diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5633 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 4000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.00125,
+      "grad_norm": 5.53125,
+      "learning_rate": 4.999980723453676e-05,
+      "loss": 0.6803,
+      "step": 5
+    },
+    {
+      "epoch": 0.0025,
+      "grad_norm": 3.0,
+      "learning_rate": 4.9999228941119745e-05,
+      "loss": 0.6176,
+      "step": 10
+    },
+    {
+      "epoch": 0.00375,
+      "grad_norm": 1.8984375,
+      "learning_rate": 4.999826512866693e-05,
+      "loss": 0.5313,
+      "step": 15
+    },
+    {
+      "epoch": 0.005,
+      "grad_norm": 1.6015625,
+      "learning_rate": 4.999691581204152e-05,
+      "loss": 0.5659,
+      "step": 20
+    },
+    {
+      "epoch": 0.00625,
+      "grad_norm": 1.5703125,
+      "learning_rate": 4.9995181012051625e-05,
+      "loss": 0.5822,
+      "step": 25
+    },
+    {
+      "epoch": 0.0075,
+      "grad_norm": 1.515625,
+      "learning_rate": 4.9993060755450015e-05,
+      "loss": 0.5849,
+      "step": 30
+    },
+    {
+      "epoch": 0.00875,
+      "grad_norm": 1.4375,
+      "learning_rate": 4.999055507493368e-05,
+      "loss": 0.55,
+      "step": 35
+    },
+    {
+      "epoch": 0.01,
+      "grad_norm": 1.515625,
+      "learning_rate": 4.998766400914329e-05,
+      "loss": 0.5323,
+      "step": 40
+    },
+    {
+      "epoch": 0.01125,
+      "grad_norm": 1.6640625,
+      "learning_rate": 4.9984387602662675e-05,
+      "loss": 0.5775,
+      "step": 45
+    },
+    {
+      "epoch": 0.0125,
+      "grad_norm": 1.8359375,
+      "learning_rate": 4.9980725906018074e-05,
+      "loss": 0.5523,
+      "step": 50
+    },
+    {
+      "epoch": 0.01375,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.9976678975677376e-05,
+      "loss": 0.6089,
+      "step": 55
+    },
+    {
+      "epoch": 0.015,
+      "grad_norm": 1.40625,
+      "learning_rate": 4.9972246874049254e-05,
+      "loss": 0.5445,
+      "step": 60
+    },
+    {
+      "epoch": 0.01625,
+      "grad_norm": 1.375,
+      "learning_rate": 4.996742966948219e-05,
+      "loss": 0.5256,
+      "step": 65
+    },
+    {
+      "epoch": 0.0175,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.9962227436263453e-05,
+      "loss": 0.5118,
+      "step": 70
+    },
+    {
+      "epoch": 0.01875,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.9956640254617906e-05,
+      "loss": 0.5458,
+      "step": 75
+    },
+    {
+      "epoch": 0.02,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.995066821070679e-05,
+      "loss": 0.5946,
+      "step": 80
+    },
+    {
+      "epoch": 0.02125,
+      "grad_norm": 1.4375,
+      "learning_rate": 4.994431139662643e-05,
+      "loss": 0.515,
+      "step": 85
+    },
+    {
+      "epoch": 0.0225,
+      "grad_norm": 1.4140625,
+      "learning_rate": 4.9937569910406756e-05,
+      "loss": 0.5501,
+      "step": 90
+    },
+    {
+      "epoch": 0.02375,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.9930443856009826e-05,
+      "loss": 0.5475,
+      "step": 95
+    },
+    {
+      "epoch": 0.025,
+      "grad_norm": 1.46875,
+      "learning_rate": 4.99229333433282e-05,
+      "loss": 0.5625,
+      "step": 100
+    },
+    {
+      "epoch": 0.02625,
+      "grad_norm": 1.578125,
+      "learning_rate": 4.9915038488183295e-05,
+      "loss": 0.5627,
+      "step": 105
+    },
+    {
+      "epoch": 0.0275,
+      "grad_norm": 1.640625,
+      "learning_rate": 4.990675941232353e-05,
+      "loss": 0.5561,
+      "step": 110
+    },
+    {
+      "epoch": 0.02875,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.989809624342251e-05,
+      "loss": 0.5254,
+      "step": 115
+    },
+    {
+      "epoch": 0.03,
+      "grad_norm": 1.0859375,
+      "learning_rate": 4.9889049115077005e-05,
+      "loss": 0.5184,
+      "step": 120
+    },
+    {
+      "epoch": 0.03125,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.987961816680492e-05,
+      "loss": 0.5563,
+      "step": 125
+    },
+    {
+      "epoch": 0.0325,
+      "grad_norm": 1.6875,
+      "learning_rate": 4.9869803544043166e-05,
+      "loss": 0.5536,
+      "step": 130
+    },
+    {
+      "epoch": 0.03375,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.985960539814535e-05,
+      "loss": 0.5544,
+      "step": 135
+    },
+    {
+      "epoch": 0.035,
+      "grad_norm": 1.453125,
+      "learning_rate": 4.98490238863795e-05,
+      "loss": 0.5117,
+      "step": 140
+    },
+    {
+      "epoch": 0.03625,
+      "grad_norm": 1.421875,
+      "learning_rate": 4.983805917192561e-05,
+      "loss": 0.5125,
+      "step": 145
+    },
+    {
+      "epoch": 0.0375,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.982671142387316e-05,
+      "loss": 0.5563,
+      "step": 150
+    },
+    {
+      "epoch": 0.03875,
+      "grad_norm": 1.375,
+      "learning_rate": 4.9814980817218447e-05,
+      "loss": 0.5408,
+      "step": 155
+    },
+    {
+      "epoch": 0.04,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.980286753286195e-05,
+      "loss": 0.5249,
+      "step": 160
+    },
+    {
+      "epoch": 0.04125,
+      "grad_norm": 1.546875,
+      "learning_rate": 4.979037175760548e-05,
+      "loss": 0.546,
+      "step": 165
+    },
+    {
+      "epoch": 0.0425,
+      "grad_norm": 1.28125,
+      "learning_rate": 4.9777493684149375e-05,
+      "loss": 0.5019,
+      "step": 170
+    },
+    {
+      "epoch": 0.04375,
+      "grad_norm": 1.453125,
+      "learning_rate": 4.976423351108943e-05,
+      "loss": 0.5364,
+      "step": 175
+    },
+    {
+      "epoch": 0.045,
+      "grad_norm": 1.5234375,
+      "learning_rate": 4.975059144291394e-05,
+      "loss": 0.5504,
+      "step": 180
+    },
+    {
+      "epoch": 0.04625,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.973656769000046e-05,
+      "loss": 0.4682,
+      "step": 185
+    },
+    {
+      "epoch": 0.0475,
+      "grad_norm": 1.328125,
+      "learning_rate": 4.972216246861262e-05,
+      "loss": 0.5262,
+      "step": 190
+    },
+    {
+      "epoch": 0.04875,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.9707376000896736e-05,
+      "loss": 0.5343,
+      "step": 195
+    },
+    {
+      "epoch": 0.05,
+      "grad_norm": 1.3671875,
+      "learning_rate": 4.9692208514878444e-05,
+      "loss": 0.5171,
+      "step": 200
+    },
+    {
+      "epoch": 0.05125,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.967666024445914e-05,
+      "loss": 0.5454,
+      "step": 205
+    },
+    {
+      "epoch": 0.0525,
+      "grad_norm": 1.25,
+      "learning_rate": 4.966073142941239e-05,
+      "loss": 0.5378,
+      "step": 210
+    },
+    {
+      "epoch": 0.05375,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.9644422315380225e-05,
+      "loss": 0.4792,
+      "step": 215
+    },
+    {
+      "epoch": 0.055,
+      "grad_norm": 1.46875,
+      "learning_rate": 4.962773315386935e-05,
+      "loss": 0.5336,
+      "step": 220
+    },
+    {
+      "epoch": 0.05625,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.9610664202247294e-05,
+      "loss": 0.5293,
+      "step": 225
+    },
+    {
+      "epoch": 0.0575,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.9593215723738404e-05,
+      "loss": 0.4678,
+      "step": 230
+    },
+    {
+      "epoch": 0.05875,
+      "grad_norm": 1.421875,
+      "learning_rate": 4.957538798741979e-05,
+      "loss": 0.549,
+      "step": 235
+    },
+    {
+      "epoch": 0.06,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.9557181268217227e-05,
+      "loss": 0.5642,
+      "step": 240
+    },
+    {
+      "epoch": 0.06125,
+      "grad_norm": 1.2734375,
+      "learning_rate": 4.953859584690082e-05,
+      "loss": 0.5544,
+      "step": 245
+    },
+    {
+      "epoch": 0.0625,
+      "grad_norm": 1.25,
+      "learning_rate": 4.951963201008076e-05,
+      "loss": 0.5604,
+      "step": 250
+    },
+    {
+      "epoch": 0.06375,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.9500290050202894e-05,
+      "loss": 0.5349,
+      "step": 255
+    },
+    {
+      "epoch": 0.065,
+      "grad_norm": 1.5234375,
+      "learning_rate": 4.9480570265544144e-05,
+      "loss": 0.5393,
+      "step": 260
+    },
+    {
+      "epoch": 0.06625,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.9460472960208e-05,
+      "loss": 0.527,
+      "step": 265
+    },
+    {
+      "epoch": 0.0675,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.943999844411977e-05,
+      "loss": 0.4982,
+      "step": 270
+    },
+    {
+      "epoch": 0.06875,
+      "grad_norm": 0.9765625,
+      "learning_rate": 4.9419147033021814e-05,
+      "loss": 0.4377,
+      "step": 275
+    },
+    {
+      "epoch": 0.07,
+      "grad_norm": 1.453125,
+      "learning_rate": 4.939791904846869e-05,
+      "loss": 0.4919,
+      "step": 280
+    },
+    {
+      "epoch": 0.07125,
+      "grad_norm": 1.7421875,
+      "learning_rate": 4.937631481782218e-05,
+      "loss": 0.5107,
+      "step": 285
+    },
+    {
+      "epoch": 0.0725,
+      "grad_norm": 1.4921875,
+      "learning_rate": 4.935433467424624e-05,
+      "loss": 0.5611,
+      "step": 290
+    },
+    {
+      "epoch": 0.07375,
+      "grad_norm": 1.8984375,
+      "learning_rate": 4.9331978956701875e-05,
+      "loss": 0.534,
+      "step": 295
+    },
+    {
+      "epoch": 0.075,
+      "grad_norm": 1.3515625,
+      "learning_rate": 4.9309248009941914e-05,
+      "loss": 0.5319,
+      "step": 300
+    },
+    {
+      "epoch": 0.07625,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.928614218450568e-05,
+      "loss": 0.4805,
+      "step": 305
+    },
+    {
+      "epoch": 0.0775,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.9262661836713564e-05,
+      "loss": 0.4656,
+      "step": 310
+    },
+    {
+      "epoch": 0.07875,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.923880732866159e-05,
+      "loss": 0.5328,
+      "step": 315
+    },
+    {
+      "epoch": 0.08,
+      "grad_norm": 1.0,
+      "learning_rate": 4.9214579028215776e-05,
+      "loss": 0.4949,
+      "step": 320
+    },
+    {
+      "epoch": 0.08125,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.9189977309006495e-05,
+      "loss": 0.5222,
+      "step": 325
+    },
+    {
+      "epoch": 0.0825,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.916500255042268e-05,
+      "loss": 0.5029,
+      "step": 330
+    },
+    {
+      "epoch": 0.08375,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.9139655137606015e-05,
+      "loss": 0.5188,
+      "step": 335
+    },
+    {
+      "epoch": 0.085,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.9113935461444955e-05,
+      "loss": 0.5651,
+      "step": 340
+    },
+    {
+      "epoch": 0.08625,
+      "grad_norm": 1.046875,
+      "learning_rate": 4.908784391856872e-05,
+      "loss": 0.4586,
+      "step": 345
+    },
+    {
+      "epoch": 0.0875,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.906138091134118e-05,
+      "loss": 0.539,
+      "step": 350
+    },
+    {
+      "epoch": 0.08875,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.9034546847854656e-05,
+      "loss": 0.5331,
+      "step": 355
+    },
+    {
+      "epoch": 0.09,
+      "grad_norm": 1.015625,
+      "learning_rate": 4.900734214192358e-05,
+      "loss": 0.4227,
+      "step": 360
+    },
+    {
+      "epoch": 0.09125,
+      "grad_norm": 1.28125,
+      "learning_rate": 4.897976721307819e-05,
+      "loss": 0.5005,
+      "step": 365
+    },
+    {
+      "epoch": 0.0925,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.8951822486557986e-05,
+      "loss": 0.5294,
+      "step": 370
+    },
+    {
+      "epoch": 0.09375,
+      "grad_norm": 1.375,
+      "learning_rate": 4.892350839330522e-05,
+      "loss": 0.5729,
+      "step": 375
+    },
+    {
+      "epoch": 0.095,
+      "grad_norm": 0.95703125,
+      "learning_rate": 4.8894825369958255e-05,
+      "loss": 0.4837,
+      "step": 380
+    },
+    {
+      "epoch": 0.09625,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.8865773858844776e-05,
+      "loss": 0.5266,
+      "step": 385
+    },
+    {
+      "epoch": 0.0975,
+      "grad_norm": 1.375,
+      "learning_rate": 4.8836354307975026e-05,
+      "loss": 0.5329,
+      "step": 390
+    },
+    {
+      "epoch": 0.09875,
+      "grad_norm": 1.4921875,
+      "learning_rate": 4.880656717103489e-05,
+      "loss": 0.5096,
+      "step": 395
+    },
+    {
+      "epoch": 0.1,
+      "grad_norm": 1.171875,
+      "learning_rate": 4.877641290737884e-05,
+      "loss": 0.4919,
+      "step": 400
+    },
+    {
+      "epoch": 0.10125,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.874589198202294e-05,
+      "loss": 0.4633,
+      "step": 405
+    },
+    {
+      "epoch": 0.1025,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.8715004865637614e-05,
+      "loss": 0.4981,
+      "step": 410
+    },
+    {
+      "epoch": 0.10375,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.868375203454041e-05,
+      "loss": 0.4699,
+      "step": 415
+    },
+    {
+      "epoch": 0.105,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.8652133970688636e-05,
+      "loss": 0.5086,
+      "step": 420
+    },
+    {
+      "epoch": 0.10625,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.862015116167196e-05,
+      "loss": 0.5406,
+      "step": 425
+    },
+    {
+      "epoch": 0.1075,
+      "grad_norm": 1.640625,
+      "learning_rate": 4.8587804100704845e-05,
+      "loss": 0.5456,
+      "step": 430
+    },
+    {
+      "epoch": 0.10875,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.8555093286618995e-05,
+      "loss": 0.5107,
+      "step": 435
+    },
+    {
+      "epoch": 0.11,
+      "grad_norm": 1.34375,
+      "learning_rate": 4.852201922385564e-05,
+      "loss": 0.4078,
+      "step": 440
+    },
+    {
+      "epoch": 0.11125,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.848858242245773e-05,
+      "loss": 0.4958,
+      "step": 445
+    },
+    {
+      "epoch": 0.1125,
+      "grad_norm": 1.4765625,
+      "learning_rate": 4.8454783398062106e-05,
+      "loss": 0.4822,
+      "step": 450
+    },
+    {
+      "epoch": 0.11375,
+      "grad_norm": 1.546875,
+      "learning_rate": 4.8420622671891533e-05,
+      "loss": 0.5489,
+      "step": 455
+    },
+    {
+      "epoch": 0.115,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.838610077074669e-05,
+      "loss": 0.4884,
+      "step": 460
+    },
+    {
+      "epoch": 0.11625,
+      "grad_norm": 1.609375,
+      "learning_rate": 4.835121822699796e-05,
+      "loss": 0.529,
+      "step": 465
+    },
+    {
+      "epoch": 0.1175,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.8315975578577355e-05,
+      "loss": 0.5414,
+      "step": 470
+    },
+    {
+      "epoch": 0.11875,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.828037336897009e-05,
+      "loss": 0.4749,
+      "step": 475
+    },
+    {
+      "epoch": 0.12,
+      "grad_norm": 0.9765625,
+      "learning_rate": 4.8244412147206284e-05,
+      "loss": 0.5205,
+      "step": 480
+    },
+    {
+      "epoch": 0.12125,
+      "grad_norm": 1.2890625,
+      "learning_rate": 4.820809246785247e-05,
+      "loss": 0.5343,
+      "step": 485
+    },
+    {
+      "epoch": 0.1225,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.817141489100302e-05,
+      "loss": 0.5324,
+      "step": 490
+    },
+    {
+      "epoch": 0.12375,
+      "grad_norm": 1.546875,
+      "learning_rate": 4.8134379982271556e-05,
+      "loss": 0.5451,
+      "step": 495
+    },
+    {
+      "epoch": 0.125,
+      "grad_norm": 1.390625,
+      "learning_rate": 4.8096988312782174e-05,
+      "loss": 0.5428,
+      "step": 500
+    },
+    {
+      "epoch": 0.12625,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.805924045916067e-05,
+      "loss": 0.5002,
+      "step": 505
+    },
+    {
+      "epoch": 0.1275,
+      "grad_norm": 1.2890625,
+      "learning_rate": 4.8021137003525664e-05,
+      "loss": 0.5277,
+      "step": 510
+    },
+    {
+      "epoch": 0.12875,
+      "grad_norm": 1.3125,
+      "learning_rate": 4.7982678533479555e-05,
+      "loss": 0.5185,
+      "step": 515
+    },
+    {
+      "epoch": 0.13,
+      "grad_norm": 1.3125,
+      "learning_rate": 4.794386564209953e-05,
+      "loss": 0.501,
+      "step": 520
+    },
+    {
+      "epoch": 0.13125,
+      "grad_norm": 1.4453125,
+      "learning_rate": 4.7904698927928406e-05,
+      "loss": 0.4903,
+      "step": 525
+    },
+    {
+      "epoch": 0.1325,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.7865178994965344e-05,
+      "loss": 0.4764,
+      "step": 530
+    },
+    {
+      "epoch": 0.13375,
+      "grad_norm": 1.3515625,
+      "learning_rate": 4.782530645265661e-05,
+      "loss": 0.5046,
+      "step": 535
+    },
+    {
+      "epoch": 0.135,
+      "grad_norm": 1.25,
+      "learning_rate": 4.7785081915886134e-05,
+      "loss": 0.4849,
+      "step": 540
+    },
+    {
+      "epoch": 0.13625,
+      "grad_norm": 1.4375,
+      "learning_rate": 4.7744506004966025e-05,
+      "loss": 0.4874,
+      "step": 545
+    },
+    {
+      "epoch": 0.1375,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.7703579345627035e-05,
+      "loss": 0.5632,
+      "step": 550
+    },
+    {
+      "epoch": 0.13875,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.766230256900887e-05,
+      "loss": 0.4894,
+      "step": 555
+    },
+    {
+      "epoch": 0.14,
+      "grad_norm": 0.9140625,
+      "learning_rate": 4.762067631165049e-05,
+      "loss": 0.4819,
+      "step": 560
+    },
+    {
+      "epoch": 0.14125,
+      "grad_norm": 1.3203125,
+      "learning_rate": 4.7578701215480284e-05,
+      "loss": 0.4872,
+      "step": 565
+    },
+    {
+      "epoch": 0.1425,
+      "grad_norm": 1.6328125,
+      "learning_rate": 4.753637792780614e-05,
+      "loss": 0.5274,
+      "step": 570
+    },
+    {
+      "epoch": 0.14375,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.749370710130554e-05,
+      "loss": 0.5052,
+      "step": 575
+    },
+    {
+      "epoch": 0.145,
+      "grad_norm": 1.2109375,
+      "learning_rate": 4.745068939401539e-05,
+      "loss": 0.4819,
+      "step": 580
+    },
+    {
+      "epoch": 0.14625,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.740732546932197e-05,
+      "loss": 0.5159,
+      "step": 585
+    },
+    {
+      "epoch": 0.1475,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.7363615995950626e-05,
+      "loss": 0.5338,
+      "step": 590
+    },
+    {
+      "epoch": 0.14875,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.7319561647955526e-05,
+      "loss": 0.4797,
+      "step": 595
+    },
+    {
+      "epoch": 0.15,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.72751631047092e-05,
+      "loss": 0.5453,
+      "step": 600
+    },
+    {
+      "epoch": 0.15125,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.7230421050892116e-05,
+      "loss": 0.5009,
+      "step": 605
+    },
+    {
+      "epoch": 0.1525,
+      "grad_norm": 1.5625,
+      "learning_rate": 4.718533617648209e-05,
+      "loss": 0.4602,
+      "step": 610
+    },
+    {
+      "epoch": 0.15375,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.713990917674365e-05,
+      "loss": 0.5399,
+      "step": 615
+    },
+    {
+      "epoch": 0.155,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.709414075221734e-05,
+      "loss": 0.5006,
+      "step": 620
+    },
+    {
+      "epoch": 0.15625,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.7048031608708876e-05,
+      "loss": 0.4784,
+      "step": 625
+    },
+    {
+      "epoch": 0.1575,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.7001582457278304e-05,
+      "loss": 0.4764,
+      "step": 630
+    },
+    {
+      "epoch": 0.15875,
+      "grad_norm": 1.03125,
+      "learning_rate": 4.695479401422898e-05,
+      "loss": 0.5003,
+      "step": 635
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.9921875,
+      "learning_rate": 4.690766700109659e-05,
+      "loss": 0.4586,
+      "step": 640
+    },
+    {
+      "epoch": 0.16125,
+      "grad_norm": 1.5234375,
+      "learning_rate": 4.686020214463798e-05,
+      "loss": 0.5272,
+      "step": 645
+    },
+    {
+      "epoch": 0.1625,
+      "grad_norm": 1.3203125,
+      "learning_rate": 4.681240017681993e-05,
+      "loss": 0.5626,
+      "step": 650
+    },
+    {
+      "epoch": 0.16375,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.676426183480794e-05,
+      "loss": 0.5696,
+      "step": 655
+    },
+    {
+      "epoch": 0.165,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.671578786095478e-05,
+      "loss": 0.5391,
+      "step": 660
+    },
+    {
+      "epoch": 0.16625,
+      "grad_norm": 1.1484375,
+      "learning_rate": 4.6666979002789105e-05,
+      "loss": 0.5195,
+      "step": 665
+    },
+    {
+      "epoch": 0.1675,
+      "grad_norm": 1.5,
+      "learning_rate": 4.661783601300388e-05,
+      "loss": 0.4973,
+      "step": 670
+    },
+    {
+      "epoch": 0.16875,
+      "grad_norm": 1.2890625,
+      "learning_rate": 4.65683596494448e-05,
+      "loss": 0.4741,
+      "step": 675
+    },
+    {
+      "epoch": 0.17,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.65185506750986e-05,
+      "loss": 0.4684,
+      "step": 680
+    },
+    {
+      "epoch": 0.17125,
+      "grad_norm": 1.5390625,
+      "learning_rate": 4.646840985808126e-05,
+      "loss": 0.5307,
+      "step": 685
+    },
+    {
+      "epoch": 0.1725,
+      "grad_norm": 1.296875,
+      "learning_rate": 4.6417937971626245e-05,
+      "loss": 0.5154,
+      "step": 690
+    },
+    {
+      "epoch": 0.17375,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.636713579407245e-05,
+      "loss": 0.5348,
+      "step": 695
+    },
+    {
+      "epoch": 0.175,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.6316004108852305e-05,
+      "loss": 0.477,
+      "step": 700
+    },
+    {
+      "epoch": 0.17625,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.6264543704479655e-05,
+      "loss": 0.4989,
+      "step": 705
+    },
+    {
+      "epoch": 0.1775,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.6212755374537596e-05,
+      "loss": 0.5109,
+      "step": 710
+    },
+    {
+      "epoch": 0.17875,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.616063991766623e-05,
+      "loss": 0.48,
+      "step": 715
+    },
+    {
+      "epoch": 0.18,
+      "grad_norm": 1.3984375,
+      "learning_rate": 4.610819813755038e-05,
+      "loss": 0.5151,
+      "step": 720
+    },
+    {
+      "epoch": 0.18125,
+      "grad_norm": 0.9453125,
+      "learning_rate": 4.6055430842907167e-05,
+      "loss": 0.5235,
+      "step": 725
+    },
+    {
+      "epoch": 0.1825,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.600233884747355e-05,
+      "loss": 0.5006,
+      "step": 730
+    },
+    {
+      "epoch": 0.18375,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.594892296999378e-05,
+      "loss": 0.479,
+      "step": 735
+    },
+    {
+      "epoch": 0.185,
+      "grad_norm": 1.015625,
+      "learning_rate": 4.5895184034206765e-05,
+      "loss": 0.4807,
+      "step": 740
+    },
+    {
+      "epoch": 0.18625,
+      "grad_norm": 1.40625,
+      "learning_rate": 4.5841122868833364e-05,
+      "loss": 0.5189,
+      "step": 745
+    },
+    {
+      "epoch": 0.1875,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.5786740307563636e-05,
+      "loss": 0.4768,
+      "step": 750
+    },
+    {
+      "epoch": 0.18875,
+      "grad_norm": 1.5234375,
+      "learning_rate": 4.573203718904394e-05,
+      "loss": 0.4747,
+      "step": 755
+    },
+    {
+      "epoch": 0.19,
+      "grad_norm": 1.5703125,
+      "learning_rate": 4.567701435686404e-05,
+      "loss": 0.4756,
+      "step": 760
+    },
+    {
+      "epoch": 0.19125,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.562167265954409e-05,
+      "loss": 0.5102,
+      "step": 765
+    },
+    {
+      "epoch": 0.1925,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.55660129505215e-05,
+      "loss": 0.5229,
+      "step": 770
+    },
+    {
+      "epoch": 0.19375,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.551003608813784e-05,
+      "loss": 0.5103,
+      "step": 775
+    },
+    {
+      "epoch": 0.195,
+      "grad_norm": 1.21875,
+      "learning_rate": 4.545374293562559e-05,
+      "loss": 0.5216,
+      "step": 780
+    },
+    {
+      "epoch": 0.19625,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.5397134361094786e-05,
+      "loss": 0.5039,
+      "step": 785
+    },
+    {
+      "epoch": 0.1975,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.534021123751968e-05,
+      "loss": 0.4834,
+      "step": 790
+    },
+    {
+      "epoch": 0.19875,
+      "grad_norm": 1.3828125,
+      "learning_rate": 4.528297444272525e-05,
+      "loss": 0.4386,
+      "step": 795
+    },
+    {
+      "epoch": 0.2,
+      "grad_norm": 1.328125,
+      "learning_rate": 4.522542485937369e-05,
+      "loss": 0.5097,
+      "step": 800
+    },
+    {
+      "epoch": 0.20125,
+      "grad_norm": 1.125,
+      "learning_rate": 4.516756337495075e-05,
+      "loss": 0.5574,
+      "step": 805
+    },
+    {
+      "epoch": 0.2025,
+      "grad_norm": 1.1953125,
+      "learning_rate": 4.5109390881752114e-05,
+      "loss": 0.5492,
+      "step": 810
+    },
+    {
+      "epoch": 0.20375,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.5050908276869586e-05,
+      "loss": 0.5281,
+      "step": 815
+    },
+    {
+      "epoch": 0.205,
+      "grad_norm": 1.6484375,
+      "learning_rate": 4.499211646217727e-05,
+      "loss": 0.5042,
+      "step": 820
+    },
+    {
+      "epoch": 0.20625,
+      "grad_norm": 0.9765625,
+      "learning_rate": 4.493301634431768e-05,
+      "loss": 0.4746,
+      "step": 825
+    },
+    {
+      "epoch": 0.2075,
+      "grad_norm": 1.3046875,
+      "learning_rate": 4.487360883468775e-05,
+      "loss": 0.5611,
+      "step": 830
+    },
+    {
+      "epoch": 0.20875,
+      "grad_norm": 1.703125,
+      "learning_rate": 4.481389484942478e-05,
+      "loss": 0.5058,
+      "step": 835
+    },
+    {
+      "epoch": 0.21,
+      "grad_norm": 1.0703125,
+      "learning_rate": 4.4753875309392266e-05,
+      "loss": 0.4352,
+      "step": 840
+    },
+    {
+      "epoch": 0.21125,
+      "grad_norm": 1.125,
+      "learning_rate": 4.469355114016577e-05,
+      "loss": 0.4849,
+      "step": 845
+    },
+    {
+      "epoch": 0.2125,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.463292327201862e-05,
+      "loss": 0.5195,
+      "step": 850
+    },
+    {
+      "epoch": 0.21375,
+      "grad_norm": 1.09375,
+      "learning_rate": 4.4571992639907545e-05,
+      "loss": 0.3864,
+      "step": 855
+    },
+    {
+      "epoch": 0.215,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.451076018345825e-05,
+      "loss": 0.4903,
+      "step": 860
+    },
+    {
+      "epoch": 0.21625,
+      "grad_norm": 1.21875,
+      "learning_rate": 4.444922684695097e-05,
+      "loss": 0.5126,
+      "step": 865
+    },
+    {
+      "epoch": 0.2175,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.4387393579305865e-05,
+      "loss": 0.4958,
+      "step": 870
+    },
+    {
+      "epoch": 0.21875,
+      "grad_norm": 1.5078125,
+      "learning_rate": 4.4325261334068426e-05,
+      "loss": 0.5307,
+      "step": 875
+    },
+    {
+      "epoch": 0.22,
+      "grad_norm": 1.5,
+      "learning_rate": 4.426283106939474e-05,
+      "loss": 0.5048,
+      "step": 880
+    },
+    {
+      "epoch": 0.22125,
+      "grad_norm": 1.0,
+      "learning_rate": 4.4200103748036695e-05,
+      "loss": 0.4757,
+      "step": 885
+    },
+    {
+      "epoch": 0.2225,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.4137080337327205e-05,
+      "loss": 0.5525,
+      "step": 890
+    },
+    {
+      "epoch": 0.22375,
+      "grad_norm": 1.5390625,
+      "learning_rate": 4.407376180916522e-05,
+      "loss": 0.4781,
+      "step": 895
+    },
+    {
+      "epoch": 0.225,
+      "grad_norm": 1.28125,
+      "learning_rate": 4.401014914000078e-05,
+      "loss": 0.4797,
+      "step": 900
+    },
+    {
+      "epoch": 0.22625,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.3946243310819926e-05,
+      "loss": 0.5529,
+      "step": 905
+    },
+    {
+      "epoch": 0.2275,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.3882045307129594e-05,
+      "loss": 0.4669,
+      "step": 910
+    },
+    {
+      "epoch": 0.22875,
+      "grad_norm": 1.234375,
+      "learning_rate": 4.3817556118942425e-05,
+      "loss": 0.5328,
+      "step": 915
+    },
+    {
+      "epoch": 0.23,
+      "grad_norm": 1.4140625,
+      "learning_rate": 4.375277674076149e-05,
+      "loss": 0.5347,
+      "step": 920
+    },
+    {
+      "epoch": 0.23125,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.3687708171564925e-05,
+      "loss": 0.4615,
+      "step": 925
+    },
+    {
+      "epoch": 0.2325,
+      "grad_norm": 1.1328125,
+      "learning_rate": 4.3622351414790554e-05,
+      "loss": 0.5132,
+      "step": 930
+    },
+    {
+      "epoch": 0.23375,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.355670747832042e-05,
+      "loss": 0.4063,
+      "step": 935
+    },
+    {
+      "epoch": 0.235,
+      "grad_norm": 1.1171875,
+      "learning_rate": 4.349077737446525e-05,
+      "loss": 0.493,
+      "step": 940
+    },
+    {
+      "epoch": 0.23625,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.3424562119948776e-05,
+      "loss": 0.4826,
+      "step": 945
+    },
+    {
+      "epoch": 0.2375,
+      "grad_norm": 1.0546875,
+      "learning_rate": 4.335806273589214e-05,
+      "loss": 0.4726,
+      "step": 950
+    },
+    {
+      "epoch": 0.23875,
+      "grad_norm": 1.59375,
+      "learning_rate": 4.329128024779812e-05,
+      "loss": 0.4672,
+      "step": 955
+    },
+    {
+      "epoch": 0.24,
+      "grad_norm": 0.91796875,
+      "learning_rate": 4.3224215685535294e-05,
+      "loss": 0.4467,
+      "step": 960
+    },
+    {
+      "epoch": 0.24125,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.315687008332217e-05,
+      "loss": 0.5019,
+      "step": 965
+    },
+    {
+      "epoch": 0.2425,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.3089244479711236e-05,
+      "loss": 0.526,
+      "step": 970
+    },
+    {
+      "epoch": 0.24375,
+      "grad_norm": 1.2890625,
+      "learning_rate": 4.302133991757297e-05,
+      "loss": 0.509,
+      "step": 975
+    },
+    {
+      "epoch": 0.245,
+      "grad_norm": 0.953125,
+      "learning_rate": 4.295315744407972e-05,
+      "loss": 0.447,
+      "step": 980
+    },
+    {
+      "epoch": 0.24625,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.2884698110689575e-05,
+      "loss": 0.4927,
+      "step": 985
+    },
+    {
+      "epoch": 0.2475,
+      "grad_norm": 1.578125,
+      "learning_rate": 4.281596297313013e-05,
+      "loss": 0.4891,
+      "step": 990
+    },
+    {
+      "epoch": 0.24875,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.274695309138226e-05,
+      "loss": 0.5046,
+      "step": 995
+    },
+    {
+      "epoch": 0.25,
+      "grad_norm": 1.1796875,
+      "learning_rate": 4.267766952966369e-05,
+      "loss": 0.4642,
+      "step": 1000
+    },
+    {
+      "epoch": 0.25125,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.260811335641266e-05,
+      "loss": 0.4396,
+      "step": 1005
+    },
+    {
+      "epoch": 0.2525,
+      "grad_norm": 1.2265625,
+      "learning_rate": 4.25382856442714e-05,
+      "loss": 0.4386,
+      "step": 1010
+    },
+    {
+      "epoch": 0.25375,
+      "grad_norm": 1.546875,
+      "learning_rate": 4.2468187470069607e-05,
+      "loss": 0.5335,
+      "step": 1015
+    },
+    {
+      "epoch": 0.255,
+      "grad_norm": 1.59375,
+      "learning_rate": 4.2397819914807856e-05,
+      "loss": 0.4703,
+      "step": 1020
+    },
+    {
+      "epoch": 0.25625,
+      "grad_norm": 1.0625,
+      "learning_rate": 4.23271840636409e-05,
+      "loss": 0.5011,
+      "step": 1025
+    },
+    {
+      "epoch": 0.2575,
+      "grad_norm": 1.046875,
+      "learning_rate": 4.225628100586093e-05,
+      "loss": 0.5234,
+      "step": 1030
+    },
+    {
+      "epoch": 0.25875,
+      "grad_norm": 1.0078125,
+      "learning_rate": 4.218511183488082e-05,
+      "loss": 0.4749,
+      "step": 1035
+    },
+    {
+      "epoch": 0.26,
+      "grad_norm": 1.34375,
+      "learning_rate": 4.211367764821722e-05,
+      "loss": 0.5361,
+      "step": 1040
+    },
+    {
+      "epoch": 0.26125,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.2041979547473665e-05,
+      "loss": 0.4458,
+      "step": 1045
+    },
+    {
+      "epoch": 0.2625,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.197001863832355e-05,
+      "loss": 0.4517,
+      "step": 1050
+    },
+    {
+      "epoch": 0.26375,
+      "grad_norm": 0.9921875,
+      "learning_rate": 4.189779603049312e-05,
+      "loss": 0.4571,
+      "step": 1055
+    },
+    {
+      "epoch": 0.265,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.182531283774434e-05,
+      "loss": 0.487,
+      "step": 1060
+    },
+    {
+      "epoch": 0.26625,
+      "grad_norm": 1.140625,
+      "learning_rate": 4.17525701778577e-05,
+      "loss": 0.5186,
+      "step": 1065
+    },
+    {
+      "epoch": 0.2675,
+      "grad_norm": 1.2578125,
+      "learning_rate": 4.1679569172614996e-05,
+      "loss": 0.4815,
+      "step": 1070
+    },
+    {
+      "epoch": 0.26875,
+      "grad_norm": 0.98046875,
+      "learning_rate": 4.1606310947782044e-05,
+      "loss": 0.4563,
+      "step": 1075
+    },
+    {
+      "epoch": 0.27,
+      "grad_norm": 1.0,
+      "learning_rate": 4.1532796633091296e-05,
+      "loss": 0.4585,
+      "step": 1080
+    },
+    {
+      "epoch": 0.27125,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.1459027362224436e-05,
+      "loss": 0.4846,
+      "step": 1085
+    },
+    {
+      "epoch": 0.2725,
+      "grad_norm": 1.03125,
+      "learning_rate": 4.138500427279485e-05,
+      "loss": 0.505,
+      "step": 1090
+    },
+    {
+      "epoch": 0.27375,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.1310728506330174e-05,
+      "loss": 0.4765,
+      "step": 1095
+    },
+    {
+      "epoch": 0.275,
+      "grad_norm": 1.5546875,
+      "learning_rate": 4.123620120825459e-05,
+      "loss": 0.5105,
+      "step": 1100
+    },
+    {
+      "epoch": 0.27625,
+      "grad_norm": 0.94140625,
+      "learning_rate": 4.116142352787125e-05,
+      "loss": 0.4193,
+      "step": 1105
+    },
+    {
+      "epoch": 0.2775,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.1086396618344476e-05,
+      "loss": 0.4953,
+      "step": 1110
+    },
+    {
+      "epoch": 0.27875,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.101112163668203e-05,
+      "loss": 0.4572,
+      "step": 1115
+    },
+    {
+      "epoch": 0.28,
+      "grad_norm": 1.1015625,
+      "learning_rate": 4.093559974371725e-05,
+      "loss": 0.4247,
+      "step": 1120
+    },
+    {
+      "epoch": 0.28125,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.085983210409114e-05,
+      "loss": 0.483,
+      "step": 1125
+    },
+    {
+      "epoch": 0.2825,
+      "grad_norm": 0.98828125,
+      "learning_rate": 4.0783819886234445e-05,
+      "loss": 0.4787,
+      "step": 1130
+    },
+    {
+      "epoch": 0.28375,
+      "grad_norm": 1.015625,
+      "learning_rate": 4.0707564262349595e-05,
+      "loss": 0.4891,
+      "step": 1135
+    },
+    {
+      "epoch": 0.285,
+      "grad_norm": 1.5859375,
+      "learning_rate": 4.063106640839264e-05,
+      "loss": 0.503,
+      "step": 1140
+    },
+    {
+      "epoch": 0.28625,
+      "grad_norm": 1.4140625,
+      "learning_rate": 4.05543275040551e-05,
+      "loss": 0.5003,
+      "step": 1145
+    },
+    {
+      "epoch": 0.2875,
+      "grad_norm": 1.328125,
+      "learning_rate": 4.047734873274586e-05,
+      "loss": 0.5444,
+      "step": 1150
+    },
+    {
+      "epoch": 0.28875,
+      "grad_norm": 0.86328125,
+      "learning_rate": 4.040013128157275e-05,
+      "loss": 0.4193,
+      "step": 1155
+    },
+    {
+      "epoch": 0.29,
+      "grad_norm": 1.1328125,
+      "learning_rate": 4.0322676341324415e-05,
+      "loss": 0.497,
+      "step": 1160
+    },
+    {
+      "epoch": 0.29125,
+      "grad_norm": 0.94140625,
+      "learning_rate": 4.024498510645185e-05,
+      "loss": 0.377,
+      "step": 1165
+    },
+    {
+      "epoch": 0.2925,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.0167058775049996e-05,
+      "loss": 0.5118,
+      "step": 1170
+    },
+    {
+      "epoch": 0.29375,
+      "grad_norm": 1.59375,
+      "learning_rate": 4.008889854883929e-05,
+      "loss": 0.4941,
+      "step": 1175
+    },
+    {
+      "epoch": 0.295,
+      "grad_norm": 1.15625,
+      "learning_rate": 4.0010505633147106e-05,
+      "loss": 0.5302,
+      "step": 1180
+    },
+    {
+      "epoch": 0.29625,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.993188123688918e-05,
+      "loss": 0.5273,
+      "step": 1185
+    },
+    {
+      "epoch": 0.2975,
+      "grad_norm": 1.4921875,
+      "learning_rate": 3.985302657255097e-05,
+      "loss": 0.463,
+      "step": 1190
+    },
+    {
+      "epoch": 0.29875,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.977394285616893e-05,
+      "loss": 0.5116,
+      "step": 1195
+    },
+    {
+      "epoch": 0.3,
+      "grad_norm": 1.0078125,
+      "learning_rate": 3.969463130731183e-05,
+      "loss": 0.5089,
+      "step": 1200
+    },
+    {
+      "epoch": 0.30125,
+      "grad_norm": 0.890625,
+      "learning_rate": 3.961509314906184e-05,
+      "loss": 0.5043,
+      "step": 1205
+    },
+    {
+      "epoch": 0.3025,
+      "grad_norm": 1.2265625,
+      "learning_rate": 3.953532960799577e-05,
+      "loss": 0.4877,
+      "step": 1210
+    },
+    {
+      "epoch": 0.30375,
+      "grad_norm": 0.953125,
+      "learning_rate": 3.9455341914166075e-05,
+      "loss": 0.5368,
+      "step": 1215
+    },
+    {
+      "epoch": 0.305,
+      "grad_norm": 1.1015625,
+      "learning_rate": 3.937513130108197e-05,
+      "loss": 0.4303,
+      "step": 1220
+    },
+    {
+      "epoch": 0.30625,
+      "grad_norm": 0.953125,
+      "learning_rate": 3.9294699005690305e-05,
+      "loss": 0.4978,
+      "step": 1225
+    },
+    {
+      "epoch": 0.3075,
+      "grad_norm": 1.4453125,
+      "learning_rate": 3.92140462683566e-05,
+      "loss": 0.4898,
+      "step": 1230
+    },
+    {
+      "epoch": 0.30875,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.913317433284582e-05,
+      "loss": 0.4307,
+      "step": 1235
+    },
+    {
+      "epoch": 0.31,
+      "grad_norm": 1.078125,
+      "learning_rate": 3.905208444630327e-05,
+      "loss": 0.4599,
+      "step": 1240
+    },
+    {
+      "epoch": 0.31125,
+      "grad_norm": 0.94921875,
+      "learning_rate": 3.897077785923529e-05,
+      "loss": 0.4449,
+      "step": 1245
+    },
+    {
+      "epoch": 0.3125,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.888925582549006e-05,
+      "loss": 0.4508,
+      "step": 1250
+    },
+    {
+      "epoch": 0.31375,
+      "grad_norm": 1.0703125,
+      "learning_rate": 3.880751960223817e-05,
+      "loss": 0.4523,
+      "step": 1255
+    },
+    {
+      "epoch": 0.315,
+      "grad_norm": 1.203125,
+      "learning_rate": 3.87255704499533e-05,
+      "loss": 0.4782,
+      "step": 1260
+    },
+    {
+      "epoch": 0.31625,
+      "grad_norm": 1.09375,
+      "learning_rate": 3.864340963239275e-05,
+      "loss": 0.4821,
+      "step": 1265
+    },
+    {
+      "epoch": 0.3175,
+      "grad_norm": 1.0625,
+      "learning_rate": 3.856103841657797e-05,
+      "loss": 0.393,
+      "step": 1270
+    },
+    {
+      "epoch": 0.31875,
+      "grad_norm": 1.1328125,
+      "learning_rate": 3.847845807277502e-05,
+      "loss": 0.4731,
+      "step": 1275
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.94140625,
+      "learning_rate": 3.8395669874474915e-05,
+      "loss": 0.4644,
+      "step": 1280
+    },
+    {
+      "epoch": 0.32125,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.831267509837414e-05,
+      "loss": 0.5069,
+      "step": 1285
+    },
+    {
+      "epoch": 0.3225,
+      "grad_norm": 1.1328125,
+      "learning_rate": 3.822947502435477e-05,
+      "loss": 0.4767,
+      "step": 1290
+    },
+    {
+      "epoch": 0.32375,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.814607093546489e-05,
+      "loss": 0.472,
+      "step": 1295
+    },
+    {
+      "epoch": 0.325,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.8062464117898724e-05,
+      "loss": 0.4598,
+      "step": 1300
+    },
+    {
+      "epoch": 0.32625,
+      "grad_norm": 1.71875,
+      "learning_rate": 3.7978655860976824e-05,
+      "loss": 0.4794,
+      "step": 1305
+    },
+    {
+      "epoch": 0.3275,
+      "grad_norm": 1.046875,
+      "learning_rate": 3.789464745712619e-05,
+      "loss": 0.4728,
+      "step": 1310
+    },
+    {
+      "epoch": 0.32875,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.7810440201860334e-05,
+      "loss": 0.4535,
+      "step": 1315
+    },
+    {
+      "epoch": 0.33,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.7726035393759285e-05,
+      "loss": 0.4646,
+      "step": 1320
+    },
+    {
+      "epoch": 0.33125,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.764143433444962e-05,
+      "loss": 0.4597,
+      "step": 1325
+    },
+    {
+      "epoch": 0.3325,
+      "grad_norm": 0.9921875,
+      "learning_rate": 3.755663832858432e-05,
+      "loss": 0.516,
+      "step": 1330
+    },
+    {
+      "epoch": 0.33375,
+      "grad_norm": 1.078125,
+      "learning_rate": 3.747164868382269e-05,
+      "loss": 0.4492,
+      "step": 1335
+    },
+    {
+      "epoch": 0.335,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.7386466710810194e-05,
+      "loss": 0.4644,
+      "step": 1340
+    },
+    {
+      "epoch": 0.33625,
+      "grad_norm": 1.03125,
+      "learning_rate": 3.730109372315822e-05,
+      "loss": 0.5028,
+      "step": 1345
+    },
+    {
+      "epoch": 0.3375,
+      "grad_norm": 1.359375,
+      "learning_rate": 3.721553103742388e-05,
+      "loss": 0.424,
+      "step": 1350
+    },
+    {
+      "epoch": 0.33875,
+      "grad_norm": 1.1015625,
+      "learning_rate": 3.71297799730896e-05,
+      "loss": 0.4592,
+      "step": 1355
+    },
+    {
+      "epoch": 0.34,
+      "grad_norm": 1.0390625,
+      "learning_rate": 3.704384185254288e-05,
+      "loss": 0.4678,
+      "step": 1360
+    },
+    {
+      "epoch": 0.34125,
+      "grad_norm": 1.0703125,
+      "learning_rate": 3.695771800105586e-05,
+      "loss": 0.4809,
+      "step": 1365
+    },
+    {
+      "epoch": 0.3425,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.6871409746764865e-05,
+      "loss": 0.5093,
+      "step": 1370
+    },
+    {
+      "epoch": 0.34375,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.678491842064995e-05,
+      "loss": 0.4937,
+      "step": 1375
+    },
+    {
+      "epoch": 0.345,
+      "grad_norm": 1.5625,
+      "learning_rate": 3.6698245356514335e-05,
+      "loss": 0.4107,
+      "step": 1380
+    },
+    {
+      "epoch": 0.34625,
+      "grad_norm": 1.0,
+      "learning_rate": 3.661139189096391e-05,
+      "loss": 0.4578,
+      "step": 1385
+    },
+    {
+      "epoch": 0.3475,
+      "grad_norm": 0.8359375,
+      "learning_rate": 3.652435936338656e-05,
+      "loss": 0.3964,
+      "step": 1390
+    },
+    {
+      "epoch": 0.34875,
+      "grad_norm": 1.1171875,
+      "learning_rate": 3.6437149115931514e-05,
+      "loss": 0.5011,
+      "step": 1395
+    },
+    {
+      "epoch": 0.35,
+      "grad_norm": 1.28125,
+      "learning_rate": 3.634976249348867e-05,
+      "loss": 0.494,
+      "step": 1400
+    },
+    {
+      "epoch": 0.35125,
+      "grad_norm": 1.125,
+      "learning_rate": 3.626220084366786e-05,
+      "loss": 0.4773,
+      "step": 1405
+    },
+    {
+      "epoch": 0.3525,
+      "grad_norm": 1.0390625,
+      "learning_rate": 3.6174465516778035e-05,
+      "loss": 0.4338,
+      "step": 1410
+    },
+    {
+      "epoch": 0.35375,
+      "grad_norm": 0.99609375,
+      "learning_rate": 3.608655786580647e-05,
+      "loss": 0.4538,
+      "step": 1415
+    },
+    {
+      "epoch": 0.355,
+      "grad_norm": 1.1875,
+      "learning_rate": 3.599847924639788e-05,
+      "loss": 0.4537,
+      "step": 1420
+    },
+    {
+      "epoch": 0.35625,
+      "grad_norm": 1.0078125,
+      "learning_rate": 3.591023101683355e-05,
+      "loss": 0.448,
+      "step": 1425
+    },
+    {
+      "epoch": 0.3575,
+      "grad_norm": 1.125,
+      "learning_rate": 3.582181453801036e-05,
+      "loss": 0.4645,
+      "step": 1430
+    },
+    {
+      "epoch": 0.35875,
+      "grad_norm": 1.6015625,
+      "learning_rate": 3.5733231173419754e-05,
+      "loss": 0.4578,
+      "step": 1435
+    },
+    {
+      "epoch": 0.36,
+      "grad_norm": 1.0,
+      "learning_rate": 3.564448228912682e-05,
+      "loss": 0.4704,
+      "step": 1440
+    },
+    {
+      "epoch": 0.36125,
+      "grad_norm": 1.8671875,
+      "learning_rate": 3.555556925374914e-05,
+      "loss": 0.4383,
+      "step": 1445
+    },
+    {
+      "epoch": 0.3625,
+      "grad_norm": 1.2734375,
+      "learning_rate": 3.54664934384357e-05,
+      "loss": 0.4192,
+      "step": 1450
+    },
+    {
+      "epoch": 0.36375,
+      "grad_norm": 0.94921875,
+      "learning_rate": 3.5377256216845785e-05,
+      "loss": 0.5063,
+      "step": 1455
+    },
+    {
+      "epoch": 0.365,
+      "grad_norm": 1.203125,
+      "learning_rate": 3.528785896512772e-05,
+      "loss": 0.4711,
+      "step": 1460
+    },
+    {
+      "epoch": 0.36625,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.519830306189773e-05,
+      "loss": 0.4494,
+      "step": 1465
+    },
+    {
+      "epoch": 0.3675,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.510858988821863e-05,
+      "loss": 0.4972,
+      "step": 1470
+    },
+    {
+      "epoch": 0.36875,
+      "grad_norm": 1.4765625,
+      "learning_rate": 3.5018720827578524e-05,
+      "loss": 0.4312,
+      "step": 1475
+    },
+    {
+      "epoch": 0.37,
+      "grad_norm": 0.99609375,
+      "learning_rate": 3.4928697265869515e-05,
+      "loss": 0.4267,
+      "step": 1480
+    },
+    {
+      "epoch": 0.37125,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.483852059136629e-05,
+      "loss": 0.4563,
+      "step": 1485
+    },
+    {
+      "epoch": 0.3725,
+      "grad_norm": 0.99609375,
+      "learning_rate": 3.474819219470471e-05,
+      "loss": 0.4642,
+      "step": 1490
+    },
+    {
+      "epoch": 0.37375,
+      "grad_norm": 1.1796875,
+      "learning_rate": 3.4657713468860405e-05,
+      "loss": 0.414,
+      "step": 1495
+    },
+    {
+      "epoch": 0.375,
+      "grad_norm": 1.140625,
+      "learning_rate": 3.456708580912725e-05,
+      "loss": 0.4919,
+      "step": 1500
+    },
+    {
+      "epoch": 0.37625,
+      "grad_norm": 1.25,
+      "learning_rate": 3.447631061309587e-05,
+      "loss": 0.5023,
+      "step": 1505
+    },
+    {
+      "epoch": 0.3775,
+      "grad_norm": 1.140625,
+      "learning_rate": 3.438538928063208e-05,
+      "loss": 0.469,
+      "step": 1510
+    },
+    {
+      "epoch": 0.37875,
+      "grad_norm": 1.0859375,
+      "learning_rate": 3.4294323213855305e-05,
+      "loss": 0.4322,
+      "step": 1515
+    },
+    {
+      "epoch": 0.38,
+      "grad_norm": 0.9296875,
+      "learning_rate": 3.4203113817116957e-05,
+      "loss": 0.4393,
+      "step": 1520
+    },
+    {
+      "epoch": 0.38125,
+      "grad_norm": 1.1015625,
+      "learning_rate": 3.411176249697875e-05,
+      "loss": 0.4005,
+      "step": 1525
+    },
+    {
+      "epoch": 0.3825,
+      "grad_norm": 1.4609375,
+      "learning_rate": 3.402027066219105e-05,
+      "loss": 0.4094,
+      "step": 1530
+    },
+    {
+      "epoch": 0.38375,
+      "grad_norm": 1.4921875,
+      "learning_rate": 3.392863972367114e-05,
+      "loss": 0.4474,
+      "step": 1535
+    },
+    {
+      "epoch": 0.385,
+      "grad_norm": 1.265625,
+      "learning_rate": 3.383687109448143e-05,
+      "loss": 0.399,
+      "step": 1540
+    },
+    {
+      "epoch": 0.38625,
+      "grad_norm": 1.3515625,
+      "learning_rate": 3.374496618980772e-05,
+      "loss": 0.4342,
+      "step": 1545
+    },
+    {
+      "epoch": 0.3875,
+      "grad_norm": 1.2890625,
+      "learning_rate": 3.365292642693732e-05,
+      "loss": 0.4847,
+      "step": 1550
+    },
+    {
+      "epoch": 0.38875,
+      "grad_norm": 1.046875,
+      "learning_rate": 3.356075322523725e-05,
+      "loss": 0.4343,
+      "step": 1555
+    },
+    {
+      "epoch": 0.39,
+      "grad_norm": 0.9921875,
+      "learning_rate": 3.346844800613229e-05,
+      "loss": 0.498,
+      "step": 1560
+    },
+    {
+      "epoch": 0.39125,
+      "grad_norm": 1.0703125,
+      "learning_rate": 3.33760121930831e-05,
+      "loss": 0.4737,
+      "step": 1565
+    },
+    {
+      "epoch": 0.3925,
+      "grad_norm": 0.99609375,
+      "learning_rate": 3.3283447211564276e-05,
+      "loss": 0.4965,
+      "step": 1570
+    },
+    {
+      "epoch": 0.39375,
+      "grad_norm": 1.3359375,
+      "learning_rate": 3.319075448904234e-05,
+      "loss": 0.4626,
+      "step": 1575
+    },
+    {
+      "epoch": 0.395,
+      "grad_norm": 1.1640625,
+      "learning_rate": 3.309793545495374e-05,
+      "loss": 0.5161,
+      "step": 1580
+    },
+    {
+      "epoch": 0.39625,
+      "grad_norm": 1.21875,
+      "learning_rate": 3.3004991540682796e-05,
+      "loss": 0.4371,
+      "step": 1585
+    },
+    {
+      "epoch": 0.3975,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.2911924179539656e-05,
+      "loss": 0.4427,
+      "step": 1590
+    },
+    {
+      "epoch": 0.39875,
+      "grad_norm": 0.93359375,
+      "learning_rate": 3.281873480673815e-05,
+      "loss": 0.4318,
+      "step": 1595
+    },
+    {
+      "epoch": 0.4,
+      "grad_norm": 1.046875,
+      "learning_rate": 3.272542485937369e-05,
+      "loss": 0.4756,
+      "step": 1600
+    },
+    {
+      "epoch": 0.40125,
+      "grad_norm": 1.0390625,
+      "learning_rate": 3.2631995776401094e-05,
+      "loss": 0.4507,
+      "step": 1605
+    },
+    {
+      "epoch": 0.4025,
+      "grad_norm": 1.03125,
+      "learning_rate": 3.253844899861239e-05,
+      "loss": 0.4444,
+      "step": 1610
+    },
+    {
+      "epoch": 0.40375,
+      "grad_norm": 1.046875,
+      "learning_rate": 3.244478596861464e-05,
+      "loss": 0.4291,
+      "step": 1615
+    },
+    {
+      "epoch": 0.405,
+      "grad_norm": 1.125,
+      "learning_rate": 3.23510081308076e-05,
+      "loss": 0.4615,
+      "step": 1620
+    },
+    {
+      "epoch": 0.40625,
+      "grad_norm": 1.0546875,
+      "learning_rate": 3.225711693136156e-05,
+      "loss": 0.4347,
+      "step": 1625
+    },
+    {
+      "epoch": 0.4075,
+      "grad_norm": 1.3046875,
+      "learning_rate": 3.2163113818194964e-05,
+      "loss": 0.4349,
+      "step": 1630
+    },
+    {
+      "epoch": 0.40875,
+      "grad_norm": 1.609375,
+      "learning_rate": 3.206900024095208e-05,
+      "loss": 0.4814,
+      "step": 1635
+    },
+    {
+      "epoch": 0.41,
+      "grad_norm": 1.3828125,
+      "learning_rate": 3.1974777650980735e-05,
+      "loss": 0.502,
+      "step": 1640
+    },
+    {
+      "epoch": 0.41125,
+      "grad_norm": 1.5234375,
+      "learning_rate": 3.188044750130979e-05,
+      "loss": 0.4457,
+      "step": 1645
+    },
+    {
+      "epoch": 0.4125,
+      "grad_norm": 1.140625,
+      "learning_rate": 3.178601124662686e-05,
+      "loss": 0.505,
+      "step": 1650
+    },
+    {
+      "epoch": 0.41375,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.169147034325582e-05,
+      "loss": 0.4941,
+      "step": 1655
+    },
+    {
+      "epoch": 0.415,
+      "grad_norm": 1.078125,
+      "learning_rate": 3.1596826249134324e-05,
+      "loss": 0.4524,
+      "step": 1660
+    },
+    {
+      "epoch": 0.41625,
+      "grad_norm": 1.1015625,
+      "learning_rate": 3.150208042379142e-05,
+      "loss": 0.4826,
+      "step": 1665
+    },
+    {
+      "epoch": 0.4175,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.140723432832492e-05,
+      "loss": 0.4101,
+      "step": 1670
+    },
+    {
+      "epoch": 0.41875,
+      "grad_norm": 1.3359375,
+      "learning_rate": 3.131228942537895e-05,
+      "loss": 0.4068,
+      "step": 1675
+    },
+    {
+      "epoch": 0.42,
+      "grad_norm": 1.53125,
+      "learning_rate": 3.121724717912138e-05,
+      "loss": 0.4341,
+      "step": 1680
+    },
+    {
+      "epoch": 0.42125,
+      "grad_norm": 1.2109375,
+      "learning_rate": 3.112210905522119e-05,
+      "loss": 0.4197,
+      "step": 1685
+    },
+    {
+      "epoch": 0.4225,
+      "grad_norm": 1.3046875,
+      "learning_rate": 3.102687652082597e-05,
+      "loss": 0.4257,
+      "step": 1690
+    },
+    {
+      "epoch": 0.42375,
+      "grad_norm": 1.25,
+      "learning_rate": 3.0931551044539194e-05,
+      "loss": 0.4513,
+      "step": 1695
+    },
+    {
+      "epoch": 0.425,
+      "grad_norm": 1.21875,
+      "learning_rate": 3.083613409639764e-05,
+      "loss": 0.4757,
+      "step": 1700
+    },
+    {
+      "epoch": 0.42625,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.0740627147848675e-05,
+      "loss": 0.441,
+      "step": 1705
+    },
+    {
+      "epoch": 0.4275,
+      "grad_norm": 1.2265625,
+      "learning_rate": 3.06450316717276e-05,
+      "loss": 0.4249,
+      "step": 1710
+    },
+    {
+      "epoch": 0.42875,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.05493491422349e-05,
+      "loss": 0.4189,
+      "step": 1715
+    },
+    {
+      "epoch": 0.43,
+      "grad_norm": 1.3203125,
+      "learning_rate": 3.045358103491357e-05,
+      "loss": 0.4315,
+      "step": 1720
+    },
+    {
+      "epoch": 0.43125,
+      "grad_norm": 1.15625,
+      "learning_rate": 3.035772882662627e-05,
+      "loss": 0.4641,
+      "step": 1725
+    },
+    {
+      "epoch": 0.4325,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.026179399553264e-05,
+      "loss": 0.4701,
+      "step": 1730
+    },
+    {
+      "epoch": 0.43375,
+      "grad_norm": 1.421875,
+      "learning_rate": 3.0165778021066453e-05,
+      "loss": 0.4827,
+      "step": 1735
+    },
+    {
+      "epoch": 0.435,
+      "grad_norm": 1.3203125,
+      "learning_rate": 3.0069682383912813e-05,
+      "loss": 0.4439,
+      "step": 1740
+    },
+    {
+      "epoch": 0.43625,
+      "grad_norm": 1.28125,
+      "learning_rate": 2.9973508565985313e-05,
+      "loss": 0.4916,
+      "step": 1745
+    },
+    {
+      "epoch": 0.4375,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.9877258050403212e-05,
+      "loss": 0.464,
+      "step": 1750
+    },
+    {
+      "epoch": 0.43875,
+      "grad_norm": 1.0078125,
+      "learning_rate": 2.9780932321468515e-05,
+      "loss": 0.4105,
+      "step": 1755
+    },
+    {
+      "epoch": 0.44,
+      "grad_norm": 1.3125,
+      "learning_rate": 2.9684532864643122e-05,
+      "loss": 0.4312,
+      "step": 1760
+    },
+    {
+      "epoch": 0.44125,
+      "grad_norm": 1.1875,
+      "learning_rate": 2.9588061166525914e-05,
+      "loss": 0.4465,
+      "step": 1765
+    },
+    {
+      "epoch": 0.4425,
+      "grad_norm": 1.5,
+      "learning_rate": 2.949151871482982e-05,
+      "loss": 0.4136,
+      "step": 1770
+    },
+    {
+      "epoch": 0.44375,
+      "grad_norm": 1.21875,
+      "learning_rate": 2.9394906998358868e-05,
+      "loss": 0.4107,
+      "step": 1775
+    },
+    {
+      "epoch": 0.445,
+      "grad_norm": 0.98828125,
+      "learning_rate": 2.929822750698524e-05,
+      "loss": 0.4327,
+      "step": 1780
+    },
+    {
+      "epoch": 0.44625,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.92014817316263e-05,
+      "loss": 0.4597,
+      "step": 1785
+    },
+    {
+      "epoch": 0.4475,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.9104671164221576e-05,
+      "loss": 0.4685,
+      "step": 1790
+    },
+    {
+      "epoch": 0.44875,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.9007797297709782e-05,
+      "loss": 0.451,
+      "step": 1795
+    },
+    {
+      "epoch": 0.45,
+      "grad_norm": 1.40625,
+      "learning_rate": 2.8910861626005776e-05,
+      "loss": 0.4101,
+      "step": 1800
+    },
+    {
+      "epoch": 0.45125,
+      "grad_norm": 1.2109375,
+      "learning_rate": 2.8813865643977526e-05,
+      "loss": 0.4775,
+      "step": 1805
+    },
+    {
+      "epoch": 0.4525,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.871681084742308e-05,
+      "loss": 0.4588,
+      "step": 1810
+    },
+    {
+      "epoch": 0.45375,
+      "grad_norm": 1.265625,
+      "learning_rate": 2.8619698733047447e-05,
+      "loss": 0.4476,
+      "step": 1815
+    },
+    {
+      "epoch": 0.455,
+      "grad_norm": 1.140625,
+      "learning_rate": 2.8522530798439567e-05,
+      "loss": 0.4375,
+      "step": 1820
+    },
+    {
+      "epoch": 0.45625,
+      "grad_norm": 1.015625,
+      "learning_rate": 2.8425308542049206e-05,
+      "loss": 0.422,
+      "step": 1825
+    },
+    {
+      "epoch": 0.4575,
+      "grad_norm": 0.94921875,
+      "learning_rate": 2.832803346316381e-05,
+      "loss": 0.4887,
+      "step": 1830
+    },
+    {
+      "epoch": 0.45875,
+      "grad_norm": 1.1796875,
+      "learning_rate": 2.8230707061885443e-05,
+      "loss": 0.4136,
+      "step": 1835
+    },
+    {
+      "epoch": 0.46,
+      "grad_norm": 1.140625,
+      "learning_rate": 2.8133330839107608e-05,
+      "loss": 0.4236,
+      "step": 1840
+    },
+    {
+      "epoch": 0.46125,
+      "grad_norm": 1.078125,
+      "learning_rate": 2.803590629649212e-05,
+      "loss": 0.4983,
+      "step": 1845
+    },
+    {
+      "epoch": 0.4625,
+      "grad_norm": 1.265625,
+      "learning_rate": 2.7938434936445945e-05,
+      "loss": 0.4988,
+      "step": 1850
+    },
+    {
+      "epoch": 0.46375,
+      "grad_norm": 0.8828125,
+      "learning_rate": 2.784091826209803e-05,
+      "loss": 0.4337,
+      "step": 1855
+    },
+    {
+      "epoch": 0.465,
+      "grad_norm": 1.1640625,
+      "learning_rate": 2.774335777727613e-05,
+      "loss": 0.4574,
+      "step": 1860
+    },
+    {
+      "epoch": 0.46625,
+      "grad_norm": 0.95703125,
+      "learning_rate": 2.764575498648362e-05,
+      "loss": 0.4606,
+      "step": 1865
+    },
+    {
+      "epoch": 0.4675,
+      "grad_norm": 1.1328125,
+      "learning_rate": 2.754811139487625e-05,
+      "loss": 0.4489,
+      "step": 1870
+    },
+    {
+      "epoch": 0.46875,
+      "grad_norm": 1.1171875,
+      "learning_rate": 2.7450428508239024e-05,
+      "loss": 0.4016,
+      "step": 1875
+    },
+    {
+      "epoch": 0.47,
+      "grad_norm": 1.1328125,
+      "learning_rate": 2.7352707832962865e-05,
+      "loss": 0.4191,
+      "step": 1880
+    },
+    {
+      "epoch": 0.47125,
+      "grad_norm": 1.2109375,
+      "learning_rate": 2.725495087602148e-05,
+      "loss": 0.5397,
+      "step": 1885
+    },
+    {
+      "epoch": 0.4725,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.7157159144948092e-05,
+      "loss": 0.4646,
+      "step": 1890
+    },
+    {
+      "epoch": 0.47375,
+      "grad_norm": 0.9453125,
+      "learning_rate": 2.7059334147812142e-05,
+      "loss": 0.4443,
+      "step": 1895
+    },
+    {
+      "epoch": 0.475,
+      "grad_norm": 0.95703125,
+      "learning_rate": 2.6961477393196126e-05,
+      "loss": 0.4943,
+      "step": 1900
+    },
+    {
+      "epoch": 0.47625,
+      "grad_norm": 0.8046875,
+      "learning_rate": 2.6863590390172243e-05,
+      "loss": 0.4654,
+      "step": 1905
+    },
+    {
+      "epoch": 0.4775,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.6765674648279172e-05,
+      "loss": 0.4517,
+      "step": 1910
+    },
+    {
+      "epoch": 0.47875,
+      "grad_norm": 1.1875,
+      "learning_rate": 2.666773167749878e-05,
+      "loss": 0.4525,
+      "step": 1915
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 1.09375,
+      "learning_rate": 2.656976298823284e-05,
+      "loss": 0.4676,
+      "step": 1920
+    },
+    {
+      "epoch": 0.48125,
+      "grad_norm": 1.4609375,
+      "learning_rate": 2.6471770091279724e-05,
+      "loss": 0.495,
+      "step": 1925
+    },
+    {
+      "epoch": 0.4825,
+      "grad_norm": 1.0234375,
+      "learning_rate": 2.637375449781115e-05,
+      "loss": 0.4322,
+      "step": 1930
+    },
+    {
+      "epoch": 0.48375,
+      "grad_norm": 1.1328125,
+      "learning_rate": 2.627571771934879e-05,
+      "loss": 0.4147,
+      "step": 1935
+    },
+    {
+      "epoch": 0.485,
+      "grad_norm": 1.078125,
+      "learning_rate": 2.6177661267741065e-05,
+      "loss": 0.4204,
+      "step": 1940
+    },
+    {
+      "epoch": 0.48625,
+      "grad_norm": 0.98828125,
+      "learning_rate": 2.607958665513976e-05,
+      "loss": 0.4245,
+      "step": 1945
+    },
+    {
+      "epoch": 0.4875,
+      "grad_norm": 0.95703125,
+      "learning_rate": 2.598149539397672e-05,
+      "loss": 0.4582,
+      "step": 1950
+    },
+    {
+      "epoch": 0.48875,
+      "grad_norm": 1.46875,
+      "learning_rate": 2.5883388996940534e-05,
+      "loss": 0.4445,
+      "step": 1955
+    },
+    {
+      "epoch": 0.49,
+      "grad_norm": 1.25,
+      "learning_rate": 2.578526897695321e-05,
+      "loss": 0.4533,
+      "step": 1960
+    },
+    {
+      "epoch": 0.49125,
+      "grad_norm": 1.0390625,
+      "learning_rate": 2.5687136847146838e-05,
+      "loss": 0.4334,
+      "step": 1965
+    },
+    {
+      "epoch": 0.4925,
+      "grad_norm": 0.89453125,
+      "learning_rate": 2.558899412084026e-05,
+      "loss": 0.434,
+      "step": 1970
+    },
+    {
+      "epoch": 0.49375,
+      "grad_norm": 1.25,
+      "learning_rate": 2.5490842311515707e-05,
+      "loss": 0.4257,
+      "step": 1975
+    },
+    {
+      "epoch": 0.495,
+      "grad_norm": 1.6953125,
+      "learning_rate": 2.539268293279552e-05,
+      "loss": 0.4503,
+      "step": 1980
+    },
+    {
+      "epoch": 0.49625,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.529451749841873e-05,
+      "loss": 0.5045,
+      "step": 1985
+    },
+    {
+      "epoch": 0.4975,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.5196347522217784e-05,
+      "loss": 0.4307,
+      "step": 1990
+    },
+    {
+      "epoch": 0.49875,
+      "grad_norm": 1.0078125,
+      "learning_rate": 2.509817451809515e-05,
+      "loss": 0.4701,
+      "step": 1995
+    },
+    {
+      "epoch": 0.5,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.5e-05,
+      "loss": 0.4573,
+      "step": 2000
+    },
+    {
+      "epoch": 0.50125,
+      "grad_norm": 1.2421875,
+      "learning_rate": 2.4901825481904855e-05,
+      "loss": 0.4304,
+      "step": 2005
+    },
+    {
+      "epoch": 0.5025,
+      "grad_norm": 1.578125,
+      "learning_rate": 2.480365247778223e-05,
+      "loss": 0.4334,
+      "step": 2010
+    },
+    {
+      "epoch": 0.50375,
+      "grad_norm": 1.0390625,
+      "learning_rate": 2.4705482501581266e-05,
+      "loss": 0.4507,
+      "step": 2015
+    },
+    {
+      "epoch": 0.505,
+      "grad_norm": 1.3984375,
+      "learning_rate": 2.460731706720449e-05,
+      "loss": 0.4555,
+      "step": 2020
+    },
+    {
+      "epoch": 0.50625,
+      "grad_norm": 1.1328125,
+      "learning_rate": 2.4509157688484295e-05,
+      "loss": 0.4791,
+      "step": 2025
+    },
+    {
+      "epoch": 0.5075,
+      "grad_norm": 1.171875,
+      "learning_rate": 2.4411005879159753e-05,
+      "loss": 0.4324,
+      "step": 2030
+    },
+    {
+      "epoch": 0.50875,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.4312863152853165e-05,
+      "loss": 0.4534,
+      "step": 2035
+    },
+    {
+      "epoch": 0.51,
+      "grad_norm": 1.5546875,
+      "learning_rate": 2.4214731023046793e-05,
+      "loss": 0.4411,
+      "step": 2040
+    },
+    {
+      "epoch": 0.51125,
+      "grad_norm": 1.3359375,
+      "learning_rate": 2.4116611003059472e-05,
+      "loss": 0.4333,
+      "step": 2045
+    },
+    {
+      "epoch": 0.5125,
+      "grad_norm": 1.46875,
+      "learning_rate": 2.4018504606023293e-05,
+      "loss": 0.4231,
+      "step": 2050
+    },
+    {
+      "epoch": 0.51375,
+      "grad_norm": 1.4140625,
+      "learning_rate": 2.392041334486024e-05,
+      "loss": 0.3752,
+      "step": 2055
+    },
+    {
+      "epoch": 0.515,
+      "grad_norm": 1.1328125,
+      "learning_rate": 2.3822338732258937e-05,
+      "loss": 0.4876,
+      "step": 2060
+    },
+    {
+      "epoch": 0.51625,
+      "grad_norm": 0.91796875,
+      "learning_rate": 2.3724282280651214e-05,
+      "loss": 0.3989,
+      "step": 2065
+    },
+    {
+      "epoch": 0.5175,
+      "grad_norm": 0.859375,
+      "learning_rate": 2.3626245502188864e-05,
+      "loss": 0.4102,
+      "step": 2070
+    },
+    {
+      "epoch": 0.51875,
+      "grad_norm": 0.9921875,
+      "learning_rate": 2.3528229908720272e-05,
+      "loss": 0.3997,
+      "step": 2075
+    },
+    {
+      "epoch": 0.52,
+      "grad_norm": 0.71484375,
+      "learning_rate": 2.3430237011767167e-05,
+      "loss": 0.4009,
+      "step": 2080
+    },
+    {
+      "epoch": 0.52125,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.3332268322501228e-05,
+      "loss": 0.4769,
+      "step": 2085
+    },
+    {
+      "epoch": 0.5225,
+      "grad_norm": 1.078125,
+      "learning_rate": 2.323432535172084e-05,
+      "loss": 0.4405,
+      "step": 2090
+    },
+    {
+      "epoch": 0.52375,
+      "grad_norm": 1.0625,
+      "learning_rate": 2.313640960982776e-05,
+      "loss": 0.4436,
+      "step": 2095
+    },
+    {
+      "epoch": 0.525,
+      "grad_norm": 1.09375,
+      "learning_rate": 2.303852260680388e-05,
+      "loss": 0.4027,
+      "step": 2100
+    },
+    {
+      "epoch": 0.52625,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.294066585218786e-05,
+      "loss": 0.4086,
+      "step": 2105
+    },
+    {
+      "epoch": 0.5275,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.284284085505192e-05,
+      "loss": 0.4262,
+      "step": 2110
+    },
+    {
+      "epoch": 0.52875,
+      "grad_norm": 1.484375,
+      "learning_rate": 2.274504912397852e-05,
+      "loss": 0.4605,
+      "step": 2115
+    },
+    {
+      "epoch": 0.53,
+      "grad_norm": 1.2734375,
+      "learning_rate": 2.2647292167037144e-05,
+      "loss": 0.4534,
+      "step": 2120
+    },
+    {
+      "epoch": 0.53125,
+      "grad_norm": 0.890625,
+      "learning_rate": 2.2549571491760986e-05,
+      "loss": 0.3628,
+      "step": 2125
+    },
+    {
+      "epoch": 0.5325,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.2451888605123754e-05,
+      "loss": 0.4879,
+      "step": 2130
+    },
+    {
+      "epoch": 0.53375,
+      "grad_norm": 0.96875,
+      "learning_rate": 2.2354245013516393e-05,
+      "loss": 0.4517,
+      "step": 2135
+    },
+    {
+      "epoch": 0.535,
+      "grad_norm": 1.1640625,
+      "learning_rate": 2.225664222272387e-05,
+      "loss": 0.4303,
+      "step": 2140
+    },
+    {
+      "epoch": 0.53625,
+      "grad_norm": 0.83203125,
+      "learning_rate": 2.2159081737901975e-05,
+      "loss": 0.4172,
+      "step": 2145
+    },
+    {
+      "epoch": 0.5375,
+      "grad_norm": 1.125,
+      "learning_rate": 2.2061565063554064e-05,
+      "loss": 0.4169,
+      "step": 2150
+    },
+    {
+      "epoch": 0.53875,
+      "grad_norm": 1.25,
+      "learning_rate": 2.1964093703507893e-05,
+      "loss": 0.4839,
+      "step": 2155
+    },
+    {
+      "epoch": 0.54,
+      "grad_norm": 0.8359375,
+      "learning_rate": 2.186666916089239e-05,
+      "loss": 0.3919,
+      "step": 2160
+    },
+    {
+      "epoch": 0.54125,
+      "grad_norm": 1.2421875,
+      "learning_rate": 2.1769292938114563e-05,
+      "loss": 0.4435,
+      "step": 2165
+    },
+    {
+      "epoch": 0.5425,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.1671966536836196e-05,
+      "loss": 0.4902,
+      "step": 2170
+    },
+    {
+      "epoch": 0.54375,
+      "grad_norm": 1.296875,
+      "learning_rate": 2.1574691457950803e-05,
+      "loss": 0.4667,
+      "step": 2175
+    },
+    {
+      "epoch": 0.545,
+      "grad_norm": 1.1015625,
+      "learning_rate": 2.1477469201560435e-05,
+      "loss": 0.3795,
+      "step": 2180
+    },
+    {
+      "epoch": 0.54625,
+      "grad_norm": 1.140625,
+      "learning_rate": 2.1380301266952556e-05,
+      "loss": 0.4658,
+      "step": 2185
+    },
+    {
+      "epoch": 0.5475,
+      "grad_norm": 1.6171875,
+      "learning_rate": 2.1283189152576925e-05,
+      "loss": 0.4589,
+      "step": 2190
+    },
+    {
+      "epoch": 0.54875,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.118613435602248e-05,
+      "loss": 0.4394,
+      "step": 2195
+    },
+    {
+      "epoch": 0.55,
+      "grad_norm": 1.0390625,
+      "learning_rate": 2.1089138373994223e-05,
+      "loss": 0.4321,
+      "step": 2200
+    },
+    {
+      "epoch": 0.55125,
+      "grad_norm": 1.3046875,
+      "learning_rate": 2.0992202702290227e-05,
+      "loss": 0.4084,
+      "step": 2205
+    },
+    {
+      "epoch": 0.5525,
+      "grad_norm": 1.2890625,
+      "learning_rate": 2.089532883577843e-05,
+      "loss": 0.4489,
+      "step": 2210
+    },
+    {
+      "epoch": 0.55375,
+      "grad_norm": 1.2421875,
+      "learning_rate": 2.0798518268373706e-05,
+      "loss": 0.4403,
+      "step": 2215
+    },
+    {
+      "epoch": 0.555,
+      "grad_norm": 1.1796875,
+      "learning_rate": 2.070177249301476e-05,
+      "loss": 0.4286,
+      "step": 2220
+    },
+    {
+      "epoch": 0.55625,
+      "grad_norm": 1.28125,
+      "learning_rate": 2.0605093001641138e-05,
+      "loss": 0.4557,
+      "step": 2225
+    },
+    {
+      "epoch": 0.5575,
+      "grad_norm": 1.171875,
+      "learning_rate": 2.0508481285170186e-05,
+      "loss": 0.4686,
+      "step": 2230
+    },
+    {
+      "epoch": 0.55875,
+      "grad_norm": 1.375,
+      "learning_rate": 2.04119388334741e-05,
+      "loss": 0.4402,
+      "step": 2235
+    },
+    {
+      "epoch": 0.56,
+      "grad_norm": 1.1953125,
+      "learning_rate": 2.031546713535688e-05,
+      "loss": 0.3973,
+      "step": 2240
+    },
+    {
+      "epoch": 0.56125,
+      "grad_norm": 1.1015625,
+      "learning_rate": 2.0219067678531494e-05,
+      "loss": 0.4349,
+      "step": 2245
+    },
+    {
+      "epoch": 0.5625,
+      "grad_norm": 1.1015625,
+      "learning_rate": 2.0122741949596797e-05,
+      "loss": 0.4329,
+      "step": 2250
+    },
+    {
+      "epoch": 0.56375,
+      "grad_norm": 1.1875,
+      "learning_rate": 2.002649143401469e-05,
+      "loss": 0.4402,
+      "step": 2255
+    },
+    {
+      "epoch": 0.565,
+      "grad_norm": 1.0546875,
+      "learning_rate": 1.9930317616087196e-05,
+      "loss": 0.4342,
+      "step": 2260
+    },
+    {
+      "epoch": 0.56625,
+      "grad_norm": 0.98046875,
+      "learning_rate": 1.9834221978933543e-05,
+      "loss": 0.4537,
+      "step": 2265
+    },
+    {
+      "epoch": 0.5675,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.9738206004467363e-05,
+      "loss": 0.4597,
+      "step": 2270
+    },
+    {
+      "epoch": 0.56875,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.9642271173373737e-05,
+      "loss": 0.4372,
+      "step": 2275
+    },
+    {
+      "epoch": 0.57,
+      "grad_norm": 1.171875,
+      "learning_rate": 1.9546418965086442e-05,
+      "loss": 0.4062,
+      "step": 2280
+    },
+    {
+      "epoch": 0.57125,
+      "grad_norm": 1.453125,
+      "learning_rate": 1.9450650857765102e-05,
+      "loss": 0.4698,
+      "step": 2285
+    },
+    {
+      "epoch": 0.5725,
+      "grad_norm": 1.140625,
+      "learning_rate": 1.935496832827241e-05,
+      "loss": 0.4312,
+      "step": 2290
+    },
+    {
+      "epoch": 0.57375,
+      "grad_norm": 1.21875,
+      "learning_rate": 1.925937285215133e-05,
+      "loss": 0.4643,
+      "step": 2295
+    },
+    {
+      "epoch": 0.575,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.9163865903602374e-05,
+      "loss": 0.4256,
+      "step": 2300
+    },
+    {
+      "epoch": 0.57625,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.9068448955460805e-05,
+      "loss": 0.3879,
+      "step": 2305
+    },
+    {
+      "epoch": 0.5775,
+      "grad_norm": 1.5390625,
+      "learning_rate": 1.897312347917404e-05,
+      "loss": 0.4048,
+      "step": 2310
+    },
+    {
+      "epoch": 0.57875,
+      "grad_norm": 1.03125,
+      "learning_rate": 1.8877890944778815e-05,
+      "loss": 0.4572,
+      "step": 2315
+    },
+    {
+      "epoch": 0.58,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.8782752820878634e-05,
+      "loss": 0.4634,
+      "step": 2320
+    },
+    {
+      "epoch": 0.58125,
+      "grad_norm": 1.2265625,
+      "learning_rate": 1.868771057462105e-05,
+      "loss": 0.3985,
+      "step": 2325
+    },
+    {
+      "epoch": 0.5825,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.8592765671675084e-05,
+      "loss": 0.4463,
+      "step": 2330
+    },
+    {
+      "epoch": 0.58375,
+      "grad_norm": 0.91015625,
+      "learning_rate": 1.8497919576208585e-05,
+      "loss": 0.4083,
+      "step": 2335
+    },
+    {
+      "epoch": 0.585,
+      "grad_norm": 1.4921875,
+      "learning_rate": 1.8403173750865685e-05,
+      "loss": 0.3929,
+      "step": 2340
+    },
+    {
+      "epoch": 0.58625,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.830852965674419e-05,
+      "loss": 0.4659,
+      "step": 2345
+    },
+    {
+      "epoch": 0.5875,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.8213988753373146e-05,
+      "loss": 0.3986,
+      "step": 2350
+    },
+    {
+      "epoch": 0.58875,
+      "grad_norm": 1.046875,
+      "learning_rate": 1.8119552498690215e-05,
+      "loss": 0.4043,
+      "step": 2355
+    },
+    {
+      "epoch": 0.59,
+      "grad_norm": 2.84375,
+      "learning_rate": 1.802522234901927e-05,
+      "loss": 0.449,
+      "step": 2360
+    },
+    {
+      "epoch": 0.59125,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.793099975904791e-05,
+      "loss": 0.4178,
+      "step": 2365
+    },
+    {
+      "epoch": 0.5925,
+      "grad_norm": 1.390625,
+      "learning_rate": 1.783688618180504e-05,
+      "loss": 0.4422,
+      "step": 2370
+    },
+    {
+      "epoch": 0.59375,
+      "grad_norm": 0.94921875,
+      "learning_rate": 1.7742883068638447e-05,
+      "loss": 0.4666,
+      "step": 2375
+    },
+    {
+      "epoch": 0.595,
+      "grad_norm": 1.015625,
+      "learning_rate": 1.7648991869192405e-05,
+      "loss": 0.4226,
+      "step": 2380
+    },
+    {
+      "epoch": 0.59625,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.7555214031385375e-05,
+      "loss": 0.408,
+      "step": 2385
+    },
+    {
+      "epoch": 0.5975,
+      "grad_norm": 1.7109375,
+      "learning_rate": 1.746155100138761e-05,
+      "loss": 0.4778,
+      "step": 2390
+    },
+    {
+      "epoch": 0.59875,
+      "grad_norm": 1.25,
+      "learning_rate": 1.7368004223598912e-05,
+      "loss": 0.4059,
+      "step": 2395
+    },
+    {
+      "epoch": 0.6,
+      "grad_norm": 1.296875,
+      "learning_rate": 1.7274575140626318e-05,
+      "loss": 0.4398,
+      "step": 2400
+    },
+    {
+      "epoch": 0.60125,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.7181265193261865e-05,
+      "loss": 0.482,
+      "step": 2405
+    },
+    {
+      "epoch": 0.6025,
+      "grad_norm": 0.84765625,
+      "learning_rate": 1.7088075820460346e-05,
+      "loss": 0.4192,
+      "step": 2410
+    },
+    {
+      "epoch": 0.60375,
+      "grad_norm": 1.3125,
+      "learning_rate": 1.6995008459317206e-05,
+      "loss": 0.4748,
+      "step": 2415
+    },
+    {
+      "epoch": 0.605,
+      "grad_norm": 1.5703125,
+      "learning_rate": 1.690206454504627e-05,
+      "loss": 0.4276,
+      "step": 2420
+    },
+    {
+      "epoch": 0.60625,
+      "grad_norm": 1.125,
+      "learning_rate": 1.6809245510957665e-05,
+      "loss": 0.3548,
+      "step": 2425
+    },
+    {
+      "epoch": 0.6075,
+      "grad_norm": 1.0078125,
+      "learning_rate": 1.6716552788435724e-05,
+      "loss": 0.4122,
+      "step": 2430
+    },
+    {
+      "epoch": 0.60875,
+      "grad_norm": 1.1484375,
+      "learning_rate": 1.66239878069169e-05,
+      "loss": 0.432,
+      "step": 2435
+    },
+    {
+      "epoch": 0.61,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.6531551993867717e-05,
+      "loss": 0.4467,
+      "step": 2440
+    },
+    {
+      "epoch": 0.61125,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.643924677476276e-05,
+      "loss": 0.4652,
+      "step": 2445
+    },
+    {
+      "epoch": 0.6125,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.6347073573062672e-05,
+      "loss": 0.4024,
+      "step": 2450
+    },
+    {
+      "epoch": 0.61375,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.6255033810192282e-05,
+      "loss": 0.3974,
+      "step": 2455
+    },
+    {
+      "epoch": 0.615,
+      "grad_norm": 1.09375,
+      "learning_rate": 1.6163128905518578e-05,
+      "loss": 0.3891,
+      "step": 2460
+    },
+    {
+      "epoch": 0.61625,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.6071360276328874e-05,
+      "loss": 0.3499,
+      "step": 2465
+    },
+    {
+      "epoch": 0.6175,
+      "grad_norm": 1.109375,
+      "learning_rate": 1.5979729337808955e-05,
+      "loss": 0.4386,
+      "step": 2470
+    },
+    {
+      "epoch": 0.61875,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.588823750302126e-05,
+      "loss": 0.4494,
+      "step": 2475
+    },
+    {
+      "epoch": 0.62,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.5796886182883053e-05,
+      "loss": 0.4025,
+      "step": 2480
+    },
+    {
+      "epoch": 0.62125,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.57056767861447e-05,
+      "loss": 0.4192,
+      "step": 2485
+    },
+    {
+      "epoch": 0.6225,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.561461071936792e-05,
+      "loss": 0.4509,
+      "step": 2490
+    },
+    {
+      "epoch": 0.62375,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.552368938690414e-05,
+      "loss": 0.3897,
+      "step": 2495
+    },
+    {
+      "epoch": 0.625,
+      "grad_norm": 1.28125,
+      "learning_rate": 1.5432914190872757e-05,
+      "loss": 0.473,
+      "step": 2500
+    },
+    {
+      "epoch": 0.62625,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.5342286531139605e-05,
+      "loss": 0.4333,
+      "step": 2505
+    },
+    {
+      "epoch": 0.6275,
+      "grad_norm": 1.0546875,
+      "learning_rate": 1.5251807805295302e-05,
+      "loss": 0.4245,
+      "step": 2510
+    },
+    {
+      "epoch": 0.62875,
+      "grad_norm": 1.1953125,
+      "learning_rate": 1.5161479408633713e-05,
+      "loss": 0.4342,
+      "step": 2515
+    },
+    {
+      "epoch": 0.63,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.5071302734130489e-05,
+      "loss": 0.3951,
+      "step": 2520
+    },
+    {
+      "epoch": 0.63125,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.498127917242148e-05,
+      "loss": 0.4367,
+      "step": 2525
+    },
+    {
+      "epoch": 0.6325,
+      "grad_norm": 1.2578125,
+      "learning_rate": 1.4891410111781378e-05,
+      "loss": 0.4766,
+      "step": 2530
+    },
+    {
+      "epoch": 0.63375,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.4801696938102272e-05,
+      "loss": 0.373,
+      "step": 2535
+    },
+    {
+      "epoch": 0.635,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.4712141034872282e-05,
+      "loss": 0.3804,
+      "step": 2540
+    },
+    {
+      "epoch": 0.63625,
+      "grad_norm": 1.09375,
+      "learning_rate": 1.4622743783154223e-05,
+      "loss": 0.4206,
+      "step": 2545
+    },
+    {
+      "epoch": 0.6375,
+      "grad_norm": 1.1875,
+      "learning_rate": 1.4533506561564306e-05,
+      "loss": 0.4585,
+      "step": 2550
+    },
+    {
+      "epoch": 0.63875,
+      "grad_norm": 0.84765625,
+      "learning_rate": 1.4444430746250867e-05,
+      "loss": 0.3796,
+      "step": 2555
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 1.0625,
+      "learning_rate": 1.4355517710873184e-05,
+      "loss": 0.4296,
+      "step": 2560
+    },
+    {
+      "epoch": 0.64125,
+      "grad_norm": 1.6796875,
+      "learning_rate": 1.4266768826580257e-05,
+      "loss": 0.5008,
+      "step": 2565
+    },
+    {
+      "epoch": 0.6425,
+      "grad_norm": 1.0625,
+      "learning_rate": 1.4178185461989662e-05,
+      "loss": 0.3952,
+      "step": 2570
+    },
+    {
+      "epoch": 0.64375,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.4089768983166444e-05,
+      "loss": 0.4494,
+      "step": 2575
+    },
+    {
+      "epoch": 0.645,
+      "grad_norm": 0.88671875,
+      "learning_rate": 1.4001520753602121e-05,
+      "loss": 0.3944,
+      "step": 2580
+    },
+    {
+      "epoch": 0.64625,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.3913442134193544e-05,
+      "loss": 0.4276,
+      "step": 2585
+    },
+    {
+      "epoch": 0.6475,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.3825534483221974e-05,
+      "loss": 0.4433,
+      "step": 2590
+    },
+    {
+      "epoch": 0.64875,
+      "grad_norm": 1.1640625,
+      "learning_rate": 1.3737799156332143e-05,
+      "loss": 0.3992,
+      "step": 2595
+    },
+    {
+      "epoch": 0.65,
+      "grad_norm": 0.96875,
+      "learning_rate": 1.3650237506511331e-05,
+      "loss": 0.4488,
+      "step": 2600
+    },
+    {
+      "epoch": 0.65125,
+      "grad_norm": 1.3828125,
+      "learning_rate": 1.3562850884068487e-05,
+      "loss": 0.4243,
+      "step": 2605
+    },
+    {
+      "epoch": 0.6525,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.3475640636613446e-05,
+      "loss": 0.3477,
+      "step": 2610
+    },
+    {
+      "epoch": 0.65375,
+      "grad_norm": 1.2734375,
+      "learning_rate": 1.3388608109036086e-05,
+      "loss": 0.4413,
+      "step": 2615
+    },
+    {
+      "epoch": 0.655,
+      "grad_norm": 1.0625,
+      "learning_rate": 1.330175464348567e-05,
+      "loss": 0.4487,
+      "step": 2620
+    },
+    {
+      "epoch": 0.65625,
+      "grad_norm": 1.28125,
+      "learning_rate": 1.3215081579350058e-05,
+      "loss": 0.4122,
+      "step": 2625
+    },
+    {
+      "epoch": 0.6575,
+      "grad_norm": 1.09375,
+      "learning_rate": 1.312859025323514e-05,
+      "loss": 0.424,
+      "step": 2630
+    },
+    {
+      "epoch": 0.65875,
+      "grad_norm": 1.078125,
+      "learning_rate": 1.3042281998944151e-05,
+      "loss": 0.4013,
+      "step": 2635
+    },
+    {
+      "epoch": 0.66,
+      "grad_norm": 1.9375,
+      "learning_rate": 1.2956158147457115e-05,
+      "loss": 0.5066,
+      "step": 2640
+    },
+    {
+      "epoch": 0.66125,
+      "grad_norm": 1.25,
+      "learning_rate": 1.2870220026910407e-05,
+      "loss": 0.3935,
+      "step": 2645
+    },
+    {
+      "epoch": 0.6625,
+      "grad_norm": 1.21875,
+      "learning_rate": 1.2784468962576136e-05,
+      "loss": 0.4039,
+      "step": 2650
+    },
+    {
+      "epoch": 0.66375,
+      "grad_norm": 1.1875,
+      "learning_rate": 1.2698906276841776e-05,
+      "loss": 0.4817,
+      "step": 2655
+    },
+    {
+      "epoch": 0.665,
+      "grad_norm": 0.99609375,
+      "learning_rate": 1.261353328918981e-05,
+      "loss": 0.3917,
+      "step": 2660
+    },
+    {
+      "epoch": 0.66625,
+      "grad_norm": 1.2421875,
+      "learning_rate": 1.2528351316177319e-05,
+      "loss": 0.425,
+      "step": 2665
+    },
+    {
+      "epoch": 0.6675,
+      "grad_norm": 1.015625,
+      "learning_rate": 1.2443361671415687e-05,
+      "loss": 0.4234,
+      "step": 2670
+    },
+    {
+      "epoch": 0.66875,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.235856566555039e-05,
+      "loss": 0.4414,
+      "step": 2675
+    },
+    {
+      "epoch": 0.67,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.2273964606240718e-05,
+      "loss": 0.4563,
+      "step": 2680
+    },
+    {
+      "epoch": 0.67125,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.2189559798139682e-05,
+      "loss": 0.4132,
+      "step": 2685
+    },
+    {
+      "epoch": 0.6725,
+      "grad_norm": 1.765625,
+      "learning_rate": 1.2105352542873815e-05,
+      "loss": 0.4317,
+      "step": 2690
+    },
+    {
+      "epoch": 0.67375,
+      "grad_norm": 0.9296875,
+      "learning_rate": 1.2021344139023186e-05,
+      "loss": 0.4073,
+      "step": 2695
+    },
+    {
+      "epoch": 0.675,
+      "grad_norm": 1.4765625,
+      "learning_rate": 1.1937535882101281e-05,
+      "loss": 0.4147,
+      "step": 2700
+    },
+    {
+      "epoch": 0.67625,
+      "grad_norm": 1.5625,
+      "learning_rate": 1.1853929064535111e-05,
+      "loss": 0.4394,
+      "step": 2705
+    },
+    {
+      "epoch": 0.6775,
+      "grad_norm": 1.40625,
+      "learning_rate": 1.1770524975645238e-05,
+      "loss": 0.461,
+      "step": 2710
+    },
+    {
+      "epoch": 0.67875,
+      "grad_norm": 1.0,
+      "learning_rate": 1.1687324901625879e-05,
+      "loss": 0.4279,
+      "step": 2715
+    },
+    {
+      "epoch": 0.68,
+      "grad_norm": 0.9765625,
+      "learning_rate": 1.1604330125525079e-05,
+      "loss": 0.4201,
+      "step": 2720
+    },
+    {
+      "epoch": 0.68125,
+      "grad_norm": 0.98828125,
+      "learning_rate": 1.1521541927224994e-05,
+      "loss": 0.4392,
+      "step": 2725
+    },
+    {
+      "epoch": 0.6825,
+      "grad_norm": 1.140625,
+      "learning_rate": 1.1438961583422037e-05,
+      "loss": 0.4064,
+      "step": 2730
+    },
+    {
+      "epoch": 0.68375,
+      "grad_norm": 1.1796875,
+      "learning_rate": 1.1356590367607252e-05,
+      "loss": 0.4081,
+      "step": 2735
+    },
+    {
+      "epoch": 0.685,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.1274429550046704e-05,
+      "loss": 0.4629,
+      "step": 2740
+    },
+    {
+      "epoch": 0.68625,
+      "grad_norm": 0.9453125,
+      "learning_rate": 1.1192480397761837e-05,
+      "loss": 0.3942,
+      "step": 2745
+    },
+    {
+      "epoch": 0.6875,
+      "grad_norm": 0.9765625,
+      "learning_rate": 1.1110744174509952e-05,
+      "loss": 0.4581,
+      "step": 2750
+    },
+    {
+      "epoch": 0.68875,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.1029222140764712e-05,
+      "loss": 0.4079,
+      "step": 2755
+    },
+    {
+      "epoch": 0.69,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.0947915553696742e-05,
+      "loss": 0.3924,
+      "step": 2760
+    },
+    {
+      "epoch": 0.69125,
+      "grad_norm": 0.94921875,
+      "learning_rate": 1.0866825667154182e-05,
+      "loss": 0.3715,
+      "step": 2765
+    },
+    {
+      "epoch": 0.6925,
+      "grad_norm": 1.09375,
+      "learning_rate": 1.07859537316434e-05,
+      "loss": 0.4238,
+      "step": 2770
+    },
+    {
+      "epoch": 0.69375,
+      "grad_norm": 1.3203125,
+      "learning_rate": 1.0705300994309697e-05,
+      "loss": 0.4465,
+      "step": 2775
+    },
+    {
+      "epoch": 0.695,
+      "grad_norm": 1.046875,
+      "learning_rate": 1.0624868698918045e-05,
+      "loss": 0.4295,
+      "step": 2780
+    },
+    {
+      "epoch": 0.69625,
+      "grad_norm": 1.140625,
+      "learning_rate": 1.0544658085833919e-05,
+      "loss": 0.4527,
+      "step": 2785
+    },
+    {
+      "epoch": 0.6975,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.0464670392004235e-05,
+      "loss": 0.4721,
+      "step": 2790
+    },
+    {
+      "epoch": 0.69875,
+      "grad_norm": 1.234375,
+      "learning_rate": 1.0384906850938166e-05,
+      "loss": 0.4632,
+      "step": 2795
+    },
+    {
+      "epoch": 0.7,
+      "grad_norm": 1.03125,
+      "learning_rate": 1.0305368692688174e-05,
+      "loss": 0.4382,
+      "step": 2800
+    },
+    {
+      "epoch": 0.70125,
+      "grad_norm": 1.046875,
+      "learning_rate": 1.0226057143831064e-05,
+      "loss": 0.4699,
+      "step": 2805
+    },
+    {
+      "epoch": 0.7025,
+      "grad_norm": 1.4296875,
+      "learning_rate": 1.0146973427449038e-05,
+      "loss": 0.4368,
+      "step": 2810
+    },
+    {
+      "epoch": 0.70375,
+      "grad_norm": 0.9765625,
+      "learning_rate": 1.0068118763110824e-05,
+      "loss": 0.4513,
+      "step": 2815
+    },
+    {
+      "epoch": 0.705,
+      "grad_norm": 1.0,
+      "learning_rate": 9.989494366852904e-06,
+      "loss": 0.3863,
+      "step": 2820
+    },
+    {
+      "epoch": 0.70625,
+      "grad_norm": 1.0390625,
+      "learning_rate": 9.911101451160715e-06,
+      "loss": 0.3907,
+      "step": 2825
+    },
+    {
+      "epoch": 0.7075,
+      "grad_norm": 1.078125,
+      "learning_rate": 9.832941224950012e-06,
+      "loss": 0.4537,
+      "step": 2830
+    },
+    {
+      "epoch": 0.70875,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.755014893548157e-06,
+      "loss": 0.4082,
+      "step": 2835
+    },
+    {
+      "epoch": 0.71,
+      "grad_norm": 1.1015625,
+      "learning_rate": 9.677323658675594e-06,
+      "loss": 0.3992,
+      "step": 2840
+    },
+    {
+      "epoch": 0.71125,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.599868718427257e-06,
+      "loss": 0.4512,
+      "step": 2845
+    },
+    {
+      "epoch": 0.7125,
+      "grad_norm": 0.8984375,
+      "learning_rate": 9.522651267254149e-06,
+      "loss": 0.419,
+      "step": 2850
+    },
+    {
+      "epoch": 0.71375,
+      "grad_norm": 1.3671875,
+      "learning_rate": 9.445672495944899e-06,
+      "loss": 0.4542,
+      "step": 2855
+    },
+    {
+      "epoch": 0.715,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.368933591607378e-06,
+      "loss": 0.4554,
+      "step": 2860
+    },
+    {
+      "epoch": 0.71625,
+      "grad_norm": 1.3359375,
+      "learning_rate": 9.292435737650407e-06,
+      "loss": 0.4158,
+      "step": 2865
+    },
+    {
+      "epoch": 0.7175,
+      "grad_norm": 0.8671875,
+      "learning_rate": 9.216180113765558e-06,
+      "loss": 0.4145,
+      "step": 2870
+    },
+    {
+      "epoch": 0.71875,
+      "grad_norm": 1.1640625,
+      "learning_rate": 9.140167895908867e-06,
+      "loss": 0.4276,
+      "step": 2875
+    },
+    {
+      "epoch": 0.72,
+      "grad_norm": 1.125,
+      "learning_rate": 9.064400256282757e-06,
+      "loss": 0.4477,
+      "step": 2880
+    },
+    {
+      "epoch": 0.72125,
+      "grad_norm": 1.1796875,
+      "learning_rate": 8.988878363317979e-06,
+      "loss": 0.4563,
+      "step": 2885
+    },
+    {
+      "epoch": 0.7225,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.913603381655528e-06,
+      "loss": 0.4396,
+      "step": 2890
+    },
+    {
+      "epoch": 0.72375,
+      "grad_norm": 1.5390625,
+      "learning_rate": 8.838576472128756e-06,
+      "loss": 0.4831,
+      "step": 2895
+    },
+    {
+      "epoch": 0.725,
+      "grad_norm": 1.03125,
+      "learning_rate": 8.763798791745411e-06,
+      "loss": 0.4437,
+      "step": 2900
+    },
+    {
+      "epoch": 0.72625,
+      "grad_norm": 1.140625,
+      "learning_rate": 8.689271493669837e-06,
+      "loss": 0.4639,
+      "step": 2905
+    },
+    {
+      "epoch": 0.7275,
+      "grad_norm": 1.421875,
+      "learning_rate": 8.614995727205156e-06,
+      "loss": 0.4215,
+      "step": 2910
+    },
+    {
+      "epoch": 0.72875,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.540972637775572e-06,
+      "loss": 0.4615,
+      "step": 2915
+    },
+    {
+      "epoch": 0.73,
+      "grad_norm": 1.0703125,
+      "learning_rate": 8.467203366908707e-06,
+      "loss": 0.4043,
+      "step": 2920
+    },
+    {
+      "epoch": 0.73125,
+      "grad_norm": 1.3046875,
+      "learning_rate": 8.393689052217966e-06,
+      "loss": 0.4634,
+      "step": 2925
+    },
+    {
+      "epoch": 0.7325,
+      "grad_norm": 1.328125,
+      "learning_rate": 8.320430827385003e-06,
+      "loss": 0.4411,
+      "step": 2930
+    },
+    {
+      "epoch": 0.73375,
+      "grad_norm": 1.2265625,
+      "learning_rate": 8.24742982214231e-06,
+      "loss": 0.3556,
+      "step": 2935
+    },
+    {
+      "epoch": 0.735,
+      "grad_norm": 1.0234375,
+      "learning_rate": 8.174687162255672e-06,
+      "loss": 0.4456,
+      "step": 2940
+    },
+    {
+      "epoch": 0.73625,
+      "grad_norm": 0.8984375,
+      "learning_rate": 8.102203969506886e-06,
+      "loss": 0.4254,
+      "step": 2945
+    },
+    {
+      "epoch": 0.7375,
+      "grad_norm": 1.234375,
+      "learning_rate": 8.029981361676456e-06,
+      "loss": 0.427,
+      "step": 2950
+    },
+    {
+      "epoch": 0.73875,
+      "grad_norm": 1.0390625,
+      "learning_rate": 7.958020452526346e-06,
+      "loss": 0.4323,
+      "step": 2955
+    },
+    {
+      "epoch": 0.74,
+      "grad_norm": 1.125,
+      "learning_rate": 7.886322351782783e-06,
+      "loss": 0.3968,
+      "step": 2960
+    },
+    {
+      "epoch": 0.74125,
+      "grad_norm": 1.0234375,
+      "learning_rate": 7.814888165119186e-06,
+      "loss": 0.4628,
+      "step": 2965
+    },
+    {
+      "epoch": 0.7425,
+      "grad_norm": 1.0078125,
+      "learning_rate": 7.743718994139071e-06,
+      "loss": 0.4388,
+      "step": 2970
+    },
+    {
+      "epoch": 0.74375,
+      "grad_norm": 1.34375,
+      "learning_rate": 7.672815936359107e-06,
+      "loss": 0.4029,
+      "step": 2975
+    },
+    {
+      "epoch": 0.745,
+      "grad_norm": 1.0703125,
+      "learning_rate": 7.602180085192143e-06,
+      "loss": 0.4214,
+      "step": 2980
+    },
+    {
+      "epoch": 0.74625,
+      "grad_norm": 1.4296875,
+      "learning_rate": 7.531812529930398e-06,
+      "loss": 0.4165,
+      "step": 2985
+    },
+    {
+      "epoch": 0.7475,
+      "grad_norm": 1.1171875,
+      "learning_rate": 7.461714355728608e-06,
+      "loss": 0.4016,
+      "step": 2990
+    },
+    {
+      "epoch": 0.74875,
+      "grad_norm": 1.203125,
+      "learning_rate": 7.391886643587342e-06,
+      "loss": 0.4527,
+      "step": 2995
+    },
+    {
+      "epoch": 0.75,
+      "grad_norm": 1.15625,
+      "learning_rate": 7.3223304703363135e-06,
+      "loss": 0.4143,
+      "step": 3000
+    },
+    {
+      "epoch": 0.75125,
+      "grad_norm": 1.328125,
+      "learning_rate": 7.253046908617747e-06,
+      "loss": 0.4667,
+      "step": 3005
+    },
+    {
+      "epoch": 0.7525,
+      "grad_norm": 1.2734375,
+      "learning_rate": 7.184037026869867e-06,
+      "loss": 0.4032,
+      "step": 3010
+    },
+    {
+      "epoch": 0.75375,
+      "grad_norm": 1.0859375,
+      "learning_rate": 7.115301889310427e-06,
+      "loss": 0.433,
+      "step": 3015
+    },
+    {
+      "epoch": 0.755,
+      "grad_norm": 1.046875,
+      "learning_rate": 7.046842555920283e-06,
+      "loss": 0.4017,
+      "step": 3020
+    },
+    {
+      "epoch": 0.75625,
+      "grad_norm": 1.0546875,
+      "learning_rate": 6.9786600824270296e-06,
+      "loss": 0.4006,
+      "step": 3025
+    },
+    {
+      "epoch": 0.7575,
+      "grad_norm": 1.3515625,
+      "learning_rate": 6.91075552028877e-06,
+      "loss": 0.4536,
+      "step": 3030
+    },
+    {
+      "epoch": 0.75875,
+      "grad_norm": 1.4609375,
+      "learning_rate": 6.84312991667784e-06,
+      "loss": 0.4295,
+      "step": 3035
+    },
+    {
+      "epoch": 0.76,
+      "grad_norm": 0.9375,
+      "learning_rate": 6.775784314464717e-06,
+      "loss": 0.4216,
+      "step": 3040
+    },
+    {
+      "epoch": 0.76125,
+      "grad_norm": 1.1328125,
+      "learning_rate": 6.708719752201884e-06,
+      "loss": 0.4071,
+      "step": 3045
+    },
+    {
+      "epoch": 0.7625,
+      "grad_norm": 0.87109375,
+      "learning_rate": 6.641937264107867e-06,
+      "loss": 0.4518,
+      "step": 3050
+    },
+    {
+      "epoch": 0.76375,
+      "grad_norm": 1.453125,
+      "learning_rate": 6.575437880051233e-06,
+      "loss": 0.4776,
+      "step": 3055
+    },
+    {
+      "epoch": 0.765,
+      "grad_norm": 1.1796875,
+      "learning_rate": 6.509222625534755e-06,
+      "loss": 0.4084,
+      "step": 3060
+    },
+    {
+      "epoch": 0.76625,
+      "grad_norm": 1.4609375,
+      "learning_rate": 6.443292521679578e-06,
+      "loss": 0.4825,
+      "step": 3065
+    },
+    {
+      "epoch": 0.7675,
+      "grad_norm": 1.171875,
+      "learning_rate": 6.377648585209456e-06,
+      "loss": 0.4788,
+      "step": 3070
+    },
+    {
+      "epoch": 0.76875,
+      "grad_norm": 1.1953125,
+      "learning_rate": 6.312291828435077e-06,
+      "loss": 0.4077,
+      "step": 3075
+    },
+    {
+      "epoch": 0.77,
+      "grad_norm": 0.98046875,
+      "learning_rate": 6.247223259238511e-06,
+      "loss": 0.4103,
+      "step": 3080
+    },
+    {
+      "epoch": 0.77125,
+      "grad_norm": 1.21875,
+      "learning_rate": 6.182443881057576e-06,
+      "loss": 0.4401,
+      "step": 3085
+    },
+    {
+      "epoch": 0.7725,
+      "grad_norm": 1.3984375,
+      "learning_rate": 6.117954692870412e-06,
+      "loss": 0.4628,
+      "step": 3090
+    },
+    {
+      "epoch": 0.77375,
+      "grad_norm": 1.15625,
+      "learning_rate": 6.053756689180082e-06,
+      "loss": 0.3789,
+      "step": 3095
+    },
+    {
+      "epoch": 0.775,
+      "grad_norm": 0.9609375,
+      "learning_rate": 5.989850859999227e-06,
+      "loss": 0.4261,
+      "step": 3100
+    },
+    {
+      "epoch": 0.77625,
+      "grad_norm": 1.3359375,
+      "learning_rate": 5.926238190834779e-06,
+      "loss": 0.4548,
+      "step": 3105
+    },
+    {
+      "epoch": 0.7775,
+      "grad_norm": 1.125,
+      "learning_rate": 5.8629196626728e-06,
+      "loss": 0.4496,
+      "step": 3110
+    },
+    {
+      "epoch": 0.77875,
+      "grad_norm": 0.984375,
+      "learning_rate": 5.7998962519633045e-06,
+      "loss": 0.3764,
+      "step": 3115
+    },
+    {
+      "epoch": 0.78,
+      "grad_norm": 1.6015625,
+      "learning_rate": 5.737168930605272e-06,
+      "loss": 0.3888,
+      "step": 3120
+    },
+    {
+      "epoch": 0.78125,
+      "grad_norm": 1.015625,
+      "learning_rate": 5.674738665931575e-06,
+      "loss": 0.4209,
+      "step": 3125
+    },
+    {
+      "epoch": 0.7825,
+      "grad_norm": 1.3203125,
+      "learning_rate": 5.612606420694141e-06,
+      "loss": 0.4727,
+      "step": 3130
+    },
+    {
+      "epoch": 0.78375,
+      "grad_norm": 1.0546875,
+      "learning_rate": 5.550773153049046e-06,
+      "loss": 0.4365,
+      "step": 3135
+    },
+    {
+      "epoch": 0.785,
+      "grad_norm": 1.2421875,
+      "learning_rate": 5.489239816541755e-06,
+      "loss": 0.4403,
+      "step": 3140
+    },
+    {
+      "epoch": 0.78625,
+      "grad_norm": 1.390625,
+      "learning_rate": 5.428007360092463e-06,
+      "loss": 0.4521,
+      "step": 3145
+    },
+    {
+      "epoch": 0.7875,
+      "grad_norm": 1.25,
+      "learning_rate": 5.367076727981382e-06,
+      "loss": 0.4657,
+      "step": 3150
+    },
+    {
+      "epoch": 0.78875,
+      "grad_norm": 1.015625,
+      "learning_rate": 5.306448859834228e-06,
+      "loss": 0.4367,
+      "step": 3155
+    },
+    {
+      "epoch": 0.79,
+      "grad_norm": 1.1015625,
+      "learning_rate": 5.24612469060774e-06,
+      "loss": 0.4053,
+      "step": 3160
+    },
+    {
+      "epoch": 0.79125,
+      "grad_norm": 1.515625,
+      "learning_rate": 5.186105150575232e-06,
+      "loss": 0.3926,
+      "step": 3165
+    },
+    {
+      "epoch": 0.7925,
+      "grad_norm": 1.34375,
+      "learning_rate": 5.12639116531225e-06,
+      "loss": 0.4534,
+      "step": 3170
+    },
+    {
+      "epoch": 0.79375,
+      "grad_norm": 1.3125,
+      "learning_rate": 5.066983655682325e-06,
+      "loss": 0.4551,
+      "step": 3175
+    },
+    {
+      "epoch": 0.795,
+      "grad_norm": 1.375,
+      "learning_rate": 5.007883537822736e-06,
+      "loss": 0.4066,
+      "step": 3180
+    },
+    {
+      "epoch": 0.79625,
+      "grad_norm": 0.75,
+      "learning_rate": 4.949091723130425e-06,
+      "loss": 0.4247,
+      "step": 3185
+    },
+    {
+      "epoch": 0.7975,
+      "grad_norm": 1.6640625,
+      "learning_rate": 4.890609118247888e-06,
+      "loss": 0.4215,
+      "step": 3190
+    },
+    {
+      "epoch": 0.79875,
+      "grad_norm": 1.4296875,
+      "learning_rate": 4.832436625049256e-06,
+      "loss": 0.4385,
+      "step": 3195
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.98046875,
+      "learning_rate": 4.7745751406263165e-06,
+      "loss": 0.4393,
+      "step": 3200
+    },
+    {
+      "epoch": 0.80125,
+      "grad_norm": 0.9140625,
+      "learning_rate": 4.717025557274749e-06,
+      "loss": 0.42,
+      "step": 3205
+    },
+    {
+      "epoch": 0.8025,
+      "grad_norm": 0.90234375,
+      "learning_rate": 4.659788762480327e-06,
+      "loss": 0.3758,
+      "step": 3210
+    },
+    {
+      "epoch": 0.80375,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.602865638905224e-06,
+      "loss": 0.4448,
+      "step": 3215
+    },
+    {
+      "epoch": 0.805,
+      "grad_norm": 1.078125,
+      "learning_rate": 4.54625706437441e-06,
+      "loss": 0.4453,
+      "step": 3220
+    },
+    {
+      "epoch": 0.80625,
+      "grad_norm": 1.3125,
+      "learning_rate": 4.48996391186216e-06,
+      "loss": 0.4359,
+      "step": 3225
+    },
+    {
+      "epoch": 0.8075,
+      "grad_norm": 1.3515625,
+      "learning_rate": 4.433987049478508e-06,
+      "loss": 0.3974,
+      "step": 3230
+    },
+    {
+      "epoch": 0.80875,
+      "grad_norm": 1.125,
+      "learning_rate": 4.378327340455915e-06,
+      "loss": 0.4194,
+      "step": 3235
+    },
+    {
+      "epoch": 0.81,
+      "grad_norm": 0.87890625,
+      "learning_rate": 4.322985643135952e-06,
+      "loss": 0.4214,
+      "step": 3240
+    },
+    {
+      "epoch": 0.81125,
+      "grad_norm": 1.1640625,
+      "learning_rate": 4.267962810956061e-06,
+      "loss": 0.3592,
+      "step": 3245
+    },
+    {
+      "epoch": 0.8125,
+      "grad_norm": 1.2421875,
+      "learning_rate": 4.213259692436367e-06,
+      "loss": 0.3997,
+      "step": 3250
+    },
+    {
+      "epoch": 0.81375,
+      "grad_norm": 1.484375,
+      "learning_rate": 4.158877131166641e-06,
+      "loss": 0.4471,
+      "step": 3255
+    },
+    {
+      "epoch": 0.815,
+      "grad_norm": 0.953125,
+      "learning_rate": 4.104815965793249e-06,
+      "loss": 0.4293,
+      "step": 3260
+    },
+    {
+      "epoch": 0.81625,
+      "grad_norm": 1.203125,
+      "learning_rate": 4.051077030006228e-06,
+      "loss": 0.4562,
+      "step": 3265
+    },
+    {
+      "epoch": 0.8175,
+      "grad_norm": 1.0703125,
+      "learning_rate": 3.9976611525264525e-06,
+      "loss": 0.434,
+      "step": 3270
+    },
+    {
+      "epoch": 0.81875,
+      "grad_norm": 0.8671875,
+      "learning_rate": 3.944569157092839e-06,
+      "loss": 0.4524,
+      "step": 3275
+    },
+    {
+      "epoch": 0.82,
+      "grad_norm": 1.296875,
+      "learning_rate": 3.891801862449629e-06,
+      "loss": 0.4498,
+      "step": 3280
+    },
+    {
+      "epoch": 0.82125,
+      "grad_norm": 1.0625,
+      "learning_rate": 3.839360082333771e-06,
+      "loss": 0.4329,
+      "step": 3285
+    },
+    {
+      "epoch": 0.8225,
+      "grad_norm": 1.375,
+      "learning_rate": 3.7872446254624104e-06,
+      "loss": 0.3884,
+      "step": 3290
+    },
+    {
+      "epoch": 0.82375,
+      "grad_norm": 1.046875,
+      "learning_rate": 3.735456295520348e-06,
+      "loss": 0.4114,
+      "step": 3295
+    },
+    {
+      "epoch": 0.825,
+      "grad_norm": 1.3125,
+      "learning_rate": 3.6839958911476957e-06,
+      "loss": 0.4404,
+      "step": 3300
+    },
+    {
+      "epoch": 0.82625,
+      "grad_norm": 1.171875,
+      "learning_rate": 3.6328642059275524e-06,
+      "loss": 0.4548,
+      "step": 3305
+    },
+    {
+      "epoch": 0.8275,
+      "grad_norm": 0.92578125,
+      "learning_rate": 3.5820620283737616e-06,
+      "loss": 0.4648,
+      "step": 3310
+    },
+    {
+      "epoch": 0.82875,
+      "grad_norm": 0.875,
+      "learning_rate": 3.5315901419187363e-06,
+      "loss": 0.4233,
+      "step": 3315
+    },
+    {
+      "epoch": 0.83,
+      "grad_norm": 0.953125,
+      "learning_rate": 3.4814493249014116e-06,
+      "loss": 0.4021,
+      "step": 3320
+    },
+    {
+      "epoch": 0.83125,
+      "grad_norm": 1.0546875,
+      "learning_rate": 3.431640350555204e-06,
+      "loss": 0.4732,
+      "step": 3325
+    },
+    {
+      "epoch": 0.8325,
+      "grad_norm": 1.0546875,
+      "learning_rate": 3.382163986996126e-06,
+      "loss": 0.4174,
+      "step": 3330
+    },
+    {
+      "epoch": 0.83375,
+      "grad_norm": 1.3984375,
+      "learning_rate": 3.3330209972108976e-06,
+      "loss": 0.4284,
+      "step": 3335
+    },
+    {
+      "epoch": 0.835,
+      "grad_norm": 1.109375,
+      "learning_rate": 3.284212139045223e-06,
+      "loss": 0.4183,
+      "step": 3340
+    },
+    {
+      "epoch": 0.83625,
+      "grad_norm": 1.390625,
+      "learning_rate": 3.2357381651920648e-06,
+      "loss": 0.3996,
+      "step": 3345
+    },
+    {
+      "epoch": 0.8375,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.187599823180071e-06,
+      "loss": 0.4317,
+      "step": 3350
+    },
+    {
+      "epoch": 0.83875,
+      "grad_norm": 1.21875,
+      "learning_rate": 3.139797855362031e-06,
+      "loss": 0.4341,
+      "step": 3355
+    },
+    {
+      "epoch": 0.84,
+      "grad_norm": 1.59375,
+      "learning_rate": 3.092332998903416e-06,
+      "loss": 0.4516,
+      "step": 3360
+    },
+    {
+      "epoch": 0.84125,
+      "grad_norm": 1.0703125,
+      "learning_rate": 3.0452059857710186e-06,
+      "loss": 0.4371,
+      "step": 3365
+    },
+    {
+      "epoch": 0.8425,
+      "grad_norm": 0.9921875,
+      "learning_rate": 2.9984175427217016e-06,
+      "loss": 0.4346,
+      "step": 3370
+    },
+    {
+      "epoch": 0.84375,
+      "grad_norm": 0.96875,
+      "learning_rate": 2.9519683912911266e-06,
+      "loss": 0.3893,
+      "step": 3375
+    },
+    {
+      "epoch": 0.845,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.9058592477826636e-06,
+      "loss": 0.4086,
+      "step": 3380
+    },
+    {
+      "epoch": 0.84625,
+      "grad_norm": 0.9375,
+      "learning_rate": 2.860090823256359e-06,
+      "loss": 0.4211,
+      "step": 3385
+    },
+    {
+      "epoch": 0.8475,
+      "grad_norm": 1.296875,
+      "learning_rate": 2.8146638235179213e-06,
+      "loss": 0.422,
+      "step": 3390
+    },
+    {
+      "epoch": 0.84875,
+      "grad_norm": 1.0234375,
+      "learning_rate": 2.769578949107893e-06,
+      "loss": 0.4117,
+      "step": 3395
+    },
+    {
+      "epoch": 0.85,
+      "grad_norm": 0.90625,
+      "learning_rate": 2.7248368952908053e-06,
+      "loss": 0.3836,
+      "step": 3400
+    },
+    {
+      "epoch": 0.85125,
+      "grad_norm": 1.046875,
+      "learning_rate": 2.6804383520444815e-06,
+      "loss": 0.3996,
+      "step": 3405
+    },
+    {
+      "epoch": 0.8525,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.6363840040493747e-06,
+      "loss": 0.4007,
+      "step": 3410
+    },
+    {
+      "epoch": 0.85375,
+      "grad_norm": 1.0625,
+      "learning_rate": 2.5926745306780324e-06,
+      "loss": 0.4431,
+      "step": 3415
+    },
+    {
+      "epoch": 0.855,
+      "grad_norm": 1.09375,
+      "learning_rate": 2.5493106059846116e-06,
+      "loss": 0.4013,
+      "step": 3420
+    },
+    {
+      "epoch": 0.85625,
+      "grad_norm": 1.3125,
+      "learning_rate": 2.506292898694468e-06,
+      "loss": 0.4748,
+      "step": 3425
+    },
+    {
+      "epoch": 0.8575,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.4636220721938554e-06,
+      "loss": 0.4454,
+      "step": 3430
+    },
+    {
+      "epoch": 0.85875,
+      "grad_norm": 1.546875,
+      "learning_rate": 2.421298784519724e-06,
+      "loss": 0.3844,
+      "step": 3435
+    },
+    {
+      "epoch": 0.86,
+      "grad_norm": 1.296875,
+      "learning_rate": 2.379323688349516e-06,
+      "loss": 0.4772,
+      "step": 3440
+    },
+    {
+      "epoch": 0.86125,
+      "grad_norm": 1.09375,
+      "learning_rate": 2.3376974309911343e-06,
+      "loss": 0.4668,
+      "step": 3445
+    },
+    {
+      "epoch": 0.8625,
+      "grad_norm": 1.359375,
+      "learning_rate": 2.296420654372966e-06,
+      "loss": 0.4191,
+      "step": 3450
+    },
+    {
+      "epoch": 0.86375,
+      "grad_norm": 0.8828125,
+      "learning_rate": 2.2554939950339747e-06,
+      "loss": 0.3971,
+      "step": 3455
+    },
+    {
+      "epoch": 0.865,
+      "grad_norm": 1.296875,
+      "learning_rate": 2.2149180841138676e-06,
+      "loss": 0.4282,
+      "step": 3460
+    },
+    {
+      "epoch": 0.86625,
+      "grad_norm": 1.0,
+      "learning_rate": 2.1746935473433928e-06,
+      "loss": 0.4406,
+      "step": 3465
+    },
+    {
+      "epoch": 0.8675,
+      "grad_norm": 1.3828125,
+      "learning_rate": 2.1348210050346595e-06,
+      "loss": 0.3914,
+      "step": 3470
+    },
+    {
+      "epoch": 0.86875,
+      "grad_norm": 1.1015625,
+      "learning_rate": 2.0953010720716037e-06,
+      "loss": 0.3676,
+      "step": 3475
+    },
+    {
+      "epoch": 0.87,
+      "grad_norm": 1.234375,
+      "learning_rate": 2.0561343579004715e-06,
+      "loss": 0.3973,
+      "step": 3480
+    },
+    {
+      "epoch": 0.87125,
+      "grad_norm": 1.4375,
+      "learning_rate": 2.0173214665204555e-06,
+      "loss": 0.4067,
+      "step": 3485
+    },
+    {
+      "epoch": 0.8725,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.9788629964743455e-06,
+      "loss": 0.4279,
+      "step": 3490
+    },
+    {
+      "epoch": 0.87375,
+      "grad_norm": 1.3359375,
+      "learning_rate": 1.940759540839329e-06,
+      "loss": 0.4449,
+      "step": 3495
+    },
+    {
+      "epoch": 0.875,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.9030116872178316e-06,
+      "loss": 0.4364,
+      "step": 3500
+    },
+    {
+      "epoch": 0.87625,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.8656200177284505e-06,
+      "loss": 0.3991,
+      "step": 3505
+    },
+    {
+      "epoch": 0.8775,
+      "grad_norm": 1.109375,
+      "learning_rate": 1.8285851089969802e-06,
+      "loss": 0.3922,
+      "step": 3510
+    },
+    {
+      "epoch": 0.87875,
+      "grad_norm": 1.015625,
+      "learning_rate": 1.7919075321475325e-06,
+      "loss": 0.4559,
+      "step": 3515
+    },
+    {
+      "epoch": 0.88,
+      "grad_norm": 1.15625,
+      "learning_rate": 1.7555878527937164e-06,
+      "loss": 0.3947,
+      "step": 3520
+    },
+    {
+      "epoch": 0.88125,
+      "grad_norm": 0.98828125,
+      "learning_rate": 1.7196266310299108e-06,
+      "loss": 0.3853,
+      "step": 3525
+    },
+    {
+      "epoch": 0.8825,
+      "grad_norm": 1.1015625,
+      "learning_rate": 1.6840244214226502e-06,
+      "loss": 0.4429,
+      "step": 3530
+    },
+    {
+      "epoch": 0.88375,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.6487817730020365e-06,
+      "loss": 0.4092,
+      "step": 3535
+    },
+    {
+      "epoch": 0.885,
+      "grad_norm": 1.046875,
+      "learning_rate": 1.6138992292533183e-06,
+      "loss": 0.4348,
+      "step": 3540
+    },
+    {
+      "epoch": 0.88625,
+      "grad_norm": 0.92578125,
+      "learning_rate": 1.579377328108464e-06,
+      "loss": 0.4362,
+      "step": 3545
+    },
+    {
+      "epoch": 0.8875,
+      "grad_norm": 1.3359375,
+      "learning_rate": 1.5452166019378989e-06,
+      "loss": 0.431,
+      "step": 3550
+    },
+    {
+      "epoch": 0.88875,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.5114175775422762e-06,
+      "loss": 0.4164,
+      "step": 3555
+    },
+    {
+      "epoch": 0.89,
+      "grad_norm": 1.296875,
+      "learning_rate": 1.4779807761443636e-06,
+      "loss": 0.4154,
+      "step": 3560
+    },
+    {
+      "epoch": 0.89125,
+      "grad_norm": 1.3046875,
+      "learning_rate": 1.4449067133810056e-06,
+      "loss": 0.4108,
+      "step": 3565
+    },
+    {
+      "epoch": 0.8925,
+      "grad_norm": 1.1796875,
+      "learning_rate": 1.4121958992951629e-06,
+      "loss": 0.4024,
+      "step": 3570
+    },
+    {
+      "epoch": 0.89375,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.379848838328049e-06,
+      "loss": 0.4041,
+      "step": 3575
+    },
+    {
+      "epoch": 0.895,
+      "grad_norm": 1.2109375,
+      "learning_rate": 1.3478660293113676e-06,
+      "loss": 0.4291,
+      "step": 3580
+    },
+    {
+      "epoch": 0.89625,
+      "grad_norm": 1.0703125,
+      "learning_rate": 1.3162479654595938e-06,
+      "loss": 0.4561,
+      "step": 3585
+    },
+    {
+      "epoch": 0.8975,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.284995134362385e-06,
+      "loss": 0.4599,
+      "step": 3590
+    },
+    {
+      "epoch": 0.89875,
+      "grad_norm": 0.92578125,
+      "learning_rate": 1.2541080179770571e-06,
+      "loss": 0.371,
+      "step": 3595
+    },
+    {
+      "epoch": 0.9,
+      "grad_norm": 1.1171875,
+      "learning_rate": 1.2235870926211619e-06,
+      "loss": 0.428,
+      "step": 3600
+    },
+    {
+      "epoch": 0.90125,
+      "grad_norm": 1.1796875,
+      "learning_rate": 1.193432828965113e-06,
+      "loss": 0.4093,
+      "step": 3605
+    },
+    {
+      "epoch": 0.9025,
+      "grad_norm": 1.0078125,
+      "learning_rate": 1.16364569202497e-06,
+      "loss": 0.4123,
+      "step": 3610
+    },
+    {
+      "epoch": 0.90375,
+      "grad_norm": 1.0625,
+      "learning_rate": 1.134226141155223e-06,
+      "loss": 0.4212,
+      "step": 3615
+    },
+    {
+      "epoch": 0.905,
+      "grad_norm": 0.99609375,
+      "learning_rate": 1.105174630041747e-06,
+      "loss": 0.4379,
+      "step": 3620
+    },
+    {
+      "epoch": 0.90625,
+      "grad_norm": 1.28125,
+      "learning_rate": 1.0764916066947794e-06,
+      "loss": 0.4788,
+      "step": 3625
+    },
+    {
+      "epoch": 0.9075,
+      "grad_norm": 1.1875,
+      "learning_rate": 1.0481775134420225e-06,
+      "loss": 0.4523,
+      "step": 3630
+    },
+    {
+      "epoch": 0.90875,
+      "grad_norm": 1.5078125,
+      "learning_rate": 1.020232786921821e-06,
+      "loss": 0.4141,
+      "step": 3635
+    },
+    {
+      "epoch": 0.91,
+      "grad_norm": 1.0078125,
+      "learning_rate": 9.926578580764234e-07,
+      "loss": 0.3628,
+      "step": 3640
+    },
+    {
+      "epoch": 0.91125,
+      "grad_norm": 1.109375,
+      "learning_rate": 9.654531521453513e-07,
+      "loss": 0.4142,
+      "step": 3645
+    },
+    {
+      "epoch": 0.9125,
+      "grad_norm": 1.3046875,
+      "learning_rate": 9.386190886588208e-07,
+      "loss": 0.4348,
+      "step": 3650
+    },
+    {
+      "epoch": 0.91375,
+      "grad_norm": 1.6796875,
+      "learning_rate": 9.121560814312813e-07,
+      "loss": 0.4138,
+      "step": 3655
+    },
+    {
+      "epoch": 0.915,
+      "grad_norm": 1.078125,
+      "learning_rate": 8.860645385550481e-07,
+      "loss": 0.4332,
+      "step": 3660
+    },
+    {
+      "epoch": 0.91625,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.603448623939858e-07,
+      "loss": 0.4577,
+      "step": 3665
+    },
+    {
+      "epoch": 0.9175,
+      "grad_norm": 1.1484375,
+      "learning_rate": 8.349974495773183e-07,
+      "loss": 0.4456,
+      "step": 3670
+    },
+    {
+      "epoch": 0.91875,
+      "grad_norm": 1.28125,
+      "learning_rate": 8.10022690993506e-07,
+      "loss": 0.3946,
+      "step": 3675
+    },
+    {
+      "epoch": 0.92,
+      "grad_norm": 1.3203125,
+      "learning_rate": 7.854209717842231e-07,
+      "loss": 0.478,
+      "step": 3680
+    },
+    {
+      "epoch": 0.92125,
+      "grad_norm": 1.1171875,
+      "learning_rate": 7.611926713384121e-07,
+      "loss": 0.3592,
+      "step": 3685
+    },
+    {
+      "epoch": 0.9225,
+      "grad_norm": 1.234375,
+      "learning_rate": 7.373381632864384e-07,
+      "loss": 0.4425,
+      "step": 3690
+    },
+    {
+      "epoch": 0.92375,
+      "grad_norm": 1.1796875,
+      "learning_rate": 7.138578154943288e-07,
+      "loss": 0.4219,
+      "step": 3695
+    },
+    {
+      "epoch": 0.925,
+      "grad_norm": 1.3125,
+      "learning_rate": 6.907519900580861e-07,
+      "loss": 0.4419,
+      "step": 3700
+    },
+    {
+      "epoch": 0.92625,
+      "grad_norm": 1.1796875,
+      "learning_rate": 6.680210432981254e-07,
+      "loss": 0.3983,
+      "step": 3705
+    },
+    {
+      "epoch": 0.9275,
+      "grad_norm": 1.046875,
+      "learning_rate": 6.456653257537665e-07,
+      "loss": 0.4417,
+      "step": 3710
+    },
+    {
+      "epoch": 0.92875,
+      "grad_norm": 1.03125,
+      "learning_rate": 6.2368518217783e-07,
+      "loss": 0.4469,
+      "step": 3715
+    },
+    {
+      "epoch": 0.93,
+      "grad_norm": 1.0234375,
+      "learning_rate": 6.020809515313142e-07,
+      "loss": 0.435,
+      "step": 3720
+    },
+    {
+      "epoch": 0.93125,
+      "grad_norm": 1.6875,
+      "learning_rate": 5.808529669781904e-07,
+      "loss": 0.3856,
+      "step": 3725
+    },
+    {
+      "epoch": 0.9325,
+      "grad_norm": 1.1875,
+      "learning_rate": 5.600015558802352e-07,
+      "loss": 0.4587,
+      "step": 3730
+    },
+    {
+      "epoch": 0.93375,
+      "grad_norm": 1.203125,
+      "learning_rate": 5.39527039792001e-07,
+      "loss": 0.4325,
+      "step": 3735
+    },
+    {
+      "epoch": 0.935,
+      "grad_norm": 1.484375,
+      "learning_rate": 5.194297344558536e-07,
+      "loss": 0.4166,
+      "step": 3740
+    },
+    {
+      "epoch": 0.93625,
+      "grad_norm": 1.2890625,
+      "learning_rate": 4.997099497971114e-07,
+      "loss": 0.4347,
+      "step": 3745
+    },
+    {
+      "epoch": 0.9375,
+      "grad_norm": 0.95703125,
+      "learning_rate": 4.803679899192392e-07,
+      "loss": 0.4252,
+      "step": 3750
+    },
+    {
+      "epoch": 0.93875,
+      "grad_norm": 1.1875,
+      "learning_rate": 4.614041530991903e-07,
+      "loss": 0.4036,
+      "step": 3755
+    },
+    {
+      "epoch": 0.94,
+      "grad_norm": 1.375,
+      "learning_rate": 4.4281873178278475e-07,
+      "loss": 0.4359,
+      "step": 3760
+    },
+    {
+      "epoch": 0.94125,
+      "grad_norm": 1.359375,
+      "learning_rate": 4.246120125802111e-07,
+      "loss": 0.4566,
+      "step": 3765
+    },
+    {
+      "epoch": 0.9425,
+      "grad_norm": 1.0078125,
+      "learning_rate": 4.067842762616014e-07,
+      "loss": 0.4226,
+      "step": 3770
+    },
+    {
+      "epoch": 0.94375,
+      "grad_norm": 1.1953125,
+      "learning_rate": 3.8933579775271013e-07,
+      "loss": 0.3903,
+      "step": 3775
+    },
+    {
+      "epoch": 0.945,
+      "grad_norm": 1.125,
+      "learning_rate": 3.7226684613065333e-07,
+      "loss": 0.4119,
+      "step": 3780
+    },
+    {
+      "epoch": 0.94625,
+      "grad_norm": 1.3515625,
+      "learning_rate": 3.555776846197817e-07,
+      "loss": 0.4268,
+      "step": 3785
+    },
+    {
+      "epoch": 0.9475,
+      "grad_norm": 0.9765625,
+      "learning_rate": 3.3926857058761417e-07,
+      "loss": 0.405,
+      "step": 3790
+    },
+    {
+      "epoch": 0.94875,
+      "grad_norm": 1.2734375,
+      "learning_rate": 3.233397555408607e-07,
+      "loss": 0.4161,
+      "step": 3795
+    },
+    {
+      "epoch": 0.95,
+      "grad_norm": 1.2578125,
+      "learning_rate": 3.077914851215585e-07,
+      "loss": 0.4293,
+      "step": 3800
+    },
+    {
+      "epoch": 0.95125,
+      "grad_norm": 1.3671875,
+      "learning_rate": 2.92623999103267e-07,
+      "loss": 0.4066,
+      "step": 3805
+    },
+    {
+      "epoch": 0.9525,
+      "grad_norm": 1.0859375,
+      "learning_rate": 2.778375313873871e-07,
+      "loss": 0.4309,
+      "step": 3810
+    },
+    {
+      "epoch": 0.95375,
+      "grad_norm": 1.2421875,
+      "learning_rate": 2.634323099995395e-07,
+      "loss": 0.4623,
+      "step": 3815
+    },
+    {
+      "epoch": 0.955,
+      "grad_norm": 1.6015625,
+      "learning_rate": 2.494085570860616e-07,
+      "loss": 0.3977,
+      "step": 3820
+    },
+    {
+      "epoch": 0.95625,
+      "grad_norm": 1.28125,
+      "learning_rate": 2.3576648891056875e-07,
+      "loss": 0.4135,
+      "step": 3825
+    },
+    {
+      "epoch": 0.9575,
+      "grad_norm": 1.1640625,
+      "learning_rate": 2.2250631585063186e-07,
+      "loss": 0.3874,
+      "step": 3830
+    },
+    {
+      "epoch": 0.95875,
+      "grad_norm": 1.21875,
+      "learning_rate": 2.0962824239451894e-07,
+      "loss": 0.4494,
+      "step": 3835
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 1.1328125,
+      "learning_rate": 1.9713246713805588e-07,
+      "loss": 0.3946,
+      "step": 3840
+    },
+    {
+      "epoch": 0.96125,
+      "grad_norm": 0.85546875,
+      "learning_rate": 1.8501918278155393e-07,
+      "loss": 0.4312,
+      "step": 3845
+    },
+    {
+      "epoch": 0.9625,
+      "grad_norm": 1.359375,
+      "learning_rate": 1.732885761268427e-07,
+      "loss": 0.4453,
+      "step": 3850
+    },
+    {
+      "epoch": 0.96375,
+      "grad_norm": 0.984375,
+      "learning_rate": 1.619408280743917e-07,
+      "loss": 0.3943,
+      "step": 3855
+    },
+    {
+      "epoch": 0.965,
+      "grad_norm": 0.90234375,
+      "learning_rate": 1.509761136205101e-07,
+      "loss": 0.4144,
+      "step": 3860
+    },
+    {
+      "epoch": 0.96625,
+      "grad_norm": 1.046875,
+      "learning_rate": 1.4039460185465703e-07,
+      "loss": 0.4026,
+      "step": 3865
+    },
+    {
+      "epoch": 0.9675,
+      "grad_norm": 1.0625,
+      "learning_rate": 1.3019645595683806e-07,
+      "loss": 0.432,
+      "step": 3870
+    },
+    {
+      "epoch": 0.96875,
+      "grad_norm": 1.2890625,
+      "learning_rate": 1.2038183319507955e-07,
+      "loss": 0.4384,
+      "step": 3875
+    },
+    {
+      "epoch": 0.97,
+      "grad_norm": 1.328125,
+      "learning_rate": 1.109508849230001e-07,
+      "loss": 0.4622,
+      "step": 3880
+    },
+    {
+      "epoch": 0.97125,
+      "grad_norm": 1.265625,
+      "learning_rate": 1.0190375657749274e-07,
+      "loss": 0.4136,
+      "step": 3885
+    },
+    {
+      "epoch": 0.9725,
+      "grad_norm": 1.359375,
+      "learning_rate": 9.324058767646859e-08,
+      "loss": 0.4417,
+      "step": 3890
+    },
+    {
+      "epoch": 0.97375,
+      "grad_norm": 1.0625,
+      "learning_rate": 8.496151181670852e-08,
+      "loss": 0.4316,
+      "step": 3895
+    },
+    {
+      "epoch": 0.975,
+      "grad_norm": 0.8515625,
+      "learning_rate": 7.706665667180091e-08,
+      "loss": 0.4246,
+      "step": 3900
+    },
+    {
+      "epoch": 0.97625,
+      "grad_norm": 1.3515625,
+      "learning_rate": 6.955614399018206e-08,
+      "loss": 0.4481,
+      "step": 3905
+    },
+    {
+      "epoch": 0.9775,
+      "grad_norm": 1.1171875,
+      "learning_rate": 6.243008959324892e-08,
+      "loss": 0.4544,
+      "step": 3910
+    },
+    {
+      "epoch": 0.97875,
+      "grad_norm": 1.3671875,
+      "learning_rate": 5.568860337357151e-08,
+      "loss": 0.4005,
+      "step": 3915
+    },
+    {
+      "epoch": 0.98,
+      "grad_norm": 1.265625,
+      "learning_rate": 4.9331789293211026e-08,
+      "loss": 0.4955,
+      "step": 3920
+    },
+    {
+      "epoch": 0.98125,
+      "grad_norm": 1.5546875,
+      "learning_rate": 4.335974538210441e-08,
+      "loss": 0.4102,
+      "step": 3925
+    },
+    {
+      "epoch": 0.9825,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.7772563736551694e-08,
+      "loss": 0.4542,
+      "step": 3930
+    },
+    {
+      "epoch": 0.98375,
+      "grad_norm": 1.2421875,
+      "learning_rate": 3.2570330517811555e-08,
+      "loss": 0.4691,
+      "step": 3935
+    },
+    {
+      "epoch": 0.985,
+      "grad_norm": 0.9140625,
+      "learning_rate": 2.7753125950752413e-08,
+      "loss": 0.4155,
+      "step": 3940
+    },
+    {
+      "epoch": 0.98625,
+      "grad_norm": 1.3125,
+      "learning_rate": 2.3321024322625617e-08,
+      "loss": 0.4268,
+      "step": 3945
+    },
+    {
+      "epoch": 0.9875,
+      "grad_norm": 1.0546875,
+      "learning_rate": 1.9274093981927478e-08,
+      "loss": 0.4098,
+      "step": 3950
+    },
+    {
+      "epoch": 0.98875,
+      "grad_norm": 1.171875,
+      "learning_rate": 1.5612397337325113e-08,
+      "loss": 0.3965,
+      "step": 3955
+    },
+    {
+      "epoch": 0.99,
+      "grad_norm": 1.2421875,
+      "learning_rate": 1.233599085671e-08,
+      "loss": 0.4047,
+      "step": 3960
+    },
+    {
+      "epoch": 0.99125,
+      "grad_norm": 1.3125,
+      "learning_rate": 9.444925066329213e-09,
+      "loss": 0.4358,
+      "step": 3965
+    },
+    {
+      "epoch": 0.9925,
+      "grad_norm": 0.88671875,
+      "learning_rate": 6.939244549986068e-09,
+      "loss": 0.465,
+      "step": 3970
+    },
+    {
+      "epoch": 0.99375,
+      "grad_norm": 1.109375,
+      "learning_rate": 4.818987948379539e-09,
+      "loss": 0.4483,
+      "step": 3975
+    },
+    {
+      "epoch": 0.995,
+      "grad_norm": 1.46875,
+      "learning_rate": 3.0841879584853073e-09,
+      "loss": 0.4986,
+      "step": 3980
+    },
+    {
+      "epoch": 0.99625,
+      "grad_norm": 1.2734375,
+      "learning_rate": 1.7348713330672671e-09,
+      "loss": 0.4121,
+      "step": 3985
+    },
+    {
+      "epoch": 0.9975,
+      "grad_norm": 1.078125,
+      "learning_rate": 7.710588802584129e-10,
+      "loss": 0.3764,
+      "step": 3990
+    },
+    {
+      "epoch": 0.99875,
+      "grad_norm": 1.125,
+      "learning_rate": 1.9276546323609978e-10,
+      "loss": 0.4443,
+      "step": 3995
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 0.8515625,
+      "learning_rate": 0.0,
+      "loss": 0.4357,
+      "step": 4000
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 4000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 400,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.184022656018907e+18,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}