diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5492 @@
+{
+  "best_metric": 0.15992629528045654,
+  "best_model_checkpoint": "checkpoints/star_plus-llama-3.1-8b-math50k/math50k/finetune-llama-3.1-8b-math-step-1/checkpoint-3456",
+  "epoch": 1.0,
+  "eval_steps": 384,
+  "global_step": 3837,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0013031013812874641,
+      "grad_norm": 21.375,
+      "learning_rate": 1.3020833333333334e-07,
+      "loss": 0.5553,
+      "step": 5
+    },
+    {
+      "epoch": 0.0026062027625749283,
+      "grad_norm": 19.125,
+      "learning_rate": 2.604166666666667e-07,
+      "loss": 0.5275,
+      "step": 10
+    },
+    {
+      "epoch": 0.003909304143862392,
+      "grad_norm": 22.5,
+      "learning_rate": 3.90625e-07,
+      "loss": 0.5672,
+      "step": 15
+    },
+    {
+      "epoch": 0.005212405525149857,
+      "grad_norm": 20.125,
+      "learning_rate": 5.208333333333334e-07,
+      "loss": 0.5298,
+      "step": 20
+    },
+    {
+      "epoch": 0.006515506906437321,
+      "grad_norm": 17.375,
+      "learning_rate": 6.510416666666668e-07,
+      "loss": 0.5131,
+      "step": 25
+    },
+    {
+      "epoch": 0.007818608287724784,
+      "grad_norm": 16.75,
+      "learning_rate": 7.8125e-07,
+      "loss": 0.4643,
+      "step": 30
+    },
+    {
+      "epoch": 0.009121709669012249,
+      "grad_norm": 14.9375,
+      "learning_rate": 9.114583333333333e-07,
+      "loss": 0.4317,
+      "step": 35
+    },
+    {
+      "epoch": 0.010424811050299713,
+      "grad_norm": 15.8125,
+      "learning_rate": 1.0416666666666667e-06,
+      "loss": 0.3643,
+      "step": 40
+    },
+    {
+      "epoch": 0.011727912431587178,
+      "grad_norm": 11.375,
+      "learning_rate": 1.1718750000000001e-06,
+      "loss": 0.2948,
+      "step": 45
+    },
+    {
+      "epoch": 0.013031013812874642,
+      "grad_norm": 6.6875,
+      "learning_rate": 1.3020833333333335e-06,
+      "loss": 0.2206,
+      "step": 50
+    },
+    {
+      "epoch": 0.014334115194162106,
+      "grad_norm": 5.0,
+      "learning_rate": 1.4322916666666667e-06,
+      "loss": 0.1895,
+      "step": 55
+    },
+    {
+      "epoch": 0.01563721657544957,
+      "grad_norm": 2.859375,
+      "learning_rate": 1.5625e-06,
+      "loss": 0.1681,
+      "step": 60
+    },
+    {
+      "epoch": 0.016940317956737033,
+      "grad_norm": 2.4375,
+      "learning_rate": 1.6927083333333335e-06,
+      "loss": 0.155,
+      "step": 65
+    },
+    {
+      "epoch": 0.018243419338024498,
+      "grad_norm": 2.390625,
+      "learning_rate": 1.8229166666666666e-06,
+      "loss": 0.1643,
+      "step": 70
+    },
+    {
+      "epoch": 0.019546520719311962,
+      "grad_norm": 2.46875,
+      "learning_rate": 1.953125e-06,
+      "loss": 0.157,
+      "step": 75
+    },
+    {
+      "epoch": 0.020849622100599426,
+      "grad_norm": 2.796875,
+      "learning_rate": 2.0833333333333334e-06,
+      "loss": 0.1477,
+      "step": 80
+    },
+    {
+      "epoch": 0.02215272348188689,
+      "grad_norm": 2.859375,
+      "learning_rate": 2.213541666666667e-06,
+      "loss": 0.1519,
+      "step": 85
+    },
+    {
+      "epoch": 0.023455824863174355,
+      "grad_norm": 2.390625,
+      "learning_rate": 2.3437500000000002e-06,
+      "loss": 0.148,
+      "step": 90
+    },
+    {
+      "epoch": 0.02475892624446182,
+      "grad_norm": 2.5,
+      "learning_rate": 2.4739583333333336e-06,
+      "loss": 0.1546,
+      "step": 95
+    },
+    {
+      "epoch": 0.026062027625749284,
+      "grad_norm": 2.25,
+      "learning_rate": 2.604166666666667e-06,
+      "loss": 0.1564,
+      "step": 100
+    },
+    {
+      "epoch": 0.027365129007036748,
+      "grad_norm": 2.84375,
+      "learning_rate": 2.7343750000000004e-06,
+      "loss": 0.1493,
+      "step": 105
+    },
+    {
+      "epoch": 0.028668230388324212,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.8645833333333334e-06,
+      "loss": 0.1525,
+      "step": 110
+    },
+    {
+      "epoch": 0.029971331769611677,
+      "grad_norm": 2.78125,
+      "learning_rate": 2.994791666666667e-06,
+      "loss": 0.1582,
+      "step": 115
+    },
+    {
+      "epoch": 0.03127443315089914,
+      "grad_norm": 2.734375,
+      "learning_rate": 3.125e-06,
+      "loss": 0.1559,
+      "step": 120
+    },
+    {
+      "epoch": 0.032577534532186606,
+      "grad_norm": 2.828125,
+      "learning_rate": 3.2552083333333335e-06,
+      "loss": 0.1514,
+      "step": 125
+    },
+    {
+      "epoch": 0.033880635913474066,
+      "grad_norm": 2.859375,
+      "learning_rate": 3.385416666666667e-06,
+      "loss": 0.1528,
+      "step": 130
+    },
+    {
+      "epoch": 0.035183737294761534,
+      "grad_norm": 2.5,
+      "learning_rate": 3.5156250000000003e-06,
+      "loss": 0.1613,
+      "step": 135
+    },
+    {
+      "epoch": 0.036486838676048995,
+      "grad_norm": 2.71875,
+      "learning_rate": 3.6458333333333333e-06,
+      "loss": 0.1503,
+      "step": 140
+    },
+    {
+      "epoch": 0.03778994005733646,
+      "grad_norm": 2.75,
+      "learning_rate": 3.776041666666667e-06,
+      "loss": 0.1454,
+      "step": 145
+    },
+    {
+      "epoch": 0.039093041438623924,
+      "grad_norm": 2.640625,
+      "learning_rate": 3.90625e-06,
+      "loss": 0.1506,
+      "step": 150
+    },
+    {
+      "epoch": 0.04039614281991139,
+      "grad_norm": 2.640625,
+      "learning_rate": 4.0364583333333335e-06,
+      "loss": 0.1523,
+      "step": 155
+    },
+    {
+      "epoch": 0.04169924420119885,
+      "grad_norm": 2.453125,
+      "learning_rate": 4.166666666666667e-06,
+      "loss": 0.1519,
+      "step": 160
+    },
+    {
+      "epoch": 0.04300234558248632,
+      "grad_norm": 3.015625,
+      "learning_rate": 4.296875e-06,
+      "loss": 0.16,
+      "step": 165
+    },
+    {
+      "epoch": 0.04430544696377378,
+      "grad_norm": 3.0,
+      "learning_rate": 4.427083333333334e-06,
+      "loss": 0.1495,
+      "step": 170
+    },
+    {
+      "epoch": 0.04560854834506125,
+      "grad_norm": 2.46875,
+      "learning_rate": 4.557291666666667e-06,
+      "loss": 0.145,
+      "step": 175
+    },
+    {
+      "epoch": 0.04691164972634871,
+      "grad_norm": 2.59375,
+      "learning_rate": 4.6875000000000004e-06,
+      "loss": 0.1398,
+      "step": 180
+    },
+    {
+      "epoch": 0.04821475110763617,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.817708333333334e-06,
+      "loss": 0.1562,
+      "step": 185
+    },
+    {
+      "epoch": 0.04951785248892364,
+      "grad_norm": 2.90625,
+      "learning_rate": 4.947916666666667e-06,
+      "loss": 0.1523,
+      "step": 190
+    },
+    {
+      "epoch": 0.0508209538702111,
+      "grad_norm": 2.578125,
+      "learning_rate": 5.078125000000001e-06,
+      "loss": 0.1494,
+      "step": 195
+    },
+    {
+      "epoch": 0.05212405525149857,
+      "grad_norm": 2.703125,
+      "learning_rate": 5.208333333333334e-06,
+      "loss": 0.1432,
+      "step": 200
+    },
+    {
+      "epoch": 0.05342715663278603,
+      "grad_norm": 2.703125,
+      "learning_rate": 5.3385416666666666e-06,
+      "loss": 0.1472,
+      "step": 205
+    },
+    {
+      "epoch": 0.054730258014073496,
+      "grad_norm": 2.59375,
+      "learning_rate": 5.468750000000001e-06,
+      "loss": 0.1471,
+      "step": 210
+    },
+    {
+      "epoch": 0.05603335939536096,
+      "grad_norm": 2.34375,
+      "learning_rate": 5.598958333333334e-06,
+      "loss": 0.1511,
+      "step": 215
+    },
+    {
+      "epoch": 0.057336460776648425,
+      "grad_norm": 2.671875,
+      "learning_rate": 5.729166666666667e-06,
+      "loss": 0.1449,
+      "step": 220
+    },
+    {
+      "epoch": 0.058639562157935886,
+      "grad_norm": 2.5625,
+      "learning_rate": 5.859375e-06,
+      "loss": 0.1502,
+      "step": 225
+    },
+    {
+      "epoch": 0.059942663539223354,
+      "grad_norm": 2.53125,
+      "learning_rate": 5.989583333333334e-06,
+      "loss": 0.1438,
+      "step": 230
+    },
+    {
+      "epoch": 0.061245764920510815,
+      "grad_norm": 2.53125,
+      "learning_rate": 6.119791666666667e-06,
+      "loss": 0.1584,
+      "step": 235
+    },
+    {
+      "epoch": 0.06254886630179828,
+      "grad_norm": 2.8125,
+      "learning_rate": 6.25e-06,
+      "loss": 0.1572,
+      "step": 240
+    },
+    {
+      "epoch": 0.06385196768308575,
+      "grad_norm": 2.828125,
+      "learning_rate": 6.3802083333333345e-06,
+      "loss": 0.1516,
+      "step": 245
+    },
+    {
+      "epoch": 0.06515506906437321,
+      "grad_norm": 2.515625,
+      "learning_rate": 6.510416666666667e-06,
+      "loss": 0.152,
+      "step": 250
+    },
+    {
+      "epoch": 0.06645817044566067,
+      "grad_norm": 2.5625,
+      "learning_rate": 6.6406250000000005e-06,
+      "loss": 0.1508,
+      "step": 255
+    },
+    {
+      "epoch": 0.06776127182694813,
+      "grad_norm": 2.75,
+      "learning_rate": 6.770833333333334e-06,
+      "loss": 0.1569,
+      "step": 260
+    },
+    {
+      "epoch": 0.06906437320823561,
+      "grad_norm": 2.390625,
+      "learning_rate": 6.901041666666667e-06,
+      "loss": 0.1426,
+      "step": 265
+    },
+    {
+      "epoch": 0.07036747458952307,
+      "grad_norm": 2.515625,
+      "learning_rate": 7.031250000000001e-06,
+      "loss": 0.1595,
+      "step": 270
+    },
+    {
+      "epoch": 0.07167057597081053,
+      "grad_norm": 2.5,
+      "learning_rate": 7.161458333333334e-06,
+      "loss": 0.1603,
+      "step": 275
+    },
+    {
+      "epoch": 0.07297367735209799,
+      "grad_norm": 2.71875,
+      "learning_rate": 7.291666666666667e-06,
+      "loss": 0.1533,
+      "step": 280
+    },
+    {
+      "epoch": 0.07427677873338545,
+      "grad_norm": 2.546875,
+      "learning_rate": 7.421875000000001e-06,
+      "loss": 0.1505,
+      "step": 285
+    },
+    {
+      "epoch": 0.07557988011467293,
+      "grad_norm": 2.390625,
+      "learning_rate": 7.552083333333334e-06,
+      "loss": 0.1499,
+      "step": 290
+    },
+    {
+      "epoch": 0.07688298149596039,
+      "grad_norm": 2.765625,
+      "learning_rate": 7.682291666666668e-06,
+      "loss": 0.1544,
+      "step": 295
+    },
+    {
+      "epoch": 0.07818608287724785,
+      "grad_norm": 2.359375,
+      "learning_rate": 7.8125e-06,
+      "loss": 0.1535,
+      "step": 300
+    },
+    {
+      "epoch": 0.07948918425853531,
+      "grad_norm": 2.53125,
+      "learning_rate": 7.942708333333334e-06,
+      "loss": 0.1497,
+      "step": 305
+    },
+    {
+      "epoch": 0.08079228563982278,
+      "grad_norm": 2.4375,
+      "learning_rate": 8.072916666666667e-06,
+      "loss": 0.1568,
+      "step": 310
+    },
+    {
+      "epoch": 0.08209538702111024,
+      "grad_norm": 2.28125,
+      "learning_rate": 8.203125000000001e-06,
+      "loss": 0.1525,
+      "step": 315
+    },
+    {
+      "epoch": 0.0833984884023977,
+      "grad_norm": 2.671875,
+      "learning_rate": 8.333333333333334e-06,
+      "loss": 0.15,
+      "step": 320
+    },
+    {
+      "epoch": 0.08470158978368517,
+      "grad_norm": 2.84375,
+      "learning_rate": 8.463541666666666e-06,
+      "loss": 0.1583,
+      "step": 325
+    },
+    {
+      "epoch": 0.08600469116497264,
+      "grad_norm": 2.5625,
+      "learning_rate": 8.59375e-06,
+      "loss": 0.1555,
+      "step": 330
+    },
+    {
+      "epoch": 0.0873077925462601,
+      "grad_norm": 2.640625,
+      "learning_rate": 8.723958333333335e-06,
+      "loss": 0.1534,
+      "step": 335
+    },
+    {
+      "epoch": 0.08861089392754756,
+      "grad_norm": 2.59375,
+      "learning_rate": 8.854166666666667e-06,
+      "loss": 0.1565,
+      "step": 340
+    },
+    {
+      "epoch": 0.08991399530883502,
+      "grad_norm": 2.5,
+      "learning_rate": 8.984375000000002e-06,
+      "loss": 0.1544,
+      "step": 345
+    },
+    {
+      "epoch": 0.0912170966901225,
+      "grad_norm": 2.671875,
+      "learning_rate": 9.114583333333334e-06,
+      "loss": 0.1543,
+      "step": 350
+    },
+    {
+      "epoch": 0.09252019807140996,
+      "grad_norm": 2.640625,
+      "learning_rate": 9.244791666666667e-06,
+      "loss": 0.1536,
+      "step": 355
+    },
+    {
+      "epoch": 0.09382329945269742,
+      "grad_norm": 2.953125,
+      "learning_rate": 9.375000000000001e-06,
+      "loss": 0.1606,
+      "step": 360
+    },
+    {
+      "epoch": 0.09512640083398488,
+      "grad_norm": 2.328125,
+      "learning_rate": 9.505208333333335e-06,
+      "loss": 0.1533,
+      "step": 365
+    },
+    {
+      "epoch": 0.09642950221527234,
+      "grad_norm": 2.453125,
+      "learning_rate": 9.635416666666668e-06,
+      "loss": 0.1616,
+      "step": 370
+    },
+    {
+      "epoch": 0.09773260359655982,
+      "grad_norm": 5.1875,
+      "learning_rate": 9.765625e-06,
+      "loss": 0.1703,
+      "step": 375
+    },
+    {
+      "epoch": 0.09903570497784728,
+      "grad_norm": 3.046875,
+      "learning_rate": 9.895833333333334e-06,
+      "loss": 0.1659,
+      "step": 380
+    },
+    {
+      "epoch": 0.10007818608287725,
+      "eval_loss": 0.18054868280887604,
+      "eval_runtime": 49.2875,
+      "eval_samples_per_second": 298.25,
+      "eval_steps_per_second": 9.333,
+      "step": 384
+    },
+    {
+      "epoch": 0.10033880635913474,
+      "grad_norm": 2.34375,
+      "learning_rate": 9.997103967564437e-06,
+      "loss": 0.1524,
+      "step": 385
+    },
+    {
+      "epoch": 0.1016419077404222,
+      "grad_norm": 3.25,
+      "learning_rate": 9.98262380538662e-06,
+      "loss": 0.1655,
+      "step": 390
+    },
+    {
+      "epoch": 0.10294500912170967,
+      "grad_norm": 2.4375,
+      "learning_rate": 9.968143643208805e-06,
+      "loss": 0.1657,
+      "step": 395
+    },
+    {
+      "epoch": 0.10424811050299713,
+      "grad_norm": 2.53125,
+      "learning_rate": 9.953663481030988e-06,
+      "loss": 0.1566,
+      "step": 400
+    },
+    {
+      "epoch": 0.1055512118842846,
+      "grad_norm": 2.546875,
+      "learning_rate": 9.939183318853171e-06,
+      "loss": 0.162,
+      "step": 405
+    },
+    {
+      "epoch": 0.10685431326557206,
+      "grad_norm": 2.9375,
+      "learning_rate": 9.924703156675356e-06,
+      "loss": 0.156,
+      "step": 410
+    },
+    {
+      "epoch": 0.10815741464685953,
+      "grad_norm": 2.40625,
+      "learning_rate": 9.910222994497538e-06,
+      "loss": 0.164,
+      "step": 415
+    },
+    {
+      "epoch": 0.10946051602814699,
+      "grad_norm": 2.53125,
+      "learning_rate": 9.895742832319723e-06,
+      "loss": 0.1557,
+      "step": 420
+    },
+    {
+      "epoch": 0.11076361740943445,
+      "grad_norm": 2.828125,
+      "learning_rate": 9.881262670141906e-06,
+      "loss": 0.1643,
+      "step": 425
+    },
+    {
+      "epoch": 0.11206671879072191,
+      "grad_norm": 2.8125,
+      "learning_rate": 9.86678250796409e-06,
+      "loss": 0.1624,
+      "step": 430
+    },
+    {
+      "epoch": 0.11336982017200939,
+      "grad_norm": 2.6875,
+      "learning_rate": 9.852302345786274e-06,
+      "loss": 0.1536,
+      "step": 435
+    },
+    {
+      "epoch": 0.11467292155329685,
+      "grad_norm": 2.859375,
+      "learning_rate": 9.837822183608458e-06,
+      "loss": 0.1638,
+      "step": 440
+    },
+    {
+      "epoch": 0.11597602293458431,
+      "grad_norm": 2.5625,
+      "learning_rate": 9.823342021430641e-06,
+      "loss": 0.1587,
+      "step": 445
+    },
+    {
+      "epoch": 0.11727912431587177,
+      "grad_norm": 2.828125,
+      "learning_rate": 9.808861859252824e-06,
+      "loss": 0.1611,
+      "step": 450
+    },
+    {
+      "epoch": 0.11858222569715923,
+      "grad_norm": 2.59375,
+      "learning_rate": 9.794381697075009e-06,
+      "loss": 0.1539,
+      "step": 455
+    },
+    {
+      "epoch": 0.11988532707844671,
+      "grad_norm": 2.78125,
+      "learning_rate": 9.779901534897192e-06,
+      "loss": 0.1607,
+      "step": 460
+    },
+    {
+      "epoch": 0.12118842845973417,
+      "grad_norm": 2.3125,
+      "learning_rate": 9.765421372719375e-06,
+      "loss": 0.1598,
+      "step": 465
+    },
+    {
+      "epoch": 0.12249152984102163,
+      "grad_norm": 2.515625,
+      "learning_rate": 9.75094121054156e-06,
+      "loss": 0.1531,
+      "step": 470
+    },
+    {
+      "epoch": 0.12379463122230909,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.736461048363742e-06,
+      "loss": 0.16,
+      "step": 475
+    },
+    {
+      "epoch": 0.12509773260359655,
+      "grad_norm": 2.53125,
+      "learning_rate": 9.721980886185927e-06,
+      "loss": 0.1572,
+      "step": 480
+    },
+    {
+      "epoch": 0.12640083398488403,
+      "grad_norm": 2.59375,
+      "learning_rate": 9.70750072400811e-06,
+      "loss": 0.1581,
+      "step": 485
+    },
+    {
+      "epoch": 0.1277039353661715,
+      "grad_norm": 2.71875,
+      "learning_rate": 9.693020561830293e-06,
+      "loss": 0.1599,
+      "step": 490
+    },
+    {
+      "epoch": 0.12900703674745895,
+      "grad_norm": 2.796875,
+      "learning_rate": 9.678540399652477e-06,
+      "loss": 0.1632,
+      "step": 495
+    },
+    {
+      "epoch": 0.13031013812874642,
+      "grad_norm": 2.5625,
+      "learning_rate": 9.66406023747466e-06,
+      "loss": 0.1602,
+      "step": 500
+    },
+    {
+      "epoch": 0.13161323951003387,
+      "grad_norm": 2.65625,
+      "learning_rate": 9.649580075296843e-06,
+      "loss": 0.1551,
+      "step": 505
+    },
+    {
+      "epoch": 0.13291634089132134,
+      "grad_norm": 2.5625,
+      "learning_rate": 9.635099913119028e-06,
+      "loss": 0.1571,
+      "step": 510
+    },
+    {
+      "epoch": 0.13421944227260882,
+      "grad_norm": 2.46875,
+      "learning_rate": 9.620619750941211e-06,
+      "loss": 0.1544,
+      "step": 515
+    },
+    {
+      "epoch": 0.13552254365389627,
+      "grad_norm": 2.59375,
+      "learning_rate": 9.606139588763394e-06,
+      "loss": 0.1594,
+      "step": 520
+    },
+    {
+      "epoch": 0.13682564503518374,
+      "grad_norm": 2.453125,
+      "learning_rate": 9.591659426585578e-06,
+      "loss": 0.1597,
+      "step": 525
+    },
+    {
+      "epoch": 0.13812874641647122,
+      "grad_norm": 2.4375,
+      "learning_rate": 9.577179264407761e-06,
+      "loss": 0.1568,
+      "step": 530
+    },
+    {
+      "epoch": 0.13943184779775866,
+      "grad_norm": 2.171875,
+      "learning_rate": 9.562699102229946e-06,
+      "loss": 0.1501,
+      "step": 535
+    },
+    {
+      "epoch": 0.14073494917904614,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.548218940052129e-06,
+      "loss": 0.154,
+      "step": 540
+    },
+    {
+      "epoch": 0.14203805056033358,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.533738777874314e-06,
+      "loss": 0.1579,
+      "step": 545
+    },
+    {
+      "epoch": 0.14334115194162106,
+      "grad_norm": 2.46875,
+      "learning_rate": 9.519258615696497e-06,
+      "loss": 0.1582,
+      "step": 550
+    },
+    {
+      "epoch": 0.14464425332290853,
+      "grad_norm": 2.703125,
+      "learning_rate": 9.504778453518681e-06,
+      "loss": 0.1622,
+      "step": 555
+    },
+    {
+      "epoch": 0.14594735470419598,
+      "grad_norm": 2.625,
+      "learning_rate": 9.490298291340864e-06,
+      "loss": 0.1558,
+      "step": 560
+    },
+    {
+      "epoch": 0.14725045608548346,
+      "grad_norm": 2.421875,
+      "learning_rate": 9.475818129163047e-06,
+      "loss": 0.1533,
+      "step": 565
+    },
+    {
+      "epoch": 0.1485535574667709,
+      "grad_norm": 2.5,
+      "learning_rate": 9.461337966985232e-06,
+      "loss": 0.1531,
+      "step": 570
+    },
+    {
+      "epoch": 0.14985665884805838,
+      "grad_norm": 2.78125,
+      "learning_rate": 9.446857804807415e-06,
+      "loss": 0.1585,
+      "step": 575
+    },
+    {
+      "epoch": 0.15115976022934585,
+      "grad_norm": 2.859375,
+      "learning_rate": 9.4323776426296e-06,
+      "loss": 0.1522,
+      "step": 580
+    },
+    {
+      "epoch": 0.1524628616106333,
+      "grad_norm": 2.40625,
+      "learning_rate": 9.417897480451782e-06,
+      "loss": 0.1572,
+      "step": 585
+    },
+    {
+      "epoch": 0.15376596299192077,
+      "grad_norm": 2.671875,
+      "learning_rate": 9.403417318273965e-06,
+      "loss": 0.1553,
+      "step": 590
+    },
+    {
+      "epoch": 0.15506906437320825,
+      "grad_norm": 3.125,
+      "learning_rate": 9.38893715609615e-06,
+      "loss": 0.1664,
+      "step": 595
+    },
+    {
+      "epoch": 0.1563721657544957,
+      "grad_norm": 2.46875,
+      "learning_rate": 9.374456993918333e-06,
+      "loss": 0.1469,
+      "step": 600
+    },
+    {
+      "epoch": 0.15767526713578317,
+      "grad_norm": 2.5,
+      "learning_rate": 9.359976831740516e-06,
+      "loss": 0.16,
+      "step": 605
+    },
+    {
+      "epoch": 0.15897836851707062,
+      "grad_norm": 2.265625,
+      "learning_rate": 9.3454966695627e-06,
+      "loss": 0.1546,
+      "step": 610
+    },
+    {
+      "epoch": 0.1602814698983581,
+      "grad_norm": 2.3125,
+      "learning_rate": 9.331016507384883e-06,
+      "loss": 0.1491,
+      "step": 615
+    },
+    {
+      "epoch": 0.16158457127964557,
+      "grad_norm": 2.46875,
+      "learning_rate": 9.316536345207066e-06,
+      "loss": 0.1484,
+      "step": 620
+    },
+    {
+      "epoch": 0.16288767266093301,
+      "grad_norm": 2.234375,
+      "learning_rate": 9.30205618302925e-06,
+      "loss": 0.1665,
+      "step": 625
+    },
+    {
+      "epoch": 0.1641907740422205,
+      "grad_norm": 2.53125,
+      "learning_rate": 9.287576020851434e-06,
+      "loss": 0.1523,
+      "step": 630
+    },
+    {
+      "epoch": 0.16549387542350794,
+      "grad_norm": 2.71875,
+      "learning_rate": 9.273095858673617e-06,
+      "loss": 0.1597,
+      "step": 635
+    },
+    {
+      "epoch": 0.1667969768047954,
+      "grad_norm": 2.703125,
+      "learning_rate": 9.258615696495801e-06,
+      "loss": 0.1563,
+      "step": 640
+    },
+    {
+      "epoch": 0.16810007818608289,
+      "grad_norm": 2.671875,
+      "learning_rate": 9.244135534317984e-06,
+      "loss": 0.1578,
+      "step": 645
+    },
+    {
+      "epoch": 0.16940317956737033,
+      "grad_norm": 2.4375,
+      "learning_rate": 9.229655372140169e-06,
+      "loss": 0.1505,
+      "step": 650
+    },
+    {
+      "epoch": 0.1707062809486578,
+      "grad_norm": 2.828125,
+      "learning_rate": 9.215175209962352e-06,
+      "loss": 0.1563,
+      "step": 655
+    },
+    {
+      "epoch": 0.17200938232994528,
+      "grad_norm": 3.140625,
+      "learning_rate": 9.200695047784537e-06,
+      "loss": 0.1584,
+      "step": 660
+    },
+    {
+      "epoch": 0.17331248371123273,
+      "grad_norm": 2.65625,
+      "learning_rate": 9.18621488560672e-06,
+      "loss": 0.1513,
+      "step": 665
+    },
+    {
+      "epoch": 0.1746155850925202,
+      "grad_norm": 2.390625,
+      "learning_rate": 9.171734723428904e-06,
+      "loss": 0.1553,
+      "step": 670
+    },
+    {
+      "epoch": 0.17591868647380765,
+      "grad_norm": 2.375,
+      "learning_rate": 9.157254561251087e-06,
+      "loss": 0.1547,
+      "step": 675
+    },
+    {
+      "epoch": 0.17722178785509513,
+      "grad_norm": 2.65625,
+      "learning_rate": 9.142774399073272e-06,
+      "loss": 0.1553,
+      "step": 680
+    },
+    {
+      "epoch": 0.1785248892363826,
+      "grad_norm": 2.40625,
+      "learning_rate": 9.128294236895455e-06,
+      "loss": 0.1527,
+      "step": 685
+    },
+    {
+      "epoch": 0.17982799061767005,
+      "grad_norm": 2.53125,
+      "learning_rate": 9.113814074717638e-06,
+      "loss": 0.1545,
+      "step": 690
+    },
+    {
+      "epoch": 0.18113109199895752,
+      "grad_norm": 2.484375,
+      "learning_rate": 9.099333912539822e-06,
+      "loss": 0.1503,
+      "step": 695
+    },
+    {
+      "epoch": 0.182434193380245,
+      "grad_norm": 2.453125,
+      "learning_rate": 9.084853750362005e-06,
+      "loss": 0.1546,
+      "step": 700
+    },
+    {
+      "epoch": 0.18373729476153244,
+      "grad_norm": 2.21875,
+      "learning_rate": 9.070373588184188e-06,
+      "loss": 0.1533,
+      "step": 705
+    },
+    {
+      "epoch": 0.18504039614281992,
+      "grad_norm": 2.6875,
+      "learning_rate": 9.055893426006373e-06,
+      "loss": 0.1499,
+      "step": 710
+    },
+    {
+      "epoch": 0.18634349752410737,
+      "grad_norm": 2.203125,
+      "learning_rate": 9.041413263828556e-06,
+      "loss": 0.1561,
+      "step": 715
+    },
+    {
+      "epoch": 0.18764659890539484,
+      "grad_norm": 2.453125,
+      "learning_rate": 9.026933101650739e-06,
+      "loss": 0.161,
+      "step": 720
+    },
+    {
+      "epoch": 0.18894970028668231,
+      "grad_norm": 2.40625,
+      "learning_rate": 9.012452939472923e-06,
+      "loss": 0.1602,
+      "step": 725
+    },
+    {
+      "epoch": 0.19025280166796976,
+      "grad_norm": 2.609375,
+      "learning_rate": 8.997972777295106e-06,
+      "loss": 0.1579,
+      "step": 730
+    },
+    {
+      "epoch": 0.19155590304925724,
+      "grad_norm": 2.4375,
+      "learning_rate": 8.983492615117289e-06,
+      "loss": 0.1537,
+      "step": 735
+    },
+    {
+      "epoch": 0.19285900443054468,
+      "grad_norm": 2.484375,
+      "learning_rate": 8.969012452939474e-06,
+      "loss": 0.1508,
+      "step": 740
+    },
+    {
+      "epoch": 0.19416210581183216,
+      "grad_norm": 2.359375,
+      "learning_rate": 8.954532290761657e-06,
+      "loss": 0.1525,
+      "step": 745
+    },
+    {
+      "epoch": 0.19546520719311963,
+      "grad_norm": 2.3125,
+      "learning_rate": 8.94005212858384e-06,
+      "loss": 0.1562,
+      "step": 750
+    },
+    {
+      "epoch": 0.19676830857440708,
+      "grad_norm": 2.34375,
+      "learning_rate": 8.925571966406024e-06,
+      "loss": 0.1566,
+      "step": 755
+    },
+    {
+      "epoch": 0.19807140995569456,
+      "grad_norm": 2.28125,
+      "learning_rate": 8.911091804228207e-06,
+      "loss": 0.1541,
+      "step": 760
+    },
+    {
+      "epoch": 0.19937451133698203,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.896611642050392e-06,
+      "loss": 0.1484,
+      "step": 765
+    },
+    {
+      "epoch": 0.2001563721657545,
+      "eval_loss": 0.17529802024364471,
+      "eval_runtime": 49.2622,
+      "eval_samples_per_second": 298.403,
+      "eval_steps_per_second": 9.338,
+      "step": 768
+    },
+    {
+      "epoch": 0.20067761271826948,
+      "grad_norm": 2.390625,
+      "learning_rate": 8.882131479872575e-06,
+      "loss": 0.1509,
+      "step": 770
+    },
+    {
+      "epoch": 0.20198071409955695,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.86765131769476e-06,
+      "loss": 0.1499,
+      "step": 775
+    },
+    {
+      "epoch": 0.2032838154808444,
+      "grad_norm": 2.25,
+      "learning_rate": 8.853171155516942e-06,
+      "loss": 0.1563,
+      "step": 780
+    },
+    {
+      "epoch": 0.20458691686213187,
+      "grad_norm": 2.453125,
+      "learning_rate": 8.838690993339127e-06,
+      "loss": 0.1493,
+      "step": 785
+    },
+    {
+      "epoch": 0.20589001824341935,
+      "grad_norm": 2.703125,
+      "learning_rate": 8.82421083116131e-06,
+      "loss": 0.1432,
+      "step": 790
+    },
+    {
+      "epoch": 0.2071931196247068,
+      "grad_norm": 2.484375,
+      "learning_rate": 8.809730668983495e-06,
+      "loss": 0.1428,
+      "step": 795
+    },
+    {
+      "epoch": 0.20849622100599427,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.795250506805678e-06,
+      "loss": 0.1597,
+      "step": 800
+    },
+    {
+      "epoch": 0.20979932238728172,
+      "grad_norm": 2.671875,
+      "learning_rate": 8.78077034462786e-06,
+      "loss": 0.1599,
+      "step": 805
+    },
+    {
+      "epoch": 0.2111024237685692,
+      "grad_norm": 2.328125,
+      "learning_rate": 8.766290182450045e-06,
+      "loss": 0.1554,
+      "step": 810
+    },
+    {
+      "epoch": 0.21240552514985667,
+      "grad_norm": 2.3125,
+      "learning_rate": 8.751810020272228e-06,
+      "loss": 0.1588,
+      "step": 815
+    },
+    {
+      "epoch": 0.2137086265311441,
+      "grad_norm": 2.390625,
+      "learning_rate": 8.737329858094411e-06,
+      "loss": 0.1498,
+      "step": 820
+    },
+    {
+      "epoch": 0.2150117279124316,
+      "grad_norm": 3.171875,
+      "learning_rate": 8.722849695916596e-06,
+      "loss": 0.1571,
+      "step": 825
+    },
+    {
+      "epoch": 0.21631482929371906,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.708369533738779e-06,
+      "loss": 0.149,
+      "step": 830
+    },
+    {
+      "epoch": 0.2176179306750065,
+      "grad_norm": 2.515625,
+      "learning_rate": 8.693889371560962e-06,
+      "loss": 0.1558,
+      "step": 835
+    },
+    {
+      "epoch": 0.21892103205629398,
+      "grad_norm": 2.203125,
+      "learning_rate": 8.679409209383146e-06,
+      "loss": 0.1543,
+      "step": 840
+    },
+    {
+      "epoch": 0.22022413343758143,
+      "grad_norm": 2.3125,
+      "learning_rate": 8.664929047205329e-06,
+      "loss": 0.146,
+      "step": 845
+    },
+    {
+      "epoch": 0.2215272348188689,
+      "grad_norm": 2.5,
+      "learning_rate": 8.650448885027512e-06,
+      "loss": 0.1457,
+      "step": 850
+    },
+    {
+      "epoch": 0.22283033620015638,
+      "grad_norm": 2.78125,
+      "learning_rate": 8.635968722849697e-06,
+      "loss": 0.156,
+      "step": 855
+    },
+    {
+      "epoch": 0.22413343758144383,
+      "grad_norm": 2.40625,
+      "learning_rate": 8.62148856067188e-06,
+      "loss": 0.1496,
+      "step": 860
+    },
+    {
+      "epoch": 0.2254365389627313,
+      "grad_norm": 2.25,
+      "learning_rate": 8.607008398494063e-06,
+      "loss": 0.1495,
+      "step": 865
+    },
+    {
+      "epoch": 0.22673964034401878,
+      "grad_norm": 2.359375,
+      "learning_rate": 8.592528236316247e-06,
+      "loss": 0.1501,
+      "step": 870
+    },
+    {
+      "epoch": 0.22804274172530622,
+      "grad_norm": 2.046875,
+      "learning_rate": 8.57804807413843e-06,
+      "loss": 0.154,
+      "step": 875
+    },
+    {
+      "epoch": 0.2293458431065937,
+      "grad_norm": 2.3125,
+      "learning_rate": 8.563567911960615e-06,
+      "loss": 0.1444,
+      "step": 880
+    },
+    {
+      "epoch": 0.23064894448788115,
+      "grad_norm": 2.453125,
+      "learning_rate": 8.549087749782798e-06,
+      "loss": 0.1505,
+      "step": 885
+    },
+    {
+      "epoch": 0.23195204586916862,
+      "grad_norm": 2.5625,
+      "learning_rate": 8.534607587604982e-06,
+      "loss": 0.1593,
+      "step": 890
+    },
+    {
+      "epoch": 0.2332551472504561,
+      "grad_norm": 2.328125,
+      "learning_rate": 8.520127425427165e-06,
+      "loss": 0.1477,
+      "step": 895
+    },
+    {
+      "epoch": 0.23455824863174354,
+      "grad_norm": 2.21875,
+      "learning_rate": 8.50564726324935e-06,
+      "loss": 0.1487,
+      "step": 900
+    },
+    {
+      "epoch": 0.23586135001303102,
+      "grad_norm": 2.1875,
+      "learning_rate": 8.491167101071533e-06,
+      "loss": 0.1434,
+      "step": 905
+    },
+    {
+      "epoch": 0.23716445139431847,
+      "grad_norm": 2.4375,
+      "learning_rate": 8.476686938893718e-06,
+      "loss": 0.1496,
+      "step": 910
+    },
+    {
+      "epoch": 0.23846755277560594,
+      "grad_norm": 2.359375,
+      "learning_rate": 8.4622067767159e-06,
+      "loss": 0.1515,
+      "step": 915
+    },
+    {
+      "epoch": 0.23977065415689341,
+      "grad_norm": 2.453125,
+      "learning_rate": 8.447726614538083e-06,
+      "loss": 0.155,
+      "step": 920
+    },
+    {
+      "epoch": 0.24107375553818086,
+      "grad_norm": 2.109375,
+      "learning_rate": 8.433246452360268e-06,
+      "loss": 0.146,
+      "step": 925
+    },
+    {
+      "epoch": 0.24237685691946834,
+      "grad_norm": 2.15625,
+      "learning_rate": 8.418766290182451e-06,
+      "loss": 0.1495,
+      "step": 930
+    },
+    {
+      "epoch": 0.2436799583007558,
+      "grad_norm": 2.203125,
+      "learning_rate": 8.404286128004634e-06,
+      "loss": 0.1536,
+      "step": 935
+    },
+    {
+      "epoch": 0.24498305968204326,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.389805965826819e-06,
+      "loss": 0.1529,
+      "step": 940
+    },
+    {
+      "epoch": 0.24628616106333073,
+      "grad_norm": 2.40625,
+      "learning_rate": 8.375325803649002e-06,
+      "loss": 0.157,
+      "step": 945
+    },
+    {
+      "epoch": 0.24758926244461818,
+      "grad_norm": 2.25,
+      "learning_rate": 8.360845641471184e-06,
+      "loss": 0.1486,
+      "step": 950
+    },
+    {
+      "epoch": 0.24889236382590565,
+      "grad_norm": 2.71875,
+      "learning_rate": 8.346365479293369e-06,
+      "loss": 0.1495,
+      "step": 955
+    },
+    {
+      "epoch": 0.2501954652071931,
+      "grad_norm": 2.234375,
+      "learning_rate": 8.331885317115552e-06,
+      "loss": 0.1415,
+      "step": 960
+    },
+    {
+      "epoch": 0.2514985665884806,
+      "grad_norm": 2.046875,
+      "learning_rate": 8.317405154937735e-06,
+      "loss": 0.1478,
+      "step": 965
+    },
+    {
+      "epoch": 0.25280166796976805,
+      "grad_norm": 2.15625,
+      "learning_rate": 8.30292499275992e-06,
+      "loss": 0.1531,
+      "step": 970
+    },
+    {
+      "epoch": 0.2541047693510555,
+      "grad_norm": 2.140625,
+      "learning_rate": 8.288444830582103e-06,
+      "loss": 0.1456,
+      "step": 975
+    },
+    {
+      "epoch": 0.255407870732343,
+      "grad_norm": 2.265625,
+      "learning_rate": 8.273964668404287e-06,
+      "loss": 0.149,
+      "step": 980
+    },
+    {
+      "epoch": 0.2567109721136304,
+      "grad_norm": 2.078125,
+      "learning_rate": 8.25948450622647e-06,
+      "loss": 0.1565,
+      "step": 985
+    },
+    {
+      "epoch": 0.2580140734949179,
+      "grad_norm": 2.421875,
+      "learning_rate": 8.245004344048653e-06,
+      "loss": 0.1463,
+      "step": 990
+    },
+    {
+      "epoch": 0.25931717487620537,
+      "grad_norm": 2.0625,
+      "learning_rate": 8.230524181870838e-06,
+      "loss": 0.1466,
+      "step": 995
+    },
+    {
+      "epoch": 0.26062027625749284,
+      "grad_norm": 2.40625,
+      "learning_rate": 8.21604401969302e-06,
+      "loss": 0.1492,
+      "step": 1000
+    },
+    {
+      "epoch": 0.2619233776387803,
+      "grad_norm": 2.234375,
+      "learning_rate": 8.201563857515205e-06,
+      "loss": 0.1538,
+      "step": 1005
+    },
+    {
+      "epoch": 0.26322647902006774,
+      "grad_norm": 2.15625,
+      "learning_rate": 8.187083695337388e-06,
+      "loss": 0.1436,
+      "step": 1010
+    },
+    {
+      "epoch": 0.2645295804013552,
+      "grad_norm": 2.171875,
+      "learning_rate": 8.172603533159573e-06,
+      "loss": 0.1543,
+      "step": 1015
+    },
+    {
+      "epoch": 0.2658326817826427,
+      "grad_norm": 2.140625,
+      "learning_rate": 8.158123370981756e-06,
+      "loss": 0.1461,
+      "step": 1020
+    },
+    {
+      "epoch": 0.26713578316393016,
+      "grad_norm": 2.453125,
+      "learning_rate": 8.14364320880394e-06,
+      "loss": 0.1554,
+      "step": 1025
+    },
+    {
+      "epoch": 0.26843888454521764,
+      "grad_norm": 2.28125,
+      "learning_rate": 8.129163046626123e-06,
+      "loss": 0.1517,
+      "step": 1030
+    },
+    {
+      "epoch": 0.26974198592650506,
+      "grad_norm": 2.09375,
+      "learning_rate": 8.114682884448306e-06,
+      "loss": 0.1467,
+      "step": 1035
+    },
+    {
+      "epoch": 0.27104508730779253,
+      "grad_norm": 2.34375,
+      "learning_rate": 8.100202722270491e-06,
+      "loss": 0.1444,
+      "step": 1040
+    },
+    {
+      "epoch": 0.27234818868908,
+      "grad_norm": 2.421875,
+      "learning_rate": 8.085722560092674e-06,
+      "loss": 0.1499,
+      "step": 1045
+    },
+    {
+      "epoch": 0.2736512900703675,
+      "grad_norm": 2.515625,
+      "learning_rate": 8.071242397914857e-06,
+      "loss": 0.1494,
+      "step": 1050
+    },
+    {
+      "epoch": 0.27495439145165496,
+      "grad_norm": 2.453125,
+      "learning_rate": 8.056762235737041e-06,
+      "loss": 0.1397,
+      "step": 1055
+    },
+    {
+      "epoch": 0.27625749283294243,
+      "grad_norm": 2.09375,
+      "learning_rate": 8.042282073559224e-06,
+      "loss": 0.146,
+      "step": 1060
+    },
+    {
+      "epoch": 0.27756059421422985,
+      "grad_norm": 2.046875,
+      "learning_rate": 8.027801911381407e-06,
+      "loss": 0.1477,
+      "step": 1065
+    },
+    {
+      "epoch": 0.2788636955955173,
+      "grad_norm": 1.90625,
+      "learning_rate": 8.013321749203592e-06,
+      "loss": 0.1517,
+      "step": 1070
+    },
+    {
+      "epoch": 0.2801667969768048,
+      "grad_norm": 2.296875,
+      "learning_rate": 7.998841587025775e-06,
+      "loss": 0.1522,
+      "step": 1075
+    },
+    {
+      "epoch": 0.2814698983580923,
+      "grad_norm": 2.1875,
+      "learning_rate": 7.984361424847958e-06,
+      "loss": 0.1571,
+      "step": 1080
+    },
+    {
+      "epoch": 0.28277299973937975,
+      "grad_norm": 2.3125,
+      "learning_rate": 7.969881262670143e-06,
+      "loss": 0.1424,
+      "step": 1085
+    },
+    {
+      "epoch": 0.28407610112066717,
+      "grad_norm": 2.265625,
+      "learning_rate": 7.955401100492325e-06,
+      "loss": 0.1458,
+      "step": 1090
+    },
+    {
+      "epoch": 0.28537920250195464,
+      "grad_norm": 2.171875,
+      "learning_rate": 7.94092093831451e-06,
+      "loss": 0.1484,
+      "step": 1095
+    },
+    {
+      "epoch": 0.2866823038832421,
+      "grad_norm": 1.984375,
+      "learning_rate": 7.926440776136693e-06,
+      "loss": 0.1493,
+      "step": 1100
+    },
+    {
+      "epoch": 0.2879854052645296,
+      "grad_norm": 2.328125,
+      "learning_rate": 7.911960613958876e-06,
+      "loss": 0.1471,
+      "step": 1105
+    },
+    {
+      "epoch": 0.28928850664581707,
+      "grad_norm": 2.078125,
+      "learning_rate": 7.89748045178106e-06,
+      "loss": 0.1472,
+      "step": 1110
+    },
+    {
+      "epoch": 0.2905916080271045,
+      "grad_norm": 2.40625,
+      "learning_rate": 7.883000289603244e-06,
+      "loss": 0.1443,
+      "step": 1115
+    },
+    {
+      "epoch": 0.29189470940839196,
+      "grad_norm": 2.0,
+      "learning_rate": 7.868520127425428e-06,
+      "loss": 0.1387,
+      "step": 1120
+    },
+    {
+      "epoch": 0.29319781078967944,
+      "grad_norm": 2.09375,
+      "learning_rate": 7.854039965247611e-06,
+      "loss": 0.1372,
+      "step": 1125
+    },
+    {
+      "epoch": 0.2945009121709669,
+      "grad_norm": 2.0625,
+      "learning_rate": 7.839559803069796e-06,
+      "loss": 0.1472,
+      "step": 1130
+    },
+    {
+      "epoch": 0.2958040135522544,
+      "grad_norm": 2.578125,
+      "learning_rate": 7.825079640891979e-06,
+      "loss": 0.1519,
+      "step": 1135
+    },
+    {
+      "epoch": 0.2971071149335418,
+      "grad_norm": 2.328125,
+      "learning_rate": 7.810599478714163e-06,
+      "loss": 0.1448,
+      "step": 1140
+    },
+    {
+      "epoch": 0.2984102163148293,
+      "grad_norm": 2.328125,
+      "learning_rate": 7.796119316536346e-06,
+      "loss": 0.1482,
+      "step": 1145
+    },
+    {
+      "epoch": 0.29971331769611675,
+      "grad_norm": 2.140625,
+      "learning_rate": 7.78163915435853e-06,
+      "loss": 0.1409,
+      "step": 1150
+    },
+    {
+      "epoch": 0.30023455824863177,
+      "eval_loss": 0.16887266933918,
+      "eval_runtime": 49.2745,
+      "eval_samples_per_second": 298.329,
+      "eval_steps_per_second": 9.335,
+      "step": 1152
+    },
+    {
+      "epoch": 0.30101641907740423,
+      "grad_norm": 2.171875,
+      "learning_rate": 7.767158992180714e-06,
+      "loss": 0.1418,
+      "step": 1155
+    },
+    {
+      "epoch": 0.3023195204586917,
+      "grad_norm": 2.375,
+      "learning_rate": 7.752678830002897e-06,
+      "loss": 0.1511,
+      "step": 1160
+    },
+    {
+      "epoch": 0.3036226218399791,
+      "grad_norm": 2.421875,
+      "learning_rate": 7.73819866782508e-06,
+      "loss": 0.1356,
+      "step": 1165
+    },
+    {
+      "epoch": 0.3049257232212666,
+      "grad_norm": 2.375,
+      "learning_rate": 7.723718505647264e-06,
+      "loss": 0.1461,
+      "step": 1170
+    },
+    {
+      "epoch": 0.3062288246025541,
+      "grad_norm": 2.25,
+      "learning_rate": 7.709238343469447e-06,
+      "loss": 0.1415,
+      "step": 1175
+    },
+    {
+      "epoch": 0.30753192598384155,
+      "grad_norm": 2.140625,
+      "learning_rate": 7.69475818129163e-06,
+      "loss": 0.1401,
+      "step": 1180
+    },
+    {
+      "epoch": 0.308835027365129,
+      "grad_norm": 2.03125,
+      "learning_rate": 7.680278019113815e-06,
+      "loss": 0.1477,
+      "step": 1185
+    },
+    {
+      "epoch": 0.3101381287464165,
+      "grad_norm": 1.921875,
+      "learning_rate": 7.665797856935998e-06,
+      "loss": 0.1526,
+      "step": 1190
+    },
+    {
+      "epoch": 0.3114412301277039,
+      "grad_norm": 2.375,
+      "learning_rate": 7.651317694758183e-06,
+      "loss": 0.143,
+      "step": 1195
+    },
+    {
+      "epoch": 0.3127443315089914,
+      "grad_norm": 2.0625,
+      "learning_rate": 7.636837532580365e-06,
+      "loss": 0.1538,
+      "step": 1200
+    },
+    {
+      "epoch": 0.31404743289027887,
+      "grad_norm": 2.40625,
+      "learning_rate": 7.622357370402549e-06,
+      "loss": 0.1497,
+      "step": 1205
+    },
+    {
+      "epoch": 0.31535053427156634,
+      "grad_norm": 2.015625,
+      "learning_rate": 7.607877208224733e-06,
+      "loss": 0.1461,
+      "step": 1210
+    },
+    {
+      "epoch": 0.3166536356528538,
+      "grad_norm": 2.15625,
+      "learning_rate": 7.593397046046917e-06,
+      "loss": 0.147,
+      "step": 1215
+    },
+    {
+      "epoch": 0.31795673703414123,
+      "grad_norm": 2.578125,
+      "learning_rate": 7.5789168838691e-06,
+      "loss": 0.1453,
+      "step": 1220
+    },
+    {
+      "epoch": 0.3192598384154287,
+      "grad_norm": 2.296875,
+      "learning_rate": 7.5644367216912836e-06,
+      "loss": 0.146,
+      "step": 1225
+    },
+    {
+      "epoch": 0.3205629397967162,
+      "grad_norm": 2.296875,
+      "learning_rate": 7.549956559513467e-06,
+      "loss": 0.1533,
+      "step": 1230
+    },
+    {
+      "epoch": 0.32186604117800366,
+      "grad_norm": 2.046875,
+      "learning_rate": 7.53547639733565e-06,
+      "loss": 0.1475,
+      "step": 1235
+    },
+    {
+      "epoch": 0.32316914255929113,
+      "grad_norm": 2.078125,
+      "learning_rate": 7.520996235157835e-06,
+      "loss": 0.1522,
+      "step": 1240
+    },
+    {
+      "epoch": 0.32447224394057855,
+      "grad_norm": 2.296875,
+      "learning_rate": 7.506516072980018e-06,
+      "loss": 0.1496,
+      "step": 1245
+    },
+    {
+      "epoch": 0.32577534532186603,
+      "grad_norm": 1.9765625,
+      "learning_rate": 7.492035910802201e-06,
+      "loss": 0.1426,
+      "step": 1250
+    },
+    {
+      "epoch": 0.3270784467031535,
+      "grad_norm": 2.515625,
+      "learning_rate": 7.4775557486243854e-06,
+      "loss": 0.1515,
+      "step": 1255
+    },
+    {
+      "epoch": 0.328381548084441,
+      "grad_norm": 2.234375,
+      "learning_rate": 7.463075586446568e-06,
+      "loss": 0.1374,
+      "step": 1260
+    },
+    {
+      "epoch": 0.32968464946572845,
+      "grad_norm": 2.1875,
+      "learning_rate": 7.448595424268752e-06,
+      "loss": 0.1424,
+      "step": 1265
+    },
+    {
+      "epoch": 0.33098775084701587,
+      "grad_norm": 2.375,
+      "learning_rate": 7.434115262090936e-06,
+      "loss": 0.1434,
+      "step": 1270
+    },
+    {
+      "epoch": 0.33229085222830335,
+      "grad_norm": 2.25,
+      "learning_rate": 7.41963509991312e-06,
+      "loss": 0.1441,
+      "step": 1275
+    },
+    {
+      "epoch": 0.3335939536095908,
+      "grad_norm": 2.078125,
+      "learning_rate": 7.405154937735303e-06,
+      "loss": 0.1419,
+      "step": 1280
+    },
+    {
+      "epoch": 0.3348970549908783,
+      "grad_norm": 2.265625,
+      "learning_rate": 7.390674775557487e-06,
+      "loss": 0.1471,
+      "step": 1285
+    },
+    {
+      "epoch": 0.33620015637216577,
+      "grad_norm": 2.140625,
+      "learning_rate": 7.37619461337967e-06,
+      "loss": 0.1478,
+      "step": 1290
+    },
+    {
+      "epoch": 0.33750325775345325,
+      "grad_norm": 2.34375,
+      "learning_rate": 7.361714451201855e-06,
+      "loss": 0.1432,
+      "step": 1295
+    },
+    {
+      "epoch": 0.33880635913474066,
+      "grad_norm": 2.53125,
+      "learning_rate": 7.347234289024038e-06,
+      "loss": 0.1434,
+      "step": 1300
+    },
+    {
+      "epoch": 0.34010946051602814,
+      "grad_norm": 1.9921875,
+      "learning_rate": 7.332754126846221e-06,
+      "loss": 0.1441,
+      "step": 1305
+    },
+    {
+      "epoch": 0.3414125618973156,
+      "grad_norm": 2.28125,
+      "learning_rate": 7.3182739646684054e-06,
+      "loss": 0.1445,
+      "step": 1310
+    },
+    {
+      "epoch": 0.3427156632786031,
+      "grad_norm": 2.28125,
+      "learning_rate": 7.303793802490588e-06,
+      "loss": 0.1419,
+      "step": 1315
+    },
+    {
+      "epoch": 0.34401876465989056,
+      "grad_norm": 2.125,
+      "learning_rate": 7.289313640312772e-06,
+      "loss": 0.1426,
+      "step": 1320
+    },
+    {
+      "epoch": 0.345321866041178,
+      "grad_norm": 2.5,
+      "learning_rate": 7.274833478134956e-06,
+      "loss": 0.1477,
+      "step": 1325
+    },
+    {
+      "epoch": 0.34662496742246546,
+      "grad_norm": 2.171875,
+      "learning_rate": 7.26035331595714e-06,
+      "loss": 0.1391,
+      "step": 1330
+    },
+    {
+      "epoch": 0.34792806880375293,
+      "grad_norm": 2.078125,
+      "learning_rate": 7.245873153779323e-06,
+      "loss": 0.1377,
+      "step": 1335
+    },
+    {
+      "epoch": 0.3492311701850404,
+      "grad_norm": 2.234375,
+      "learning_rate": 7.2313929916015065e-06,
+      "loss": 0.1455,
+      "step": 1340
+    },
+    {
+      "epoch": 0.3505342715663279,
+      "grad_norm": 2.3125,
+      "learning_rate": 7.21691282942369e-06,
+      "loss": 0.1449,
+      "step": 1345
+    },
+    {
+      "epoch": 0.3518373729476153,
+      "grad_norm": 2.0625,
+      "learning_rate": 7.202432667245873e-06,
+      "loss": 0.15,
+      "step": 1350
+    },
+    {
+      "epoch": 0.3531404743289028,
+      "grad_norm": 1.8828125,
+      "learning_rate": 7.187952505068058e-06,
+      "loss": 0.1456,
+      "step": 1355
+    },
+    {
+      "epoch": 0.35444357571019025,
+      "grad_norm": 2.203125,
+      "learning_rate": 7.173472342890241e-06,
+      "loss": 0.146,
+      "step": 1360
+    },
+    {
+      "epoch": 0.3557466770914777,
+      "grad_norm": 2.3125,
+      "learning_rate": 7.158992180712424e-06,
+      "loss": 0.1426,
+      "step": 1365
+    },
+    {
+      "epoch": 0.3570497784727652,
+      "grad_norm": 2.078125,
+      "learning_rate": 7.144512018534608e-06,
+      "loss": 0.139,
+      "step": 1370
+    },
+    {
+      "epoch": 0.3583528798540526,
+      "grad_norm": 1.9609375,
+      "learning_rate": 7.130031856356791e-06,
+      "loss": 0.1387,
+      "step": 1375
+    },
+    {
+      "epoch": 0.3596559812353401,
+      "grad_norm": 2.5625,
+      "learning_rate": 7.115551694178975e-06,
+      "loss": 0.1463,
+      "step": 1380
+    },
+    {
+      "epoch": 0.36095908261662757,
+      "grad_norm": 2.015625,
+      "learning_rate": 7.101071532001159e-06,
+      "loss": 0.1407,
+      "step": 1385
+    },
+    {
+      "epoch": 0.36226218399791504,
+      "grad_norm": 2.3125,
+      "learning_rate": 7.086591369823343e-06,
+      "loss": 0.14,
+      "step": 1390
+    },
+    {
+      "epoch": 0.3635652853792025,
+      "grad_norm": 1.9921875,
+      "learning_rate": 7.0721112076455265e-06,
+      "loss": 0.1442,
+      "step": 1395
+    },
+    {
+      "epoch": 0.36486838676049,
+      "grad_norm": 2.40625,
+      "learning_rate": 7.05763104546771e-06,
+      "loss": 0.1497,
+      "step": 1400
+    },
+    {
+      "epoch": 0.3661714881417774,
+      "grad_norm": 1.890625,
+      "learning_rate": 7.043150883289893e-06,
+      "loss": 0.1404,
+      "step": 1405
+    },
+    {
+      "epoch": 0.3674745895230649,
+      "grad_norm": 2.046875,
+      "learning_rate": 7.028670721112078e-06,
+      "loss": 0.1327,
+      "step": 1410
+    },
+    {
+      "epoch": 0.36877769090435236,
+      "grad_norm": 2.046875,
+      "learning_rate": 7.014190558934261e-06,
+      "loss": 0.1447,
+      "step": 1415
+    },
+    {
+      "epoch": 0.37008079228563984,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.999710396756444e-06,
+      "loss": 0.1513,
+      "step": 1420
+    },
+    {
+      "epoch": 0.3713838936669273,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.985230234578628e-06,
+      "loss": 0.1429,
+      "step": 1425
+    },
+    {
+      "epoch": 0.37268699504821473,
+      "grad_norm": 2.015625,
+      "learning_rate": 6.970750072400811e-06,
+      "loss": 0.1424,
+      "step": 1430
+    },
+    {
+      "epoch": 0.3739900964295022,
+      "grad_norm": 2.109375,
+      "learning_rate": 6.956269910222995e-06,
+      "loss": 0.1382,
+      "step": 1435
+    },
+    {
+      "epoch": 0.3752931978107897,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.941789748045179e-06,
+      "loss": 0.1472,
+      "step": 1440
+    },
+    {
+      "epoch": 0.37659629919207716,
+      "grad_norm": 2.078125,
+      "learning_rate": 6.927309585867362e-06,
+      "loss": 0.1379,
+      "step": 1445
+    },
+    {
+      "epoch": 0.37789940057336463,
+      "grad_norm": 2.3125,
+      "learning_rate": 6.912829423689546e-06,
+      "loss": 0.1397,
+      "step": 1450
+    },
+    {
+      "epoch": 0.37920250195465205,
+      "grad_norm": 2.140625,
+      "learning_rate": 6.898349261511729e-06,
+      "loss": 0.1443,
+      "step": 1455
+    },
+    {
+      "epoch": 0.3805056033359395,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.883869099333913e-06,
+      "loss": 0.14,
+      "step": 1460
+    },
+    {
+      "epoch": 0.381808704717227,
+      "grad_norm": 2.03125,
+      "learning_rate": 6.869388937156096e-06,
+      "loss": 0.1392,
+      "step": 1465
+    },
+    {
+      "epoch": 0.3831118060985145,
+      "grad_norm": 2.1875,
+      "learning_rate": 6.854908774978281e-06,
+      "loss": 0.1447,
+      "step": 1470
+    },
+    {
+      "epoch": 0.38441490747980195,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.840428612800464e-06,
+      "loss": 0.1419,
+      "step": 1475
+    },
+    {
+      "epoch": 0.38571800886108937,
+      "grad_norm": 2.234375,
+      "learning_rate": 6.825948450622647e-06,
+      "loss": 0.1408,
+      "step": 1480
+    },
+    {
+      "epoch": 0.38702111024237684,
+      "grad_norm": 2.203125,
+      "learning_rate": 6.811468288444831e-06,
+      "loss": 0.1482,
+      "step": 1485
+    },
+    {
+      "epoch": 0.3883242116236643,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.796988126267014e-06,
+      "loss": 0.139,
+      "step": 1490
+    },
+    {
+      "epoch": 0.3896273130049518,
+      "grad_norm": 2.265625,
+      "learning_rate": 6.782507964089199e-06,
+      "loss": 0.1472,
+      "step": 1495
+    },
+    {
+      "epoch": 0.39093041438623927,
+      "grad_norm": 1.9453125,
+      "learning_rate": 6.768027801911382e-06,
+      "loss": 0.1412,
+      "step": 1500
+    },
+    {
+      "epoch": 0.3922335157675267,
+      "grad_norm": 2.046875,
+      "learning_rate": 6.753547639733566e-06,
+      "loss": 0.14,
+      "step": 1505
+    },
+    {
+      "epoch": 0.39353661714881416,
+      "grad_norm": 2.265625,
+      "learning_rate": 6.739067477555749e-06,
+      "loss": 0.145,
+      "step": 1510
+    },
+    {
+      "epoch": 0.39483971853010164,
+      "grad_norm": 2.0625,
+      "learning_rate": 6.724587315377933e-06,
+      "loss": 0.1397,
+      "step": 1515
+    },
+    {
+      "epoch": 0.3961428199113891,
+      "grad_norm": 2.1875,
+      "learning_rate": 6.710107153200116e-06,
+      "loss": 0.15,
+      "step": 1520
+    },
+    {
+      "epoch": 0.3974459212926766,
+      "grad_norm": 2.203125,
+      "learning_rate": 6.695626991022301e-06,
+      "loss": 0.1401,
+      "step": 1525
+    },
+    {
+      "epoch": 0.39874902267396406,
+      "grad_norm": 1.9453125,
+      "learning_rate": 6.681146828844484e-06,
+      "loss": 0.1353,
+      "step": 1530
+    },
+    {
+      "epoch": 0.4000521240552515,
+      "grad_norm": 2.46875,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.1436,
+      "step": 1535
+    },
+    {
+      "epoch": 0.400312744331509,
+      "eval_loss": 0.16504338383674622,
+      "eval_runtime": 49.3084,
+      "eval_samples_per_second": 298.123,
+      "eval_steps_per_second": 9.329,
+      "step": 1536
+    },
+    {
+      "epoch": 0.40135522543653895,
+      "grad_norm": 2.0625,
+      "learning_rate": 6.652186504488851e-06,
+      "loss": 0.1405,
+      "step": 1540
+    },
+    {
+      "epoch": 0.40265832681782643,
+      "grad_norm": 1.984375,
+      "learning_rate": 6.637706342311034e-06,
+      "loss": 0.1397,
+      "step": 1545
+    },
+    {
+      "epoch": 0.4039614281991139,
+      "grad_norm": 2.390625,
+      "learning_rate": 6.623226180133218e-06,
+      "loss": 0.1429,
+      "step": 1550
+    },
+    {
+      "epoch": 0.4052645295804014,
+      "grad_norm": 2.078125,
+      "learning_rate": 6.608746017955402e-06,
+      "loss": 0.1476,
+      "step": 1555
+    },
+    {
+      "epoch": 0.4065676309616888,
+      "grad_norm": 2.671875,
+      "learning_rate": 6.594265855777585e-06,
+      "loss": 0.1463,
+      "step": 1560
+    },
+    {
+      "epoch": 0.40787073234297627,
+      "grad_norm": 2.15625,
+      "learning_rate": 6.5797856935997685e-06,
+      "loss": 0.1373,
+      "step": 1565
+    },
+    {
+      "epoch": 0.40917383372426375,
+      "grad_norm": 1.9609375,
+      "learning_rate": 6.565305531421952e-06,
+      "loss": 0.1356,
+      "step": 1570
+    },
+    {
+      "epoch": 0.4104769351055512,
+      "grad_norm": 1.984375,
+      "learning_rate": 6.550825369244136e-06,
+      "loss": 0.1422,
+      "step": 1575
+    },
+    {
+      "epoch": 0.4117800364868387,
+      "grad_norm": 2.109375,
+      "learning_rate": 6.536345207066319e-06,
+      "loss": 0.1392,
+      "step": 1580
+    },
+    {
+      "epoch": 0.4130831378681261,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.521865044888504e-06,
+      "loss": 0.1446,
+      "step": 1585
+    },
+    {
+      "epoch": 0.4143862392494136,
+      "grad_norm": 2.0625,
+      "learning_rate": 6.507384882710687e-06,
+      "loss": 0.1353,
+      "step": 1590
+    },
+    {
+      "epoch": 0.41568934063070107,
+      "grad_norm": 2.21875,
+      "learning_rate": 6.492904720532871e-06,
+      "loss": 0.1417,
+      "step": 1595
+    },
+    {
+      "epoch": 0.41699244201198854,
+      "grad_norm": 1.9140625,
+      "learning_rate": 6.478424558355054e-06,
+      "loss": 0.1359,
+      "step": 1600
+    },
+    {
+      "epoch": 0.418295543393276,
+      "grad_norm": 1.9140625,
+      "learning_rate": 6.463944396177237e-06,
+      "loss": 0.141,
+      "step": 1605
+    },
+    {
+      "epoch": 0.41959864477456343,
+      "grad_norm": 1.8828125,
+      "learning_rate": 6.449464233999422e-06,
+      "loss": 0.144,
+      "step": 1610
+    },
+    {
+      "epoch": 0.4209017461558509,
+      "grad_norm": 2.109375,
+      "learning_rate": 6.434984071821605e-06,
+      "loss": 0.1424,
+      "step": 1615
+    },
+    {
+      "epoch": 0.4222048475371384,
+      "grad_norm": 2.328125,
+      "learning_rate": 6.4205039096437885e-06,
+      "loss": 0.1374,
+      "step": 1620
+    },
+    {
+      "epoch": 0.42350794891842586,
+      "grad_norm": 2.359375,
+      "learning_rate": 6.406023747465972e-06,
+      "loss": 0.1332,
+      "step": 1625
+    },
+    {
+      "epoch": 0.42481105029971333,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.391543585288156e-06,
+      "loss": 0.1363,
+      "step": 1630
+    },
+    {
+      "epoch": 0.4261141516810008,
+      "grad_norm": 2.171875,
+      "learning_rate": 6.377063423110339e-06,
+      "loss": 0.1446,
+      "step": 1635
+    },
+    {
+      "epoch": 0.4274172530622882,
+      "grad_norm": 2.203125,
+      "learning_rate": 6.362583260932524e-06,
+      "loss": 0.1428,
+      "step": 1640
+    },
+    {
+      "epoch": 0.4287203544435757,
+      "grad_norm": 1.9453125,
+      "learning_rate": 6.348103098754707e-06,
+      "loss": 0.1331,
+      "step": 1645
+    },
+    {
+      "epoch": 0.4300234558248632,
+      "grad_norm": 2.265625,
+      "learning_rate": 6.3336229365768896e-06,
+      "loss": 0.1429,
+      "step": 1650
+    },
+    {
+      "epoch": 0.43132655720615065,
+      "grad_norm": 2.125,
+      "learning_rate": 6.319142774399074e-06,
+      "loss": 0.1396,
+      "step": 1655
+    },
+    {
+      "epoch": 0.4326296585874381,
+      "grad_norm": 2.03125,
+      "learning_rate": 6.304662612221257e-06,
+      "loss": 0.1464,
+      "step": 1660
+    },
+    {
+      "epoch": 0.43393275996872555,
+      "grad_norm": 1.90625,
+      "learning_rate": 6.290182450043441e-06,
+      "loss": 0.1417,
+      "step": 1665
+    },
+    {
+      "epoch": 0.435235861350013,
+      "grad_norm": 1.765625,
+      "learning_rate": 6.275702287865625e-06,
+      "loss": 0.141,
+      "step": 1670
+    },
+    {
+      "epoch": 0.4365389627313005,
+      "grad_norm": 2.875,
+      "learning_rate": 6.261222125687808e-06,
+      "loss": 0.1432,
+      "step": 1675
+    },
+    {
+      "epoch": 0.43784206411258797,
+      "grad_norm": 2.109375,
+      "learning_rate": 6.2467419635099915e-06,
+      "loss": 0.135,
+      "step": 1680
+    },
+    {
+      "epoch": 0.43914516549387544,
+      "grad_norm": 2.203125,
+      "learning_rate": 6.232261801332175e-06,
+      "loss": 0.1428,
+      "step": 1685
+    },
+    {
+      "epoch": 0.44044826687516286,
+      "grad_norm": 2.140625,
+      "learning_rate": 6.217781639154359e-06,
+      "loss": 0.1415,
+      "step": 1690
+    },
+    {
+      "epoch": 0.44175136825645034,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.203301476976543e-06,
+      "loss": 0.13,
+      "step": 1695
+    },
+    {
+      "epoch": 0.4430544696377378,
+      "grad_norm": 2.15625,
+      "learning_rate": 6.188821314798727e-06,
+      "loss": 0.1388,
+      "step": 1700
+    },
+    {
+      "epoch": 0.4443575710190253,
+      "grad_norm": 1.9140625,
+      "learning_rate": 6.1743411526209096e-06,
+      "loss": 0.1409,
+      "step": 1705
+    },
+    {
+      "epoch": 0.44566067240031276,
+      "grad_norm": 1.84375,
+      "learning_rate": 6.159860990443094e-06,
+      "loss": 0.1308,
+      "step": 1710
+    },
+    {
+      "epoch": 0.4469637737816002,
+      "grad_norm": 2.046875,
+      "learning_rate": 6.145380828265277e-06,
+      "loss": 0.1417,
+      "step": 1715
+    },
+    {
+      "epoch": 0.44826687516288766,
+      "grad_norm": 2.046875,
+      "learning_rate": 6.13090066608746e-06,
+      "loss": 0.1456,
+      "step": 1720
+    },
+    {
+      "epoch": 0.44956997654417513,
+      "grad_norm": 1.8671875,
+      "learning_rate": 6.116420503909645e-06,
+      "loss": 0.1413,
+      "step": 1725
+    },
+    {
+      "epoch": 0.4508730779254626,
+      "grad_norm": 2.265625,
+      "learning_rate": 6.101940341731828e-06,
+      "loss": 0.1384,
+      "step": 1730
+    },
+    {
+      "epoch": 0.4521761793067501,
+      "grad_norm": 1.9921875,
+      "learning_rate": 6.0874601795540115e-06,
+      "loss": 0.1406,
+      "step": 1735
+    },
+    {
+      "epoch": 0.45347928068803756,
+      "grad_norm": 2.1875,
+      "learning_rate": 6.072980017376195e-06,
+      "loss": 0.141,
+      "step": 1740
+    },
+    {
+      "epoch": 0.454782382069325,
+      "grad_norm": 2.046875,
+      "learning_rate": 6.058499855198379e-06,
+      "loss": 0.1478,
+      "step": 1745
+    },
+    {
+      "epoch": 0.45608548345061245,
+      "grad_norm": 2.109375,
+      "learning_rate": 6.044019693020562e-06,
+      "loss": 0.1412,
+      "step": 1750
+    },
+    {
+      "epoch": 0.4573885848318999,
+      "grad_norm": 1.9453125,
+      "learning_rate": 6.029539530842747e-06,
+      "loss": 0.1407,
+      "step": 1755
+    },
+    {
+      "epoch": 0.4586916862131874,
+      "grad_norm": 1.9296875,
+      "learning_rate": 6.0150593686649296e-06,
+      "loss": 0.1398,
+      "step": 1760
+    },
+    {
+      "epoch": 0.4599947875944749,
+      "grad_norm": 2.53125,
+      "learning_rate": 6.0005792064871125e-06,
+      "loss": 0.1397,
+      "step": 1765
+    },
+    {
+      "epoch": 0.4612978889757623,
+      "grad_norm": 2.328125,
+      "learning_rate": 5.986099044309297e-06,
+      "loss": 0.1458,
+      "step": 1770
+    },
+    {
+      "epoch": 0.46260099035704977,
+      "grad_norm": 2.25,
+      "learning_rate": 5.97161888213148e-06,
+      "loss": 0.1421,
+      "step": 1775
+    },
+    {
+      "epoch": 0.46390409173833724,
+      "grad_norm": 2.09375,
+      "learning_rate": 5.957138719953663e-06,
+      "loss": 0.1394,
+      "step": 1780
+    },
+    {
+      "epoch": 0.4652071931196247,
+      "grad_norm": 2.328125,
+      "learning_rate": 5.942658557775848e-06,
+      "loss": 0.1426,
+      "step": 1785
+    },
+    {
+      "epoch": 0.4665102945009122,
+      "grad_norm": 2.4375,
+      "learning_rate": 5.928178395598031e-06,
+      "loss": 0.1427,
+      "step": 1790
+    },
+    {
+      "epoch": 0.4678133958821996,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.913698233420215e-06,
+      "loss": 0.14,
+      "step": 1795
+    },
+    {
+      "epoch": 0.4691164972634871,
+      "grad_norm": 2.0,
+      "learning_rate": 5.899218071242398e-06,
+      "loss": 0.1403,
+      "step": 1800
+    },
+    {
+      "epoch": 0.47041959864477456,
+      "grad_norm": 2.609375,
+      "learning_rate": 5.884737909064582e-06,
+      "loss": 0.1472,
+      "step": 1805
+    },
+    {
+      "epoch": 0.47172270002606204,
+      "grad_norm": 1.8203125,
+      "learning_rate": 5.870257746886766e-06,
+      "loss": 0.1431,
+      "step": 1810
+    },
+    {
+      "epoch": 0.4730258014073495,
+      "grad_norm": 2.03125,
+      "learning_rate": 5.8557775847089495e-06,
+      "loss": 0.1376,
+      "step": 1815
+    },
+    {
+      "epoch": 0.47432890278863693,
+      "grad_norm": 2.484375,
+      "learning_rate": 5.8412974225311325e-06,
+      "loss": 0.1437,
+      "step": 1820
+    },
+    {
+      "epoch": 0.4756320041699244,
+      "grad_norm": 1.8515625,
+      "learning_rate": 5.826817260353317e-06,
+      "loss": 0.1352,
+      "step": 1825
+    },
+    {
+      "epoch": 0.4769351055512119,
+      "grad_norm": 2.078125,
+      "learning_rate": 5.8123370981755e-06,
+      "loss": 0.1385,
+      "step": 1830
+    },
+    {
+      "epoch": 0.47823820693249935,
+      "grad_norm": 2.703125,
+      "learning_rate": 5.797856935997683e-06,
+      "loss": 0.1408,
+      "step": 1835
+    },
+    {
+      "epoch": 0.47954130831378683,
+      "grad_norm": 2.078125,
+      "learning_rate": 5.783376773819868e-06,
+      "loss": 0.1452,
+      "step": 1840
+    },
+    {
+      "epoch": 0.4808444096950743,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.768896611642051e-06,
+      "loss": 0.1372,
+      "step": 1845
+    },
+    {
+      "epoch": 0.4821475110763617,
+      "grad_norm": 1.9921875,
+      "learning_rate": 5.754416449464234e-06,
+      "loss": 0.1416,
+      "step": 1850
+    },
+    {
+      "epoch": 0.4834506124576492,
+      "grad_norm": 2.125,
+      "learning_rate": 5.739936287286418e-06,
+      "loss": 0.1502,
+      "step": 1855
+    },
+    {
+      "epoch": 0.4847537138389367,
+      "grad_norm": 2.171875,
+      "learning_rate": 5.725456125108602e-06,
+      "loss": 0.1389,
+      "step": 1860
+    },
+    {
+      "epoch": 0.48605681522022415,
+      "grad_norm": 2.296875,
+      "learning_rate": 5.710975962930785e-06,
+      "loss": 0.1355,
+      "step": 1865
+    },
+    {
+      "epoch": 0.4873599166015116,
+      "grad_norm": 2.15625,
+      "learning_rate": 5.6964958007529695e-06,
+      "loss": 0.1404,
+      "step": 1870
+    },
+    {
+      "epoch": 0.48866301798279904,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.6820156385751525e-06,
+      "loss": 0.1399,
+      "step": 1875
+    },
+    {
+      "epoch": 0.4899661193640865,
+      "grad_norm": 2.0,
+      "learning_rate": 5.667535476397335e-06,
+      "loss": 0.1404,
+      "step": 1880
+    },
+    {
+      "epoch": 0.491269220745374,
+      "grad_norm": 2.0625,
+      "learning_rate": 5.65305531421952e-06,
+      "loss": 0.1399,
+      "step": 1885
+    },
+    {
+      "epoch": 0.49257232212666147,
+      "grad_norm": 1.7421875,
+      "learning_rate": 5.638575152041703e-06,
+      "loss": 0.1385,
+      "step": 1890
+    },
+    {
+      "epoch": 0.49387542350794894,
+      "grad_norm": 2.125,
+      "learning_rate": 5.624094989863888e-06,
+      "loss": 0.1362,
+      "step": 1895
+    },
+    {
+      "epoch": 0.49517852488923636,
+      "grad_norm": 2.0625,
+      "learning_rate": 5.609614827686071e-06,
+      "loss": 0.1394,
+      "step": 1900
+    },
+    {
+      "epoch": 0.49648162627052383,
+      "grad_norm": 2.125,
+      "learning_rate": 5.5951346655082535e-06,
+      "loss": 0.1378,
+      "step": 1905
+    },
+    {
+      "epoch": 0.4977847276518113,
+      "grad_norm": 2.671875,
+      "learning_rate": 5.580654503330438e-06,
+      "loss": 0.1399,
+      "step": 1910
+    },
+    {
+      "epoch": 0.4990878290330988,
+      "grad_norm": 2.125,
+      "learning_rate": 5.566174341152621e-06,
+      "loss": 0.1367,
+      "step": 1915
+    },
+    {
+      "epoch": 0.5003909304143862,
+      "grad_norm": 2.328125,
+      "learning_rate": 5.551694178974805e-06,
+      "loss": 0.1381,
+      "step": 1920
+    },
+    {
+      "epoch": 0.5003909304143862,
+      "eval_loss": 0.16250096261501312,
+      "eval_runtime": 49.2724,
+      "eval_samples_per_second": 298.342,
+      "eval_steps_per_second": 9.336,
+      "step": 1920
+    },
+    {
+      "epoch": 0.5016940317956737,
+      "grad_norm": 2.21875,
+      "learning_rate": 5.537214016796989e-06,
+      "loss": 0.1415,
+      "step": 1925
+    },
+    {
+      "epoch": 0.5029971331769612,
+      "grad_norm": 2.375,
+      "learning_rate": 5.5227338546191725e-06,
+      "loss": 0.1413,
+      "step": 1930
+    },
+    {
+      "epoch": 0.5043002345582487,
+      "grad_norm": 2.3125,
+      "learning_rate": 5.508253692441355e-06,
+      "loss": 0.1418,
+      "step": 1935
+    },
+    {
+      "epoch": 0.5056033359395361,
+      "grad_norm": 2.03125,
+      "learning_rate": 5.49377353026354e-06,
+      "loss": 0.1403,
+      "step": 1940
+    },
+    {
+      "epoch": 0.5069064373208235,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.479293368085723e-06,
+      "loss": 0.1344,
+      "step": 1945
+    },
+    {
+      "epoch": 0.508209538702111,
+      "grad_norm": 2.234375,
+      "learning_rate": 5.464813205907906e-06,
+      "loss": 0.1362,
+      "step": 1950
+    },
+    {
+      "epoch": 0.5095126400833985,
+      "grad_norm": 2.328125,
+      "learning_rate": 5.4503330437300906e-06,
+      "loss": 0.1374,
+      "step": 1955
+    },
+    {
+      "epoch": 0.510815741464686,
+      "grad_norm": 2.265625,
+      "learning_rate": 5.4358528815522735e-06,
+      "loss": 0.1389,
+      "step": 1960
+    },
+    {
+      "epoch": 0.5121188428459734,
+      "grad_norm": 2.171875,
+      "learning_rate": 5.421372719374457e-06,
+      "loss": 0.1414,
+      "step": 1965
+    },
+    {
+      "epoch": 0.5134219442272608,
+      "grad_norm": 2.09375,
+      "learning_rate": 5.406892557196641e-06,
+      "loss": 0.1366,
+      "step": 1970
+    },
+    {
+      "epoch": 0.5147250456085484,
+      "grad_norm": 2.265625,
+      "learning_rate": 5.392412395018825e-06,
+      "loss": 0.1424,
+      "step": 1975
+    },
+    {
+      "epoch": 0.5160281469898358,
+      "grad_norm": 2.125,
+      "learning_rate": 5.377932232841008e-06,
+      "loss": 0.1387,
+      "step": 1980
+    },
+    {
+      "epoch": 0.5173312483711233,
+      "grad_norm": 1.984375,
+      "learning_rate": 5.3634520706631925e-06,
+      "loss": 0.1393,
+      "step": 1985
+    },
+    {
+      "epoch": 0.5186343497524107,
+      "grad_norm": 2.1875,
+      "learning_rate": 5.348971908485375e-06,
+      "loss": 0.1339,
+      "step": 1990
+    },
+    {
+      "epoch": 0.5199374511336982,
+      "grad_norm": 2.296875,
+      "learning_rate": 5.33449174630756e-06,
+      "loss": 0.1366,
+      "step": 1995
+    },
+    {
+      "epoch": 0.5212405525149857,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.320011584129743e-06,
+      "loss": 0.1333,
+      "step": 2000
+    },
+    {
+      "epoch": 0.5225436538962731,
+      "grad_norm": 2.140625,
+      "learning_rate": 5.305531421951926e-06,
+      "loss": 0.1352,
+      "step": 2005
+    },
+    {
+      "epoch": 0.5238467552775606,
+      "grad_norm": 2.0,
+      "learning_rate": 5.2910512597741106e-06,
+      "loss": 0.1384,
+      "step": 2010
+    },
+    {
+      "epoch": 0.5251498566588481,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.2765710975962935e-06,
+      "loss": 0.137,
+      "step": 2015
+    },
+    {
+      "epoch": 0.5264529580401355,
+      "grad_norm": 1.9296875,
+      "learning_rate": 5.2620909354184764e-06,
+      "loss": 0.132,
+      "step": 2020
+    },
+    {
+      "epoch": 0.527756059421423,
+      "grad_norm": 2.03125,
+      "learning_rate": 5.247610773240661e-06,
+      "loss": 0.1414,
+      "step": 2025
+    },
+    {
+      "epoch": 0.5290591608027104,
+      "grad_norm": 2.15625,
+      "learning_rate": 5.233130611062844e-06,
+      "loss": 0.1389,
+      "step": 2030
+    },
+    {
+      "epoch": 0.530362262183998,
+      "grad_norm": 1.8828125,
+      "learning_rate": 5.218650448885028e-06,
+      "loss": 0.1405,
+      "step": 2035
+    },
+    {
+      "epoch": 0.5316653635652854,
+      "grad_norm": 1.90625,
+      "learning_rate": 5.204170286707212e-06,
+      "loss": 0.138,
+      "step": 2040
+    },
+    {
+      "epoch": 0.5329684649465728,
+      "grad_norm": 1.859375,
+      "learning_rate": 5.189690124529395e-06,
+      "loss": 0.1382,
+      "step": 2045
+    },
+    {
+      "epoch": 0.5342715663278603,
+      "grad_norm": 1.984375,
+      "learning_rate": 5.175209962351578e-06,
+      "loss": 0.1332,
+      "step": 2050
+    },
+    {
+      "epoch": 0.5355746677091477,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.160729800173763e-06,
+      "loss": 0.1303,
+      "step": 2055
+    },
+    {
+      "epoch": 0.5368777690904353,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.146249637995946e-06,
+      "loss": 0.1356,
+      "step": 2060
+    },
+    {
+      "epoch": 0.5381808704717227,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.131769475818129e-06,
+      "loss": 0.1397,
+      "step": 2065
+    },
+    {
+      "epoch": 0.5394839718530101,
+      "grad_norm": 2.265625,
+      "learning_rate": 5.1172893136403135e-06,
+      "loss": 0.1384,
+      "step": 2070
+    },
+    {
+      "epoch": 0.5407870732342976,
+      "grad_norm": 2.28125,
+      "learning_rate": 5.1028091514624964e-06,
+      "loss": 0.1385,
+      "step": 2075
+    },
+    {
+      "epoch": 0.5420901746155851,
+      "grad_norm": 2.09375,
+      "learning_rate": 5.08832898928468e-06,
+      "loss": 0.1349,
+      "step": 2080
+    },
+    {
+      "epoch": 0.5433932759968726,
+      "grad_norm": 2.234375,
+      "learning_rate": 5.073848827106864e-06,
+      "loss": 0.1324,
+      "step": 2085
+    },
+    {
+      "epoch": 0.54469637737816,
+      "grad_norm": 2.0,
+      "learning_rate": 5.059368664929048e-06,
+      "loss": 0.1352,
+      "step": 2090
+    },
+    {
+      "epoch": 0.5459994787594474,
+      "grad_norm": 2.125,
+      "learning_rate": 5.044888502751232e-06,
+      "loss": 0.1326,
+      "step": 2095
+    },
+    {
+      "epoch": 0.547302580140735,
+      "grad_norm": 2.171875,
+      "learning_rate": 5.030408340573415e-06,
+      "loss": 0.1363,
+      "step": 2100
+    },
+    {
+      "epoch": 0.5486056815220224,
+      "grad_norm": 2.359375,
+      "learning_rate": 5.015928178395598e-06,
+      "loss": 0.1309,
+      "step": 2105
+    },
+    {
+      "epoch": 0.5499087829033099,
+      "grad_norm": 1.8671875,
+      "learning_rate": 5.001448016217782e-06,
+      "loss": 0.1398,
+      "step": 2110
+    },
+    {
+      "epoch": 0.5512118842845973,
+      "grad_norm": 2.21875,
+      "learning_rate": 4.986967854039966e-06,
+      "loss": 0.1307,
+      "step": 2115
+    },
+    {
+      "epoch": 0.5525149856658849,
+      "grad_norm": 2.0,
+      "learning_rate": 4.97248769186215e-06,
+      "loss": 0.1334,
+      "step": 2120
+    },
+    {
+      "epoch": 0.5538180870471723,
+      "grad_norm": 1.9375,
+      "learning_rate": 4.958007529684333e-06,
+      "loss": 0.1336,
+      "step": 2125
+    },
+    {
+      "epoch": 0.5551211884284597,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.9435273675065164e-06,
+      "loss": 0.1363,
+      "step": 2130
+    },
+    {
+      "epoch": 0.5564242898097472,
+      "grad_norm": 1.984375,
+      "learning_rate": 4.9290472053287e-06,
+      "loss": 0.144,
+      "step": 2135
+    },
+    {
+      "epoch": 0.5577273911910346,
+      "grad_norm": 2.0,
+      "learning_rate": 4.914567043150883e-06,
+      "loss": 0.1312,
+      "step": 2140
+    },
+    {
+      "epoch": 0.5590304925723222,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.900086880973067e-06,
+      "loss": 0.1339,
+      "step": 2145
+    },
+    {
+      "epoch": 0.5603335939536096,
+      "grad_norm": 2.53125,
+      "learning_rate": 4.885606718795251e-06,
+      "loss": 0.1363,
+      "step": 2150
+    },
+    {
+      "epoch": 0.561636695334897,
+      "grad_norm": 2.28125,
+      "learning_rate": 4.8711265566174345e-06,
+      "loss": 0.141,
+      "step": 2155
+    },
+    {
+      "epoch": 0.5629397967161845,
+      "grad_norm": 1.9140625,
+      "learning_rate": 4.856646394439618e-06,
+      "loss": 0.1312,
+      "step": 2160
+    },
+    {
+      "epoch": 0.564242898097472,
+      "grad_norm": 2.109375,
+      "learning_rate": 4.842166232261802e-06,
+      "loss": 0.1418,
+      "step": 2165
+    },
+    {
+      "epoch": 0.5655459994787595,
+      "grad_norm": 2.640625,
+      "learning_rate": 4.827686070083986e-06,
+      "loss": 0.1391,
+      "step": 2170
+    },
+    {
+      "epoch": 0.5668491008600469,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.813205907906169e-06,
+      "loss": 0.1379,
+      "step": 2175
+    },
+    {
+      "epoch": 0.5681522022413343,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.798725745728353e-06,
+      "loss": 0.1337,
+      "step": 2180
+    },
+    {
+      "epoch": 0.5694553036226219,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.784245583550536e-06,
+      "loss": 0.1339,
+      "step": 2185
+    },
+    {
+      "epoch": 0.5707584050039093,
+      "grad_norm": 2.5625,
+      "learning_rate": 4.769765421372719e-06,
+      "loss": 0.1413,
+      "step": 2190
+    },
+    {
+      "epoch": 0.5720615063851968,
+      "grad_norm": 2.109375,
+      "learning_rate": 4.755285259194903e-06,
+      "loss": 0.1394,
+      "step": 2195
+    },
+    {
+      "epoch": 0.5733646077664842,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.740805097017087e-06,
+      "loss": 0.1383,
+      "step": 2200
+    },
+    {
+      "epoch": 0.5746677091477717,
+      "grad_norm": 2.125,
+      "learning_rate": 4.726324934839271e-06,
+      "loss": 0.1409,
+      "step": 2205
+    },
+    {
+      "epoch": 0.5759708105290592,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.7118447726614545e-06,
+      "loss": 0.1359,
+      "step": 2210
+    },
+    {
+      "epoch": 0.5772739119103466,
+      "grad_norm": 1.9453125,
+      "learning_rate": 4.697364610483638e-06,
+      "loss": 0.1325,
+      "step": 2215
+    },
+    {
+      "epoch": 0.5785770132916341,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.682884448305822e-06,
+      "loss": 0.1383,
+      "step": 2220
+    },
+    {
+      "epoch": 0.5798801146729216,
+      "grad_norm": 1.9375,
+      "learning_rate": 4.668404286128005e-06,
+      "loss": 0.138,
+      "step": 2225
+    },
+    {
+      "epoch": 0.581183216054209,
+      "grad_norm": 2.046875,
+      "learning_rate": 4.653924123950189e-06,
+      "loss": 0.1275,
+      "step": 2230
+    },
+    {
+      "epoch": 0.5824863174354965,
+      "grad_norm": 2.140625,
+      "learning_rate": 4.639443961772373e-06,
+      "loss": 0.1408,
+      "step": 2235
+    },
+    {
+      "epoch": 0.5837894188167839,
+      "grad_norm": 1.8671875,
+      "learning_rate": 4.6249637995945556e-06,
+      "loss": 0.1286,
+      "step": 2240
+    },
+    {
+      "epoch": 0.5850925201980715,
+      "grad_norm": 2.109375,
+      "learning_rate": 4.610483637416739e-06,
+      "loss": 0.1424,
+      "step": 2245
+    },
+    {
+      "epoch": 0.5863956215793589,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.596003475238923e-06,
+      "loss": 0.1355,
+      "step": 2250
+    },
+    {
+      "epoch": 0.5876987229606463,
+      "grad_norm": 2.25,
+      "learning_rate": 4.581523313061106e-06,
+      "loss": 0.1409,
+      "step": 2255
+    },
+    {
+      "epoch": 0.5890018243419338,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.56704315088329e-06,
+      "loss": 0.1419,
+      "step": 2260
+    },
+    {
+      "epoch": 0.5903049257232212,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.552562988705474e-06,
+      "loss": 0.1317,
+      "step": 2265
+    },
+    {
+      "epoch": 0.5916080271045088,
+      "grad_norm": 1.921875,
+      "learning_rate": 4.5380828265276575e-06,
+      "loss": 0.1379,
+      "step": 2270
+    },
+    {
+      "epoch": 0.5929111284857962,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.523602664349841e-06,
+      "loss": 0.1337,
+      "step": 2275
+    },
+    {
+      "epoch": 0.5942142298670836,
+      "grad_norm": 1.90625,
+      "learning_rate": 4.509122502172025e-06,
+      "loss": 0.1293,
+      "step": 2280
+    },
+    {
+      "epoch": 0.5955173312483711,
+      "grad_norm": 2.109375,
+      "learning_rate": 4.494642339994209e-06,
+      "loss": 0.1414,
+      "step": 2285
+    },
+    {
+      "epoch": 0.5968204326296586,
+      "grad_norm": 2.15625,
+      "learning_rate": 4.480162177816392e-06,
+      "loss": 0.1327,
+      "step": 2290
+    },
+    {
+      "epoch": 0.5981235340109461,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.4656820156385756e-06,
+      "loss": 0.1383,
+      "step": 2295
+    },
+    {
+      "epoch": 0.5994266353922335,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.451201853460759e-06,
+      "loss": 0.1349,
+      "step": 2300
+    },
+    {
+      "epoch": 0.6004691164972635,
+      "eval_loss": 0.1611909121274948,
+      "eval_runtime": 49.2444,
+      "eval_samples_per_second": 298.511,
+      "eval_steps_per_second": 9.341,
+      "step": 2304
+    },
+    {
+      "epoch": 0.6007297367735209,
+      "grad_norm": 2.390625,
+      "learning_rate": 4.436721691282942e-06,
+      "loss": 0.1374,
+      "step": 2305
+    },
+    {
+      "epoch": 0.6020328381548085,
+      "grad_norm": 2.046875,
+      "learning_rate": 4.422241529105126e-06,
+      "loss": 0.1303,
+      "step": 2310
+    },
+    {
+      "epoch": 0.6033359395360959,
+      "grad_norm": 1.9609375,
+      "learning_rate": 4.40776136692731e-06,
+      "loss": 0.1292,
+      "step": 2315
+    },
+    {
+      "epoch": 0.6046390409173834,
+      "grad_norm": 1.953125,
+      "learning_rate": 4.393281204749494e-06,
+      "loss": 0.1412,
+      "step": 2320
+    },
+    {
+      "epoch": 0.6059421422986708,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.3788010425716774e-06,
+      "loss": 0.1396,
+      "step": 2325
+    },
+    {
+      "epoch": 0.6072452436799582,
+      "grad_norm": 2.109375,
+      "learning_rate": 4.364320880393861e-06,
+      "loss": 0.1325,
+      "step": 2330
+    },
+    {
+      "epoch": 0.6085483450612458,
+      "grad_norm": 2.125,
+      "learning_rate": 4.349840718216044e-06,
+      "loss": 0.1413,
+      "step": 2335
+    },
+    {
+      "epoch": 0.6098514464425332,
+      "grad_norm": 1.9453125,
+      "learning_rate": 4.335360556038228e-06,
+      "loss": 0.138,
+      "step": 2340
+    },
+    {
+      "epoch": 0.6111545478238207,
+      "grad_norm": 2.0625,
+      "learning_rate": 4.320880393860412e-06,
+      "loss": 0.1325,
+      "step": 2345
+    },
+    {
+      "epoch": 0.6124576492051081,
+      "grad_norm": 1.9140625,
+      "learning_rate": 4.3064002316825955e-06,
+      "loss": 0.1382,
+      "step": 2350
+    },
+    {
+      "epoch": 0.6137607505863957,
+      "grad_norm": 2.078125,
+      "learning_rate": 4.2919200695047785e-06,
+      "loss": 0.1403,
+      "step": 2355
+    },
+    {
+      "epoch": 0.6150638519676831,
+      "grad_norm": 2.234375,
+      "learning_rate": 4.277439907326962e-06,
+      "loss": 0.1308,
+      "step": 2360
+    },
+    {
+      "epoch": 0.6163669533489705,
+      "grad_norm": 2.125,
+      "learning_rate": 4.262959745149146e-06,
+      "loss": 0.138,
+      "step": 2365
+    },
+    {
+      "epoch": 0.617670054730258,
+      "grad_norm": 2.984375,
+      "learning_rate": 4.24847958297133e-06,
+      "loss": 0.1346,
+      "step": 2370
+    },
+    {
+      "epoch": 0.6189731561115455,
+      "grad_norm": 2.046875,
+      "learning_rate": 4.233999420793513e-06,
+      "loss": 0.1342,
+      "step": 2375
+    },
+    {
+      "epoch": 0.620276257492833,
+      "grad_norm": 1.8984375,
+      "learning_rate": 4.219519258615697e-06,
+      "loss": 0.1367,
+      "step": 2380
+    },
+    {
+      "epoch": 0.6215793588741204,
+      "grad_norm": 2.296875,
+      "learning_rate": 4.20503909643788e-06,
+      "loss": 0.1313,
+      "step": 2385
+    },
+    {
+      "epoch": 0.6228824602554078,
+      "grad_norm": 2.0625,
+      "learning_rate": 4.190558934260064e-06,
+      "loss": 0.1299,
+      "step": 2390
+    },
+    {
+      "epoch": 0.6241855616366954,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.176078772082248e-06,
+      "loss": 0.1286,
+      "step": 2395
+    },
+    {
+      "epoch": 0.6254886630179828,
+      "grad_norm": 2.25,
+      "learning_rate": 4.161598609904432e-06,
+      "loss": 0.1368,
+      "step": 2400
+    },
+    {
+      "epoch": 0.6267917643992703,
+      "grad_norm": 2.0,
+      "learning_rate": 4.147118447726615e-06,
+      "loss": 0.1316,
+      "step": 2405
+    },
+    {
+      "epoch": 0.6280948657805577,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.1326382855487985e-06,
+      "loss": 0.1346,
+      "step": 2410
+    },
+    {
+      "epoch": 0.6293979671618452,
+      "grad_norm": 1.921875,
+      "learning_rate": 4.118158123370982e-06,
+      "loss": 0.1375,
+      "step": 2415
+    },
+    {
+      "epoch": 0.6307010685431327,
+      "grad_norm": 2.59375,
+      "learning_rate": 4.103677961193166e-06,
+      "loss": 0.1322,
+      "step": 2420
+    },
+    {
+      "epoch": 0.6320041699244201,
+      "grad_norm": 1.9296875,
+      "learning_rate": 4.089197799015349e-06,
+      "loss": 0.1275,
+      "step": 2425
+    },
+    {
+      "epoch": 0.6333072713057076,
+      "grad_norm": 2.046875,
+      "learning_rate": 4.074717636837533e-06,
+      "loss": 0.1364,
+      "step": 2430
+    },
+    {
+      "epoch": 0.634610372686995,
+      "grad_norm": 1.96875,
+      "learning_rate": 4.0602374746597166e-06,
+      "loss": 0.1381,
+      "step": 2435
+    },
+    {
+      "epoch": 0.6359134740682825,
+      "grad_norm": 2.203125,
+      "learning_rate": 4.0457573124819e-06,
+      "loss": 0.1428,
+      "step": 2440
+    },
+    {
+      "epoch": 0.63721657544957,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.031277150304083e-06,
+      "loss": 0.1371,
+      "step": 2445
+    },
+    {
+      "epoch": 0.6385196768308574,
+      "grad_norm": 2.125,
+      "learning_rate": 4.016796988126267e-06,
+      "loss": 0.1384,
+      "step": 2450
+    },
+    {
+      "epoch": 0.639822778212145,
+      "grad_norm": 2.0,
+      "learning_rate": 4.002316825948451e-06,
+      "loss": 0.1389,
+      "step": 2455
+    },
+    {
+      "epoch": 0.6411258795934324,
+      "grad_norm": 2.234375,
+      "learning_rate": 3.987836663770635e-06,
+      "loss": 0.1322,
+      "step": 2460
+    },
+    {
+      "epoch": 0.6424289809747198,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.9733565015928185e-06,
+      "loss": 0.1454,
+      "step": 2465
+    },
+    {
+      "epoch": 0.6437320823560073,
+      "grad_norm": 1.7734375,
+      "learning_rate": 3.958876339415002e-06,
+      "loss": 0.1276,
+      "step": 2470
+    },
+    {
+      "epoch": 0.6450351837372947,
+      "grad_norm": 2.375,
+      "learning_rate": 3.944396177237185e-06,
+      "loss": 0.1334,
+      "step": 2475
+    },
+    {
+      "epoch": 0.6463382851185823,
+      "grad_norm": 2.703125,
+      "learning_rate": 3.929916015059369e-06,
+      "loss": 0.1355,
+      "step": 2480
+    },
+    {
+      "epoch": 0.6476413864998697,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.915435852881553e-06,
+      "loss": 0.1338,
+      "step": 2485
+    },
+    {
+      "epoch": 0.6489444878811571,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.900955690703736e-06,
+      "loss": 0.1335,
+      "step": 2490
+    },
+    {
+      "epoch": 0.6502475892624446,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.8864755285259195e-06,
+      "loss": 0.1353,
+      "step": 2495
+    },
+    {
+      "epoch": 0.6515506906437321,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.871995366348103e-06,
+      "loss": 0.1358,
+      "step": 2500
+    },
+    {
+      "epoch": 0.6528537920250196,
+      "grad_norm": 1.9140625,
+      "learning_rate": 3.857515204170287e-06,
+      "loss": 0.1299,
+      "step": 2505
+    },
+    {
+      "epoch": 0.654156893406307,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.843035041992471e-06,
+      "loss": 0.1346,
+      "step": 2510
+    },
+    {
+      "epoch": 0.6554599947875944,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.828554879814655e-06,
+      "loss": 0.1322,
+      "step": 2515
+    },
+    {
+      "epoch": 0.656763096168882,
+      "grad_norm": 1.953125,
+      "learning_rate": 3.814074717636838e-06,
+      "loss": 0.1324,
+      "step": 2520
+    },
+    {
+      "epoch": 0.6580661975501694,
+      "grad_norm": 1.953125,
+      "learning_rate": 3.7995945554590214e-06,
+      "loss": 0.1301,
+      "step": 2525
+    },
+    {
+      "epoch": 0.6593692989314569,
+      "grad_norm": 1.9765625,
+      "learning_rate": 3.785114393281205e-06,
+      "loss": 0.1365,
+      "step": 2530
+    },
+    {
+      "epoch": 0.6606724003127443,
+      "grad_norm": 1.8984375,
+      "learning_rate": 3.770634231103389e-06,
+      "loss": 0.1357,
+      "step": 2535
+    },
+    {
+      "epoch": 0.6619755016940317,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.756154068925572e-06,
+      "loss": 0.1352,
+      "step": 2540
+    },
+    {
+      "epoch": 0.6632786030753193,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.7416739067477557e-06,
+      "loss": 0.1278,
+      "step": 2545
+    },
+    {
+      "epoch": 0.6645817044566067,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.7271937445699395e-06,
+      "loss": 0.139,
+      "step": 2550
+    },
+    {
+      "epoch": 0.6658848058378942,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.712713582392123e-06,
+      "loss": 0.1453,
+      "step": 2555
+    },
+    {
+      "epoch": 0.6671879072191816,
+      "grad_norm": 2.0,
+      "learning_rate": 3.6982334202143067e-06,
+      "loss": 0.1366,
+      "step": 2560
+    },
+    {
+      "epoch": 0.6684910086004691,
+      "grad_norm": 2.5625,
+      "learning_rate": 3.6837532580364904e-06,
+      "loss": 0.1348,
+      "step": 2565
+    },
+    {
+      "epoch": 0.6697941099817566,
+      "grad_norm": 2.515625,
+      "learning_rate": 3.6692730958586742e-06,
+      "loss": 0.136,
+      "step": 2570
+    },
+    {
+      "epoch": 0.671097211363044,
+      "grad_norm": 1.921875,
+      "learning_rate": 3.6547929336808576e-06,
+      "loss": 0.1458,
+      "step": 2575
+    },
+    {
+      "epoch": 0.6724003127443315,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.6403127715030414e-06,
+      "loss": 0.1393,
+      "step": 2580
+    },
+    {
+      "epoch": 0.673703414125619,
+      "grad_norm": 2.125,
+      "learning_rate": 3.6258326093252248e-06,
+      "loss": 0.1326,
+      "step": 2585
+    },
+    {
+      "epoch": 0.6750065155069065,
+      "grad_norm": 2.1875,
+      "learning_rate": 3.611352447147408e-06,
+      "loss": 0.1305,
+      "step": 2590
+    },
+    {
+      "epoch": 0.6763096168881939,
+      "grad_norm": 2.296875,
+      "learning_rate": 3.596872284969592e-06,
+      "loss": 0.1385,
+      "step": 2595
+    },
+    {
+      "epoch": 0.6776127182694813,
+      "grad_norm": 1.9765625,
+      "learning_rate": 3.5823921227917757e-06,
+      "loss": 0.1295,
+      "step": 2600
+    },
+    {
+      "epoch": 0.6789158196507689,
+      "grad_norm": 2.140625,
+      "learning_rate": 3.567911960613959e-06,
+      "loss": 0.1316,
+      "step": 2605
+    },
+    {
+      "epoch": 0.6802189210320563,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.553431798436143e-06,
+      "loss": 0.1299,
+      "step": 2610
+    },
+    {
+      "epoch": 0.6815220224133438,
+      "grad_norm": 2.078125,
+      "learning_rate": 3.5389516362583266e-06,
+      "loss": 0.1307,
+      "step": 2615
+    },
+    {
+      "epoch": 0.6828251237946312,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.5244714740805096e-06,
+      "loss": 0.1329,
+      "step": 2620
+    },
+    {
+      "epoch": 0.6841282251759186,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.5099913119026934e-06,
+      "loss": 0.1314,
+      "step": 2625
+    },
+    {
+      "epoch": 0.6854313265572062,
+      "grad_norm": 2.03125,
+      "learning_rate": 3.495511149724877e-06,
+      "loss": 0.1286,
+      "step": 2630
+    },
+    {
+      "epoch": 0.6867344279384936,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.481030987547061e-06,
+      "loss": 0.136,
+      "step": 2635
+    },
+    {
+      "epoch": 0.6880375293197811,
+      "grad_norm": 2.34375,
+      "learning_rate": 3.4665508253692443e-06,
+      "loss": 0.1382,
+      "step": 2640
+    },
+    {
+      "epoch": 0.6893406307010685,
+      "grad_norm": 2.265625,
+      "learning_rate": 3.452070663191428e-06,
+      "loss": 0.1378,
+      "step": 2645
+    },
+    {
+      "epoch": 0.690643732082356,
+      "grad_norm": 1.9375,
+      "learning_rate": 3.437590501013612e-06,
+      "loss": 0.1323,
+      "step": 2650
+    },
+    {
+      "epoch": 0.6919468334636435,
+      "grad_norm": 1.9765625,
+      "learning_rate": 3.423110338835795e-06,
+      "loss": 0.1323,
+      "step": 2655
+    },
+    {
+      "epoch": 0.6932499348449309,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.4086301766579786e-06,
+      "loss": 0.1302,
+      "step": 2660
+    },
+    {
+      "epoch": 0.6945530362262184,
+      "grad_norm": 1.96875,
+      "learning_rate": 3.3941500144801624e-06,
+      "loss": 0.1344,
+      "step": 2665
+    },
+    {
+      "epoch": 0.6958561376075059,
+      "grad_norm": 2.015625,
+      "learning_rate": 3.379669852302346e-06,
+      "loss": 0.1321,
+      "step": 2670
+    },
+    {
+      "epoch": 0.6971592389887933,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.3651896901245296e-06,
+      "loss": 0.1349,
+      "step": 2675
+    },
+    {
+      "epoch": 0.6984623403700808,
+      "grad_norm": 2.25,
+      "learning_rate": 3.3507095279467134e-06,
+      "loss": 0.1409,
+      "step": 2680
+    },
+    {
+      "epoch": 0.6997654417513682,
+      "grad_norm": 1.9921875,
+      "learning_rate": 3.336229365768897e-06,
+      "loss": 0.1352,
+      "step": 2685
+    },
+    {
+      "epoch": 0.7005473025801408,
+      "eval_loss": 0.16054348647594452,
+      "eval_runtime": 49.2246,
+      "eval_samples_per_second": 298.631,
+      "eval_steps_per_second": 9.345,
+      "step": 2688
+    },
+    {
+      "epoch": 0.7010685431326558,
+      "grad_norm": 1.875,
+      "learning_rate": 3.3217492035910805e-06,
+      "loss": 0.1307,
+      "step": 2690
+    },
+    {
+      "epoch": 0.7023716445139432,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.3072690414132643e-06,
+      "loss": 0.1413,
+      "step": 2695
+    },
+    {
+      "epoch": 0.7036747458952306,
+      "grad_norm": 2.09375,
+      "learning_rate": 3.2927888792354477e-06,
+      "loss": 0.1426,
+      "step": 2700
+    },
+    {
+      "epoch": 0.7049778472765181,
+      "grad_norm": 2.21875,
+      "learning_rate": 3.278308717057631e-06,
+      "loss": 0.1334,
+      "step": 2705
+    },
+    {
+      "epoch": 0.7062809486578056,
+      "grad_norm": 1.875,
+      "learning_rate": 3.263828554879815e-06,
+      "loss": 0.1329,
+      "step": 2710
+    },
+    {
+      "epoch": 0.7075840500390931,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.2493483927019986e-06,
+      "loss": 0.1359,
+      "step": 2715
+    },
+    {
+      "epoch": 0.7088871514203805,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.234868230524182e-06,
+      "loss": 0.1361,
+      "step": 2720
+    },
+    {
+      "epoch": 0.7101902528016679,
+      "grad_norm": 1.9375,
+      "learning_rate": 3.2203880683463658e-06,
+      "loss": 0.1377,
+      "step": 2725
+    },
+    {
+      "epoch": 0.7114933541829555,
+      "grad_norm": 1.8984375,
+      "learning_rate": 3.2059079061685496e-06,
+      "loss": 0.1301,
+      "step": 2730
+    },
+    {
+      "epoch": 0.7127964555642429,
+      "grad_norm": 2.0,
+      "learning_rate": 3.1914277439907334e-06,
+      "loss": 0.1315,
+      "step": 2735
+    },
+    {
+      "epoch": 0.7140995569455304,
+      "grad_norm": 1.984375,
+      "learning_rate": 3.1769475818129163e-06,
+      "loss": 0.1309,
+      "step": 2740
+    },
+    {
+      "epoch": 0.7154026583268178,
+      "grad_norm": 1.9765625,
+      "learning_rate": 3.1624674196351e-06,
+      "loss": 0.1359,
+      "step": 2745
+    },
+    {
+      "epoch": 0.7167057597081052,
+      "grad_norm": 2.109375,
+      "learning_rate": 3.147987257457284e-06,
+      "loss": 0.148,
+      "step": 2750
+    },
+    {
+      "epoch": 0.7180088610893928,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.1335070952794672e-06,
+      "loss": 0.1287,
+      "step": 2755
+    },
+    {
+      "epoch": 0.7193119624706802,
+      "grad_norm": 2.171875,
+      "learning_rate": 3.119026933101651e-06,
+      "loss": 0.137,
+      "step": 2760
+    },
+    {
+      "epoch": 0.7206150638519677,
+      "grad_norm": 2.09375,
+      "learning_rate": 3.104546770923835e-06,
+      "loss": 0.1322,
+      "step": 2765
+    },
+    {
+      "epoch": 0.7219181652332551,
+      "grad_norm": 2.140625,
+      "learning_rate": 3.0900666087460178e-06,
+      "loss": 0.1354,
+      "step": 2770
+    },
+    {
+      "epoch": 0.7232212666145426,
+      "grad_norm": 2.15625,
+      "learning_rate": 3.0755864465682016e-06,
+      "loss": 0.1353,
+      "step": 2775
+    },
+    {
+      "epoch": 0.7245243679958301,
+      "grad_norm": 2.140625,
+      "learning_rate": 3.0611062843903853e-06,
+      "loss": 0.1287,
+      "step": 2780
+    },
+    {
+      "epoch": 0.7258274693771175,
+      "grad_norm": 1.8828125,
+      "learning_rate": 3.046626122212569e-06,
+      "loss": 0.1334,
+      "step": 2785
+    },
+    {
+      "epoch": 0.727130570758405,
+      "grad_norm": 2.046875,
+      "learning_rate": 3.0321459600347525e-06,
+      "loss": 0.1314,
+      "step": 2790
+    },
+    {
+      "epoch": 0.7284336721396925,
+      "grad_norm": 2.0625,
+      "learning_rate": 3.0176657978569363e-06,
+      "loss": 0.1381,
+      "step": 2795
+    },
+    {
+      "epoch": 0.72973677352098,
+      "grad_norm": 2.140625,
+      "learning_rate": 3.00318563567912e-06,
+      "loss": 0.1377,
+      "step": 2800
+    },
+    {
+      "epoch": 0.7310398749022674,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.9887054735013034e-06,
+      "loss": 0.1361,
+      "step": 2805
+    },
+    {
+      "epoch": 0.7323429762835548,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.974225311323487e-06,
+      "loss": 0.1361,
+      "step": 2810
+    },
+    {
+      "epoch": 0.7336460776648424,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.9597451491456706e-06,
+      "loss": 0.1273,
+      "step": 2815
+    },
+    {
+      "epoch": 0.7349491790461298,
+      "grad_norm": 1.890625,
+      "learning_rate": 2.945264986967854e-06,
+      "loss": 0.1328,
+      "step": 2820
+    },
+    {
+      "epoch": 0.7362522804274173,
+      "grad_norm": 2.265625,
+      "learning_rate": 2.9307848247900378e-06,
+      "loss": 0.1322,
+      "step": 2825
+    },
+    {
+      "epoch": 0.7375553818087047,
+      "grad_norm": 2.1875,
+      "learning_rate": 2.9163046626122215e-06,
+      "loss": 0.1322,
+      "step": 2830
+    },
+    {
+      "epoch": 0.7388584831899921,
+      "grad_norm": 2.3125,
+      "learning_rate": 2.9018245004344053e-06,
+      "loss": 0.1296,
+      "step": 2835
+    },
+    {
+      "epoch": 0.7401615845712797,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2.8873443382565887e-06,
+      "loss": 0.1352,
+      "step": 2840
+    },
+    {
+      "epoch": 0.7414646859525671,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.8728641760787725e-06,
+      "loss": 0.1346,
+      "step": 2845
+    },
+    {
+      "epoch": 0.7427677873338546,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.8583840139009563e-06,
+      "loss": 0.1325,
+      "step": 2850
+    },
+    {
+      "epoch": 0.744070888715142,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.8439038517231392e-06,
+      "loss": 0.1311,
+      "step": 2855
+    },
+    {
+      "epoch": 0.7453739900964295,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.829423689545323e-06,
+      "loss": 0.1342,
+      "step": 2860
+    },
+    {
+      "epoch": 0.746677091477717,
+      "grad_norm": 2.21875,
+      "learning_rate": 2.814943527367507e-06,
+      "loss": 0.1369,
+      "step": 2865
+    },
+    {
+      "epoch": 0.7479801928590044,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.80046336518969e-06,
+      "loss": 0.1378,
+      "step": 2870
+    },
+    {
+      "epoch": 0.7492832942402919,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.785983203011874e-06,
+      "loss": 0.1246,
+      "step": 2875
+    },
+    {
+      "epoch": 0.7505863956215794,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.7715030408340578e-06,
+      "loss": 0.1377,
+      "step": 2880
+    },
+    {
+      "epoch": 0.7518894970028668,
+      "grad_norm": 2.171875,
+      "learning_rate": 2.7570228786562415e-06,
+      "loss": 0.1345,
+      "step": 2885
+    },
+    {
+      "epoch": 0.7531925983841543,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.7425427164784245e-06,
+      "loss": 0.1344,
+      "step": 2890
+    },
+    {
+      "epoch": 0.7544956997654417,
+      "grad_norm": 2.109375,
+      "learning_rate": 2.7280625543006083e-06,
+      "loss": 0.1319,
+      "step": 2895
+    },
+    {
+      "epoch": 0.7557988011467293,
+      "grad_norm": 1.9375,
+      "learning_rate": 2.713582392122792e-06,
+      "loss": 0.1286,
+      "step": 2900
+    },
+    {
+      "epoch": 0.7571019025280167,
+      "grad_norm": 1.984375,
+      "learning_rate": 2.6991022299449754e-06,
+      "loss": 0.1281,
+      "step": 2905
+    },
+    {
+      "epoch": 0.7584050039093041,
+      "grad_norm": 2.0,
+      "learning_rate": 2.6846220677671592e-06,
+      "loss": 0.1318,
+      "step": 2910
+    },
+    {
+      "epoch": 0.7597081052905916,
+      "grad_norm": 2.015625,
+      "learning_rate": 2.670141905589343e-06,
+      "loss": 0.1317,
+      "step": 2915
+    },
+    {
+      "epoch": 0.761011206671879,
+      "grad_norm": 2.703125,
+      "learning_rate": 2.655661743411526e-06,
+      "loss": 0.1419,
+      "step": 2920
+    },
+    {
+      "epoch": 0.7623143080531666,
+      "grad_norm": 2.40625,
+      "learning_rate": 2.6411815812337097e-06,
+      "loss": 0.1354,
+      "step": 2925
+    },
+    {
+      "epoch": 0.763617409434454,
+      "grad_norm": 3.21875,
+      "learning_rate": 2.6267014190558935e-06,
+      "loss": 0.1425,
+      "step": 2930
+    },
+    {
+      "epoch": 0.7649205108157414,
+      "grad_norm": 2.0625,
+      "learning_rate": 2.6122212568780773e-06,
+      "loss": 0.133,
+      "step": 2935
+    },
+    {
+      "epoch": 0.766223612197029,
+      "grad_norm": 1.8984375,
+      "learning_rate": 2.5977410947002607e-06,
+      "loss": 0.1315,
+      "step": 2940
+    },
+    {
+      "epoch": 0.7675267135783164,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.5832609325224445e-06,
+      "loss": 0.1355,
+      "step": 2945
+    },
+    {
+      "epoch": 0.7688298149596039,
+      "grad_norm": 2.234375,
+      "learning_rate": 2.5687807703446283e-06,
+      "loss": 0.139,
+      "step": 2950
+    },
+    {
+      "epoch": 0.7701329163408913,
+      "grad_norm": 2.203125,
+      "learning_rate": 2.5543006081668116e-06,
+      "loss": 0.1372,
+      "step": 2955
+    },
+    {
+      "epoch": 0.7714360177221787,
+      "grad_norm": 2.28125,
+      "learning_rate": 2.5398204459889954e-06,
+      "loss": 0.1359,
+      "step": 2960
+    },
+    {
+      "epoch": 0.7727391191034663,
+      "grad_norm": 1.921875,
+      "learning_rate": 2.525340283811179e-06,
+      "loss": 0.1332,
+      "step": 2965
+    },
+    {
+      "epoch": 0.7740422204847537,
+      "grad_norm": 1.9453125,
+      "learning_rate": 2.510860121633362e-06,
+      "loss": 0.1375,
+      "step": 2970
+    },
+    {
+      "epoch": 0.7753453218660412,
+      "grad_norm": 2.0,
+      "learning_rate": 2.496379959455546e-06,
+      "loss": 0.1336,
+      "step": 2975
+    },
+    {
+      "epoch": 0.7766484232473286,
+      "grad_norm": 1.9609375,
+      "learning_rate": 2.4818997972777297e-06,
+      "loss": 0.1326,
+      "step": 2980
+    },
+    {
+      "epoch": 0.777951524628616,
+      "grad_norm": 1.8359375,
+      "learning_rate": 2.467419635099913e-06,
+      "loss": 0.131,
+      "step": 2985
+    },
+    {
+      "epoch": 0.7792546260099036,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.452939472922097e-06,
+      "loss": 0.1254,
+      "step": 2990
+    },
+    {
+      "epoch": 0.780557727391191,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2.4384593107442807e-06,
+      "loss": 0.1367,
+      "step": 2995
+    },
+    {
+      "epoch": 0.7818608287724785,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.423979148566464e-06,
+      "loss": 0.1301,
+      "step": 3000
+    },
+    {
+      "epoch": 0.783163930153766,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.409498986388648e-06,
+      "loss": 0.1346,
+      "step": 3005
+    },
+    {
+      "epoch": 0.7844670315350534,
+      "grad_norm": 1.859375,
+      "learning_rate": 2.395018824210831e-06,
+      "loss": 0.1285,
+      "step": 3010
+    },
+    {
+      "epoch": 0.7857701329163409,
+      "grad_norm": 2.375,
+      "learning_rate": 2.380538662033015e-06,
+      "loss": 0.1374,
+      "step": 3015
+    },
+    {
+      "epoch": 0.7870732342976283,
+      "grad_norm": 2.140625,
+      "learning_rate": 2.3660584998551988e-06,
+      "loss": 0.1382,
+      "step": 3020
+    },
+    {
+      "epoch": 0.7883763356789159,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.351578337677382e-06,
+      "loss": 0.1389,
+      "step": 3025
+    },
+    {
+      "epoch": 0.7896794370602033,
+      "grad_norm": 2.140625,
+      "learning_rate": 2.337098175499566e-06,
+      "loss": 0.1273,
+      "step": 3030
+    },
+    {
+      "epoch": 0.7909825384414908,
+      "grad_norm": 2.125,
+      "learning_rate": 2.3226180133217493e-06,
+      "loss": 0.1323,
+      "step": 3035
+    },
+    {
+      "epoch": 0.7922856398227782,
+      "grad_norm": 1.96875,
+      "learning_rate": 2.308137851143933e-06,
+      "loss": 0.1342,
+      "step": 3040
+    },
+    {
+      "epoch": 0.7935887412040656,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.2936576889661165e-06,
+      "loss": 0.133,
+      "step": 3045
+    },
+    {
+      "epoch": 0.7948918425853532,
+      "grad_norm": 2.609375,
+      "learning_rate": 2.2791775267883002e-06,
+      "loss": 0.1462,
+      "step": 3050
+    },
+    {
+      "epoch": 0.7961949439666406,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.264697364610484e-06,
+      "loss": 0.1412,
+      "step": 3055
+    },
+    {
+      "epoch": 0.7974980453479281,
+      "grad_norm": 2.1875,
+      "learning_rate": 2.2502172024326674e-06,
+      "loss": 0.144,
+      "step": 3060
+    },
+    {
+      "epoch": 0.7988011467292155,
+      "grad_norm": 2.078125,
+      "learning_rate": 2.235737040254851e-06,
+      "loss": 0.1328,
+      "step": 3065
+    },
+    {
+      "epoch": 0.800104248110503,
+      "grad_norm": 2.265625,
+      "learning_rate": 2.2212568780770346e-06,
+      "loss": 0.1342,
+      "step": 3070
+    },
+    {
+      "epoch": 0.800625488663018,
+      "eval_loss": 0.1600351482629776,
+      "eval_runtime": 49.2423,
+      "eval_samples_per_second": 298.524,
+      "eval_steps_per_second": 9.342,
+      "step": 3072
+    },
+    {
+      "epoch": 0.8014073494917905,
+      "grad_norm": 2.03125,
+      "learning_rate": 2.2067767158992183e-06,
+      "loss": 0.1334,
+      "step": 3075
+    },
+    {
+      "epoch": 0.8027104508730779,
+      "grad_norm": 2.125,
+      "learning_rate": 2.192296553721402e-06,
+      "loss": 0.1283,
+      "step": 3080
+    },
+    {
+      "epoch": 0.8040135522543654,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.1778163915435855e-06,
+      "loss": 0.1341,
+      "step": 3085
+    },
+    {
+      "epoch": 0.8053166536356529,
+      "grad_norm": 2.234375,
+      "learning_rate": 2.1633362293657693e-06,
+      "loss": 0.1435,
+      "step": 3090
+    },
+    {
+      "epoch": 0.8066197550169403,
+      "grad_norm": 1.9765625,
+      "learning_rate": 2.1488560671879527e-06,
+      "loss": 0.1358,
+      "step": 3095
+    },
+    {
+      "epoch": 0.8079228563982278,
+      "grad_norm": 2.0,
+      "learning_rate": 2.134375905010136e-06,
+      "loss": 0.128,
+      "step": 3100
+    },
+    {
+      "epoch": 0.8092259577795152,
+      "grad_norm": 2.28125,
+      "learning_rate": 2.11989574283232e-06,
+      "loss": 0.1376,
+      "step": 3105
+    },
+    {
+      "epoch": 0.8105290591608028,
+      "grad_norm": 2.046875,
+      "learning_rate": 2.1054155806545036e-06,
+      "loss": 0.1307,
+      "step": 3110
+    },
+    {
+      "epoch": 0.8118321605420902,
+      "grad_norm": 1.9296875,
+      "learning_rate": 2.0909354184766874e-06,
+      "loss": 0.1371,
+      "step": 3115
+    },
+    {
+      "epoch": 0.8131352619233776,
+      "grad_norm": 2.1875,
+      "learning_rate": 2.0764552562988708e-06,
+      "loss": 0.1276,
+      "step": 3120
+    },
+    {
+      "epoch": 0.8144383633046651,
+      "grad_norm": 1.921875,
+      "learning_rate": 2.061975094121054e-06,
+      "loss": 0.1335,
+      "step": 3125
+    },
+    {
+      "epoch": 0.8157414646859525,
+      "grad_norm": 2.171875,
+      "learning_rate": 2.047494931943238e-06,
+      "loss": 0.1319,
+      "step": 3130
+    },
+    {
+      "epoch": 0.8170445660672401,
+      "grad_norm": 2.203125,
+      "learning_rate": 2.0330147697654217e-06,
+      "loss": 0.1343,
+      "step": 3135
+    },
+    {
+      "epoch": 0.8183476674485275,
+      "grad_norm": 1.8203125,
+      "learning_rate": 2.018534607587605e-06,
+      "loss": 0.1317,
+      "step": 3140
+    },
+    {
+      "epoch": 0.8196507688298149,
+      "grad_norm": 2.0625,
+      "learning_rate": 2.004054445409789e-06,
+      "loss": 0.1276,
+      "step": 3145
+    },
+    {
+      "epoch": 0.8209538702111024,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.9895742832319722e-06,
+      "loss": 0.1308,
+      "step": 3150
+    },
+    {
+      "epoch": 0.8222569715923899,
+      "grad_norm": 2.125,
+      "learning_rate": 1.975094121054156e-06,
+      "loss": 0.1354,
+      "step": 3155
+    },
+    {
+      "epoch": 0.8235600729736774,
+      "grad_norm": 1.890625,
+      "learning_rate": 1.9606139588763394e-06,
+      "loss": 0.1256,
+      "step": 3160
+    },
+    {
+      "epoch": 0.8248631743549648,
+      "grad_norm": 1.9140625,
+      "learning_rate": 1.946133796698523e-06,
+      "loss": 0.1308,
+      "step": 3165
+    },
+    {
+      "epoch": 0.8261662757362522,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.931653634520707e-06,
+      "loss": 0.1347,
+      "step": 3170
+    },
+    {
+      "epoch": 0.8274693771175398,
+      "grad_norm": 2.015625,
+      "learning_rate": 1.9171734723428903e-06,
+      "loss": 0.1348,
+      "step": 3175
+    },
+    {
+      "epoch": 0.8287724784988272,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.9026933101650741e-06,
+      "loss": 0.1355,
+      "step": 3180
+    },
+    {
+      "epoch": 0.8300755798801147,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.8882131479872575e-06,
+      "loss": 0.1356,
+      "step": 3185
+    },
+    {
+      "epoch": 0.8313786812614021,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.8737329858094413e-06,
+      "loss": 0.1331,
+      "step": 3190
+    },
+    {
+      "epoch": 0.8326817826426895,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.8592528236316248e-06,
+      "loss": 0.1335,
+      "step": 3195
+    },
+    {
+      "epoch": 0.8339848840239771,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.8447726614538084e-06,
+      "loss": 0.1325,
+      "step": 3200
+    },
+    {
+      "epoch": 0.8352879854052645,
+      "grad_norm": 2.109375,
+      "learning_rate": 1.8302924992759922e-06,
+      "loss": 0.1282,
+      "step": 3205
+    },
+    {
+      "epoch": 0.836591086786552,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.8158123370981756e-06,
+      "loss": 0.1385,
+      "step": 3210
+    },
+    {
+      "epoch": 0.8378941881678394,
+      "grad_norm": 2.328125,
+      "learning_rate": 1.8013321749203594e-06,
+      "loss": 0.1377,
+      "step": 3215
+    },
+    {
+      "epoch": 0.8391972895491269,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.786852012742543e-06,
+      "loss": 0.1344,
+      "step": 3220
+    },
+    {
+      "epoch": 0.8405003909304144,
+      "grad_norm": 2.1875,
+      "learning_rate": 1.7723718505647263e-06,
+      "loss": 0.1362,
+      "step": 3225
+    },
+    {
+      "epoch": 0.8418034923117018,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.75789168838691e-06,
+      "loss": 0.1292,
+      "step": 3230
+    },
+    {
+      "epoch": 0.8431065936929893,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.7434115262090937e-06,
+      "loss": 0.1394,
+      "step": 3235
+    },
+    {
+      "epoch": 0.8444096950742768,
+      "grad_norm": 2.125,
+      "learning_rate": 1.7289313640312775e-06,
+      "loss": 0.1246,
+      "step": 3240
+    },
+    {
+      "epoch": 0.8457127964555643,
+      "grad_norm": 1.96875,
+      "learning_rate": 1.7144512018534608e-06,
+      "loss": 0.1382,
+      "step": 3245
+    },
+    {
+      "epoch": 0.8470158978368517,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.6999710396756444e-06,
+      "loss": 0.1289,
+      "step": 3250
+    },
+    {
+      "epoch": 0.8483189992181391,
+      "grad_norm": 1.8984375,
+      "learning_rate": 1.6854908774978282e-06,
+      "loss": 0.1302,
+      "step": 3255
+    },
+    {
+      "epoch": 0.8496221005994267,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.6710107153200118e-06,
+      "loss": 0.1298,
+      "step": 3260
+    },
+    {
+      "epoch": 0.8509252019807141,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.6565305531421956e-06,
+      "loss": 0.1345,
+      "step": 3265
+    },
+    {
+      "epoch": 0.8522283033620016,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.642050390964379e-06,
+      "loss": 0.1278,
+      "step": 3270
+    },
+    {
+      "epoch": 0.853531404743289,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.6275702287865625e-06,
+      "loss": 0.1346,
+      "step": 3275
+    },
+    {
+      "epoch": 0.8548345061245765,
+      "grad_norm": 2.28125,
+      "learning_rate": 1.6130900666087463e-06,
+      "loss": 0.1381,
+      "step": 3280
+    },
+    {
+      "epoch": 0.856137607505864,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.5986099044309297e-06,
+      "loss": 0.1323,
+      "step": 3285
+    },
+    {
+      "epoch": 0.8574407088871514,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.5841297422531135e-06,
+      "loss": 0.1279,
+      "step": 3290
+    },
+    {
+      "epoch": 0.8587438102684389,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.569649580075297e-06,
+      "loss": 0.1309,
+      "step": 3295
+    },
+    {
+      "epoch": 0.8600469116497264,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.5551694178974804e-06,
+      "loss": 0.1343,
+      "step": 3300
+    },
+    {
+      "epoch": 0.8613500130310138,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.5406892557196642e-06,
+      "loss": 0.1292,
+      "step": 3305
+    },
+    {
+      "epoch": 0.8626531144123013,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.5262090935418478e-06,
+      "loss": 0.1388,
+      "step": 3310
+    },
+    {
+      "epoch": 0.8639562157935887,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.5117289313640316e-06,
+      "loss": 0.1333,
+      "step": 3315
+    },
+    {
+      "epoch": 0.8652593171748763,
+      "grad_norm": 1.9140625,
+      "learning_rate": 1.4972487691862151e-06,
+      "loss": 0.136,
+      "step": 3320
+    },
+    {
+      "epoch": 0.8665624185561637,
+      "grad_norm": 2.09375,
+      "learning_rate": 1.4827686070083985e-06,
+      "loss": 0.1324,
+      "step": 3325
+    },
+    {
+      "epoch": 0.8678655199374511,
+      "grad_norm": 1.96875,
+      "learning_rate": 1.4682884448305823e-06,
+      "loss": 0.1288,
+      "step": 3330
+    },
+    {
+      "epoch": 0.8691686213187386,
+      "grad_norm": 2.046875,
+      "learning_rate": 1.4538082826527659e-06,
+      "loss": 0.1311,
+      "step": 3335
+    },
+    {
+      "epoch": 0.870471722700026,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.4393281204749497e-06,
+      "loss": 0.1431,
+      "step": 3340
+    },
+    {
+      "epoch": 0.8717748240813136,
+      "grad_norm": 2.375,
+      "learning_rate": 1.424847958297133e-06,
+      "loss": 0.1322,
+      "step": 3345
+    },
+    {
+      "epoch": 0.873077925462601,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.4103677961193166e-06,
+      "loss": 0.1348,
+      "step": 3350
+    },
+    {
+      "epoch": 0.8743810268438884,
+      "grad_norm": 2.296875,
+      "learning_rate": 1.3958876339415004e-06,
+      "loss": 0.1336,
+      "step": 3355
+    },
+    {
+      "epoch": 0.8756841282251759,
+      "grad_norm": 1.9453125,
+      "learning_rate": 1.3814074717636838e-06,
+      "loss": 0.132,
+      "step": 3360
+    },
+    {
+      "epoch": 0.8769872296064634,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.3669273095858675e-06,
+      "loss": 0.1257,
+      "step": 3365
+    },
+    {
+      "epoch": 0.8782903309877509,
+      "grad_norm": 2.46875,
+      "learning_rate": 1.3524471474080511e-06,
+      "loss": 0.1399,
+      "step": 3370
+    },
+    {
+      "epoch": 0.8795934323690383,
+      "grad_norm": 2.375,
+      "learning_rate": 1.3379669852302347e-06,
+      "loss": 0.1376,
+      "step": 3375
+    },
+    {
+      "epoch": 0.8808965337503257,
+      "grad_norm": 1.953125,
+      "learning_rate": 1.3234868230524183e-06,
+      "loss": 0.126,
+      "step": 3380
+    },
+    {
+      "epoch": 0.8821996351316133,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.3090066608746019e-06,
+      "loss": 0.1342,
+      "step": 3385
+    },
+    {
+      "epoch": 0.8835027365129007,
+      "grad_norm": 2.3125,
+      "learning_rate": 1.2945264986967854e-06,
+      "loss": 0.1317,
+      "step": 3390
+    },
+    {
+      "epoch": 0.8848058378941882,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.2800463365189692e-06,
+      "loss": 0.1317,
+      "step": 3395
+    },
+    {
+      "epoch": 0.8861089392754756,
+      "grad_norm": 2.375,
+      "learning_rate": 1.2655661743411526e-06,
+      "loss": 0.1353,
+      "step": 3400
+    },
+    {
+      "epoch": 0.887412040656763,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.2510860121633364e-06,
+      "loss": 0.1363,
+      "step": 3405
+    },
+    {
+      "epoch": 0.8887151420380506,
+      "grad_norm": 2.203125,
+      "learning_rate": 1.23660584998552e-06,
+      "loss": 0.1385,
+      "step": 3410
+    },
+    {
+      "epoch": 0.890018243419338,
+      "grad_norm": 2.015625,
+      "learning_rate": 1.2221256878077035e-06,
+      "loss": 0.1245,
+      "step": 3415
+    },
+    {
+      "epoch": 0.8913213448006255,
+      "grad_norm": 2.0,
+      "learning_rate": 1.2076455256298871e-06,
+      "loss": 0.1322,
+      "step": 3420
+    },
+    {
+      "epoch": 0.892624446181913,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.193165363452071e-06,
+      "loss": 0.1378,
+      "step": 3425
+    },
+    {
+      "epoch": 0.8939275475632004,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.1786852012742543e-06,
+      "loss": 0.1324,
+      "step": 3430
+    },
+    {
+      "epoch": 0.8952306489444879,
+      "grad_norm": 2.234375,
+      "learning_rate": 1.1642050390964378e-06,
+      "loss": 0.1369,
+      "step": 3435
+    },
+    {
+      "epoch": 0.8965337503257753,
+      "grad_norm": 2.265625,
+      "learning_rate": 1.1497248769186216e-06,
+      "loss": 0.1333,
+      "step": 3440
+    },
+    {
+      "epoch": 0.8978368517070628,
+      "grad_norm": 1.9765625,
+      "learning_rate": 1.1352447147408052e-06,
+      "loss": 0.1281,
+      "step": 3445
+    },
+    {
+      "epoch": 0.8991399530883503,
+      "grad_norm": 2.21875,
+      "learning_rate": 1.1207645525629888e-06,
+      "loss": 0.1431,
+      "step": 3450
+    },
+    {
+      "epoch": 0.9004430544696377,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.1062843903851724e-06,
+      "loss": 0.1332,
+      "step": 3455
+    },
+    {
+      "epoch": 0.9007036747458952,
+      "eval_loss": 0.15992629528045654,
+      "eval_runtime": 49.2173,
+      "eval_samples_per_second": 298.675,
+      "eval_steps_per_second": 9.346,
+      "step": 3456
+    },
+    {
+      "epoch": 0.9017461558509252,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.091804228207356e-06,
+      "loss": 0.1307,
+      "step": 3460
+    },
+    {
+      "epoch": 0.9030492572322126,
+      "grad_norm": 2.140625,
+      "learning_rate": 1.0773240660295395e-06,
+      "loss": 0.1314,
+      "step": 3465
+    },
+    {
+      "epoch": 0.9043523586135002,
+      "grad_norm": 1.9296875,
+      "learning_rate": 1.0628439038517233e-06,
+      "loss": 0.1371,
+      "step": 3470
+    },
+    {
+      "epoch": 0.9056554599947876,
+      "grad_norm": 2.15625,
+      "learning_rate": 1.0483637416739069e-06,
+      "loss": 0.1296,
+      "step": 3475
+    },
+    {
+      "epoch": 0.9069585613760751,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.0338835794960905e-06,
+      "loss": 0.126,
+      "step": 3480
+    },
+    {
+      "epoch": 0.9082616627573625,
+      "grad_norm": 2.0,
+      "learning_rate": 1.019403417318274e-06,
+      "loss": 0.1379,
+      "step": 3485
+    },
+    {
+      "epoch": 0.90956476413865,
+      "grad_norm": 2.171875,
+      "learning_rate": 1.0049232551404576e-06,
+      "loss": 0.1267,
+      "step": 3490
+    },
+    {
+      "epoch": 0.9108678655199375,
+      "grad_norm": 2.046875,
+      "learning_rate": 9.904430929626412e-07,
+      "loss": 0.1342,
+      "step": 3495
+    },
+    {
+      "epoch": 0.9121709669012249,
+      "grad_norm": 2.15625,
+      "learning_rate": 9.75962930784825e-07,
+      "loss": 0.1278,
+      "step": 3500
+    },
+    {
+      "epoch": 0.9134740682825124,
+      "grad_norm": 2.109375,
+      "learning_rate": 9.614827686070084e-07,
+      "loss": 0.1326,
+      "step": 3505
+    },
+    {
+      "epoch": 0.9147771696637998,
+      "grad_norm": 2.078125,
+      "learning_rate": 9.47002606429192e-07,
+      "loss": 0.1331,
+      "step": 3510
+    },
+    {
+      "epoch": 0.9160802710450873,
+      "grad_norm": 1.84375,
+      "learning_rate": 9.325224442513757e-07,
+      "loss": 0.1308,
+      "step": 3515
+    },
+    {
+      "epoch": 0.9173833724263748,
+      "grad_norm": 2.125,
+      "learning_rate": 9.180422820735593e-07,
+      "loss": 0.1354,
+      "step": 3520
+    },
+    {
+      "epoch": 0.9186864738076622,
+      "grad_norm": 1.8046875,
+      "learning_rate": 9.03562119895743e-07,
+      "loss": 0.1276,
+      "step": 3525
+    },
+    {
+      "epoch": 0.9199895751889497,
+      "grad_norm": 1.9296875,
+      "learning_rate": 8.890819577179265e-07,
+      "loss": 0.1336,
+      "step": 3530
+    },
+    {
+      "epoch": 0.9212926765702372,
+      "grad_norm": 2.09375,
+      "learning_rate": 8.746017955401101e-07,
+      "loss": 0.1311,
+      "step": 3535
+    },
+    {
+      "epoch": 0.9225957779515246,
+      "grad_norm": 2.125,
+      "learning_rate": 8.601216333622937e-07,
+      "loss": 0.1289,
+      "step": 3540
+    },
+    {
+      "epoch": 0.9238988793328121,
+      "grad_norm": 1.9921875,
+      "learning_rate": 8.456414711844774e-07,
+      "loss": 0.1339,
+      "step": 3545
+    },
+    {
+      "epoch": 0.9252019807140995,
+      "grad_norm": 2.09375,
+      "learning_rate": 8.31161309006661e-07,
+      "loss": 0.1354,
+      "step": 3550
+    },
+    {
+      "epoch": 0.9265050820953871,
+      "grad_norm": 2.21875,
+      "learning_rate": 8.166811468288445e-07,
+      "loss": 0.1395,
+      "step": 3555
+    },
+    {
+      "epoch": 0.9278081834766745,
+      "grad_norm": 2.09375,
+      "learning_rate": 8.022009846510281e-07,
+      "loss": 0.1365,
+      "step": 3560
+    },
+    {
+      "epoch": 0.9291112848579619,
+      "grad_norm": 1.90625,
+      "learning_rate": 7.877208224732118e-07,
+      "loss": 0.1335,
+      "step": 3565
+    },
+    {
+      "epoch": 0.9304143862392494,
+      "grad_norm": 2.515625,
+      "learning_rate": 7.732406602953954e-07,
+      "loss": 0.1387,
+      "step": 3570
+    },
+    {
+      "epoch": 0.9317174876205369,
+      "grad_norm": 2.25,
+      "learning_rate": 7.587604981175791e-07,
+      "loss": 0.1338,
+      "step": 3575
+    },
+    {
+      "epoch": 0.9330205890018244,
+      "grad_norm": 2.328125,
+      "learning_rate": 7.442803359397626e-07,
+      "loss": 0.1328,
+      "step": 3580
+    },
+    {
+      "epoch": 0.9343236903831118,
+      "grad_norm": 2.078125,
+      "learning_rate": 7.298001737619461e-07,
+      "loss": 0.1293,
+      "step": 3585
+    },
+    {
+      "epoch": 0.9356267917643992,
+      "grad_norm": 2.34375,
+      "learning_rate": 7.153200115841298e-07,
+      "loss": 0.1384,
+      "step": 3590
+    },
+    {
+      "epoch": 0.9369298931456868,
+      "grad_norm": 1.953125,
+      "learning_rate": 7.008398494063134e-07,
+      "loss": 0.1228,
+      "step": 3595
+    },
+    {
+      "epoch": 0.9382329945269742,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.863596872284971e-07,
+      "loss": 0.1286,
+      "step": 3600
+    },
+    {
+      "epoch": 0.9395360959082617,
+      "grad_norm": 2.03125,
+      "learning_rate": 6.718795250506806e-07,
+      "loss": 0.1314,
+      "step": 3605
+    },
+    {
+      "epoch": 0.9408391972895491,
+      "grad_norm": 1.921875,
+      "learning_rate": 6.573993628728642e-07,
+      "loss": 0.1359,
+      "step": 3610
+    },
+    {
+      "epoch": 0.9421422986708365,
+      "grad_norm": 2.09375,
+      "learning_rate": 6.429192006950478e-07,
+      "loss": 0.1332,
+      "step": 3615
+    },
+    {
+      "epoch": 0.9434454000521241,
+      "grad_norm": 1.9921875,
+      "learning_rate": 6.284390385172315e-07,
+      "loss": 0.1312,
+      "step": 3620
+    },
+    {
+      "epoch": 0.9447485014334115,
+      "grad_norm": 1.921875,
+      "learning_rate": 6.139588763394151e-07,
+      "loss": 0.1248,
+      "step": 3625
+    },
+    {
+      "epoch": 0.946051602814699,
+      "grad_norm": 2.078125,
+      "learning_rate": 5.994787141615987e-07,
+      "loss": 0.133,
+      "step": 3630
+    },
+    {
+      "epoch": 0.9473547041959864,
+      "grad_norm": 1.765625,
+      "learning_rate": 5.849985519837822e-07,
+      "loss": 0.1249,
+      "step": 3635
+    },
+    {
+      "epoch": 0.9486578055772739,
+      "grad_norm": 1.8125,
+      "learning_rate": 5.705183898059659e-07,
+      "loss": 0.1331,
+      "step": 3640
+    },
+    {
+      "epoch": 0.9499609069585614,
+      "grad_norm": 2.015625,
+      "learning_rate": 5.560382276281495e-07,
+      "loss": 0.1279,
+      "step": 3645
+    },
+    {
+      "epoch": 0.9512640083398488,
+      "grad_norm": 2.109375,
+      "learning_rate": 5.415580654503331e-07,
+      "loss": 0.1303,
+      "step": 3650
+    },
+    {
+      "epoch": 0.9525671097211363,
+      "grad_norm": 2.03125,
+      "learning_rate": 5.270779032725168e-07,
+      "loss": 0.1343,
+      "step": 3655
+    },
+    {
+      "epoch": 0.9538702111024238,
+      "grad_norm": 2.15625,
+      "learning_rate": 5.125977410947003e-07,
+      "loss": 0.1325,
+      "step": 3660
+    },
+    {
+      "epoch": 0.9551733124837112,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.981175789168839e-07,
+      "loss": 0.1268,
+      "step": 3665
+    },
+    {
+      "epoch": 0.9564764138649987,
+      "grad_norm": 2.265625,
+      "learning_rate": 4.836374167390675e-07,
+      "loss": 0.1324,
+      "step": 3670
+    },
+    {
+      "epoch": 0.9577795152462861,
+      "grad_norm": 1.859375,
+      "learning_rate": 4.691572545612511e-07,
+      "loss": 0.1284,
+      "step": 3675
+    },
+    {
+      "epoch": 0.9590826166275737,
+      "grad_norm": 2.03125,
+      "learning_rate": 4.5467709238343475e-07,
+      "loss": 0.1338,
+      "step": 3680
+    },
+    {
+      "epoch": 0.9603857180088611,
+      "grad_norm": 1.8984375,
+      "learning_rate": 4.401969302056183e-07,
+      "loss": 0.1334,
+      "step": 3685
+    },
+    {
+      "epoch": 0.9616888193901486,
+      "grad_norm": 1.84375,
+      "learning_rate": 4.2571676802780196e-07,
+      "loss": 0.1302,
+      "step": 3690
+    },
+    {
+      "epoch": 0.962991920771436,
+      "grad_norm": 2.125,
+      "learning_rate": 4.1123660584998553e-07,
+      "loss": 0.1298,
+      "step": 3695
+    },
+    {
+      "epoch": 0.9642950221527234,
+      "grad_norm": 2.234375,
+      "learning_rate": 3.9675644367216916e-07,
+      "loss": 0.1301,
+      "step": 3700
+    },
+    {
+      "epoch": 0.965598123534011,
+      "grad_norm": 2.234375,
+      "learning_rate": 3.822762814943528e-07,
+      "loss": 0.1336,
+      "step": 3705
+    },
+    {
+      "epoch": 0.9669012249152984,
+      "grad_norm": 2.3125,
+      "learning_rate": 3.6779611931653637e-07,
+      "loss": 0.1421,
+      "step": 3710
+    },
+    {
+      "epoch": 0.9682043262965859,
+      "grad_norm": 2.1875,
+      "learning_rate": 3.5331595713872e-07,
+      "loss": 0.1362,
+      "step": 3715
+    },
+    {
+      "epoch": 0.9695074276778733,
+      "grad_norm": 2.34375,
+      "learning_rate": 3.388357949609036e-07,
+      "loss": 0.1281,
+      "step": 3720
+    },
+    {
+      "epoch": 0.9708105290591608,
+      "grad_norm": 2.140625,
+      "learning_rate": 3.243556327830872e-07,
+      "loss": 0.1343,
+      "step": 3725
+    },
+    {
+      "epoch": 0.9721136304404483,
+      "grad_norm": 2.078125,
+      "learning_rate": 3.098754706052708e-07,
+      "loss": 0.1373,
+      "step": 3730
+    },
+    {
+      "epoch": 0.9734167318217357,
+      "grad_norm": 2.921875,
+      "learning_rate": 2.953953084274544e-07,
+      "loss": 0.1359,
+      "step": 3735
+    },
+    {
+      "epoch": 0.9747198332030232,
+      "grad_norm": 2.34375,
+      "learning_rate": 2.8091514624963805e-07,
+      "loss": 0.1399,
+      "step": 3740
+    },
+    {
+      "epoch": 0.9760229345843107,
+      "grad_norm": 1.9921875,
+      "learning_rate": 2.6643498407182163e-07,
+      "loss": 0.1358,
+      "step": 3745
+    },
+    {
+      "epoch": 0.9773260359655981,
+      "grad_norm": 2.296875,
+      "learning_rate": 2.519548218940052e-07,
+      "loss": 0.1323,
+      "step": 3750
+    },
+    {
+      "epoch": 0.9786291373468856,
+      "grad_norm": 1.9140625,
+      "learning_rate": 2.3747465971618884e-07,
+      "loss": 0.1328,
+      "step": 3755
+    },
+    {
+      "epoch": 0.979932238728173,
+      "grad_norm": 2.15625,
+      "learning_rate": 2.2299449753837244e-07,
+      "loss": 0.1325,
+      "step": 3760
+    },
+    {
+      "epoch": 0.9812353401094606,
+      "grad_norm": 2.21875,
+      "learning_rate": 2.0851433536055607e-07,
+      "loss": 0.1329,
+      "step": 3765
+    },
+    {
+      "epoch": 0.982538441490748,
+      "grad_norm": 1.875,
+      "learning_rate": 1.9403417318273968e-07,
+      "loss": 0.1266,
+      "step": 3770
+    },
+    {
+      "epoch": 0.9838415428720354,
+      "grad_norm": 2.296875,
+      "learning_rate": 1.7955401100492328e-07,
+      "loss": 0.1333,
+      "step": 3775
+    },
+    {
+      "epoch": 0.9851446442533229,
+      "grad_norm": 2.03125,
+      "learning_rate": 1.6507384882710686e-07,
+      "loss": 0.1293,
+      "step": 3780
+    },
+    {
+      "epoch": 0.9864477456346104,
+      "grad_norm": 2.078125,
+      "learning_rate": 1.505936866492905e-07,
+      "loss": 0.127,
+      "step": 3785
+    },
+    {
+      "epoch": 0.9877508470158979,
+      "grad_norm": 1.9609375,
+      "learning_rate": 1.361135244714741e-07,
+      "loss": 0.1404,
+      "step": 3790
+    },
+    {
+      "epoch": 0.9890539483971853,
+      "grad_norm": 2.0625,
+      "learning_rate": 1.216333622936577e-07,
+      "loss": 0.1283,
+      "step": 3795
+    },
+    {
+      "epoch": 0.9903570497784727,
+      "grad_norm": 1.984375,
+      "learning_rate": 1.071532001158413e-07,
+      "loss": 0.1372,
+      "step": 3800
+    },
+    {
+      "epoch": 0.9916601511597603,
+      "grad_norm": 2.0,
+      "learning_rate": 9.26730379380249e-08,
+      "loss": 0.1318,
+      "step": 3805
+    },
+    {
+      "epoch": 0.9929632525410477,
+      "grad_norm": 1.921875,
+      "learning_rate": 7.819287576020852e-08,
+      "loss": 0.1273,
+      "step": 3810
+    },
+    {
+      "epoch": 0.9942663539223352,
+      "grad_norm": 1.9921875,
+      "learning_rate": 6.371271358239213e-08,
+      "loss": 0.1377,
+      "step": 3815
+    },
+    {
+      "epoch": 0.9955694553036226,
+      "grad_norm": 2.171875,
+      "learning_rate": 4.923255140457574e-08,
+      "loss": 0.1361,
+      "step": 3820
+    },
+    {
+      "epoch": 0.99687255668491,
+      "grad_norm": 1.953125,
+      "learning_rate": 3.475238922675934e-08,
+      "loss": 0.1349,
+      "step": 3825
+    },
+    {
+      "epoch": 0.9981756580661976,
+      "grad_norm": 1.953125,
+      "learning_rate": 2.027222704894295e-08,
+      "loss": 0.1376,
+      "step": 3830
+    },
+    {
+      "epoch": 0.999478759447485,
+      "grad_norm": 1.890625,
+      "learning_rate": 5.792064871126557e-09,
+      "loss": 0.1353,
+      "step": 3835
+    },
+    {
+      "epoch": 1.0,
+      "step": 3837,
+      "total_flos": 3.632852192307708e+18,
+      "train_loss": 0.14543206988600105,
+      "train_runtime": 4317.1955,
+      "train_samples_per_second": 56.879,
+      "train_steps_per_second": 0.889
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 3837,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 384,
+  "stateful_callbacks": {
+    "EarlyStoppingCallback": {
+      "args": {
+        "early_stopping_patience": 5,
+        "early_stopping_threshold": 0.0
+      },
+      "attributes": {
+        "early_stopping_patience_counter": 0
+      }
+    },
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.632852192307708e+18,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}