
+    {
+      "epoch": 2.6,
+      "grad_norm": 2.625557424121124,
+      "learning_rate": 4.316493048596787e-07,
+      "loss": 0.2014,
+      "step": 2995
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 2.513699023438469,
+      "learning_rate": 4.2234978223637367e-07,
+      "loss": 0.2012,
+      "step": 3000
+    },
+    {
+      "epoch": 2.61,
+      "grad_norm": 2.438028070603665,
+      "learning_rate": 4.1314710973618e-07,
+      "loss": 0.201,
+      "step": 3005
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 2.4740778831225283,
+      "learning_rate": 4.0404148206160067e-07,
+      "loss": 0.2156,
+      "step": 3010
+    },
+    {
+      "epoch": 2.62,
+      "grad_norm": 2.5708421401806585,
+      "learning_rate": 3.9503309186194883e-07,
+      "loss": 0.2168,
+      "step": 3015
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 2.532678893774958,
+      "learning_rate": 3.861221297292656e-07,
+      "loss": 0.2096,
+      "step": 3020
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 2.4931735699463276,
+      "learning_rate": 3.7730878419429074e-07,
+      "loss": 0.212,
+      "step": 3025
+    },
+    {
+      "epoch": 2.63,
+      "grad_norm": 2.5099900710577647,
+      "learning_rate": 3.6859324172247024e-07,
+      "loss": 0.2219,
+      "step": 3030
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 2.4782098135793653,
+      "learning_rate": 3.599756867100185e-07,
+      "loss": 0.2155,
+      "step": 3035
+    },
+    {
+      "epoch": 2.64,
+      "grad_norm": 2.4326512505183318,
+      "learning_rate": 3.5145630148000984e-07,
+      "loss": 0.2107,
+      "step": 3040
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 2.3409116852369958,
+      "learning_rate": 3.4303526627852467e-07,
+      "loss": 0.1904,
+      "step": 3045
+    },
+    {
+      "epoch": 2.65,
+      "grad_norm": 2.4986649496444366,
+      "learning_rate": 3.3471275927083435e-07,
+      "loss": 0.1994,
+      "step": 3050
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 2.5621370225131526,
+      "learning_rate": 3.264889565376339e-07,
+      "loss": 0.2106,
+      "step": 3055
+    },
+    {
+      "epoch": 2.66,
+      "grad_norm": 2.397329927320774,
+      "learning_rate": 3.1836403207131384e-07,
+      "loss": 0.2039,
+      "step": 3060
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 2.5215478578818873,
+      "learning_rate": 3.103381577722814e-07,
+      "loss": 0.202,
+      "step": 3065
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 2.5156352432911664,
+      "learning_rate": 3.0241150344532235e-07,
+      "loss": 0.2179,
+      "step": 3070
+    },
+    {
+      "epoch": 2.67,
+      "grad_norm": 2.4892379086828145,
+      "learning_rate": 2.945842367960083e-07,
+      "loss": 0.1955,
+      "step": 3075
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 2.509401893058839,
+      "learning_rate": 2.8685652342714866e-07,
+      "loss": 0.1975,
+      "step": 3080
+    },
+    {
+      "epoch": 2.68,
+      "grad_norm": 2.4590377223696787,
+      "learning_rate": 2.79228526835289e-07,
+      "loss": 0.1955,
+      "step": 3085
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 2.508892056107057,
+      "learning_rate": 2.717004084072472e-07,
+      "loss": 0.211,
+      "step": 3090
+    },
+    {
+      "epoch": 2.69,
+      "grad_norm": 2.531986161518802,
+      "learning_rate": 2.642723274167036e-07,
+      "loss": 0.2136,
+      "step": 3095
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 2.4284067477323905,
+      "learning_rate": 2.5694444102082937e-07,
+      "loss": 0.2064,
+      "step": 3100
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 2.52242961694727,
+      "learning_rate": 2.4971690425696304e-07,
+      "loss": 0.2142,
+      "step": 3105
+    },
+    {
+      "epoch": 2.7,
+      "grad_norm": 2.3649373408065157,
+      "learning_rate": 2.425898700393253e-07,
+      "loss": 0.209,
+      "step": 3110
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 2.638699254233002,
+      "learning_rate": 2.3556348915579064e-07,
+      "loss": 0.2156,
+      "step": 3115
+    },
+    {
+      "epoch": 2.71,
+      "grad_norm": 2.5635088985639176,
+      "learning_rate": 2.286379102646924e-07,
+      "loss": 0.2032,
+      "step": 3120
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 2.524195852568805,
+      "learning_rate": 2.2181327989168e-07,
+      "loss": 0.2089,
+      "step": 3125
+    },
+    {
+      "epoch": 2.72,
+      "grad_norm": 2.5965127248630555,
+      "learning_rate": 2.150897424266163e-07,
+      "loss": 0.2225,
+      "step": 3130
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 2.5503575009918946,
+      "learning_rate": 2.084674401205261e-07,
+      "loss": 0.2067,
+      "step": 3135
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 2.3639921883945583,
+      "learning_rate": 2.0194651308258374e-07,
+      "loss": 0.2134,
+      "step": 3140
+    },
+    {
+      "epoch": 2.73,
+      "grad_norm": 2.3630908811493687,
+      "learning_rate": 1.9552709927715073e-07,
+      "loss": 0.1948,
+      "step": 3145
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 2.414848836181585,
+      "learning_rate": 1.8920933452085398e-07,
+      "loss": 0.2089,
+      "step": 3150
+    },
+    {
+      "epoch": 2.74,
+      "grad_norm": 2.5198185776025115,
+      "learning_rate": 1.829933524797156e-07,
+      "loss": 0.1925,
+      "step": 3155
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 2.3410805418099154,
+      "learning_rate": 1.7687928466632421e-07,
+      "loss": 0.2095,
+      "step": 3160
+    },
+    {
+      "epoch": 2.75,
+      "grad_norm": 2.412427602576252,
+      "learning_rate": 1.7086726043705093e-07,
+      "loss": 0.2007,
+      "step": 3165
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 2.4666092611278554,
+      "learning_rate": 1.6495740698931283e-07,
+      "loss": 0.2079,
+      "step": 3170
+    },
+    {
+      "epoch": 2.76,
+      "grad_norm": 2.4668785907422617,
+      "learning_rate": 1.5914984935888278e-07,
+      "loss": 0.2088,
+      "step": 3175
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 2.409143504334115,
+      "learning_rate": 1.5344471041724485e-07,
+      "loss": 0.2154,
+      "step": 3180
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 2.476783017513207,
+      "learning_rate": 1.4784211086899147e-07,
+      "loss": 0.2035,
+      "step": 3185
+    },
+    {
+      "epoch": 2.77,
+      "grad_norm": 2.3968478523564904,
+      "learning_rate": 1.423421692492738e-07,
+      "loss": 0.2123,
+      "step": 3190
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 2.6123943502190996,
+      "learning_rate": 1.369450019212898e-07,
+      "loss": 0.199,
+      "step": 3195
+    },
+    {
+      "epoch": 2.78,
+      "grad_norm": 2.5622025492517975,
+      "learning_rate": 1.316507230738262e-07,
+      "loss": 0.2148,
+      "step": 3200
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 2.4812824980330186,
+      "learning_rate": 1.2645944471883997e-07,
+      "loss": 0.2164,
+      "step": 3205
+    },
+    {
+      "epoch": 2.79,
+      "grad_norm": 2.560828799518852,
+      "learning_rate": 1.2137127668908733e-07,
+      "loss": 0.2007,
+      "step": 3210
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 2.4257116665392915,
+      "learning_rate": 1.1638632663580452e-07,
+      "loss": 0.1984,
+      "step": 3215
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 2.391682340371937,
+      "learning_rate": 1.1150470002642689e-07,
+      "loss": 0.203,
+      "step": 3220
+    },
+    {
+      "epoch": 2.8,
+      "grad_norm": 2.545462773299088,
+      "learning_rate": 1.067265001423573e-07,
+      "loss": 0.2189,
+      "step": 3225
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 2.5517528532547398,
+      "learning_rate": 1.0205182807678182e-07,
+      "loss": 0.2092,
+      "step": 3230
+    },
+    {
+      "epoch": 2.81,
+      "grad_norm": 2.6890101372722524,
+      "learning_rate": 9.748078273253137e-08,
+      "loss": 0.1998,
+      "step": 3235
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 2.5064597040050955,
+      "learning_rate": 9.30134608199884e-08,
+      "loss": 0.2126,
+      "step": 3240
+    },
+    {
+      "epoch": 2.82,
+      "grad_norm": 2.442464216341298,
+      "learning_rate": 8.864995685504252e-08,
+      "loss": 0.2036,
+      "step": 3245
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 2.4567338439277835,
+      "learning_rate": 8.439036315708693e-08,
+      "loss": 0.2089,
+      "step": 3250
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 2.4443306161494647,
+      "learning_rate": 8.023476984706957e-08,
+      "loss": 0.2079,
+      "step": 3255
+    },
+    {
+      "epoch": 2.83,
+      "grad_norm": 2.4195988191572555,
+      "learning_rate": 7.618326484558402e-08,
+      "loss": 0.2104,
+      "step": 3260
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 2.5824717459642725,
+      "learning_rate": 7.22359338710088e-08,
+      "loss": 0.2132,
+      "step": 3265
+    },
+    {
+      "epoch": 2.84,
+      "grad_norm": 2.4494090419728174,
+      "learning_rate": 6.839286043769655e-08,
+      "loss": 0.2028,
+      "step": 3270
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 2.515432606332294,
+      "learning_rate": 6.465412585420439e-08,
+      "loss": 0.2038,
+      "step": 3275
+    },
+    {
+      "epoch": 2.85,
+      "grad_norm": 2.426424214341233,
+      "learning_rate": 6.101980922157524e-08,
+      "loss": 0.2005,
+      "step": 3280
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 2.636637920666506,
+      "learning_rate": 5.748998743166256e-08,
+      "loss": 0.215,
+      "step": 3285
+    },
+    {
+      "epoch": 2.86,
+      "grad_norm": 2.467390908234526,
+      "learning_rate": 5.4064735165506035e-08,
+      "loss": 0.2026,
+      "step": 3290
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 2.4543170202618123,
+      "learning_rate": 5.0744124891748956e-08,
+      "loss": 0.201,
+      "step": 3295
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 2.5570061460145825,
+      "learning_rate": 4.7528226865107274e-08,
+      "loss": 0.2116,
+      "step": 3300
+    },
+    {
+      "epoch": 2.87,
+      "grad_norm": 2.414663905945939,
+      "learning_rate": 4.4417109124882394e-08,
+      "loss": 0.2074,
+      "step": 3305
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 2.480556770645684,
+      "learning_rate": 4.1410837493519595e-08,
+      "loss": 0.2197,
+      "step": 3310
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 2.320041349020439,
+      "learning_rate": 3.8509475575219115e-08,
+      "loss": 0.1922,
+      "step": 3315
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 2.460634533640002,
+      "learning_rate": 3.571308475458723e-08,
+      "loss": 0.205,
+      "step": 3320
+    },
+    {
+      "epoch": 2.89,
+      "grad_norm": 2.4951723691188183,
+      "learning_rate": 3.302172419534011e-08,
+      "loss": 0.2005,
+      "step": 3325
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 2.626459593341851,
+      "learning_rate": 3.04354508390492e-08,
+      "loss": 0.2073,
+      "step": 3330
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 2.484067935611207,
+      "learning_rate": 2.7954319403940555e-08,
+      "loss": 0.2045,
+      "step": 3335
+    },
+    {
+      "epoch": 2.9,
+      "grad_norm": 2.6177507146061743,
+      "learning_rate": 2.5578382383732446e-08,
+      "loss": 0.2014,
+      "step": 3340
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 2.3815239798499728,
+      "learning_rate": 2.3307690046527887e-08,
+      "loss": 0.206,
+      "step": 3345
+    },
+    {
+      "epoch": 2.91,
+      "grad_norm": 2.4449110836459993,
+      "learning_rate": 2.114229043375049e-08,
+      "loss": 0.2001,
+      "step": 3350
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 2.5672009191881933,
+      "learning_rate": 1.9082229359127512e-08,
+      "loss": 0.2051,
+      "step": 3355
+    },
+    {
+      "epoch": 2.92,
+      "grad_norm": 2.3860476035663334,
+      "learning_rate": 1.7127550407721184e-08,
+      "loss": 0.2014,
+      "step": 3360
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 2.4564198065905374,
+      "learning_rate": 1.52782949350061e-08,
+      "loss": 0.2012,
+      "step": 3365
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 2.5449218424263855,
+      "learning_rate": 1.3534502065993826e-08,
+      "loss": 0.2168,
+      "step": 3370
+    },
+    {
+      "epoch": 2.93,
+      "grad_norm": 2.464276506097749,
+      "learning_rate": 1.1896208694406886e-08,
+      "loss": 0.2058,
+      "step": 3375
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 2.5854288843754825,
+      "learning_rate": 1.0363449481896604e-08,
+      "loss": 0.2157,
+      "step": 3380
+    },
+    {
+      "epoch": 2.94,
+      "grad_norm": 2.4558545798988454,
+      "learning_rate": 8.936256857308701e-09,
+      "loss": 0.2058,
+      "step": 3385
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 2.4507285622247257,
+      "learning_rate": 7.614661016001056e-09,
+      "loss": 0.2081,
+      "step": 3390
+    },
+    {
+      "epoch": 2.95,
+      "grad_norm": 2.3645967407276123,
+      "learning_rate": 6.398689919201451e-09,
+      "loss": 0.1973,
+      "step": 3395
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 2.389789682646647,
+      "learning_rate": 5.288369293415807e-09,
+      "loss": 0.1935,
+      "step": 3400
+    },
+    {
+      "epoch": 2.96,
+      "grad_norm": 2.4222018767284457,
+      "learning_rate": 4.283722629887521e-09,
+      "loss": 0.2011,
+      "step": 3405
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.5358338089932335,
+      "learning_rate": 3.3847711840950813e-09,
+      "loss": 0.2005,
+      "step": 3410
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.5006583397784663,
+      "learning_rate": 2.5915339753085355e-09,
+      "loss": 0.2115,
+      "step": 3415
+    },
+    {
+      "epoch": 2.97,
+      "grad_norm": 2.432870342062548,
+      "learning_rate": 1.9040277861814836e-09,
+      "loss": 0.2138,
+      "step": 3420
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.4313911378435606,
+      "learning_rate": 1.3222671623991379e-09,
+      "loss": 0.2131,
+      "step": 3425
+    },
+    {
+      "epoch": 2.98,
+      "grad_norm": 2.510956565946373,
+      "learning_rate": 8.462644123696794e-10,
+      "loss": 0.2008,
+      "step": 3430
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.44361011474203,
+      "learning_rate": 4.760296069639125e-10,
+      "loss": 0.1937,
+      "step": 3435
+    },
+    {
+      "epoch": 2.99,
+      "grad_norm": 2.6119833553202962,
+      "learning_rate": 2.1157057930321079e-10,
+      "loss": 0.2067,
+      "step": 3440
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.4761340158997815,
+      "learning_rate": 5.289292459187412e-11,
+      "loss": 0.205,
+      "step": 3445
+    },
+    {
+      "epoch": 3.0,
+      "grad_norm": 2.4290543490047325,
+      "learning_rate": 0.0,
+      "loss": 0.2061,
+      "step": 3450
+    },
+    {
+      "epoch": 3.0,
+      "step": 3450,
+      "total_flos": 722333639639040.0,
+      "train_loss": 0.45598918040593467,
+      "train_runtime": 87085.3133,
+      "train_samples_per_second": 2.535,
+      "train_steps_per_second": 0.04
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3450,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "total_flos": 722333639639040.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}