diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5251 @@
+{
+  "best_metric": 0.9022757697456493,
+  "best_model_checkpoint": "swin-tiny-patch4-window7-224-finetuned-woody_LeftGR_clean_130epochs/checkpoint-5044",
+  "epoch": 129.9904761904762,
+  "global_step": 6760,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.19,
+      "learning_rate": 7.396449704142012e-07,
+      "loss": 0.712,
+      "step": 10
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.4792899408284024e-06,
+      "loss": 0.6939,
+      "step": 20
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.2189349112426035e-06,
+      "loss": 0.6894,
+      "step": 30
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.9585798816568047e-06,
+      "loss": 0.6701,
+      "step": 40
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 3.6982248520710063e-06,
+      "loss": 0.6569,
+      "step": 50
+    },
+    {
+      "epoch": 0.99,
+      "eval_accuracy": 0.6720214190093708,
+      "eval_loss": 0.6226735711097717,
+      "eval_runtime": 147.5189,
+      "eval_samples_per_second": 5.064,
+      "eval_steps_per_second": 0.163,
+      "step": 52
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 4.437869822485207e-06,
+      "loss": 0.6668,
+      "step": 60
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 5.177514792899409e-06,
+      "loss": 0.6304,
+      "step": 70
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 5.917159763313609e-06,
+      "loss": 0.6201,
+      "step": 80
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 6.656804733727812e-06,
+      "loss": 0.624,
+      "step": 90
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 7.396449704142013e-06,
+      "loss": 0.6069,
+      "step": 100
+    },
+    {
+      "epoch": 1.99,
+      "eval_accuracy": 0.6934404283801874,
+      "eval_loss": 0.5891051888465881,
+      "eval_runtime": 14.2328,
+      "eval_samples_per_second": 52.484,
+      "eval_steps_per_second": 1.686,
+      "step": 104
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 8.136094674556213e-06,
+      "loss": 0.6461,
+      "step": 110
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 8.875739644970414e-06,
+      "loss": 0.6048,
+      "step": 120
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.5978,
+      "step": 130
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 1.0355029585798817e-05,
+      "loss": 0.5923,
+      "step": 140
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 1.1094674556213017e-05,
+      "loss": 0.6044,
+      "step": 150
+    },
+    {
+      "epoch": 2.99,
+      "eval_accuracy": 0.7202141900937081,
+      "eval_loss": 0.5543065667152405,
+      "eval_runtime": 14.1904,
+      "eval_samples_per_second": 52.641,
+      "eval_steps_per_second": 1.691,
+      "step": 156
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 1.1834319526627219e-05,
+      "loss": 0.6134,
+      "step": 160
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 1.257396449704142e-05,
+      "loss": 0.6026,
+      "step": 170
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 1.3313609467455624e-05,
+      "loss": 0.5861,
+      "step": 180
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 1.4053254437869822e-05,
+      "loss": 0.5692,
+      "step": 190
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 1.4792899408284025e-05,
+      "loss": 0.5898,
+      "step": 200
+    },
+    {
+      "epoch": 3.99,
+      "eval_accuracy": 0.7228915662650602,
+      "eval_loss": 0.5439929962158203,
+      "eval_runtime": 14.2818,
+      "eval_samples_per_second": 52.305,
+      "eval_steps_per_second": 1.68,
+      "step": 208
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 1.5532544378698223e-05,
+      "loss": 0.6072,
+      "step": 210
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 1.6272189349112425e-05,
+      "loss": 0.5797,
+      "step": 220
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 1.701183431952663e-05,
+      "loss": 0.5903,
+      "step": 230
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 1.7751479289940828e-05,
+      "loss": 0.5606,
+      "step": 240
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 1.849112426035503e-05,
+      "loss": 0.5876,
+      "step": 250
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 1.923076923076923e-05,
+      "loss": 0.5774,
+      "step": 260
+    },
+    {
+      "epoch": 4.99,
+      "eval_accuracy": 0.7282463186077643,
+      "eval_loss": 0.5360084772109985,
+      "eval_runtime": 14.1927,
+      "eval_samples_per_second": 52.633,
+      "eval_steps_per_second": 1.691,
+      "step": 260
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 1.9970414201183433e-05,
+      "loss": 0.5981,
+      "step": 270
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 2.0710059171597635e-05,
+      "loss": 0.5702,
+      "step": 280
+    },
+    {
+      "epoch": 5.57,
+      "learning_rate": 2.1449704142011836e-05,
+      "loss": 0.568,
+      "step": 290
+    },
+    {
+      "epoch": 5.76,
+      "learning_rate": 2.2189349112426034e-05,
+      "loss": 0.5665,
+      "step": 300
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 2.292899408284024e-05,
+      "loss": 0.5912,
+      "step": 310
+    },
+    {
+      "epoch": 5.99,
+      "eval_accuracy": 0.7188755020080321,
+      "eval_loss": 0.5466371774673462,
+      "eval_runtime": 14.2646,
+      "eval_samples_per_second": 52.368,
+      "eval_steps_per_second": 1.682,
+      "step": 312
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 2.3668639053254438e-05,
+      "loss": 0.5625,
+      "step": 320
+    },
+    {
+      "epoch": 6.34,
+      "learning_rate": 2.440828402366864e-05,
+      "loss": 0.597,
+      "step": 330
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 2.514792899408284e-05,
+      "loss": 0.5738,
+      "step": 340
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 2.5887573964497042e-05,
+      "loss": 0.5549,
+      "step": 350
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 2.6627218934911247e-05,
+      "loss": 0.5685,
+      "step": 360
+    },
+    {
+      "epoch": 6.99,
+      "eval_accuracy": 0.7336010709504686,
+      "eval_loss": 0.5183846354484558,
+      "eval_runtime": 14.2562,
+      "eval_samples_per_second": 52.398,
+      "eval_steps_per_second": 1.683,
+      "step": 364
+    },
+    {
+      "epoch": 7.11,
+      "learning_rate": 2.7366863905325446e-05,
+      "loss": 0.5851,
+      "step": 370
+    },
+    {
+      "epoch": 7.3,
+      "learning_rate": 2.8106508875739644e-05,
+      "loss": 0.5335,
+      "step": 380
+    },
+    {
+      "epoch": 7.5,
+      "learning_rate": 2.8846153846153845e-05,
+      "loss": 0.5746,
+      "step": 390
+    },
+    {
+      "epoch": 7.69,
+      "learning_rate": 2.958579881656805e-05,
+      "loss": 0.5636,
+      "step": 400
+    },
+    {
+      "epoch": 7.88,
+      "learning_rate": 3.032544378698225e-05,
+      "loss": 0.5604,
+      "step": 410
+    },
+    {
+      "epoch": 7.99,
+      "eval_accuracy": 0.7550200803212851,
+      "eval_loss": 0.5138036012649536,
+      "eval_runtime": 14.7813,
+      "eval_samples_per_second": 50.537,
+      "eval_steps_per_second": 1.624,
+      "step": 416
+    },
+    {
+      "epoch": 8.08,
+      "learning_rate": 3.106508875739645e-05,
+      "loss": 0.5641,
+      "step": 420
+    },
+    {
+      "epoch": 8.27,
+      "learning_rate": 3.1804733727810655e-05,
+      "loss": 0.57,
+      "step": 430
+    },
+    {
+      "epoch": 8.46,
+      "learning_rate": 3.254437869822485e-05,
+      "loss": 0.5806,
+      "step": 440
+    },
+    {
+      "epoch": 8.65,
+      "learning_rate": 3.328402366863905e-05,
+      "loss": 0.5778,
+      "step": 450
+    },
+    {
+      "epoch": 8.84,
+      "learning_rate": 3.402366863905326e-05,
+      "loss": 0.5455,
+      "step": 460
+    },
+    {
+      "epoch": 8.99,
+      "eval_accuracy": 0.7376171352074966,
+      "eval_loss": 0.5156762599945068,
+      "eval_runtime": 14.3058,
+      "eval_samples_per_second": 52.217,
+      "eval_steps_per_second": 1.678,
+      "step": 468
+    },
+    {
+      "epoch": 9.04,
+      "learning_rate": 3.4763313609467455e-05,
+      "loss": 0.5755,
+      "step": 470
+    },
+    {
+      "epoch": 9.23,
+      "learning_rate": 3.5502958579881656e-05,
+      "loss": 0.5821,
+      "step": 480
+    },
+    {
+      "epoch": 9.42,
+      "learning_rate": 3.6242603550295865e-05,
+      "loss": 0.5589,
+      "step": 490
+    },
+    {
+      "epoch": 9.61,
+      "learning_rate": 3.698224852071006e-05,
+      "loss": 0.5548,
+      "step": 500
+    },
+    {
+      "epoch": 9.8,
+      "learning_rate": 3.772189349112426e-05,
+      "loss": 0.5345,
+      "step": 510
+    },
+    {
+      "epoch": 9.99,
+      "learning_rate": 3.846153846153846e-05,
+      "loss": 0.5462,
+      "step": 520
+    },
+    {
+      "epoch": 9.99,
+      "eval_accuracy": 0.7657295850066934,
+      "eval_loss": 0.5077853798866272,
+      "eval_runtime": 14.4311,
+      "eval_samples_per_second": 51.763,
+      "eval_steps_per_second": 1.663,
+      "step": 520
+    },
+    {
+      "epoch": 10.19,
+      "learning_rate": 3.9201183431952664e-05,
+      "loss": 0.5747,
+      "step": 530
+    },
+    {
+      "epoch": 10.38,
+      "learning_rate": 3.9940828402366866e-05,
+      "loss": 0.5437,
+      "step": 540
+    },
+    {
+      "epoch": 10.57,
+      "learning_rate": 4.068047337278107e-05,
+      "loss": 0.5404,
+      "step": 550
+    },
+    {
+      "epoch": 10.76,
+      "learning_rate": 4.142011834319527e-05,
+      "loss": 0.5742,
+      "step": 560
+    },
+    {
+      "epoch": 10.95,
+      "learning_rate": 4.215976331360947e-05,
+      "loss": 0.5729,
+      "step": 570
+    },
+    {
+      "epoch": 10.99,
+      "eval_accuracy": 0.7523427041499331,
+      "eval_loss": 0.4956744313240051,
+      "eval_runtime": 14.3076,
+      "eval_samples_per_second": 52.21,
+      "eval_steps_per_second": 1.677,
+      "step": 572
+    },
+    {
+      "epoch": 11.15,
+      "learning_rate": 4.289940828402367e-05,
+      "loss": 0.5899,
+      "step": 580
+    },
+    {
+      "epoch": 11.34,
+      "learning_rate": 4.3639053254437874e-05,
+      "loss": 0.531,
+      "step": 590
+    },
+    {
+      "epoch": 11.53,
+      "learning_rate": 4.437869822485207e-05,
+      "loss": 0.5446,
+      "step": 600
+    },
+    {
+      "epoch": 11.72,
+      "learning_rate": 4.511834319526627e-05,
+      "loss": 0.5302,
+      "step": 610
+    },
+    {
+      "epoch": 11.91,
+      "learning_rate": 4.585798816568048e-05,
+      "loss": 0.5555,
+      "step": 620
+    },
+    {
+      "epoch": 11.99,
+      "eval_accuracy": 0.7563587684069611,
+      "eval_loss": 0.5016053915023804,
+      "eval_runtime": 14.3917,
+      "eval_samples_per_second": 51.905,
+      "eval_steps_per_second": 1.668,
+      "step": 624
+    },
+    {
+      "epoch": 12.11,
+      "learning_rate": 4.6597633136094674e-05,
+      "loss": 0.563,
+      "step": 630
+    },
+    {
+      "epoch": 12.3,
+      "learning_rate": 4.7337278106508875e-05,
+      "loss": 0.5364,
+      "step": 640
+    },
+    {
+      "epoch": 12.5,
+      "learning_rate": 4.8076923076923084e-05,
+      "loss": 0.5284,
+      "step": 650
+    },
+    {
+      "epoch": 12.69,
+      "learning_rate": 4.881656804733728e-05,
+      "loss": 0.5233,
+      "step": 660
+    },
+    {
+      "epoch": 12.88,
+      "learning_rate": 4.955621301775148e-05,
+      "loss": 0.5291,
+      "step": 670
+    },
+    {
+      "epoch": 12.99,
+      "eval_accuracy": 0.7322623828647925,
+      "eval_loss": 0.5664923787117004,
+      "eval_runtime": 14.2349,
+      "eval_samples_per_second": 52.477,
+      "eval_steps_per_second": 1.686,
+      "step": 676
+    },
+    {
+      "epoch": 13.08,
+      "learning_rate": 4.9967126890203814e-05,
+      "loss": 0.5765,
+      "step": 680
+    },
+    {
+      "epoch": 13.27,
+      "learning_rate": 4.988494411571335e-05,
+      "loss": 0.51,
+      "step": 690
+    },
+    {
+      "epoch": 13.46,
+      "learning_rate": 4.980276134122288e-05,
+      "loss": 0.5123,
+      "step": 700
+    },
+    {
+      "epoch": 13.65,
+      "learning_rate": 4.972057856673242e-05,
+      "loss": 0.5359,
+      "step": 710
+    },
+    {
+      "epoch": 13.84,
+      "learning_rate": 4.9638395792241945e-05,
+      "loss": 0.524,
+      "step": 720
+    },
+    {
+      "epoch": 13.99,
+      "eval_accuracy": 0.7389558232931727,
+      "eval_loss": 0.5430611371994019,
+      "eval_runtime": 14.375,
+      "eval_samples_per_second": 51.965,
+      "eval_steps_per_second": 1.67,
+      "step": 728
+    },
+    {
+      "epoch": 14.04,
+      "learning_rate": 4.955621301775148e-05,
+      "loss": 0.5595,
+      "step": 730
+    },
+    {
+      "epoch": 14.23,
+      "learning_rate": 4.9474030243261015e-05,
+      "loss": 0.5417,
+      "step": 740
+    },
+    {
+      "epoch": 14.42,
+      "learning_rate": 4.939184746877055e-05,
+      "loss": 0.5481,
+      "step": 750
+    },
+    {
+      "epoch": 14.61,
+      "learning_rate": 4.9309664694280084e-05,
+      "loss": 0.5408,
+      "step": 760
+    },
+    {
+      "epoch": 14.8,
+      "learning_rate": 4.922748191978961e-05,
+      "loss": 0.5045,
+      "step": 770
+    },
+    {
+      "epoch": 14.99,
+      "learning_rate": 4.9145299145299147e-05,
+      "loss": 0.5194,
+      "step": 780
+    },
+    {
+      "epoch": 14.99,
+      "eval_accuracy": 0.7429718875502008,
+      "eval_loss": 0.5018919110298157,
+      "eval_runtime": 14.3243,
+      "eval_samples_per_second": 52.149,
+      "eval_steps_per_second": 1.675,
+      "step": 780
+    },
+    {
+      "epoch": 15.19,
+      "learning_rate": 4.906311637080868e-05,
+      "loss": 0.5516,
+      "step": 790
+    },
+    {
+      "epoch": 15.38,
+      "learning_rate": 4.898093359631821e-05,
+      "loss": 0.5012,
+      "step": 800
+    },
+    {
+      "epoch": 15.57,
+      "learning_rate": 4.889875082182775e-05,
+      "loss": 0.5493,
+      "step": 810
+    },
+    {
+      "epoch": 15.76,
+      "learning_rate": 4.881656804733728e-05,
+      "loss": 0.5188,
+      "step": 820
+    },
+    {
+      "epoch": 15.95,
+      "learning_rate": 4.873438527284681e-05,
+      "loss": 0.5368,
+      "step": 830
+    },
+    {
+      "epoch": 15.99,
+      "eval_accuracy": 0.7724230254350736,
+      "eval_loss": 0.48103711009025574,
+      "eval_runtime": 14.4508,
+      "eval_samples_per_second": 51.693,
+      "eval_steps_per_second": 1.661,
+      "step": 832
+    },
+    {
+      "epoch": 16.15,
+      "learning_rate": 4.865220249835635e-05,
+      "loss": 0.5409,
+      "step": 840
+    },
+    {
+      "epoch": 16.34,
+      "learning_rate": 4.8570019723865876e-05,
+      "loss": 0.5263,
+      "step": 850
+    },
+    {
+      "epoch": 16.53,
+      "learning_rate": 4.848783694937542e-05,
+      "loss": 0.5,
+      "step": 860
+    },
+    {
+      "epoch": 16.72,
+      "learning_rate": 4.8405654174884945e-05,
+      "loss": 0.5098,
+      "step": 870
+    },
+    {
+      "epoch": 16.91,
+      "learning_rate": 4.832347140039448e-05,
+      "loss": 0.4917,
+      "step": 880
+    },
+    {
+      "epoch": 16.99,
+      "eval_accuracy": 0.7710843373493976,
+      "eval_loss": 0.4792889952659607,
+      "eval_runtime": 14.2838,
+      "eval_samples_per_second": 52.297,
+      "eval_steps_per_second": 1.68,
+      "step": 884
+    },
+    {
+      "epoch": 17.11,
+      "learning_rate": 4.8241288625904014e-05,
+      "loss": 0.5218,
+      "step": 890
+    },
+    {
+      "epoch": 17.3,
+      "learning_rate": 4.815910585141354e-05,
+      "loss": 0.5073,
+      "step": 900
+    },
+    {
+      "epoch": 17.5,
+      "learning_rate": 4.8076923076923084e-05,
+      "loss": 0.4931,
+      "step": 910
+    },
+    {
+      "epoch": 17.69,
+      "learning_rate": 4.799474030243261e-05,
+      "loss": 0.4988,
+      "step": 920
+    },
+    {
+      "epoch": 17.88,
+      "learning_rate": 4.7912557527942146e-05,
+      "loss": 0.4892,
+      "step": 930
+    },
+    {
+      "epoch": 17.99,
+      "eval_accuracy": 0.7630522088353414,
+      "eval_loss": 0.49813756346702576,
+      "eval_runtime": 14.4507,
+      "eval_samples_per_second": 51.693,
+      "eval_steps_per_second": 1.661,
+      "step": 936
+    },
+    {
+      "epoch": 18.08,
+      "learning_rate": 4.783037475345168e-05,
+      "loss": 0.5266,
+      "step": 940
+    },
+    {
+      "epoch": 18.27,
+      "learning_rate": 4.774819197896121e-05,
+      "loss": 0.5022,
+      "step": 950
+    },
+    {
+      "epoch": 18.46,
+      "learning_rate": 4.7666009204470743e-05,
+      "loss": 0.4954,
+      "step": 960
+    },
+    {
+      "epoch": 18.65,
+      "learning_rate": 4.758382642998028e-05,
+      "loss": 0.4796,
+      "step": 970
+    },
+    {
+      "epoch": 18.84,
+      "learning_rate": 4.750164365548981e-05,
+      "loss": 0.5117,
+      "step": 980
+    },
+    {
+      "epoch": 18.99,
+      "eval_accuracy": 0.751004016064257,
+      "eval_loss": 0.49685606360435486,
+      "eval_runtime": 14.3461,
+      "eval_samples_per_second": 52.07,
+      "eval_steps_per_second": 1.673,
+      "step": 988
+    },
+    {
+      "epoch": 19.04,
+      "learning_rate": 4.741946088099935e-05,
+      "loss": 0.5338,
+      "step": 990
+    },
+    {
+      "epoch": 19.23,
+      "learning_rate": 4.7337278106508875e-05,
+      "loss": 0.4975,
+      "step": 1000
+    },
+    {
+      "epoch": 19.42,
+      "learning_rate": 4.725509533201841e-05,
+      "loss": 0.4804,
+      "step": 1010
+    },
+    {
+      "epoch": 19.61,
+      "learning_rate": 4.7172912557527945e-05,
+      "loss": 0.501,
+      "step": 1020
+    },
+    {
+      "epoch": 19.8,
+      "learning_rate": 4.709072978303748e-05,
+      "loss": 0.4773,
+      "step": 1030
+    },
+    {
+      "epoch": 19.99,
+      "learning_rate": 4.700854700854701e-05,
+      "loss": 0.5033,
+      "step": 1040
+    },
+    {
+      "epoch": 19.99,
+      "eval_accuracy": 0.7670682730923695,
+      "eval_loss": 0.4711051285266876,
+      "eval_runtime": 14.405,
+      "eval_samples_per_second": 51.857,
+      "eval_steps_per_second": 1.666,
+      "step": 1040
+    },
+    {
+      "epoch": 20.19,
+      "learning_rate": 4.692636423405654e-05,
+      "loss": 0.4881,
+      "step": 1050
+    },
+    {
+      "epoch": 20.38,
+      "learning_rate": 4.6844181459566076e-05,
+      "loss": 0.4856,
+      "step": 1060
+    },
+    {
+      "epoch": 20.57,
+      "learning_rate": 4.676199868507561e-05,
+      "loss": 0.4976,
+      "step": 1070
+    },
+    {
+      "epoch": 20.76,
+      "learning_rate": 4.6679815910585146e-05,
+      "loss": 0.4921,
+      "step": 1080
+    },
+    {
+      "epoch": 20.95,
+      "learning_rate": 4.6597633136094674e-05,
+      "loss": 0.4807,
+      "step": 1090
+    },
+    {
+      "epoch": 20.99,
+      "eval_accuracy": 0.7724230254350736,
+      "eval_loss": 0.49587923288345337,
+      "eval_runtime": 14.3112,
+      "eval_samples_per_second": 52.197,
+      "eval_steps_per_second": 1.677,
+      "step": 1092
+    },
+    {
+      "epoch": 21.15,
+      "learning_rate": 4.6515450361604215e-05,
+      "loss": 0.5151,
+      "step": 1100
+    },
+    {
+      "epoch": 21.34,
+      "learning_rate": 4.643326758711374e-05,
+      "loss": 0.4682,
+      "step": 1110
+    },
+    {
+      "epoch": 21.53,
+      "learning_rate": 4.635108481262328e-05,
+      "loss": 0.4919,
+      "step": 1120
+    },
+    {
+      "epoch": 21.72,
+      "learning_rate": 4.626890203813281e-05,
+      "loss": 0.4596,
+      "step": 1130
+    },
+    {
+      "epoch": 21.91,
+      "learning_rate": 4.618671926364234e-05,
+      "loss": 0.493,
+      "step": 1140
+    },
+    {
+      "epoch": 21.99,
+      "eval_accuracy": 0.7898259705488622,
+      "eval_loss": 0.45088011026382446,
+      "eval_runtime": 14.2874,
+      "eval_samples_per_second": 52.284,
+      "eval_steps_per_second": 1.68,
+      "step": 1144
+    },
+    {
+      "epoch": 22.11,
+      "learning_rate": 4.610453648915188e-05,
+      "loss": 0.479,
+      "step": 1150
+    },
+    {
+      "epoch": 22.3,
+      "learning_rate": 4.602235371466141e-05,
+      "loss": 0.433,
+      "step": 1160
+    },
+    {
+      "epoch": 22.5,
+      "learning_rate": 4.594017094017094e-05,
+      "loss": 0.4991,
+      "step": 1170
+    },
+    {
+      "epoch": 22.69,
+      "learning_rate": 4.585798816568048e-05,
+      "loss": 0.4586,
+      "step": 1180
+    },
+    {
+      "epoch": 22.88,
+      "learning_rate": 4.577580539119001e-05,
+      "loss": 0.4887,
+      "step": 1190
+    },
+    {
+      "epoch": 22.99,
+      "eval_accuracy": 0.7737617135207496,
+      "eval_loss": 0.4791219234466553,
+      "eval_runtime": 14.4586,
+      "eval_samples_per_second": 51.665,
+      "eval_steps_per_second": 1.66,
+      "step": 1196
+    },
+    {
+      "epoch": 23.08,
+      "learning_rate": 4.569362261669954e-05,
+      "loss": 0.5188,
+      "step": 1200
+    },
+    {
+      "epoch": 23.27,
+      "learning_rate": 4.5611439842209076e-05,
+      "loss": 0.4379,
+      "step": 1210
+    },
+    {
+      "epoch": 23.46,
+      "learning_rate": 4.5529257067718604e-05,
+      "loss": 0.4638,
+      "step": 1220
+    },
+    {
+      "epoch": 23.65,
+      "learning_rate": 4.5447074293228145e-05,
+      "loss": 0.4636,
+      "step": 1230
+    },
+    {
+      "epoch": 23.84,
+      "learning_rate": 4.536489151873767e-05,
+      "loss": 0.4517,
+      "step": 1240
+    },
+    {
+      "epoch": 23.99,
+      "eval_accuracy": 0.7831325301204819,
+      "eval_loss": 0.47215166687965393,
+      "eval_runtime": 14.2374,
+      "eval_samples_per_second": 52.467,
+      "eval_steps_per_second": 1.686,
+      "step": 1248
+    },
+    {
+      "epoch": 24.04,
+      "learning_rate": 4.528270874424721e-05,
+      "loss": 0.4972,
+      "step": 1250
+    },
+    {
+      "epoch": 24.23,
+      "learning_rate": 4.520052596975674e-05,
+      "loss": 0.4576,
+      "step": 1260
+    },
+    {
+      "epoch": 24.42,
+      "learning_rate": 4.511834319526627e-05,
+      "loss": 0.4933,
+      "step": 1270
+    },
+    {
+      "epoch": 24.61,
+      "learning_rate": 4.5036160420775805e-05,
+      "loss": 0.4628,
+      "step": 1280
+    },
+    {
+      "epoch": 24.8,
+      "learning_rate": 4.495397764628534e-05,
+      "loss": 0.4678,
+      "step": 1290
+    },
+    {
+      "epoch": 24.99,
+      "learning_rate": 4.4871794871794874e-05,
+      "loss": 0.4617,
+      "step": 1300
+    },
+    {
+      "epoch": 24.99,
+      "eval_accuracy": 0.7991967871485943,
+      "eval_loss": 0.4344043731689453,
+      "eval_runtime": 14.3383,
+      "eval_samples_per_second": 52.098,
+      "eval_steps_per_second": 1.674,
+      "step": 1300
+    },
+    {
+      "epoch": 25.19,
+      "learning_rate": 4.478961209730441e-05,
+      "loss": 0.4566,
+      "step": 1310
+    },
+    {
+      "epoch": 25.38,
+      "learning_rate": 4.4707429322813944e-05,
+      "loss": 0.4721,
+      "step": 1320
+    },
+    {
+      "epoch": 25.57,
+      "learning_rate": 4.462524654832347e-05,
+      "loss": 0.5009,
+      "step": 1330
+    },
+    {
+      "epoch": 25.76,
+      "learning_rate": 4.4543063773833006e-05,
+      "loss": 0.4524,
+      "step": 1340
+    },
+    {
+      "epoch": 25.95,
+      "learning_rate": 4.446088099934254e-05,
+      "loss": 0.4609,
+      "step": 1350
+    },
+    {
+      "epoch": 25.99,
+      "eval_accuracy": 0.7951807228915663,
+      "eval_loss": 0.4647318124771118,
+      "eval_runtime": 14.4063,
+      "eval_samples_per_second": 51.852,
+      "eval_steps_per_second": 1.666,
+      "step": 1352
+    },
+    {
+      "epoch": 26.15,
+      "learning_rate": 4.437869822485207e-05,
+      "loss": 0.4554,
+      "step": 1360
+    },
+    {
+      "epoch": 26.34,
+      "learning_rate": 4.429651545036161e-05,
+      "loss": 0.4377,
+      "step": 1370
+    },
+    {
+      "epoch": 26.53,
+      "learning_rate": 4.421433267587114e-05,
+      "loss": 0.4439,
+      "step": 1380
+    },
+    {
+      "epoch": 26.72,
+      "learning_rate": 4.413214990138067e-05,
+      "loss": 0.4278,
+      "step": 1390
+    },
+    {
+      "epoch": 26.91,
+      "learning_rate": 4.404996712689021e-05,
+      "loss": 0.4365,
+      "step": 1400
+    },
+    {
+      "epoch": 26.99,
+      "eval_accuracy": 0.7911646586345381,
+      "eval_loss": 0.4459080696105957,
+      "eval_runtime": 14.4688,
+      "eval_samples_per_second": 51.628,
+      "eval_steps_per_second": 1.659,
+      "step": 1404
+    },
+    {
+      "epoch": 27.11,
+      "learning_rate": 4.3967784352399735e-05,
+      "loss": 0.4811,
+      "step": 1410
+    },
+    {
+      "epoch": 27.3,
+      "learning_rate": 4.388560157790928e-05,
+      "loss": 0.4798,
+      "step": 1420
+    },
+    {
+      "epoch": 27.5,
+      "learning_rate": 4.3803418803418805e-05,
+      "loss": 0.4471,
+      "step": 1430
+    },
+    {
+      "epoch": 27.69,
+      "learning_rate": 4.372123602892834e-05,
+      "loss": 0.4232,
+      "step": 1440
+    },
+    {
+      "epoch": 27.88,
+      "learning_rate": 4.3639053254437874e-05,
+      "loss": 0.4515,
+      "step": 1450
+    },
+    {
+      "epoch": 27.99,
+      "eval_accuracy": 0.7643908969210174,
+      "eval_loss": 0.5216541886329651,
+      "eval_runtime": 14.293,
+      "eval_samples_per_second": 52.263,
+      "eval_steps_per_second": 1.679,
+      "step": 1456
+    },
+    {
+      "epoch": 28.08,
+      "learning_rate": 4.35568704799474e-05,
+      "loss": 0.5067,
+      "step": 1460
+    },
+    {
+      "epoch": 28.27,
+      "learning_rate": 4.347468770545694e-05,
+      "loss": 0.4546,
+      "step": 1470
+    },
+    {
+      "epoch": 28.46,
+      "learning_rate": 4.339250493096647e-05,
+      "loss": 0.4077,
+      "step": 1480
+    },
+    {
+      "epoch": 28.65,
+      "learning_rate": 4.3310322156476e-05,
+      "loss": 0.4388,
+      "step": 1490
+    },
+    {
+      "epoch": 28.84,
+      "learning_rate": 4.322813938198554e-05,
+      "loss": 0.4538,
+      "step": 1500
+    },
+    {
+      "epoch": 28.99,
+      "eval_accuracy": 0.8165997322623829,
+      "eval_loss": 0.43745294213294983,
+      "eval_runtime": 14.2411,
+      "eval_samples_per_second": 52.454,
+      "eval_steps_per_second": 1.685,
+      "step": 1508
+    },
+    {
+      "epoch": 29.04,
+      "learning_rate": 4.314595660749507e-05,
+      "loss": 0.4386,
+      "step": 1510
+    },
+    {
+      "epoch": 29.23,
+      "learning_rate": 4.30637738330046e-05,
+      "loss": 0.4062,
+      "step": 1520
+    },
+    {
+      "epoch": 29.42,
+      "learning_rate": 4.298159105851414e-05,
+      "loss": 0.4247,
+      "step": 1530
+    },
+    {
+      "epoch": 29.61,
+      "learning_rate": 4.289940828402367e-05,
+      "loss": 0.3959,
+      "step": 1540
+    },
+    {
+      "epoch": 29.8,
+      "learning_rate": 4.281722550953321e-05,
+      "loss": 0.4228,
+      "step": 1550
+    },
+    {
+      "epoch": 29.99,
+      "learning_rate": 4.2735042735042735e-05,
+      "loss": 0.4371,
+      "step": 1560
+    },
+    {
+      "epoch": 29.99,
+      "eval_accuracy": 0.8005354752342704,
+      "eval_loss": 0.4406176805496216,
+      "eval_runtime": 14.3071,
+      "eval_samples_per_second": 52.212,
+      "eval_steps_per_second": 1.677,
+      "step": 1560
+    },
+    {
+      "epoch": 30.19,
+      "learning_rate": 4.265285996055227e-05,
+      "loss": 0.4524,
+      "step": 1570
+    },
+    {
+      "epoch": 30.38,
+      "learning_rate": 4.2570677186061804e-05,
+      "loss": 0.4256,
+      "step": 1580
+    },
+    {
+      "epoch": 30.57,
+      "learning_rate": 4.248849441157134e-05,
+      "loss": 0.4055,
+      "step": 1590
+    },
+    {
+      "epoch": 30.76,
+      "learning_rate": 4.240631163708087e-05,
+      "loss": 0.3963,
+      "step": 1600
+    },
+    {
+      "epoch": 30.95,
+      "learning_rate": 4.23241288625904e-05,
+      "loss": 0.4228,
+      "step": 1610
+    },
+    {
+      "epoch": 30.99,
+      "eval_accuracy": 0.7911646586345381,
+      "eval_loss": 0.4383251667022705,
+      "eval_runtime": 14.4146,
+      "eval_samples_per_second": 51.823,
+      "eval_steps_per_second": 1.665,
+      "step": 1612
+    },
+    {
+      "epoch": 31.15,
+      "learning_rate": 4.2241946088099936e-05,
+      "loss": 0.4038,
+      "step": 1620
+    },
+    {
+      "epoch": 31.34,
+      "learning_rate": 4.215976331360947e-05,
+      "loss": 0.4107,
+      "step": 1630
+    },
+    {
+      "epoch": 31.53,
+      "learning_rate": 4.2077580539119006e-05,
+      "loss": 0.4107,
+      "step": 1640
+    },
+    {
+      "epoch": 31.72,
+      "learning_rate": 4.1995397764628533e-05,
+      "loss": 0.4373,
+      "step": 1650
+    },
+    {
+      "epoch": 31.91,
+      "learning_rate": 4.191321499013807e-05,
+      "loss": 0.4347,
+      "step": 1660
+    },
+    {
+      "epoch": 31.99,
+      "eval_accuracy": 0.8152610441767069,
+      "eval_loss": 0.4245714843273163,
+      "eval_runtime": 14.4311,
+      "eval_samples_per_second": 51.763,
+      "eval_steps_per_second": 1.663,
+      "step": 1664
+    },
+    {
+      "epoch": 32.11,
+      "learning_rate": 4.18310322156476e-05,
+      "loss": 0.4316,
+      "step": 1670
+    },
+    {
+      "epoch": 32.3,
+      "learning_rate": 4.174884944115714e-05,
+      "loss": 0.4283,
+      "step": 1680
+    },
+    {
+      "epoch": 32.5,
+      "learning_rate": 4.166666666666667e-05,
+      "loss": 0.4287,
+      "step": 1690
+    },
+    {
+      "epoch": 32.69,
+      "learning_rate": 4.15844838921762e-05,
+      "loss": 0.4118,
+      "step": 1700
+    },
+    {
+      "epoch": 32.88,
+      "learning_rate": 4.1502301117685735e-05,
+      "loss": 0.4354,
+      "step": 1710
+    },
+    {
+      "epoch": 32.99,
+      "eval_accuracy": 0.8112449799196787,
+      "eval_loss": 0.4605826437473297,
+      "eval_runtime": 14.4124,
+      "eval_samples_per_second": 51.83,
+      "eval_steps_per_second": 1.665,
+      "step": 1716
+    },
+    {
+      "epoch": 33.08,
+      "learning_rate": 4.142011834319527e-05,
+      "loss": 0.4527,
+      "step": 1720
+    },
+    {
+      "epoch": 33.27,
+      "learning_rate": 4.13379355687048e-05,
+      "loss": 0.3774,
+      "step": 1730
+    },
+    {
+      "epoch": 33.46,
+      "learning_rate": 4.125575279421434e-05,
+      "loss": 0.3807,
+      "step": 1740
+    },
+    {
+      "epoch": 33.65,
+      "learning_rate": 4.1173570019723866e-05,
+      "loss": 0.3782,
+      "step": 1750
+    },
+    {
+      "epoch": 33.84,
+      "learning_rate": 4.10913872452334e-05,
+      "loss": 0.4194,
+      "step": 1760
+    },
+    {
+      "epoch": 33.99,
+      "eval_accuracy": 0.8112449799196787,
+      "eval_loss": 0.4371412694454193,
+      "eval_runtime": 14.3715,
+      "eval_samples_per_second": 51.978,
+      "eval_steps_per_second": 1.67,
+      "step": 1768
+    },
+    {
+      "epoch": 34.04,
+      "learning_rate": 4.1009204470742936e-05,
+      "loss": 0.4226,
+      "step": 1770
+    },
+    {
+      "epoch": 34.23,
+      "learning_rate": 4.0927021696252464e-05,
+      "loss": 0.381,
+      "step": 1780
+    },
+    {
+      "epoch": 34.42,
+      "learning_rate": 4.0844838921762005e-05,
+      "loss": 0.3928,
+      "step": 1790
+    },
+    {
+      "epoch": 34.61,
+      "learning_rate": 4.076265614727153e-05,
+      "loss": 0.3948,
+      "step": 1800
+    },
+    {
+      "epoch": 34.8,
+      "learning_rate": 4.068047337278107e-05,
+      "loss": 0.3834,
+      "step": 1810
+    },
+    {
+      "epoch": 34.99,
+      "learning_rate": 4.05982905982906e-05,
+      "loss": 0.4073,
+      "step": 1820
+    },
+    {
+      "epoch": 34.99,
+      "eval_accuracy": 0.8125836680053548,
+      "eval_loss": 0.44359979033470154,
+      "eval_runtime": 14.5174,
+      "eval_samples_per_second": 51.455,
+      "eval_steps_per_second": 1.653,
+      "step": 1820
+    },
+    {
+      "epoch": 35.19,
+      "learning_rate": 4.051610782380013e-05,
+      "loss": 0.3967,
+      "step": 1830
+    },
+    {
+      "epoch": 35.38,
+      "learning_rate": 4.0433925049309665e-05,
+      "loss": 0.39,
+      "step": 1840
+    },
+    {
+      "epoch": 35.57,
+      "learning_rate": 4.03517422748192e-05,
+      "loss": 0.4094,
+      "step": 1850
+    },
+    {
+      "epoch": 35.76,
+      "learning_rate": 4.0269559500328734e-05,
+      "loss": 0.3782,
+      "step": 1860
+    },
+    {
+      "epoch": 35.95,
+      "learning_rate": 4.018737672583827e-05,
+      "loss": 0.3935,
+      "step": 1870
+    },
+    {
+      "epoch": 35.99,
+      "eval_accuracy": 0.8273092369477911,
+      "eval_loss": 0.4254615604877472,
+      "eval_runtime": 14.4583,
+      "eval_samples_per_second": 51.666,
+      "eval_steps_per_second": 1.66,
+      "step": 1872
+    },
+    {
+      "epoch": 36.15,
+      "learning_rate": 4.01051939513478e-05,
+      "loss": 0.393,
+      "step": 1880
+    },
+    {
+      "epoch": 36.34,
+      "learning_rate": 4.002301117685733e-05,
+      "loss": 0.3581,
+      "step": 1890
+    },
+    {
+      "epoch": 36.53,
+      "learning_rate": 3.9940828402366866e-05,
+      "loss": 0.3688,
+      "step": 1900
+    },
+    {
+      "epoch": 36.72,
+      "learning_rate": 3.98586456278764e-05,
+      "loss": 0.3789,
+      "step": 1910
+    },
+    {
+      "epoch": 36.91,
+      "learning_rate": 3.977646285338593e-05,
+      "loss": 0.3862,
+      "step": 1920
+    },
+    {
+      "epoch": 36.99,
+      "eval_accuracy": 0.8232931726907631,
+      "eval_loss": 0.40541553497314453,
+      "eval_runtime": 14.2377,
+      "eval_samples_per_second": 52.466,
+      "eval_steps_per_second": 1.686,
+      "step": 1924
+    },
+    {
+      "epoch": 37.11,
+      "learning_rate": 3.969428007889546e-05,
+      "loss": 0.3843,
+      "step": 1930
+    },
+    {
+      "epoch": 37.3,
+      "learning_rate": 3.9612097304405e-05,
+      "loss": 0.372,
+      "step": 1940
+    },
+    {
+      "epoch": 37.5,
+      "learning_rate": 3.952991452991453e-05,
+      "loss": 0.3925,
+      "step": 1950
+    },
+    {
+      "epoch": 37.69,
+      "learning_rate": 3.944773175542407e-05,
+      "loss": 0.4041,
+      "step": 1960
+    },
+    {
+      "epoch": 37.88,
+      "learning_rate": 3.9365548980933595e-05,
+      "loss": 0.3739,
+      "step": 1970
+    },
+    {
+      "epoch": 37.99,
+      "eval_accuracy": 0.8125836680053548,
+      "eval_loss": 0.4205840229988098,
+      "eval_runtime": 14.3208,
+      "eval_samples_per_second": 52.162,
+      "eval_steps_per_second": 1.676,
+      "step": 1976
+    },
+    {
+      "epoch": 38.08,
+      "learning_rate": 3.9283366206443137e-05,
+      "loss": 0.4033,
+      "step": 1980
+    },
+    {
+      "epoch": 38.27,
+      "learning_rate": 3.9201183431952664e-05,
+      "loss": 0.3604,
+      "step": 1990
+    },
+    {
+      "epoch": 38.46,
+      "learning_rate": 3.91190006574622e-05,
+      "loss": 0.3727,
+      "step": 2000
+    },
+    {
+      "epoch": 38.65,
+      "learning_rate": 3.9036817882971734e-05,
+      "loss": 0.3636,
+      "step": 2010
+    },
+    {
+      "epoch": 38.84,
+      "learning_rate": 3.895463510848126e-05,
+      "loss": 0.3794,
+      "step": 2020
+    },
+    {
+      "epoch": 38.99,
+      "eval_accuracy": 0.821954484605087,
+      "eval_loss": 0.40752872824668884,
+      "eval_runtime": 14.3125,
+      "eval_samples_per_second": 52.192,
+      "eval_steps_per_second": 1.677,
+      "step": 2028
+    },
+    {
+      "epoch": 39.04,
+      "learning_rate": 3.88724523339908e-05,
+      "loss": 0.3886,
+      "step": 2030
+    },
+    {
+      "epoch": 39.23,
+      "learning_rate": 3.879026955950033e-05,
+      "loss": 0.3606,
+      "step": 2040
+    },
+    {
+      "epoch": 39.42,
+      "learning_rate": 3.870808678500986e-05,
+      "loss": 0.3748,
+      "step": 2050
+    },
+    {
+      "epoch": 39.61,
+      "learning_rate": 3.86259040105194e-05,
+      "loss": 0.3653,
+      "step": 2060
+    },
+    {
+      "epoch": 39.8,
+      "learning_rate": 3.854372123602893e-05,
+      "loss": 0.3797,
+      "step": 2070
+    },
+    {
+      "epoch": 39.99,
+      "learning_rate": 3.846153846153846e-05,
+      "loss": 0.3713,
+      "step": 2080
+    },
+    {
+      "epoch": 39.99,
+      "eval_accuracy": 0.8353413654618473,
+      "eval_loss": 0.37869057059288025,
+      "eval_runtime": 14.2945,
+      "eval_samples_per_second": 52.258,
+      "eval_steps_per_second": 1.679,
+      "step": 2080
+    },
+    {
+      "epoch": 40.19,
+      "learning_rate": 3.8379355687048e-05,
+      "loss": 0.356,
+      "step": 2090
+    },
+    {
+      "epoch": 40.38,
+      "learning_rate": 3.8297172912557525e-05,
+      "loss": 0.3248,
+      "step": 2100
+    },
+    {
+      "epoch": 40.57,
+      "learning_rate": 3.821499013806707e-05,
+      "loss": 0.3429,
+      "step": 2110
+    },
+    {
+      "epoch": 40.76,
+      "learning_rate": 3.8132807363576595e-05,
+      "loss": 0.3698,
+      "step": 2120
+    },
+    {
+      "epoch": 40.95,
+      "learning_rate": 3.805062458908613e-05,
+      "loss": 0.3901,
+      "step": 2130
+    },
+    {
+      "epoch": 40.99,
+      "eval_accuracy": 0.8246318607764391,
+      "eval_loss": 0.38403695821762085,
+      "eval_runtime": 14.2775,
+      "eval_samples_per_second": 52.32,
+      "eval_steps_per_second": 1.681,
+      "step": 2132
+    },
+    {
+      "epoch": 41.15,
+      "learning_rate": 3.7968441814595664e-05,
+      "loss": 0.3539,
+      "step": 2140
+    },
+    {
+      "epoch": 41.34,
+      "learning_rate": 3.788625904010519e-05,
+      "loss": 0.3521,
+      "step": 2150
+    },
+    {
+      "epoch": 41.53,
+      "learning_rate": 3.7804076265614727e-05,
+      "loss": 0.3476,
+      "step": 2160
+    },
+    {
+      "epoch": 41.72,
+      "learning_rate": 3.772189349112426e-05,
+      "loss": 0.3323,
+      "step": 2170
+    },
+    {
+      "epoch": 41.91,
+      "learning_rate": 3.7639710716633796e-05,
+      "loss": 0.3514,
+      "step": 2180
+    },
+    {
+      "epoch": 41.99,
+      "eval_accuracy": 0.8366800535475234,
+      "eval_loss": 0.41364696621894836,
+      "eval_runtime": 14.4321,
+      "eval_samples_per_second": 51.76,
+      "eval_steps_per_second": 1.663,
+      "step": 2184
+    },
+    {
+      "epoch": 42.11,
+      "learning_rate": 3.755752794214333e-05,
+      "loss": 0.3471,
+      "step": 2190
+    },
+    {
+      "epoch": 42.3,
+      "learning_rate": 3.7475345167652865e-05,
+      "loss": 0.3523,
+      "step": 2200
+    },
+    {
+      "epoch": 42.5,
+      "learning_rate": 3.739316239316239e-05,
+      "loss": 0.3514,
+      "step": 2210
+    },
+    {
+      "epoch": 42.69,
+      "learning_rate": 3.731097961867193e-05,
+      "loss": 0.3154,
+      "step": 2220
+    },
+    {
+      "epoch": 42.88,
+      "learning_rate": 3.722879684418146e-05,
+      "loss": 0.3718,
+      "step": 2230
+    },
+    {
+      "epoch": 42.99,
+      "eval_accuracy": 0.8393574297188755,
+      "eval_loss": 0.38667845726013184,
+      "eval_runtime": 14.3098,
+      "eval_samples_per_second": 52.202,
+      "eval_steps_per_second": 1.677,
+      "step": 2236
+    },
+    {
+      "epoch": 43.08,
+      "learning_rate": 3.714661406969099e-05,
+      "loss": 0.3761,
+      "step": 2240
+    },
+    {
+      "epoch": 43.27,
+      "learning_rate": 3.706443129520053e-05,
+      "loss": 0.3561,
+      "step": 2250
+    },
+    {
+      "epoch": 43.46,
+      "learning_rate": 3.698224852071006e-05,
+      "loss": 0.3463,
+      "step": 2260
+    },
+    {
+      "epoch": 43.65,
+      "learning_rate": 3.6900065746219594e-05,
+      "loss": 0.3357,
+      "step": 2270
+    },
+    {
+      "epoch": 43.84,
+      "learning_rate": 3.681788297172913e-05,
+      "loss": 0.3699,
+      "step": 2280
+    },
+    {
+      "epoch": 43.99,
+      "eval_accuracy": 0.8487282463186078,
+      "eval_loss": 0.3736788034439087,
+      "eval_runtime": 14.3125,
+      "eval_samples_per_second": 52.192,
+      "eval_steps_per_second": 1.677,
+      "step": 2288
+    },
+    {
+      "epoch": 44.04,
+      "learning_rate": 3.673570019723866e-05,
+      "loss": 0.3349,
+      "step": 2290
+    },
+    {
+      "epoch": 44.23,
+      "learning_rate": 3.66535174227482e-05,
+      "loss": 0.3276,
+      "step": 2300
+    },
+    {
+      "epoch": 44.42,
+      "learning_rate": 3.6571334648257726e-05,
+      "loss": 0.3015,
+      "step": 2310
+    },
+    {
+      "epoch": 44.61,
+      "learning_rate": 3.648915187376726e-05,
+      "loss": 0.3534,
+      "step": 2320
+    },
+    {
+      "epoch": 44.8,
+      "learning_rate": 3.6406969099276796e-05,
+      "loss": 0.3393,
+      "step": 2330
+    },
+    {
+      "epoch": 44.99,
+      "learning_rate": 3.6324786324786323e-05,
+      "loss": 0.3314,
+      "step": 2340
+    },
+    {
+      "epoch": 44.99,
+      "eval_accuracy": 0.8527443105756358,
+      "eval_loss": 0.3756033778190613,
+      "eval_runtime": 14.4135,
+      "eval_samples_per_second": 51.826,
+      "eval_steps_per_second": 1.665,
+      "step": 2340
+    },
+    {
+      "epoch": 45.19,
+      "learning_rate": 3.6242603550295865e-05,
+      "loss": 0.3474,
+      "step": 2350
+    },
+    {
+      "epoch": 45.38,
+      "learning_rate": 3.616042077580539e-05,
+      "loss": 0.3122,
+      "step": 2360
+    },
+    {
+      "epoch": 45.57,
+      "learning_rate": 3.607823800131492e-05,
+      "loss": 0.3145,
+      "step": 2370
+    },
+    {
+      "epoch": 45.76,
+      "learning_rate": 3.599605522682446e-05,
+      "loss": 0.3414,
+      "step": 2380
+    },
+    {
+      "epoch": 45.95,
+      "learning_rate": 3.591387245233399e-05,
+      "loss": 0.3167,
+      "step": 2390
+    },
+    {
+      "epoch": 45.99,
+      "eval_accuracy": 0.8473895582329317,
+      "eval_loss": 0.42111995816230774,
+      "eval_runtime": 14.3243,
+      "eval_samples_per_second": 52.149,
+      "eval_steps_per_second": 1.675,
+      "step": 2392
+    },
+    {
+      "epoch": 46.15,
+      "learning_rate": 3.5831689677843525e-05,
+      "loss": 0.3121,
+      "step": 2400
+    },
+    {
+      "epoch": 46.34,
+      "learning_rate": 3.574950690335306e-05,
+      "loss": 0.2994,
+      "step": 2410
+    },
+    {
+      "epoch": 46.53,
+      "learning_rate": 3.5667324128862594e-05,
+      "loss": 0.3104,
+      "step": 2420
+    },
+    {
+      "epoch": 46.72,
+      "learning_rate": 3.558514135437213e-05,
+      "loss": 0.3121,
+      "step": 2430
+    },
+    {
+      "epoch": 46.91,
+      "learning_rate": 3.5502958579881656e-05,
+      "loss": 0.301,
+      "step": 2440
+    },
+    {
+      "epoch": 46.99,
+      "eval_accuracy": 0.8433734939759037,
+      "eval_loss": 0.3869791030883789,
+      "eval_runtime": 14.4243,
+      "eval_samples_per_second": 51.788,
+      "eval_steps_per_second": 1.664,
+      "step": 2444
+    },
+    {
+      "epoch": 47.11,
+      "learning_rate": 3.542077580539119e-05,
+      "loss": 0.3623,
+      "step": 2450
+    },
+    {
+      "epoch": 47.3,
+      "learning_rate": 3.5338593030900726e-05,
+      "loss": 0.3325,
+      "step": 2460
+    },
+    {
+      "epoch": 47.5,
+      "learning_rate": 3.525641025641026e-05,
+      "loss": 0.3222,
+      "step": 2470
+    },
+    {
+      "epoch": 47.69,
+      "learning_rate": 3.517422748191979e-05,
+      "loss": 0.3344,
+      "step": 2480
+    },
+    {
+      "epoch": 47.88,
+      "learning_rate": 3.509204470742932e-05,
+      "loss": 0.3048,
+      "step": 2490
+    },
+    {
+      "epoch": 47.99,
+      "eval_accuracy": 0.8460508701472557,
+      "eval_loss": 0.42355024814605713,
+      "eval_runtime": 14.3302,
+      "eval_samples_per_second": 52.128,
+      "eval_steps_per_second": 1.675,
+      "step": 2496
+    },
+    {
+      "epoch": 48.08,
+      "learning_rate": 3.500986193293886e-05,
+      "loss": 0.3252,
+      "step": 2500
+    },
+    {
+      "epoch": 48.27,
+      "learning_rate": 3.492767915844839e-05,
+      "loss": 0.3043,
+      "step": 2510
+    },
+    {
+      "epoch": 48.46,
+      "learning_rate": 3.484549638395793e-05,
+      "loss": 0.3061,
+      "step": 2520
+    },
+    {
+      "epoch": 48.65,
+      "learning_rate": 3.4763313609467455e-05,
+      "loss": 0.3165,
+      "step": 2530
+    },
+    {
+      "epoch": 48.84,
+      "learning_rate": 3.468113083497699e-05,
+      "loss": 0.2735,
+      "step": 2540
+    },
+    {
+      "epoch": 48.99,
+      "eval_accuracy": 0.8380187416331994,
+      "eval_loss": 0.41224488615989685,
+      "eval_runtime": 14.374,
+      "eval_samples_per_second": 51.969,
+      "eval_steps_per_second": 1.67,
+      "step": 2548
+    },
+    {
+      "epoch": 49.04,
+      "learning_rate": 3.4598948060486524e-05,
+      "loss": 0.3231,
+      "step": 2550
+    },
+    {
+      "epoch": 49.23,
+      "learning_rate": 3.451676528599606e-05,
+      "loss": 0.2713,
+      "step": 2560
+    },
+    {
+      "epoch": 49.42,
+      "learning_rate": 3.4434582511505594e-05,
+      "loss": 0.3097,
+      "step": 2570
+    },
+    {
+      "epoch": 49.61,
+      "learning_rate": 3.435239973701512e-05,
+      "loss": 0.2974,
+      "step": 2580
+    },
+    {
+      "epoch": 49.8,
+      "learning_rate": 3.4270216962524656e-05,
+      "loss": 0.2918,
+      "step": 2590
+    },
+    {
+      "epoch": 49.99,
+      "learning_rate": 3.418803418803419e-05,
+      "loss": 0.3003,
+      "step": 2600
+    },
+    {
+      "epoch": 49.99,
+      "eval_accuracy": 0.856760374832664,
+      "eval_loss": 0.3608503043651581,
+      "eval_runtime": 14.447,
+      "eval_samples_per_second": 51.706,
+      "eval_steps_per_second": 1.661,
+      "step": 2600
+    },
+    {
+      "epoch": 50.19,
+      "learning_rate": 3.410585141354372e-05,
+      "loss": 0.2885,
+      "step": 2610
+    },
+    {
+      "epoch": 50.38,
+      "learning_rate": 3.402366863905326e-05,
+      "loss": 0.3132,
+      "step": 2620
+    },
+    {
+      "epoch": 50.57,
+      "learning_rate": 3.394148586456279e-05,
+      "loss": 0.307,
+      "step": 2630
+    },
+    {
+      "epoch": 50.76,
+      "learning_rate": 3.385930309007232e-05,
+      "loss": 0.2998,
+      "step": 2640
+    },
+    {
+      "epoch": 50.95,
+      "learning_rate": 3.377712031558186e-05,
+      "loss": 0.3147,
+      "step": 2650
+    },
+    {
+      "epoch": 50.99,
+      "eval_accuracy": 0.8366800535475234,
+      "eval_loss": 0.42577147483825684,
+      "eval_runtime": 14.3525,
+      "eval_samples_per_second": 52.047,
+      "eval_steps_per_second": 1.672,
+      "step": 2652
+    },
+    {
+      "epoch": 51.15,
+      "learning_rate": 3.3694937541091385e-05,
+      "loss": 0.3026,
+      "step": 2660
+    },
+    {
+      "epoch": 51.34,
+      "learning_rate": 3.3612754766600927e-05,
+      "loss": 0.3149,
+      "step": 2670
+    },
+    {
+      "epoch": 51.53,
+      "learning_rate": 3.3530571992110454e-05,
+      "loss": 0.2929,
+      "step": 2680
+    },
+    {
+      "epoch": 51.72,
+      "learning_rate": 3.344838921761999e-05,
+      "loss": 0.2882,
+      "step": 2690
+    },
+    {
+      "epoch": 51.91,
+      "learning_rate": 3.3366206443129524e-05,
+      "loss": 0.288,
+      "step": 2700
+    },
+    {
+      "epoch": 51.99,
+      "eval_accuracy": 0.8393574297188755,
+      "eval_loss": 0.38547390699386597,
+      "eval_runtime": 14.452,
+      "eval_samples_per_second": 51.688,
+      "eval_steps_per_second": 1.661,
+      "step": 2704
+    },
+    {
+      "epoch": 52.11,
+      "learning_rate": 3.328402366863905e-05,
+      "loss": 0.306,
+      "step": 2710
+    },
+    {
+      "epoch": 52.3,
+      "learning_rate": 3.3201840894148586e-05,
+      "loss": 0.278,
+      "step": 2720
+    },
+    {
+      "epoch": 52.5,
+      "learning_rate": 3.311965811965812e-05,
+      "loss": 0.2647,
+      "step": 2730
+    },
+    {
+      "epoch": 52.69,
+      "learning_rate": 3.3037475345167656e-05,
+      "loss": 0.2867,
+      "step": 2740
+    },
+    {
+      "epoch": 52.88,
+      "learning_rate": 3.295529257067719e-05,
+      "loss": 0.2895,
+      "step": 2750
+    },
+    {
+      "epoch": 52.99,
+      "eval_accuracy": 0.8527443105756358,
+      "eval_loss": 0.35428228974342346,
+      "eval_runtime": 14.3955,
+      "eval_samples_per_second": 51.891,
+      "eval_steps_per_second": 1.667,
+      "step": 2756
+    },
+    {
+      "epoch": 53.08,
+      "learning_rate": 3.287310979618672e-05,
+      "loss": 0.2792,
+      "step": 2760
+    },
+    {
+      "epoch": 53.27,
+      "learning_rate": 3.279092702169625e-05,
+      "loss": 0.2956,
+      "step": 2770
+    },
+    {
+      "epoch": 53.46,
+      "learning_rate": 3.270874424720579e-05,
+      "loss": 0.2937,
+      "step": 2780
+    },
+    {
+      "epoch": 53.65,
+      "learning_rate": 3.262656147271532e-05,
+      "loss": 0.2926,
+      "step": 2790
+    },
+    {
+      "epoch": 53.84,
+      "learning_rate": 3.254437869822485e-05,
+      "loss": 0.2685,
+      "step": 2800
+    },
+    {
+      "epoch": 53.99,
+      "eval_accuracy": 0.8540829986613119,
+      "eval_loss": 0.3668169677257538,
+      "eval_runtime": 14.32,
+      "eval_samples_per_second": 52.165,
+      "eval_steps_per_second": 1.676,
+      "step": 2808
+    },
+    {
+      "epoch": 54.04,
+      "learning_rate": 3.2462195923734385e-05,
+      "loss": 0.2803,
+      "step": 2810
+    },
+    {
+      "epoch": 54.23,
+      "learning_rate": 3.238001314924392e-05,
+      "loss": 0.2769,
+      "step": 2820
+    },
+    {
+      "epoch": 54.42,
+      "learning_rate": 3.2297830374753454e-05,
+      "loss": 0.282,
+      "step": 2830
+    },
+    {
+      "epoch": 54.61,
+      "learning_rate": 3.221564760026299e-05,
+      "loss": 0.2794,
+      "step": 2840
+    },
+    {
+      "epoch": 54.8,
+      "learning_rate": 3.2133464825772517e-05,
+      "loss": 0.2785,
+      "step": 2850
+    },
+    {
+      "epoch": 54.99,
+      "learning_rate": 3.205128205128206e-05,
+      "loss": 0.2931,
+      "step": 2860
+    },
+    {
+      "epoch": 54.99,
+      "eval_accuracy": 0.8540829986613119,
+      "eval_loss": 0.3564589023590088,
+      "eval_runtime": 14.6649,
+      "eval_samples_per_second": 50.938,
+      "eval_steps_per_second": 1.637,
+      "step": 2860
+    },
+    {
+      "epoch": 55.19,
+      "learning_rate": 3.1969099276791586e-05,
+      "loss": 0.2864,
+      "step": 2870
+    },
+    {
+      "epoch": 55.38,
+      "learning_rate": 3.188691650230112e-05,
+      "loss": 0.2737,
+      "step": 2880
+    },
+    {
+      "epoch": 55.57,
+      "learning_rate": 3.1804733727810655e-05,
+      "loss": 0.2673,
+      "step": 2890
+    },
+    {
+      "epoch": 55.76,
+      "learning_rate": 3.172255095332018e-05,
+      "loss": 0.2553,
+      "step": 2900
+    },
+    {
+      "epoch": 55.95,
+      "learning_rate": 3.1640368178829725e-05,
+      "loss": 0.2966,
+      "step": 2910
+    },
+    {
+      "epoch": 55.99,
+      "eval_accuracy": 0.856760374832664,
+      "eval_loss": 0.39847540855407715,
+      "eval_runtime": 14.3559,
+      "eval_samples_per_second": 52.034,
+      "eval_steps_per_second": 1.672,
+      "step": 2912
+    },
+    {
+      "epoch": 56.15,
+      "learning_rate": 3.155818540433925e-05,
+      "loss": 0.2765,
+      "step": 2920
+    },
+    {
+      "epoch": 56.34,
+      "learning_rate": 3.147600262984878e-05,
+      "loss": 0.2776,
+      "step": 2930
+    },
+    {
+      "epoch": 56.53,
+      "learning_rate": 3.139381985535832e-05,
+      "loss": 0.2676,
+      "step": 2940
+    },
+    {
+      "epoch": 56.72,
+      "learning_rate": 3.131163708086785e-05,
+      "loss": 0.263,
+      "step": 2950
+    },
+    {
+      "epoch": 56.91,
+      "learning_rate": 3.1229454306377384e-05,
+      "loss": 0.2737,
+      "step": 2960
+    },
+    {
+      "epoch": 56.99,
+      "eval_accuracy": 0.85809906291834,
+      "eval_loss": 0.4099937677383423,
+      "eval_runtime": 14.2828,
+      "eval_samples_per_second": 52.301,
+      "eval_steps_per_second": 1.68,
+      "step": 2964
+    },
+    {
+      "epoch": 57.11,
+      "learning_rate": 3.114727153188692e-05,
+      "loss": 0.2919,
+      "step": 2970
+    },
+    {
+      "epoch": 57.3,
+      "learning_rate": 3.106508875739645e-05,
+      "loss": 0.2548,
+      "step": 2980
+    },
+    {
+      "epoch": 57.5,
+      "learning_rate": 3.098290598290599e-05,
+      "loss": 0.254,
+      "step": 2990
+    },
+    {
+      "epoch": 57.69,
+      "learning_rate": 3.0900723208415516e-05,
+      "loss": 0.2778,
+      "step": 3000
+    },
+    {
+      "epoch": 57.88,
+      "learning_rate": 3.081854043392505e-05,
+      "loss": 0.2892,
+      "step": 3010
+    },
+    {
+      "epoch": 57.99,
+      "eval_accuracy": 0.8768406961178046,
+      "eval_loss": 0.34804263710975647,
+      "eval_runtime": 14.3118,
+      "eval_samples_per_second": 52.195,
+      "eval_steps_per_second": 1.677,
+      "step": 3016
+    },
+    {
+      "epoch": 58.08,
+      "learning_rate": 3.0736357659434586e-05,
+      "loss": 0.2615,
+      "step": 3020
+    },
+    {
+      "epoch": 58.27,
+      "learning_rate": 3.0654174884944113e-05,
+      "loss": 0.2705,
+      "step": 3030
+    },
+    {
+      "epoch": 58.46,
+      "learning_rate": 3.057199211045365e-05,
+      "loss": 0.2645,
+      "step": 3040
+    },
+    {
+      "epoch": 58.65,
+      "learning_rate": 3.0489809335963183e-05,
+      "loss": 0.2458,
+      "step": 3050
+    },
+    {
+      "epoch": 58.84,
+      "learning_rate": 3.0407626561472717e-05,
+      "loss": 0.2753,
+      "step": 3060
+    },
+    {
+      "epoch": 58.99,
+      "eval_accuracy": 0.8661311914323963,
+      "eval_loss": 0.3725629150867462,
+      "eval_runtime": 14.3684,
+      "eval_samples_per_second": 51.989,
+      "eval_steps_per_second": 1.67,
+      "step": 3068
+    },
+    {
+      "epoch": 59.04,
+      "learning_rate": 3.032544378698225e-05,
+      "loss": 0.2897,
+      "step": 3070
+    },
+    {
+      "epoch": 59.23,
+      "learning_rate": 3.0243261012491787e-05,
+      "loss": 0.2479,
+      "step": 3080
+    },
+    {
+      "epoch": 59.42,
+      "learning_rate": 3.0161078238001318e-05,
+      "loss": 0.2633,
+      "step": 3090
+    },
+    {
+      "epoch": 59.61,
+      "learning_rate": 3.007889546351085e-05,
+      "loss": 0.2519,
+      "step": 3100
+    },
+    {
+      "epoch": 59.8,
+      "learning_rate": 2.9996712689020384e-05,
+      "loss": 0.249,
+      "step": 3110
+    },
+    {
+      "epoch": 59.99,
+      "learning_rate": 2.9914529914529915e-05,
+      "loss": 0.2831,
+      "step": 3120
+    },
+    {
+      "epoch": 59.99,
+      "eval_accuracy": 0.8634538152610441,
+      "eval_loss": 0.3981005549430847,
+      "eval_runtime": 14.2903,
+      "eval_samples_per_second": 52.273,
+      "eval_steps_per_second": 1.679,
+      "step": 3120
+    },
+    {
+      "epoch": 60.19,
+      "learning_rate": 2.9832347140039453e-05,
+      "loss": 0.2496,
+      "step": 3130
+    },
+    {
+      "epoch": 60.38,
+      "learning_rate": 2.975016436554898e-05,
+      "loss": 0.2567,
+      "step": 3140
+    },
+    {
+      "epoch": 60.57,
+      "learning_rate": 2.9667981591058512e-05,
+      "loss": 0.2429,
+      "step": 3150
+    },
+    {
+      "epoch": 60.76,
+      "learning_rate": 2.958579881656805e-05,
+      "loss": 0.2487,
+      "step": 3160
+    },
+    {
+      "epoch": 60.95,
+      "learning_rate": 2.9503616042077582e-05,
+      "loss": 0.261,
+      "step": 3170
+    },
+    {
+      "epoch": 60.99,
+      "eval_accuracy": 0.8634538152610441,
+      "eval_loss": 0.4217076301574707,
+      "eval_runtime": 14.4646,
+      "eval_samples_per_second": 51.643,
+      "eval_steps_per_second": 1.659,
+      "step": 3172
+    },
+    {
+      "epoch": 61.15,
+      "learning_rate": 2.9421433267587116e-05,
+      "loss": 0.2444,
+      "step": 3180
+    },
+    {
+      "epoch": 61.34,
+      "learning_rate": 2.9339250493096648e-05,
+      "loss": 0.2665,
+      "step": 3190
+    },
+    {
+      "epoch": 61.53,
+      "learning_rate": 2.925706771860618e-05,
+      "loss": 0.2708,
+      "step": 3200
+    },
+    {
+      "epoch": 61.72,
+      "learning_rate": 2.9174884944115717e-05,
+      "loss": 0.2532,
+      "step": 3210
+    },
+    {
+      "epoch": 61.91,
+      "learning_rate": 2.9092702169625248e-05,
+      "loss": 0.2662,
+      "step": 3220
+    },
+    {
+      "epoch": 61.99,
+      "eval_accuracy": 0.8728246318607764,
+      "eval_loss": 0.3516206741333008,
+      "eval_runtime": 14.3663,
+      "eval_samples_per_second": 51.997,
+      "eval_steps_per_second": 1.671,
+      "step": 3224
+    },
+    {
+      "epoch": 62.11,
+      "learning_rate": 2.9010519395134783e-05,
+      "loss": 0.2558,
+      "step": 3230
+    },
+    {
+      "epoch": 62.3,
+      "learning_rate": 2.8928336620644314e-05,
+      "loss": 0.238,
+      "step": 3240
+    },
+    {
+      "epoch": 62.5,
+      "learning_rate": 2.8846153846153845e-05,
+      "loss": 0.2623,
+      "step": 3250
+    },
+    {
+      "epoch": 62.69,
+      "learning_rate": 2.876397107166338e-05,
+      "loss": 0.2469,
+      "step": 3260
+    },
+    {
+      "epoch": 62.88,
+      "learning_rate": 2.868178829717291e-05,
+      "loss": 0.2464,
+      "step": 3270
+    },
+    {
+      "epoch": 62.99,
+      "eval_accuracy": 0.8647925033467202,
+      "eval_loss": 0.38206276297569275,
+      "eval_runtime": 14.38,
+      "eval_samples_per_second": 51.947,
+      "eval_steps_per_second": 1.669,
+      "step": 3276
+    },
+    {
+      "epoch": 63.08,
+      "learning_rate": 2.859960552268245e-05,
+      "loss": 0.2376,
+      "step": 3280
+    },
+    {
+      "epoch": 63.27,
+      "learning_rate": 2.851742274819198e-05,
+      "loss": 0.2485,
+      "step": 3290
+    },
+    {
+      "epoch": 63.46,
+      "learning_rate": 2.8435239973701515e-05,
+      "loss": 0.2183,
+      "step": 3300
+    },
+    {
+      "epoch": 63.65,
+      "learning_rate": 2.8353057199211047e-05,
+      "loss": 0.2656,
+      "step": 3310
+    },
+    {
+      "epoch": 63.84,
+      "learning_rate": 2.8270874424720578e-05,
+      "loss": 0.256,
+      "step": 3320
+    },
+    {
+      "epoch": 63.99,
+      "eval_accuracy": 0.8688085676037484,
+      "eval_loss": 0.3970297574996948,
+      "eval_runtime": 14.4009,
+      "eval_samples_per_second": 51.872,
+      "eval_steps_per_second": 1.667,
+      "step": 3328
+    },
+    {
+      "epoch": 64.04,
+      "learning_rate": 2.8188691650230116e-05,
+      "loss": 0.2246,
+      "step": 3330
+    },
+    {
+      "epoch": 64.23,
+      "learning_rate": 2.8106508875739644e-05,
+      "loss": 0.2161,
+      "step": 3340
+    },
+    {
+      "epoch": 64.42,
+      "learning_rate": 2.8024326101249182e-05,
+      "loss": 0.2288,
+      "step": 3350
+    },
+    {
+      "epoch": 64.61,
+      "learning_rate": 2.7942143326758713e-05,
+      "loss": 0.2379,
+      "step": 3360
+    },
+    {
+      "epoch": 64.8,
+      "learning_rate": 2.7859960552268244e-05,
+      "loss": 0.218,
+      "step": 3370
+    },
+    {
+      "epoch": 64.99,
+      "learning_rate": 2.777777777777778e-05,
+      "loss": 0.2755,
+      "step": 3380
+    },
+    {
+      "epoch": 64.99,
+      "eval_accuracy": 0.8540829986613119,
+      "eval_loss": 0.4765224754810333,
+      "eval_runtime": 14.4281,
+      "eval_samples_per_second": 51.774,
+      "eval_steps_per_second": 1.663,
+      "step": 3380
+    },
+    {
+      "epoch": 65.19,
+      "learning_rate": 2.769559500328731e-05,
+      "loss": 0.2374,
+      "step": 3390
+    },
+    {
+      "epoch": 65.38,
+      "learning_rate": 2.761341222879685e-05,
+      "loss": 0.2671,
+      "step": 3400
+    },
+    {
+      "epoch": 65.57,
+      "learning_rate": 2.753122945430638e-05,
+      "loss": 0.2532,
+      "step": 3410
+    },
+    {
+      "epoch": 65.76,
+      "learning_rate": 2.744904667981591e-05,
+      "loss": 0.2281,
+      "step": 3420
+    },
+    {
+      "epoch": 65.95,
+      "learning_rate": 2.7366863905325446e-05,
+      "loss": 0.2339,
+      "step": 3430
+    },
+    {
+      "epoch": 65.99,
+      "eval_accuracy": 0.8540829986613119,
+      "eval_loss": 0.5616469979286194,
+      "eval_runtime": 14.3532,
+      "eval_samples_per_second": 52.044,
+      "eval_steps_per_second": 1.672,
+      "step": 3432
+    },
+    {
+      "epoch": 66.15,
+      "learning_rate": 2.7284681130834977e-05,
+      "loss": 0.2344,
+      "step": 3440
+    },
+    {
+      "epoch": 66.34,
+      "learning_rate": 2.7202498356344515e-05,
+      "loss": 0.2564,
+      "step": 3450
+    },
+    {
+      "epoch": 66.53,
+      "learning_rate": 2.7120315581854043e-05,
+      "loss": 0.2479,
+      "step": 3460
+    },
+    {
+      "epoch": 66.72,
+      "learning_rate": 2.7038132807363574e-05,
+      "loss": 0.2426,
+      "step": 3470
+    },
+    {
+      "epoch": 66.91,
+      "learning_rate": 2.6955950032873112e-05,
+      "loss": 0.2344,
+      "step": 3480
+    },
+    {
+      "epoch": 66.99,
+      "eval_accuracy": 0.8647925033467202,
+      "eval_loss": 0.38865524530410767,
+      "eval_runtime": 14.4504,
+      "eval_samples_per_second": 51.694,
+      "eval_steps_per_second": 1.661,
+      "step": 3484
+    },
+    {
+      "epoch": 67.11,
+      "learning_rate": 2.6873767258382643e-05,
+      "loss": 0.231,
+      "step": 3490
+    },
+    {
+      "epoch": 67.3,
+      "learning_rate": 2.6791584483892178e-05,
+      "loss": 0.214,
+      "step": 3500
+    },
+    {
+      "epoch": 67.5,
+      "learning_rate": 2.670940170940171e-05,
+      "loss": 0.2251,
+      "step": 3510
+    },
+    {
+      "epoch": 67.69,
+      "learning_rate": 2.6627218934911247e-05,
+      "loss": 0.2116,
+      "step": 3520
+    },
+    {
+      "epoch": 67.88,
+      "learning_rate": 2.654503616042078e-05,
+      "loss": 0.1995,
+      "step": 3530
+    },
+    {
+      "epoch": 67.99,
+      "eval_accuracy": 0.8674698795180723,
+      "eval_loss": 0.44000309705734253,
+      "eval_runtime": 14.4518,
+      "eval_samples_per_second": 51.689,
+      "eval_steps_per_second": 1.661,
+      "step": 3536
+    },
+    {
+      "epoch": 68.08,
+      "learning_rate": 2.646285338593031e-05,
+      "loss": 0.2246,
+      "step": 3540
+    },
+    {
+      "epoch": 68.27,
+      "learning_rate": 2.6380670611439845e-05,
+      "loss": 0.2298,
+      "step": 3550
+    },
+    {
+      "epoch": 68.46,
+      "learning_rate": 2.6298487836949376e-05,
+      "loss": 0.2148,
+      "step": 3560
+    },
+    {
+      "epoch": 68.65,
+      "learning_rate": 2.6216305062458914e-05,
+      "loss": 0.212,
+      "step": 3570
+    },
+    {
+      "epoch": 68.84,
+      "learning_rate": 2.6134122287968442e-05,
+      "loss": 0.2297,
+      "step": 3580
+    },
+    {
+      "epoch": 68.99,
+      "eval_accuracy": 0.8688085676037484,
+      "eval_loss": 0.4290466606616974,
+      "eval_runtime": 14.4921,
+      "eval_samples_per_second": 51.545,
+      "eval_steps_per_second": 1.656,
+      "step": 3588
+    },
+    {
+      "epoch": 69.04,
+      "learning_rate": 2.6051939513477973e-05,
+      "loss": 0.2219,
+      "step": 3590
+    },
+    {
+      "epoch": 69.23,
+      "learning_rate": 2.596975673898751e-05,
+      "loss": 0.2417,
+      "step": 3600
+    },
+    {
+      "epoch": 69.42,
+      "learning_rate": 2.5887573964497042e-05,
+      "loss": 0.2328,
+      "step": 3610
+    },
+    {
+      "epoch": 69.61,
+      "learning_rate": 2.5805391190006577e-05,
+      "loss": 0.2376,
+      "step": 3620
+    },
+    {
+      "epoch": 69.8,
+      "learning_rate": 2.572320841551611e-05,
+      "loss": 0.2056,
+      "step": 3630
+    },
+    {
+      "epoch": 69.99,
+      "learning_rate": 2.564102564102564e-05,
+      "loss": 0.227,
+      "step": 3640
+    },
+    {
+      "epoch": 69.99,
+      "eval_accuracy": 0.8701472556894244,
+      "eval_loss": 0.4521399736404419,
+      "eval_runtime": 14.3262,
+      "eval_samples_per_second": 52.142,
+      "eval_steps_per_second": 1.675,
+      "step": 3640
+    },
+    {
+      "epoch": 70.19,
+      "learning_rate": 2.5558842866535178e-05,
+      "loss": 0.2167,
+      "step": 3650
+    },
+    {
+      "epoch": 70.38,
+      "learning_rate": 2.547666009204471e-05,
+      "loss": 0.2373,
+      "step": 3660
+    },
+    {
+      "epoch": 70.57,
+      "learning_rate": 2.5394477317554244e-05,
+      "loss": 0.2247,
+      "step": 3670
+    },
+    {
+      "epoch": 70.76,
+      "learning_rate": 2.5312294543063775e-05,
+      "loss": 0.2233,
+      "step": 3680
+    },
+    {
+      "epoch": 70.95,
+      "learning_rate": 2.5230111768573306e-05,
+      "loss": 0.2084,
+      "step": 3690
+    },
+    {
+      "epoch": 70.99,
+      "eval_accuracy": 0.8781793842034806,
+      "eval_loss": 0.38549551367759705,
+      "eval_runtime": 14.3394,
+      "eval_samples_per_second": 52.094,
+      "eval_steps_per_second": 1.674,
+      "step": 3692
+    },
+    {
+      "epoch": 71.15,
+      "learning_rate": 2.514792899408284e-05,
+      "loss": 0.2336,
+      "step": 3700
+    },
+    {
+      "epoch": 71.34,
+      "learning_rate": 2.5065746219592372e-05,
+      "loss": 0.2179,
+      "step": 3710
+    },
+    {
+      "epoch": 71.53,
+      "learning_rate": 2.4983563445101907e-05,
+      "loss": 0.196,
+      "step": 3720
+    },
+    {
+      "epoch": 71.72,
+      "learning_rate": 2.490138067061144e-05,
+      "loss": 0.2402,
+      "step": 3730
+    },
+    {
+      "epoch": 71.91,
+      "learning_rate": 2.4819197896120973e-05,
+      "loss": 0.2225,
+      "step": 3740
+    },
+    {
+      "epoch": 71.99,
+      "eval_accuracy": 0.8741633199464525,
+      "eval_loss": 0.4200989603996277,
+      "eval_runtime": 14.3806,
+      "eval_samples_per_second": 51.945,
+      "eval_steps_per_second": 1.669,
+      "step": 3744
+    },
+    {
+      "epoch": 72.11,
+      "learning_rate": 2.4737015121630507e-05,
+      "loss": 0.2211,
+      "step": 3750
+    },
+    {
+      "epoch": 72.3,
+      "learning_rate": 2.4654832347140042e-05,
+      "loss": 0.2082,
+      "step": 3760
+    },
+    {
+      "epoch": 72.5,
+      "learning_rate": 2.4572649572649573e-05,
+      "loss": 0.2283,
+      "step": 3770
+    },
+    {
+      "epoch": 72.69,
+      "learning_rate": 2.4490466798159105e-05,
+      "loss": 0.1981,
+      "step": 3780
+    },
+    {
+      "epoch": 72.88,
+      "learning_rate": 2.440828402366864e-05,
+      "loss": 0.1897,
+      "step": 3790
+    },
+    {
+      "epoch": 72.99,
+      "eval_accuracy": 0.8500669344042838,
+      "eval_loss": 0.5137650966644287,
+      "eval_runtime": 14.5819,
+      "eval_samples_per_second": 51.228,
+      "eval_steps_per_second": 1.646,
+      "step": 3796
+    },
+    {
+      "epoch": 73.08,
+      "learning_rate": 2.4326101249178174e-05,
+      "loss": 0.2377,
+      "step": 3800
+    },
+    {
+      "epoch": 73.27,
+      "learning_rate": 2.424391847468771e-05,
+      "loss": 0.244,
+      "step": 3810
+    },
+    {
+      "epoch": 73.46,
+      "learning_rate": 2.416173570019724e-05,
+      "loss": 0.214,
+      "step": 3820
+    },
+    {
+      "epoch": 73.65,
+      "learning_rate": 2.407955292570677e-05,
+      "loss": 0.2246,
+      "step": 3830
+    },
+    {
+      "epoch": 73.84,
+      "learning_rate": 2.3997370151216306e-05,
+      "loss": 0.2136,
+      "step": 3840
+    },
+    {
+      "epoch": 73.99,
+      "eval_accuracy": 0.8848728246318608,
+      "eval_loss": 0.4110630750656128,
+      "eval_runtime": 14.3865,
+      "eval_samples_per_second": 51.924,
+      "eval_steps_per_second": 1.668,
+      "step": 3848
+    },
+    {
+      "epoch": 74.04,
+      "learning_rate": 2.391518737672584e-05,
+      "loss": 0.2235,
+      "step": 3850
+    },
+    {
+      "epoch": 74.23,
+      "learning_rate": 2.3833004602235372e-05,
+      "loss": 0.2022,
+      "step": 3860
+    },
+    {
+      "epoch": 74.42,
+      "learning_rate": 2.3750821827744906e-05,
+      "loss": 0.2024,
+      "step": 3870
+    },
+    {
+      "epoch": 74.61,
+      "learning_rate": 2.3668639053254438e-05,
+      "loss": 0.2087,
+      "step": 3880
+    },
+    {
+      "epoch": 74.8,
+      "learning_rate": 2.3586456278763972e-05,
+      "loss": 0.2099,
+      "step": 3890
+    },
+    {
+      "epoch": 74.99,
+      "learning_rate": 2.3504273504273504e-05,
+      "loss": 0.2155,
+      "step": 3900
+    },
+    {
+      "epoch": 74.99,
+      "eval_accuracy": 0.8862115127175368,
+      "eval_loss": 0.3800347149372101,
+      "eval_runtime": 14.444,
+      "eval_samples_per_second": 51.717,
+      "eval_steps_per_second": 1.662,
+      "step": 3900
+    },
+    {
+      "epoch": 75.19,
+      "learning_rate": 2.3422090729783038e-05,
+      "loss": 0.2004,
+      "step": 3910
+    },
+    {
+      "epoch": 75.38,
+      "learning_rate": 2.3339907955292573e-05,
+      "loss": 0.1997,
+      "step": 3920
+    },
+    {
+      "epoch": 75.57,
+      "learning_rate": 2.3257725180802108e-05,
+      "loss": 0.2044,
+      "step": 3930
+    },
+    {
+      "epoch": 75.76,
+      "learning_rate": 2.317554240631164e-05,
+      "loss": 0.228,
+      "step": 3940
+    },
+    {
+      "epoch": 75.95,
+      "learning_rate": 2.309335963182117e-05,
+      "loss": 0.2338,
+      "step": 3950
+    },
+    {
+      "epoch": 75.99,
+      "eval_accuracy": 0.8835341365461847,
+      "eval_loss": 0.4013587236404419,
+      "eval_runtime": 14.294,
+      "eval_samples_per_second": 52.26,
+      "eval_steps_per_second": 1.679,
+      "step": 3952
+    },
+    {
+      "epoch": 76.15,
+      "learning_rate": 2.3011176857330705e-05,
+      "loss": 0.2232,
+      "step": 3960
+    },
+    {
+      "epoch": 76.34,
+      "learning_rate": 2.292899408284024e-05,
+      "loss": 0.2088,
+      "step": 3970
+    },
+    {
+      "epoch": 76.53,
+      "learning_rate": 2.284681130834977e-05,
+      "loss": 0.1995,
+      "step": 3980
+    },
+    {
+      "epoch": 76.72,
+      "learning_rate": 2.2764628533859302e-05,
+      "loss": 0.2123,
+      "step": 3990
+    },
+    {
+      "epoch": 76.91,
+      "learning_rate": 2.2682445759368837e-05,
+      "loss": 0.2021,
+      "step": 4000
+    },
+    {
+      "epoch": 76.99,
+      "eval_accuracy": 0.892904953145917,
+      "eval_loss": 0.4213900566101074,
+      "eval_runtime": 14.3872,
+      "eval_samples_per_second": 51.921,
+      "eval_steps_per_second": 1.668,
+      "step": 4004
+    },
+    {
+      "epoch": 77.11,
+      "learning_rate": 2.260026298487837e-05,
+      "loss": 0.2131,
+      "step": 4010
+    },
+    {
+      "epoch": 77.3,
+      "learning_rate": 2.2518080210387903e-05,
+      "loss": 0.1928,
+      "step": 4020
+    },
+    {
+      "epoch": 77.5,
+      "learning_rate": 2.2435897435897437e-05,
+      "loss": 0.2112,
+      "step": 4030
+    },
+    {
+      "epoch": 77.69,
+      "learning_rate": 2.2353714661406972e-05,
+      "loss": 0.221,
+      "step": 4040
+    },
+    {
+      "epoch": 77.88,
+      "learning_rate": 2.2271531886916503e-05,
+      "loss": 0.2028,
+      "step": 4050
+    },
+    {
+      "epoch": 77.99,
+      "eval_accuracy": 0.8795180722891566,
+      "eval_loss": 0.3996668756008148,
+      "eval_runtime": 14.4474,
+      "eval_samples_per_second": 51.705,
+      "eval_steps_per_second": 1.661,
+      "step": 4056
+    },
+    {
+      "epoch": 78.08,
+      "learning_rate": 2.2189349112426034e-05,
+      "loss": 0.2217,
+      "step": 4060
+    },
+    {
+      "epoch": 78.27,
+      "learning_rate": 2.210716633793557e-05,
+      "loss": 0.2028,
+      "step": 4070
+    },
+    {
+      "epoch": 78.46,
+      "learning_rate": 2.2024983563445104e-05,
+      "loss": 0.2083,
+      "step": 4080
+    },
+    {
+      "epoch": 78.65,
+      "learning_rate": 2.194280078895464e-05,
+      "loss": 0.2049,
+      "step": 4090
+    },
+    {
+      "epoch": 78.84,
+      "learning_rate": 2.186061801446417e-05,
+      "loss": 0.2162,
+      "step": 4100
+    },
+    {
+      "epoch": 78.99,
+      "eval_accuracy": 0.8781793842034806,
+      "eval_loss": 0.4911276400089264,
+      "eval_runtime": 14.3841,
+      "eval_samples_per_second": 51.932,
+      "eval_steps_per_second": 1.669,
+      "step": 4108
+    },
+    {
+      "epoch": 79.04,
+      "learning_rate": 2.17784352399737e-05,
+      "loss": 0.2189,
+      "step": 4110
+    },
+    {
+      "epoch": 79.23,
+      "learning_rate": 2.1696252465483236e-05,
+      "loss": 0.1966,
+      "step": 4120
+    },
+    {
+      "epoch": 79.42,
+      "learning_rate": 2.161406969099277e-05,
+      "loss": 0.185,
+      "step": 4130
+    },
+    {
+      "epoch": 79.61,
+      "learning_rate": 2.15318869165023e-05,
+      "loss": 0.2232,
+      "step": 4140
+    },
+    {
+      "epoch": 79.8,
+      "learning_rate": 2.1449704142011836e-05,
+      "loss": 0.2279,
+      "step": 4150
+    },
+    {
+      "epoch": 79.99,
+      "learning_rate": 2.1367521367521368e-05,
+      "loss": 0.1889,
+      "step": 4160
+    },
+    {
+      "epoch": 79.99,
+      "eval_accuracy": 0.8701472556894244,
+      "eval_loss": 0.46514207124710083,
+      "eval_runtime": 14.5902,
+      "eval_samples_per_second": 51.199,
+      "eval_steps_per_second": 1.645,
+      "step": 4160
+    },
+    {
+      "epoch": 80.19,
+      "learning_rate": 2.1285338593030902e-05,
+      "loss": 0.1741,
+      "step": 4170
+    },
+    {
+      "epoch": 80.38,
+      "learning_rate": 2.1203155818540433e-05,
+      "loss": 0.1774,
+      "step": 4180
+    },
+    {
+      "epoch": 80.57,
+      "learning_rate": 2.1120973044049968e-05,
+      "loss": 0.1965,
+      "step": 4190
+    },
+    {
+      "epoch": 80.76,
+      "learning_rate": 2.1038790269559503e-05,
+      "loss": 0.2063,
+      "step": 4200
+    },
+    {
+      "epoch": 80.95,
+      "learning_rate": 2.0956607495069034e-05,
+      "loss": 0.2056,
+      "step": 4210
+    },
+    {
+      "epoch": 80.99,
+      "eval_accuracy": 0.8862115127175368,
+      "eval_loss": 0.41560399532318115,
+      "eval_runtime": 14.3901,
+      "eval_samples_per_second": 51.911,
+      "eval_steps_per_second": 1.668,
+      "step": 4212
+    },
+    {
+      "epoch": 81.15,
+      "learning_rate": 2.087442472057857e-05,
+      "loss": 0.2066,
+      "step": 4220
+    },
+    {
+      "epoch": 81.34,
+      "learning_rate": 2.07922419460881e-05,
+      "loss": 0.2227,
+      "step": 4230
+    },
+    {
+      "epoch": 81.53,
+      "learning_rate": 2.0710059171597635e-05,
+      "loss": 0.1966,
+      "step": 4240
+    },
+    {
+      "epoch": 81.72,
+      "learning_rate": 2.062787639710717e-05,
+      "loss": 0.2094,
+      "step": 4250
+    },
+    {
+      "epoch": 81.91,
+      "learning_rate": 2.05456936226167e-05,
+      "loss": 0.206,
+      "step": 4260
+    },
+    {
+      "epoch": 81.99,
+      "eval_accuracy": 0.8741633199464525,
+      "eval_loss": 0.43300920724868774,
+      "eval_runtime": 14.3191,
+      "eval_samples_per_second": 52.168,
+      "eval_steps_per_second": 1.676,
+      "step": 4264
+    },
+    {
+      "epoch": 82.11,
+      "learning_rate": 2.0463510848126232e-05,
+      "loss": 0.1885,
+      "step": 4270
+    },
+    {
+      "epoch": 82.3,
+      "learning_rate": 2.0381328073635766e-05,
+      "loss": 0.187,
+      "step": 4280
+    },
+    {
+      "epoch": 82.5,
+      "learning_rate": 2.02991452991453e-05,
+      "loss": 0.1819,
+      "step": 4290
+    },
+    {
+      "epoch": 82.69,
+      "learning_rate": 2.0216962524654832e-05,
+      "loss": 0.1761,
+      "step": 4300
+    },
+    {
+      "epoch": 82.88,
+      "learning_rate": 2.0134779750164367e-05,
+      "loss": 0.1919,
+      "step": 4310
+    },
+    {
+      "epoch": 82.99,
+      "eval_accuracy": 0.8955823293172691,
+      "eval_loss": 0.4198724031448364,
+      "eval_runtime": 14.4916,
+      "eval_samples_per_second": 51.547,
+      "eval_steps_per_second": 1.656,
+      "step": 4316
+    },
+    {
+      "epoch": 83.08,
+      "learning_rate": 2.00525969756739e-05,
+      "loss": 0.2072,
+      "step": 4320
+    },
+    {
+      "epoch": 83.27,
+      "learning_rate": 1.9970414201183433e-05,
+      "loss": 0.1948,
+      "step": 4330
+    },
+    {
+      "epoch": 83.46,
+      "learning_rate": 1.9888231426692964e-05,
+      "loss": 0.2043,
+      "step": 4340
+    },
+    {
+      "epoch": 83.65,
+      "learning_rate": 1.98060486522025e-05,
+      "loss": 0.186,
+      "step": 4350
+    },
+    {
+      "epoch": 83.84,
+      "learning_rate": 1.9723865877712034e-05,
+      "loss": 0.1967,
+      "step": 4360
+    },
+    {
+      "epoch": 83.99,
+      "eval_accuracy": 0.8821954484605087,
+      "eval_loss": 0.4615110754966736,
+      "eval_runtime": 14.4489,
+      "eval_samples_per_second": 51.7,
+      "eval_steps_per_second": 1.661,
+      "step": 4368
+    },
+    {
+      "epoch": 84.04,
+      "learning_rate": 1.9641683103221568e-05,
+      "loss": 0.1988,
+      "step": 4370
+    },
+    {
+      "epoch": 84.23,
+      "learning_rate": 1.95595003287311e-05,
+      "loss": 0.1745,
+      "step": 4380
+    },
+    {
+      "epoch": 84.42,
+      "learning_rate": 1.947731755424063e-05,
+      "loss": 0.1667,
+      "step": 4390
+    },
+    {
+      "epoch": 84.61,
+      "learning_rate": 1.9395134779750165e-05,
+      "loss": 0.1913,
+      "step": 4400
+    },
+    {
+      "epoch": 84.8,
+      "learning_rate": 1.93129520052597e-05,
+      "loss": 0.1929,
+      "step": 4410
+    },
+    {
+      "epoch": 84.99,
+      "learning_rate": 1.923076923076923e-05,
+      "loss": 0.2083,
+      "step": 4420
+    },
+    {
+      "epoch": 84.99,
+      "eval_accuracy": 0.8714859437751004,
+      "eval_loss": 0.45853391289711,
+      "eval_runtime": 14.4145,
+      "eval_samples_per_second": 51.823,
+      "eval_steps_per_second": 1.665,
+      "step": 4420
+    },
+    {
+      "epoch": 85.19,
+      "learning_rate": 1.9148586456278763e-05,
+      "loss": 0.1994,
+      "step": 4430
+    },
+    {
+      "epoch": 85.38,
+      "learning_rate": 1.9066403681788297e-05,
+      "loss": 0.1789,
+      "step": 4440
+    },
+    {
+      "epoch": 85.57,
+      "learning_rate": 1.8984220907297832e-05,
+      "loss": 0.1914,
+      "step": 4450
+    },
+    {
+      "epoch": 85.76,
+      "learning_rate": 1.8902038132807363e-05,
+      "loss": 0.2083,
+      "step": 4460
+    },
+    {
+      "epoch": 85.95,
+      "learning_rate": 1.8819855358316898e-05,
+      "loss": 0.1888,
+      "step": 4470
+    },
+    {
+      "epoch": 85.99,
+      "eval_accuracy": 0.8728246318607764,
+      "eval_loss": 0.5747922658920288,
+      "eval_runtime": 14.4726,
+      "eval_samples_per_second": 51.615,
+      "eval_steps_per_second": 1.658,
+      "step": 4472
+    },
+    {
+      "epoch": 86.15,
+      "learning_rate": 1.8737672583826433e-05,
+      "loss": 0.1921,
+      "step": 4480
+    },
+    {
+      "epoch": 86.34,
+      "learning_rate": 1.8655489809335964e-05,
+      "loss": 0.2068,
+      "step": 4490
+    },
+    {
+      "epoch": 86.53,
+      "learning_rate": 1.8573307034845495e-05,
+      "loss": 0.1789,
+      "step": 4500
+    },
+    {
+      "epoch": 86.72,
+      "learning_rate": 1.849112426035503e-05,
+      "loss": 0.1988,
+      "step": 4510
+    },
+    {
+      "epoch": 86.91,
+      "learning_rate": 1.8408941485864564e-05,
+      "loss": 0.1744,
+      "step": 4520
+    },
+    {
+      "epoch": 86.99,
+      "eval_accuracy": 0.8902275769745649,
+      "eval_loss": 0.44582998752593994,
+      "eval_runtime": 14.3306,
+      "eval_samples_per_second": 52.126,
+      "eval_steps_per_second": 1.675,
+      "step": 4524
+    },
+    {
+      "epoch": 87.11,
+      "learning_rate": 1.83267587113741e-05,
+      "loss": 0.1924,
+      "step": 4530
+    },
+    {
+      "epoch": 87.3,
+      "learning_rate": 1.824457593688363e-05,
+      "loss": 0.181,
+      "step": 4540
+    },
+    {
+      "epoch": 87.5,
+      "learning_rate": 1.8162393162393162e-05,
+      "loss": 0.1751,
+      "step": 4550
+    },
+    {
+      "epoch": 87.69,
+      "learning_rate": 1.8080210387902696e-05,
+      "loss": 0.1869,
+      "step": 4560
+    },
+    {
+      "epoch": 87.88,
+      "learning_rate": 1.799802761341223e-05,
+      "loss": 0.1789,
+      "step": 4570
+    },
+    {
+      "epoch": 87.99,
+      "eval_accuracy": 0.8688085676037484,
+      "eval_loss": 0.48577430844306946,
+      "eval_runtime": 14.4246,
+      "eval_samples_per_second": 51.787,
+      "eval_steps_per_second": 1.664,
+      "step": 4576
+    },
+    {
+      "epoch": 88.08,
+      "learning_rate": 1.7915844838921762e-05,
+      "loss": 0.1847,
+      "step": 4580
+    },
+    {
+      "epoch": 88.27,
+      "learning_rate": 1.7833662064431297e-05,
+      "loss": 0.1704,
+      "step": 4590
+    },
+    {
+      "epoch": 88.46,
+      "learning_rate": 1.7751479289940828e-05,
+      "loss": 0.1856,
+      "step": 4600
+    },
+    {
+      "epoch": 88.65,
+      "learning_rate": 1.7669296515450363e-05,
+      "loss": 0.1853,
+      "step": 4610
+    },
+    {
+      "epoch": 88.84,
+      "learning_rate": 1.7587113740959894e-05,
+      "loss": 0.1992,
+      "step": 4620
+    },
+    {
+      "epoch": 88.99,
+      "eval_accuracy": 0.8714859437751004,
+      "eval_loss": 0.5018042325973511,
+      "eval_runtime": 14.3941,
+      "eval_samples_per_second": 51.896,
+      "eval_steps_per_second": 1.667,
+      "step": 4628
+    },
+    {
+      "epoch": 89.04,
+      "learning_rate": 1.750493096646943e-05,
+      "loss": 0.1952,
+      "step": 4630
+    },
+    {
+      "epoch": 89.23,
+      "learning_rate": 1.7422748191978963e-05,
+      "loss": 0.1749,
+      "step": 4640
+    },
+    {
+      "epoch": 89.42,
+      "learning_rate": 1.7340565417488495e-05,
+      "loss": 0.1816,
+      "step": 4650
+    },
+    {
+      "epoch": 89.61,
+      "learning_rate": 1.725838264299803e-05,
+      "loss": 0.1763,
+      "step": 4660
+    },
+    {
+      "epoch": 89.8,
+      "learning_rate": 1.717619986850756e-05,
+      "loss": 0.1953,
+      "step": 4670
+    },
+    {
+      "epoch": 89.99,
+      "learning_rate": 1.7094017094017095e-05,
+      "loss": 0.1742,
+      "step": 4680
+    },
+    {
+      "epoch": 89.99,
+      "eval_accuracy": 0.8755020080321285,
+      "eval_loss": 0.5065856575965881,
+      "eval_runtime": 14.4975,
+      "eval_samples_per_second": 51.526,
+      "eval_steps_per_second": 1.655,
+      "step": 4680
+    },
+    {
+      "epoch": 90.19,
+      "learning_rate": 1.701183431952663e-05,
+      "loss": 0.1871,
+      "step": 4690
+    },
+    {
+      "epoch": 90.38,
+      "learning_rate": 1.692965154503616e-05,
+      "loss": 0.1718,
+      "step": 4700
+    },
+    {
+      "epoch": 90.57,
+      "learning_rate": 1.6847468770545693e-05,
+      "loss": 0.1746,
+      "step": 4710
+    },
+    {
+      "epoch": 90.76,
+      "learning_rate": 1.6765285996055227e-05,
+      "loss": 0.1809,
+      "step": 4720
+    },
+    {
+      "epoch": 90.95,
+      "learning_rate": 1.6683103221564762e-05,
+      "loss": 0.1822,
+      "step": 4730
+    },
+    {
+      "epoch": 90.99,
+      "eval_accuracy": 0.892904953145917,
+      "eval_loss": 0.4268772304058075,
+      "eval_runtime": 14.4608,
+      "eval_samples_per_second": 51.657,
+      "eval_steps_per_second": 1.66,
+      "step": 4732
+    },
+    {
+      "epoch": 91.15,
+      "learning_rate": 1.6600920447074293e-05,
+      "loss": 0.1968,
+      "step": 4740
+    },
+    {
+      "epoch": 91.34,
+      "learning_rate": 1.6518737672583828e-05,
+      "loss": 0.1653,
+      "step": 4750
+    },
+    {
+      "epoch": 91.53,
+      "learning_rate": 1.643655489809336e-05,
+      "loss": 0.1712,
+      "step": 4760
+    },
+    {
+      "epoch": 91.72,
+      "learning_rate": 1.6354372123602894e-05,
+      "loss": 0.178,
+      "step": 4770
+    },
+    {
+      "epoch": 91.91,
+      "learning_rate": 1.6272189349112425e-05,
+      "loss": 0.1883,
+      "step": 4780
+    },
+    {
+      "epoch": 91.99,
+      "eval_accuracy": 0.8795180722891566,
+      "eval_loss": 0.45498567819595337,
+      "eval_runtime": 14.3398,
+      "eval_samples_per_second": 52.093,
+      "eval_steps_per_second": 1.674,
+      "step": 4784
+    },
+    {
+      "epoch": 92.11,
+      "learning_rate": 1.619000657462196e-05,
+      "loss": 0.1778,
+      "step": 4790
+    },
+    {
+      "epoch": 92.3,
+      "learning_rate": 1.6107823800131494e-05,
+      "loss": 0.1659,
+      "step": 4800
+    },
+    {
+      "epoch": 92.5,
+      "learning_rate": 1.602564102564103e-05,
+      "loss": 0.1674,
+      "step": 4810
+    },
+    {
+      "epoch": 92.69,
+      "learning_rate": 1.594345825115056e-05,
+      "loss": 0.1803,
+      "step": 4820
+    },
+    {
+      "epoch": 92.88,
+      "learning_rate": 1.586127547666009e-05,
+      "loss": 0.1741,
+      "step": 4830
+    },
+    {
+      "epoch": 92.99,
+      "eval_accuracy": 0.8942436412315931,
+      "eval_loss": 0.41073036193847656,
+      "eval_runtime": 14.426,
+      "eval_samples_per_second": 51.781,
+      "eval_steps_per_second": 1.664,
+      "step": 4836
+    },
+    {
+      "epoch": 93.08,
+      "learning_rate": 1.5779092702169626e-05,
+      "loss": 0.1953,
+      "step": 4840
+    },
+    {
+      "epoch": 93.27,
+      "learning_rate": 1.569690992767916e-05,
+      "loss": 0.1548,
+      "step": 4850
+    },
+    {
+      "epoch": 93.46,
+      "learning_rate": 1.5614727153188692e-05,
+      "loss": 0.1648,
+      "step": 4860
+    },
+    {
+      "epoch": 93.65,
+      "learning_rate": 1.5532544378698223e-05,
+      "loss": 0.1618,
+      "step": 4870
+    },
+    {
+      "epoch": 93.84,
+      "learning_rate": 1.5450361604207758e-05,
+      "loss": 0.1574,
+      "step": 4880
+    },
+    {
+      "epoch": 93.99,
+      "eval_accuracy": 0.8808567603748326,
+      "eval_loss": 0.5604187250137329,
+      "eval_runtime": 14.3999,
+      "eval_samples_per_second": 51.876,
+      "eval_steps_per_second": 1.667,
+      "step": 4888
+    },
+    {
+      "epoch": 94.04,
+      "learning_rate": 1.5368178829717293e-05,
+      "loss": 0.1697,
+      "step": 4890
+    },
+    {
+      "epoch": 94.23,
+      "learning_rate": 1.5285996055226824e-05,
+      "loss": 0.1568,
+      "step": 4900
+    },
+    {
+      "epoch": 94.42,
+      "learning_rate": 1.5203813280736359e-05,
+      "loss": 0.1557,
+      "step": 4910
+    },
+    {
+      "epoch": 94.61,
+      "learning_rate": 1.5121630506245893e-05,
+      "loss": 0.1502,
+      "step": 4920
+    },
+    {
+      "epoch": 94.8,
+      "learning_rate": 1.5039447731755425e-05,
+      "loss": 0.1831,
+      "step": 4930
+    },
+    {
+      "epoch": 94.99,
+      "learning_rate": 1.4957264957264958e-05,
+      "loss": 0.193,
+      "step": 4940
+    },
+    {
+      "epoch": 94.99,
+      "eval_accuracy": 0.8888888888888888,
+      "eval_loss": 0.47749215364456177,
+      "eval_runtime": 14.4207,
+      "eval_samples_per_second": 51.8,
+      "eval_steps_per_second": 1.664,
+      "step": 4940
+    },
+    {
+      "epoch": 95.19,
+      "learning_rate": 1.487508218277449e-05,
+      "loss": 0.1535,
+      "step": 4950
+    },
+    {
+      "epoch": 95.38,
+      "learning_rate": 1.4792899408284025e-05,
+      "loss": 0.1545,
+      "step": 4960
+    },
+    {
+      "epoch": 95.57,
+      "learning_rate": 1.4710716633793558e-05,
+      "loss": 0.165,
+      "step": 4970
+    },
+    {
+      "epoch": 95.76,
+      "learning_rate": 1.462853385930309e-05,
+      "loss": 0.1769,
+      "step": 4980
+    },
+    {
+      "epoch": 95.95,
+      "learning_rate": 1.4546351084812624e-05,
+      "loss": 0.2018,
+      "step": 4990
+    },
+    {
+      "epoch": 95.99,
+      "eval_accuracy": 0.8995983935742972,
+      "eval_loss": 0.42000728845596313,
+      "eval_runtime": 14.3924,
+      "eval_samples_per_second": 51.903,
+      "eval_steps_per_second": 1.668,
+      "step": 4992
+    },
+    {
+      "epoch": 96.15,
+      "learning_rate": 1.4464168310322157e-05,
+      "loss": 0.182,
+      "step": 5000
+    },
+    {
+      "epoch": 96.34,
+      "learning_rate": 1.438198553583169e-05,
+      "loss": 0.1789,
+      "step": 5010
+    },
+    {
+      "epoch": 96.53,
+      "learning_rate": 1.4299802761341225e-05,
+      "loss": 0.154,
+      "step": 5020
+    },
+    {
+      "epoch": 96.72,
+      "learning_rate": 1.4217619986850758e-05,
+      "loss": 0.1978,
+      "step": 5030
+    },
+    {
+      "epoch": 96.91,
+      "learning_rate": 1.4135437212360289e-05,
+      "loss": 0.1832,
+      "step": 5040
+    },
+    {
+      "epoch": 96.99,
+      "eval_accuracy": 0.9022757697456493,
+      "eval_loss": 0.4504472613334656,
+      "eval_runtime": 14.5188,
+      "eval_samples_per_second": 51.451,
+      "eval_steps_per_second": 1.653,
+      "step": 5044
+    },
+    {
+      "epoch": 97.11,
+      "learning_rate": 1.4053254437869822e-05,
+      "loss": 0.1728,
+      "step": 5050
+    },
+    {
+      "epoch": 97.3,
+      "learning_rate": 1.3971071663379357e-05,
+      "loss": 0.1612,
+      "step": 5060
+    },
+    {
+      "epoch": 97.5,
+      "learning_rate": 1.388888888888889e-05,
+      "loss": 0.149,
+      "step": 5070
+    },
+    {
+      "epoch": 97.69,
+      "learning_rate": 1.3806706114398424e-05,
+      "loss": 0.1458,
+      "step": 5080
+    },
+    {
+      "epoch": 97.88,
+      "learning_rate": 1.3724523339907955e-05,
+      "loss": 0.1624,
+      "step": 5090
+    },
+    {
+      "epoch": 97.99,
+      "eval_accuracy": 0.8888888888888888,
+      "eval_loss": 0.48589327931404114,
+      "eval_runtime": 14.4154,
+      "eval_samples_per_second": 51.82,
+      "eval_steps_per_second": 1.665,
+      "step": 5096
+    },
+    {
+      "epoch": 98.08,
+      "learning_rate": 1.3642340565417488e-05,
+      "loss": 0.1915,
+      "step": 5100
+    },
+    {
+      "epoch": 98.27,
+      "learning_rate": 1.3560157790927021e-05,
+      "loss": 0.1808,
+      "step": 5110
+    },
+    {
+      "epoch": 98.46,
+      "learning_rate": 1.3477975016436556e-05,
+      "loss": 0.1841,
+      "step": 5120
+    },
+    {
+      "epoch": 98.65,
+      "learning_rate": 1.3395792241946089e-05,
+      "loss": 0.1592,
+      "step": 5130
+    },
+    {
+      "epoch": 98.84,
+      "learning_rate": 1.3313609467455624e-05,
+      "loss": 0.1739,
+      "step": 5140
+    },
+    {
+      "epoch": 98.99,
+      "eval_accuracy": 0.8848728246318608,
+      "eval_loss": 0.495495080947876,
+      "eval_runtime": 14.3705,
+      "eval_samples_per_second": 51.982,
+      "eval_steps_per_second": 1.67,
+      "step": 5148
+    },
+    {
+      "epoch": 99.04,
+      "learning_rate": 1.3231426692965155e-05,
+      "loss": 0.1888,
+      "step": 5150
+    },
+    {
+      "epoch": 99.23,
+      "learning_rate": 1.3149243918474688e-05,
+      "loss": 0.1766,
+      "step": 5160
+    },
+    {
+      "epoch": 99.42,
+      "learning_rate": 1.3067061143984221e-05,
+      "loss": 0.1681,
+      "step": 5170
+    },
+    {
+      "epoch": 99.61,
+      "learning_rate": 1.2984878369493756e-05,
+      "loss": 0.154,
+      "step": 5180
+    },
+    {
+      "epoch": 99.8,
+      "learning_rate": 1.2902695595003289e-05,
+      "loss": 0.1544,
+      "step": 5190
+    },
+    {
+      "epoch": 99.99,
+      "learning_rate": 1.282051282051282e-05,
+      "loss": 0.1439,
+      "step": 5200
+    },
+    {
+      "epoch": 99.99,
+      "eval_accuracy": 0.8942436412315931,
+      "eval_loss": 0.4792201817035675,
+      "eval_runtime": 14.3961,
+      "eval_samples_per_second": 51.889,
+      "eval_steps_per_second": 1.667,
+      "step": 5200
+    },
+    {
+      "epoch": 100.19,
+      "learning_rate": 1.2738330046022354e-05,
+      "loss": 0.1685,
+      "step": 5210
+    },
+    {
+      "epoch": 100.38,
+      "learning_rate": 1.2656147271531887e-05,
+      "loss": 0.1708,
+      "step": 5220
+    },
+    {
+      "epoch": 100.57,
+      "learning_rate": 1.257396449704142e-05,
+      "loss": 0.1411,
+      "step": 5230
+    },
+    {
+      "epoch": 100.76,
+      "learning_rate": 1.2491781722550953e-05,
+      "loss": 0.1652,
+      "step": 5240
+    },
+    {
+      "epoch": 100.95,
+      "learning_rate": 1.2409598948060486e-05,
+      "loss": 0.1716,
+      "step": 5250
+    },
+    {
+      "epoch": 100.99,
+      "eval_accuracy": 0.8862115127175368,
+      "eval_loss": 0.5112172365188599,
+      "eval_runtime": 14.5235,
+      "eval_samples_per_second": 51.434,
+      "eval_steps_per_second": 1.652,
+      "step": 5252
+    },
+    {
+      "epoch": 101.15,
+      "learning_rate": 1.2327416173570021e-05,
+      "loss": 0.1744,
+      "step": 5260
+    },
+    {
+      "epoch": 101.34,
+      "learning_rate": 1.2245233399079552e-05,
+      "loss": 0.1639,
+      "step": 5270
+    },
+    {
+      "epoch": 101.53,
+      "learning_rate": 1.2163050624589087e-05,
+      "loss": 0.1618,
+      "step": 5280
+    },
+    {
+      "epoch": 101.72,
+      "learning_rate": 1.208086785009862e-05,
+      "loss": 0.1677,
+      "step": 5290
+    },
+    {
+      "epoch": 101.91,
+      "learning_rate": 1.1998685075608153e-05,
+      "loss": 0.1537,
+      "step": 5300
+    },
+    {
+      "epoch": 101.99,
+      "eval_accuracy": 0.891566265060241,
+      "eval_loss": 0.45718035101890564,
+      "eval_runtime": 14.3384,
+      "eval_samples_per_second": 52.098,
+      "eval_steps_per_second": 1.674,
+      "step": 5304
+    },
+    {
+      "epoch": 102.11,
+      "learning_rate": 1.1916502301117686e-05,
+      "loss": 0.1923,
+      "step": 5310
+    },
+    {
+      "epoch": 102.3,
+      "learning_rate": 1.1834319526627219e-05,
+      "loss": 0.1562,
+      "step": 5320
+    },
+    {
+      "epoch": 102.5,
+      "learning_rate": 1.1752136752136752e-05,
+      "loss": 0.148,
+      "step": 5330
+    },
+    {
+      "epoch": 102.69,
+      "learning_rate": 1.1669953977646286e-05,
+      "loss": 0.1553,
+      "step": 5340
+    },
+    {
+      "epoch": 102.88,
+      "learning_rate": 1.158777120315582e-05,
+      "loss": 0.1655,
+      "step": 5350
+    },
+    {
+      "epoch": 102.99,
+      "eval_accuracy": 0.8808567603748326,
+      "eval_loss": 0.477417528629303,
+      "eval_runtime": 14.4044,
+      "eval_samples_per_second": 51.859,
+      "eval_steps_per_second": 1.666,
+      "step": 5356
+    },
+    {
+      "epoch": 103.08,
+      "learning_rate": 1.1505588428665352e-05,
+      "loss": 0.185,
+      "step": 5360
+    },
+    {
+      "epoch": 103.27,
+      "learning_rate": 1.1423405654174885e-05,
+      "loss": 0.1461,
+      "step": 5370
+    },
+    {
+      "epoch": 103.46,
+      "learning_rate": 1.1341222879684418e-05,
+      "loss": 0.1652,
+      "step": 5380
+    },
+    {
+      "epoch": 103.65,
+      "learning_rate": 1.1259040105193951e-05,
+      "loss": 0.1673,
+      "step": 5390
+    },
+    {
+      "epoch": 103.84,
+      "learning_rate": 1.1176857330703486e-05,
+      "loss": 0.1515,
+      "step": 5400
+    },
+    {
+      "epoch": 103.99,
+      "eval_accuracy": 0.8888888888888888,
+      "eval_loss": 0.46347784996032715,
+      "eval_runtime": 14.5532,
+      "eval_samples_per_second": 51.329,
+      "eval_steps_per_second": 1.649,
+      "step": 5408
+    },
+    {
+      "epoch": 104.04,
+      "learning_rate": 1.1094674556213017e-05,
+      "loss": 0.1752,
+      "step": 5410
+    },
+    {
+      "epoch": 104.23,
+      "learning_rate": 1.1012491781722552e-05,
+      "loss": 0.1553,
+      "step": 5420
+    },
+    {
+      "epoch": 104.42,
+      "learning_rate": 1.0930309007232085e-05,
+      "loss": 0.1523,
+      "step": 5430
+    },
+    {
+      "epoch": 104.61,
+      "learning_rate": 1.0848126232741618e-05,
+      "loss": 0.141,
+      "step": 5440
+    },
+    {
+      "epoch": 104.8,
+      "learning_rate": 1.076594345825115e-05,
+      "loss": 0.1563,
+      "step": 5450
+    },
+    {
+      "epoch": 104.99,
+      "learning_rate": 1.0683760683760684e-05,
+      "loss": 0.1594,
+      "step": 5460
+    },
+    {
+      "epoch": 104.99,
+      "eval_accuracy": 0.892904953145917,
+      "eval_loss": 0.4794405996799469,
+      "eval_runtime": 14.4485,
+      "eval_samples_per_second": 51.701,
+      "eval_steps_per_second": 1.661,
+      "step": 5460
+    },
+    {
+      "epoch": 105.19,
+      "learning_rate": 1.0601577909270217e-05,
+      "loss": 0.1537,
+      "step": 5470
+    },
+    {
+      "epoch": 105.38,
+      "learning_rate": 1.0519395134779751e-05,
+      "loss": 0.1662,
+      "step": 5480
+    },
+    {
+      "epoch": 105.57,
+      "learning_rate": 1.0437212360289284e-05,
+      "loss": 0.1549,
+      "step": 5490
+    },
+    {
+      "epoch": 105.76,
+      "learning_rate": 1.0355029585798817e-05,
+      "loss": 0.1496,
+      "step": 5500
+    },
+    {
+      "epoch": 105.95,
+      "learning_rate": 1.027284681130835e-05,
+      "loss": 0.1488,
+      "step": 5510
+    },
+    {
+      "epoch": 105.99,
+      "eval_accuracy": 0.8969210174029452,
+      "eval_loss": 0.4940612018108368,
+      "eval_runtime": 14.3875,
+      "eval_samples_per_second": 51.92,
+      "eval_steps_per_second": 1.668,
+      "step": 5512
+    },
+    {
+      "epoch": 106.15,
+      "learning_rate": 1.0190664036817883e-05,
+      "loss": 0.1763,
+      "step": 5520
+    },
+    {
+      "epoch": 106.34,
+      "learning_rate": 1.0108481262327416e-05,
+      "loss": 0.153,
+      "step": 5530
+    },
+    {
+      "epoch": 106.53,
+      "learning_rate": 1.002629848783695e-05,
+      "loss": 0.1537,
+      "step": 5540
+    },
+    {
+      "epoch": 106.72,
+      "learning_rate": 9.944115713346482e-06,
+      "loss": 0.1519,
+      "step": 5550
+    },
+    {
+      "epoch": 106.91,
+      "learning_rate": 9.861932938856017e-06,
+      "loss": 0.1634,
+      "step": 5560
+    },
+    {
+      "epoch": 106.99,
+      "eval_accuracy": 0.891566265060241,
+      "eval_loss": 0.4841405749320984,
+      "eval_runtime": 14.3082,
+      "eval_samples_per_second": 52.208,
+      "eval_steps_per_second": 1.677,
+      "step": 5564
+    },
+    {
+      "epoch": 107.11,
+      "learning_rate": 9.77975016436555e-06,
+      "loss": 0.1776,
+      "step": 5570
+    },
+    {
+      "epoch": 107.3,
+      "learning_rate": 9.697567389875083e-06,
+      "loss": 0.1385,
+      "step": 5580
+    },
+    {
+      "epoch": 107.5,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.1511,
+      "step": 5590
+    },
+    {
+      "epoch": 107.69,
+      "learning_rate": 9.533201840894149e-06,
+      "loss": 0.1309,
+      "step": 5600
+    },
+    {
+      "epoch": 107.88,
+      "learning_rate": 9.451019066403682e-06,
+      "loss": 0.1471,
+      "step": 5610
+    },
+    {
+      "epoch": 107.99,
+      "eval_accuracy": 0.9009370816599732,
+      "eval_loss": 0.4919324815273285,
+      "eval_runtime": 14.3442,
+      "eval_samples_per_second": 52.077,
+      "eval_steps_per_second": 1.673,
+      "step": 5616
+    },
+    {
+      "epoch": 108.08,
+      "learning_rate": 9.368836291913216e-06,
+      "loss": 0.1708,
+      "step": 5620
+    },
+    {
+      "epoch": 108.27,
+      "learning_rate": 9.286653517422748e-06,
+      "loss": 0.1486,
+      "step": 5630
+    },
+    {
+      "epoch": 108.46,
+      "learning_rate": 9.204470742932282e-06,
+      "loss": 0.1555,
+      "step": 5640
+    },
+    {
+      "epoch": 108.65,
+      "learning_rate": 9.122287968441815e-06,
+      "loss": 0.1642,
+      "step": 5650
+    },
+    {
+      "epoch": 108.84,
+      "learning_rate": 9.040105193951348e-06,
+      "loss": 0.1453,
+      "step": 5660
+    },
+    {
+      "epoch": 108.99,
+      "eval_accuracy": 0.9009370816599732,
+      "eval_loss": 0.4617402255535126,
+      "eval_runtime": 14.3481,
+      "eval_samples_per_second": 52.063,
+      "eval_steps_per_second": 1.673,
+      "step": 5668
+    },
+    {
+      "epoch": 109.04,
+      "learning_rate": 8.957922419460881e-06,
+      "loss": 0.185,
+      "step": 5670
+    },
+    {
+      "epoch": 109.23,
+      "learning_rate": 8.875739644970414e-06,
+      "loss": 0.1667,
+      "step": 5680
+    },
+    {
+      "epoch": 109.42,
+      "learning_rate": 8.793556870479947e-06,
+      "loss": 0.1661,
+      "step": 5690
+    },
+    {
+      "epoch": 109.61,
+      "learning_rate": 8.711374095989482e-06,
+      "loss": 0.1493,
+      "step": 5700
+    },
+    {
+      "epoch": 109.8,
+      "learning_rate": 8.629191321499015e-06,
+      "loss": 0.145,
+      "step": 5710
+    },
+    {
+      "epoch": 109.99,
+      "learning_rate": 8.547008547008548e-06,
+      "loss": 0.1578,
+      "step": 5720
+    },
+    {
+      "epoch": 109.99,
+      "eval_accuracy": 0.9009370816599732,
+      "eval_loss": 0.43282124400138855,
+      "eval_runtime": 14.5073,
+      "eval_samples_per_second": 51.491,
+      "eval_steps_per_second": 1.654,
+      "step": 5720
+    },
+    {
+      "epoch": 110.19,
+      "learning_rate": 8.46482577251808e-06,
+      "loss": 0.1675,
+      "step": 5730
+    },
+    {
+      "epoch": 110.38,
+      "learning_rate": 8.382642998027614e-06,
+      "loss": 0.1344,
+      "step": 5740
+    },
+    {
+      "epoch": 110.57,
+      "learning_rate": 8.300460223537147e-06,
+      "loss": 0.1351,
+      "step": 5750
+    },
+    {
+      "epoch": 110.76,
+      "learning_rate": 8.21827744904668e-06,
+      "loss": 0.1266,
+      "step": 5760
+    },
+    {
+      "epoch": 110.95,
+      "learning_rate": 8.136094674556213e-06,
+      "loss": 0.1754,
+      "step": 5770
+    },
+    {
+      "epoch": 110.99,
+      "eval_accuracy": 0.8955823293172691,
+      "eval_loss": 0.5239934921264648,
+      "eval_runtime": 14.4204,
+      "eval_samples_per_second": 51.802,
+      "eval_steps_per_second": 1.664,
+      "step": 5772
+    },
+    {
+      "epoch": 111.15,
+      "learning_rate": 8.053911900065747e-06,
+      "loss": 0.1602,
+      "step": 5780
+    },
+    {
+      "epoch": 111.34,
+      "learning_rate": 7.97172912557528e-06,
+      "loss": 0.1794,
+      "step": 5790
+    },
+    {
+      "epoch": 111.53,
+      "learning_rate": 7.889546351084813e-06,
+      "loss": 0.1308,
+      "step": 5800
+    },
+    {
+      "epoch": 111.72,
+      "learning_rate": 7.807363576594346e-06,
+      "loss": 0.1377,
+      "step": 5810
+    },
+    {
+      "epoch": 111.91,
+      "learning_rate": 7.725180802103879e-06,
+      "loss": 0.1657,
+      "step": 5820
+    },
+    {
+      "epoch": 111.99,
+      "eval_accuracy": 0.8969210174029452,
+      "eval_loss": 0.4820827841758728,
+      "eval_runtime": 14.3674,
+      "eval_samples_per_second": 51.993,
+      "eval_steps_per_second": 1.67,
+      "step": 5824
+    },
+    {
+      "epoch": 112.11,
+      "learning_rate": 7.642998027613412e-06,
+      "loss": 0.1633,
+      "step": 5830
+    },
+    {
+      "epoch": 112.3,
+      "learning_rate": 7.560815253122947e-06,
+      "loss": 0.1622,
+      "step": 5840
+    },
+    {
+      "epoch": 112.5,
+      "learning_rate": 7.478632478632479e-06,
+      "loss": 0.1481,
+      "step": 5850
+    },
+    {
+      "epoch": 112.69,
+      "learning_rate": 7.396449704142013e-06,
+      "loss": 0.1603,
+      "step": 5860
+    },
+    {
+      "epoch": 112.88,
+      "learning_rate": 7.314266929651545e-06,
+      "loss": 0.1516,
+      "step": 5870
+    },
+    {
+      "epoch": 112.99,
+      "eval_accuracy": 0.9022757697456493,
+      "eval_loss": 0.4410625100135803,
+      "eval_runtime": 14.3892,
+      "eval_samples_per_second": 51.914,
+      "eval_steps_per_second": 1.668,
+      "step": 5876
+    },
+    {
+      "epoch": 113.08,
+      "learning_rate": 7.2320841551610785e-06,
+      "loss": 0.1658,
+      "step": 5880
+    },
+    {
+      "epoch": 113.27,
+      "learning_rate": 7.149901380670612e-06,
+      "loss": 0.1637,
+      "step": 5890
+    },
+    {
+      "epoch": 113.46,
+      "learning_rate": 7.0677186061801445e-06,
+      "loss": 0.1622,
+      "step": 5900
+    },
+    {
+      "epoch": 113.65,
+      "learning_rate": 6.985535831689678e-06,
+      "loss": 0.1532,
+      "step": 5910
+    },
+    {
+      "epoch": 113.84,
+      "learning_rate": 6.903353057199212e-06,
+      "loss": 0.1542,
+      "step": 5920
+    },
+    {
+      "epoch": 113.99,
+      "eval_accuracy": 0.8821954484605087,
+      "eval_loss": 0.5312647223472595,
+      "eval_runtime": 14.4151,
+      "eval_samples_per_second": 51.821,
+      "eval_steps_per_second": 1.665,
+      "step": 5928
+    },
+    {
+      "epoch": 114.04,
+      "learning_rate": 6.821170282708744e-06,
+      "loss": 0.1792,
+      "step": 5930
+    },
+    {
+      "epoch": 114.23,
+      "learning_rate": 6.738987508218278e-06,
+      "loss": 0.1453,
+      "step": 5940
+    },
+    {
+      "epoch": 114.42,
+      "learning_rate": 6.656804733727812e-06,
+      "loss": 0.1754,
+      "step": 5950
+    },
+    {
+      "epoch": 114.61,
+      "learning_rate": 6.574621959237344e-06,
+      "loss": 0.1489,
+      "step": 5960
+    },
+    {
+      "epoch": 114.8,
+      "learning_rate": 6.492439184746878e-06,
+      "loss": 0.1284,
+      "step": 5970
+    },
+    {
+      "epoch": 114.99,
+      "learning_rate": 6.41025641025641e-06,
+      "loss": 0.1496,
+      "step": 5980
+    },
+    {
+      "epoch": 114.99,
+      "eval_accuracy": 0.8862115127175368,
+      "eval_loss": 0.5038286447525024,
+      "eval_runtime": 14.3257,
+      "eval_samples_per_second": 52.144,
+      "eval_steps_per_second": 1.675,
+      "step": 5980
+    },
+    {
+      "epoch": 115.19,
+      "learning_rate": 6.328073635765944e-06,
+      "loss": 0.1514,
+      "step": 5990
+    },
+    {
+      "epoch": 115.38,
+      "learning_rate": 6.245890861275477e-06,
+      "loss": 0.152,
+      "step": 6000
+    },
+    {
+      "epoch": 115.57,
+      "learning_rate": 6.1637080867850105e-06,
+      "loss": 0.1339,
+      "step": 6010
+    },
+    {
+      "epoch": 115.76,
+      "learning_rate": 6.0815253122945435e-06,
+      "loss": 0.1391,
+      "step": 6020
+    },
+    {
+      "epoch": 115.95,
+      "learning_rate": 5.9993425378040764e-06,
+      "loss": 0.1597,
+      "step": 6030
+    },
+    {
+      "epoch": 115.99,
+      "eval_accuracy": 0.8875502008032129,
+      "eval_loss": 0.49084627628326416,
+      "eval_runtime": 14.6158,
+      "eval_samples_per_second": 51.109,
+      "eval_steps_per_second": 1.642,
+      "step": 6032
+    },
+    {
+      "epoch": 116.15,
+      "learning_rate": 5.917159763313609e-06,
+      "loss": 0.1652,
+      "step": 6040
+    },
+    {
+      "epoch": 116.34,
+      "learning_rate": 5.834976988823143e-06,
+      "loss": 0.1499,
+      "step": 6050
+    },
+    {
+      "epoch": 116.53,
+      "learning_rate": 5.752794214332676e-06,
+      "loss": 0.1431,
+      "step": 6060
+    },
+    {
+      "epoch": 116.72,
+      "learning_rate": 5.670611439842209e-06,
+      "loss": 0.1359,
+      "step": 6070
+    },
+    {
+      "epoch": 116.91,
+      "learning_rate": 5.588428665351743e-06,
+      "loss": 0.1175,
+      "step": 6080
+    },
+    {
+      "epoch": 116.99,
+      "eval_accuracy": 0.8862115127175368,
+      "eval_loss": 0.5503976345062256,
+      "eval_runtime": 14.5303,
+      "eval_samples_per_second": 51.41,
+      "eval_steps_per_second": 1.652,
+      "step": 6084
+    },
+    {
+      "epoch": 117.11,
+      "learning_rate": 5.506245890861276e-06,
+      "loss": 0.1472,
+      "step": 6090
+    },
+    {
+      "epoch": 117.3,
+      "learning_rate": 5.424063116370809e-06,
+      "loss": 0.1446,
+      "step": 6100
+    },
+    {
+      "epoch": 117.5,
+      "learning_rate": 5.341880341880342e-06,
+      "loss": 0.1455,
+      "step": 6110
+    },
+    {
+      "epoch": 117.69,
+      "learning_rate": 5.259697567389876e-06,
+      "loss": 0.1539,
+      "step": 6120
+    },
+    {
+      "epoch": 117.88,
+      "learning_rate": 5.177514792899409e-06,
+      "loss": 0.1415,
+      "step": 6130
+    },
+    {
+      "epoch": 117.99,
+      "eval_accuracy": 0.891566265060241,
+      "eval_loss": 0.5018065571784973,
+      "eval_runtime": 14.4961,
+      "eval_samples_per_second": 51.531,
+      "eval_steps_per_second": 1.656,
+      "step": 6136
+    },
+    {
+      "epoch": 118.08,
+      "learning_rate": 5.095332018408942e-06,
+      "loss": 0.1494,
+      "step": 6140
+    },
+    {
+      "epoch": 118.27,
+      "learning_rate": 5.013149243918475e-06,
+      "loss": 0.1438,
+      "step": 6150
+    },
+    {
+      "epoch": 118.46,
+      "learning_rate": 4.930966469428008e-06,
+      "loss": 0.1479,
+      "step": 6160
+    },
+    {
+      "epoch": 118.65,
+      "learning_rate": 4.848783694937541e-06,
+      "loss": 0.1418,
+      "step": 6170
+    },
+    {
+      "epoch": 118.84,
+      "learning_rate": 4.766600920447074e-06,
+      "loss": 0.1614,
+      "step": 6180
+    },
+    {
+      "epoch": 118.99,
+      "eval_accuracy": 0.8902275769745649,
+      "eval_loss": 0.5221453309059143,
+      "eval_runtime": 14.4293,
+      "eval_samples_per_second": 51.77,
+      "eval_steps_per_second": 1.663,
+      "step": 6188
+    },
+    {
+      "epoch": 119.04,
+      "learning_rate": 4.684418145956608e-06,
+      "loss": 0.1587,
+      "step": 6190
+    },
+    {
+      "epoch": 119.23,
+      "learning_rate": 4.602235371466141e-06,
+      "loss": 0.1543,
+      "step": 6200
+    },
+    {
+      "epoch": 119.42,
+      "learning_rate": 4.520052596975674e-06,
+      "loss": 0.1657,
+      "step": 6210
+    },
+    {
+      "epoch": 119.61,
+      "learning_rate": 4.437869822485207e-06,
+      "loss": 0.1471,
+      "step": 6220
+    },
+    {
+      "epoch": 119.8,
+      "learning_rate": 4.355687047994741e-06,
+      "loss": 0.1346,
+      "step": 6230
+    },
+    {
+      "epoch": 119.99,
+      "learning_rate": 4.273504273504274e-06,
+      "loss": 0.1396,
+      "step": 6240
+    },
+    {
+      "epoch": 119.99,
+      "eval_accuracy": 0.8902275769745649,
+      "eval_loss": 0.504159688949585,
+      "eval_runtime": 14.5651,
+      "eval_samples_per_second": 51.287,
+      "eval_steps_per_second": 1.648,
+      "step": 6240
+    },
+    {
+      "epoch": 120.19,
+      "learning_rate": 4.191321499013807e-06,
+      "loss": 0.1553,
+      "step": 6250
+    },
+    {
+      "epoch": 120.38,
+      "learning_rate": 4.10913872452334e-06,
+      "loss": 0.1491,
+      "step": 6260
+    },
+    {
+      "epoch": 120.57,
+      "learning_rate": 4.026955950032874e-06,
+      "loss": 0.1587,
+      "step": 6270
+    },
+    {
+      "epoch": 120.76,
+      "learning_rate": 3.9447731755424066e-06,
+      "loss": 0.1511,
+      "step": 6280
+    },
+    {
+      "epoch": 120.95,
+      "learning_rate": 3.8625904010519395e-06,
+      "loss": 0.1673,
+      "step": 6290
+    },
+    {
+      "epoch": 120.99,
+      "eval_accuracy": 0.8875502008032129,
+      "eval_loss": 0.5078461766242981,
+      "eval_runtime": 14.484,
+      "eval_samples_per_second": 51.574,
+      "eval_steps_per_second": 1.657,
+      "step": 6292
+    },
+    {
+      "epoch": 121.15,
+      "learning_rate": 3.7804076265614733e-06,
+      "loss": 0.1589,
+      "step": 6300
+    },
+    {
+      "epoch": 121.34,
+      "learning_rate": 3.6982248520710063e-06,
+      "loss": 0.1227,
+      "step": 6310
+    },
+    {
+      "epoch": 121.53,
+      "learning_rate": 3.6160420775805393e-06,
+      "loss": 0.1313,
+      "step": 6320
+    },
+    {
+      "epoch": 121.72,
+      "learning_rate": 3.5338593030900722e-06,
+      "loss": 0.1529,
+      "step": 6330
+    },
+    {
+      "epoch": 121.91,
+      "learning_rate": 3.451676528599606e-06,
+      "loss": 0.1303,
+      "step": 6340
+    },
+    {
+      "epoch": 121.99,
+      "eval_accuracy": 0.8942436412315931,
+      "eval_loss": 0.49939635396003723,
+      "eval_runtime": 14.469,
+      "eval_samples_per_second": 51.627,
+      "eval_steps_per_second": 1.659,
+      "step": 6344
+    },
+    {
+      "epoch": 122.11,
+      "learning_rate": 3.369493754109139e-06,
+      "loss": 0.1597,
+      "step": 6350
+    },
+    {
+      "epoch": 122.3,
+      "learning_rate": 3.287310979618672e-06,
+      "loss": 0.1285,
+      "step": 6360
+    },
+    {
+      "epoch": 122.5,
+      "learning_rate": 3.205128205128205e-06,
+      "loss": 0.1402,
+      "step": 6370
+    },
+    {
+      "epoch": 122.69,
+      "learning_rate": 3.1229454306377383e-06,
+      "loss": 0.1647,
+      "step": 6380
+    },
+    {
+      "epoch": 122.88,
+      "learning_rate": 3.0407626561472717e-06,
+      "loss": 0.1355,
+      "step": 6390
+    },
+    {
+      "epoch": 122.99,
+      "eval_accuracy": 0.8942436412315931,
+      "eval_loss": 0.48336493968963623,
+      "eval_runtime": 14.3883,
+      "eval_samples_per_second": 51.917,
+      "eval_steps_per_second": 1.668,
+      "step": 6396
+    },
+    {
+      "epoch": 123.08,
+      "learning_rate": 2.9585798816568047e-06,
+      "loss": 0.132,
+      "step": 6400
+    },
+    {
+      "epoch": 123.27,
+      "learning_rate": 2.876397107166338e-06,
+      "loss": 0.1126,
+      "step": 6410
+    },
+    {
+      "epoch": 123.46,
+      "learning_rate": 2.7942143326758715e-06,
+      "loss": 0.136,
+      "step": 6420
+    },
+    {
+      "epoch": 123.65,
+      "learning_rate": 2.7120315581854045e-06,
+      "loss": 0.1539,
+      "step": 6430
+    },
+    {
+      "epoch": 123.84,
+      "learning_rate": 2.629848783694938e-06,
+      "loss": 0.1452,
+      "step": 6440
+    },
+    {
+      "epoch": 123.99,
+      "eval_accuracy": 0.8888888888888888,
+      "eval_loss": 0.5145061612129211,
+      "eval_runtime": 14.4808,
+      "eval_samples_per_second": 51.585,
+      "eval_steps_per_second": 1.657,
+      "step": 6448
+    },
+    {
+      "epoch": 124.04,
+      "learning_rate": 2.547666009204471e-06,
+      "loss": 0.1283,
+      "step": 6450
+    },
+    {
+      "epoch": 124.23,
+      "learning_rate": 2.465483234714004e-06,
+      "loss": 0.1339,
+      "step": 6460
+    },
+    {
+      "epoch": 124.42,
+      "learning_rate": 2.383300460223537e-06,
+      "loss": 0.1426,
+      "step": 6470
+    },
+    {
+      "epoch": 124.61,
+      "learning_rate": 2.3011176857330706e-06,
+      "loss": 0.124,
+      "step": 6480
+    },
+    {
+      "epoch": 124.8,
+      "learning_rate": 2.2189349112426035e-06,
+      "loss": 0.1396,
+      "step": 6490
+    },
+    {
+      "epoch": 124.99,
+      "learning_rate": 2.136752136752137e-06,
+      "loss": 0.142,
+      "step": 6500
+    },
+    {
+      "epoch": 124.99,
+      "eval_accuracy": 0.8821954484605087,
+      "eval_loss": 0.5479710102081299,
+      "eval_runtime": 14.4675,
+      "eval_samples_per_second": 51.633,
+      "eval_steps_per_second": 1.659,
+      "step": 6500
+    },
+    {
+      "epoch": 125.19,
+      "learning_rate": 2.05456936226167e-06,
+      "loss": 0.1376,
+      "step": 6510
+    },
+    {
+      "epoch": 125.38,
+      "learning_rate": 1.9723865877712033e-06,
+      "loss": 0.1416,
+      "step": 6520
+    },
+    {
+      "epoch": 125.57,
+      "learning_rate": 1.8902038132807367e-06,
+      "loss": 0.1457,
+      "step": 6530
+    },
+    {
+      "epoch": 125.76,
+      "learning_rate": 1.8080210387902696e-06,
+      "loss": 0.1428,
+      "step": 6540
+    },
+    {
+      "epoch": 125.95,
+      "learning_rate": 1.725838264299803e-06,
+      "loss": 0.1318,
+      "step": 6550
+    },
+    {
+      "epoch": 125.99,
+      "eval_accuracy": 0.891566265060241,
+      "eval_loss": 0.5098896026611328,
+      "eval_runtime": 14.5131,
+      "eval_samples_per_second": 51.471,
+      "eval_steps_per_second": 1.654,
+      "step": 6552
+    },
+    {
+      "epoch": 126.15,
+      "learning_rate": 1.643655489809336e-06,
+      "loss": 0.1526,
+      "step": 6560
+    },
+    {
+      "epoch": 126.34,
+      "learning_rate": 1.5614727153188692e-06,
+      "loss": 0.124,
+      "step": 6570
+    },
+    {
+      "epoch": 126.53,
+      "learning_rate": 1.4792899408284024e-06,
+      "loss": 0.1514,
+      "step": 6580
+    },
+    {
+      "epoch": 126.72,
+      "learning_rate": 1.3971071663379357e-06,
+      "loss": 0.1486,
+      "step": 6590
+    },
+    {
+      "epoch": 126.91,
+      "learning_rate": 1.314924391847469e-06,
+      "loss": 0.122,
+      "step": 6600
+    },
+    {
+      "epoch": 126.99,
+      "eval_accuracy": 0.8875502008032129,
+      "eval_loss": 0.5158522129058838,
+      "eval_runtime": 14.3975,
+      "eval_samples_per_second": 51.884,
+      "eval_steps_per_second": 1.667,
+      "step": 6604
+    },
+    {
+      "epoch": 127.11,
+      "learning_rate": 1.232741617357002e-06,
+      "loss": 0.1516,
+      "step": 6610
+    },
+    {
+      "epoch": 127.3,
+      "learning_rate": 1.1505588428665353e-06,
+      "loss": 0.1588,
+      "step": 6620
+    },
+    {
+      "epoch": 127.5,
+      "learning_rate": 1.0683760683760685e-06,
+      "loss": 0.1294,
+      "step": 6630
+    },
+    {
+      "epoch": 127.69,
+      "learning_rate": 9.861932938856016e-07,
+      "loss": 0.1319,
+      "step": 6640
+    },
+    {
+      "epoch": 127.88,
+      "learning_rate": 9.040105193951348e-07,
+      "loss": 0.1678,
+      "step": 6650
+    },
+    {
+      "epoch": 127.99,
+      "eval_accuracy": 0.891566265060241,
+      "eval_loss": 0.5079752802848816,
+      "eval_runtime": 14.4137,
+      "eval_samples_per_second": 51.826,
+      "eval_steps_per_second": 1.665,
+      "step": 6656
+    },
+    {
+      "epoch": 128.08,
+      "learning_rate": 8.21827744904668e-07,
+      "loss": 0.1269,
+      "step": 6660
+    },
+    {
+      "epoch": 128.27,
+      "learning_rate": 7.396449704142012e-07,
+      "loss": 0.1452,
+      "step": 6670
+    },
+    {
+      "epoch": 128.46,
+      "learning_rate": 6.574621959237345e-07,
+      "loss": 0.1304,
+      "step": 6680
+    },
+    {
+      "epoch": 128.65,
+      "learning_rate": 5.752794214332676e-07,
+      "loss": 0.1292,
+      "step": 6690
+    },
+    {
+      "epoch": 128.84,
+      "learning_rate": 4.930966469428008e-07,
+      "loss": 0.1444,
+      "step": 6700
+    },
+    {
+      "epoch": 128.99,
+      "eval_accuracy": 0.8902275769745649,
+      "eval_loss": 0.5113563537597656,
+      "eval_runtime": 14.4827,
+      "eval_samples_per_second": 51.579,
+      "eval_steps_per_second": 1.657,
+      "step": 6708
+    },
+    {
+      "epoch": 129.04,
+      "learning_rate": 4.10913872452334e-07,
+      "loss": 0.1478,
+      "step": 6710
+    },
+    {
+      "epoch": 129.23,
+      "learning_rate": 3.2873109796186723e-07,
+      "loss": 0.1669,
+      "step": 6720
+    },
+    {
+      "epoch": 129.42,
+      "learning_rate": 2.465483234714004e-07,
+      "loss": 0.1218,
+      "step": 6730
+    },
+    {
+      "epoch": 129.61,
+      "learning_rate": 1.6436554898093362e-07,
+      "loss": 0.1202,
+      "step": 6740
+    },
+    {
+      "epoch": 129.8,
+      "learning_rate": 8.218277449046681e-08,
+      "loss": 0.1318,
+      "step": 6750
+    },
+    {
+      "epoch": 129.99,
+      "learning_rate": 0.0,
+      "loss": 0.1282,
+      "step": 6760
+    },
+    {
+      "epoch": 129.99,
+      "eval_accuracy": 0.8888888888888888,
+      "eval_loss": 0.5223875045776367,
+      "eval_runtime": 14.5035,
+      "eval_samples_per_second": 51.505,
+      "eval_steps_per_second": 1.655,
+      "step": 6760
+    },
+    {
+      "epoch": 129.99,
+      "step": 6760,
+      "total_flos": 2.1712601194748707e+19,
+      "train_loss": 0.2964387069439747,
+      "train_runtime": 31374.7564,
+      "train_samples_per_second": 27.844,
+      "train_steps_per_second": 0.215
+    }
+  ],
+  "max_steps": 6760,
+  "num_train_epochs": 130,
+  "total_flos": 2.1712601194748707e+19,
+  "trial_name": null,
+  "trial_params": null
+}