diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,18016 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 10.0,
+  "global_step": 24000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 1e-06,
+      "loss": 1.3946,
+      "step": 20
+    },
+    {
+      "epoch": 0.01,
+      "eval_accuracy": 0.24016666666666667,
+      "eval_loss": 1.3855726718902588,
+      "eval_runtime": 15.834,
+      "eval_samples_per_second": 1515.727,
+      "eval_steps_per_second": 4.737,
+      "step": 20
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 1e-06,
+      "loss": 1.3801,
+      "step": 40
+    },
+    {
+      "epoch": 0.02,
+      "eval_accuracy": 0.2665416666666667,
+      "eval_loss": 1.3742554187774658,
+      "eval_runtime": 15.8222,
+      "eval_samples_per_second": 1516.855,
+      "eval_steps_per_second": 4.74,
+      "step": 40
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 1e-06,
+      "loss": 1.37,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "eval_accuracy": 0.299875,
+      "eval_loss": 1.362697958946228,
+      "eval_runtime": 15.4839,
+      "eval_samples_per_second": 1549.994,
+      "eval_steps_per_second": 4.844,
+      "step": 60
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 1e-06,
+      "loss": 1.3559,
+      "step": 80
+    },
+    {
+      "epoch": 0.03,
+      "eval_accuracy": 0.3585,
+      "eval_loss": 1.3499468564987183,
+      "eval_runtime": 15.9453,
+      "eval_samples_per_second": 1505.143,
+      "eval_steps_per_second": 4.704,
+      "step": 80
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 1e-06,
+      "loss": 1.3425,
+      "step": 100
+    },
+    {
+      "epoch": 0.04,
+      "eval_accuracy": 0.43916666666666665,
+      "eval_loss": 1.3347731828689575,
+      "eval_runtime": 15.6185,
+      "eval_samples_per_second": 1536.635,
+      "eval_steps_per_second": 4.802,
+      "step": 100
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1e-06,
+      "loss": 1.326,
+      "step": 120
+    },
+    {
+      "epoch": 0.05,
+      "eval_accuracy": 0.5287083333333333,
+      "eval_loss": 1.3159666061401367,
+      "eval_runtime": 16.2205,
+      "eval_samples_per_second": 1479.607,
+      "eval_steps_per_second": 4.624,
+      "step": 120
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 1e-06,
+      "loss": 1.3043,
+      "step": 140
+    },
+    {
+      "epoch": 0.06,
+      "eval_accuracy": 0.58625,
+      "eval_loss": 1.2926729917526245,
+      "eval_runtime": 15.8692,
+      "eval_samples_per_second": 1512.368,
+      "eval_steps_per_second": 4.726,
+      "step": 140
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1e-06,
+      "loss": 1.283,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "eval_accuracy": 0.6705833333333333,
+      "eval_loss": 1.2648732662200928,
+      "eval_runtime": 15.5702,
+      "eval_samples_per_second": 1541.406,
+      "eval_steps_per_second": 4.817,
+      "step": 160
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1e-06,
+      "loss": 1.2501,
+      "step": 180
+    },
+    {
+      "epoch": 0.07,
+      "eval_accuracy": 0.7289583333333334,
+      "eval_loss": 1.2298475503921509,
+      "eval_runtime": 15.3558,
+      "eval_samples_per_second": 1562.932,
+      "eval_steps_per_second": 4.884,
+      "step": 180
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 1e-06,
+      "loss": 1.2218,
+      "step": 200
+    },
+    {
+      "epoch": 0.08,
+      "eval_accuracy": 0.7775,
+      "eval_loss": 1.1907247304916382,
+      "eval_runtime": 15.6618,
+      "eval_samples_per_second": 1532.387,
+      "eval_steps_per_second": 4.789,
+      "step": 200
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1e-06,
+      "loss": 1.1819,
+      "step": 220
+    },
+    {
+      "epoch": 0.09,
+      "eval_accuracy": 0.8095833333333333,
+      "eval_loss": 1.1489461660385132,
+      "eval_runtime": 16.7595,
+      "eval_samples_per_second": 1432.027,
+      "eval_steps_per_second": 4.475,
+      "step": 220
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1e-06,
+      "loss": 1.1375,
+      "step": 240
+    },
+    {
+      "epoch": 0.1,
+      "eval_accuracy": 0.8254583333333333,
+      "eval_loss": 1.1064685583114624,
+      "eval_runtime": 15.9246,
+      "eval_samples_per_second": 1507.098,
+      "eval_steps_per_second": 4.71,
+      "step": 240
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 1e-06,
+      "loss": 1.0918,
+      "step": 260
+    },
+    {
+      "epoch": 0.11,
+      "eval_accuracy": 0.83575,
+      "eval_loss": 1.0633411407470703,
+      "eval_runtime": 15.4403,
+      "eval_samples_per_second": 1554.376,
+      "eval_steps_per_second": 4.857,
+      "step": 260
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1e-06,
+      "loss": 1.0493,
+      "step": 280
+    },
+    {
+      "epoch": 0.12,
+      "eval_accuracy": 0.8455833333333334,
+      "eval_loss": 1.0193486213684082,
+      "eval_runtime": 15.5383,
+      "eval_samples_per_second": 1544.57,
+      "eval_steps_per_second": 4.827,
+      "step": 280
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1e-06,
+      "loss": 1.0137,
+      "step": 300
+    },
+    {
+      "epoch": 0.12,
+      "eval_accuracy": 0.8547916666666666,
+      "eval_loss": 0.9759964942932129,
+      "eval_runtime": 15.7415,
+      "eval_samples_per_second": 1524.631,
+      "eval_steps_per_second": 4.764,
+      "step": 300
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 1e-06,
+      "loss": 0.9799,
+      "step": 320
+    },
+    {
+      "epoch": 0.13,
+      "eval_accuracy": 0.856875,
+      "eval_loss": 0.9349467754364014,
+      "eval_runtime": 15.9509,
+      "eval_samples_per_second": 1504.613,
+      "eval_steps_per_second": 4.702,
+      "step": 320
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1e-06,
+      "loss": 0.9269,
+      "step": 340
+    },
+    {
+      "epoch": 0.14,
+      "eval_accuracy": 0.862125,
+      "eval_loss": 0.8932501077651978,
+      "eval_runtime": 15.5524,
+      "eval_samples_per_second": 1543.167,
+      "eval_steps_per_second": 4.822,
+      "step": 340
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 1e-06,
+      "loss": 0.8982,
+      "step": 360
+    },
+    {
+      "epoch": 0.15,
+      "eval_accuracy": 0.866125,
+      "eval_loss": 0.8538751006126404,
+      "eval_runtime": 16.1674,
+      "eval_samples_per_second": 1484.473,
+      "eval_steps_per_second": 4.639,
+      "step": 360
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1e-06,
+      "loss": 0.854,
+      "step": 380
+    },
+    {
+      "epoch": 0.16,
+      "eval_accuracy": 0.8698333333333333,
+      "eval_loss": 0.8160046935081482,
+      "eval_runtime": 15.675,
+      "eval_samples_per_second": 1531.105,
+      "eval_steps_per_second": 4.785,
+      "step": 380
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1e-06,
+      "loss": 0.8087,
+      "step": 400
+    },
+    {
+      "epoch": 0.17,
+      "eval_accuracy": 0.8705416666666667,
+      "eval_loss": 0.7807948589324951,
+      "eval_runtime": 15.6603,
+      "eval_samples_per_second": 1532.542,
+      "eval_steps_per_second": 4.789,
+      "step": 400
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1e-06,
+      "loss": 0.7941,
+      "step": 420
+    },
+    {
+      "epoch": 0.17,
+      "eval_accuracy": 0.8717916666666666,
+      "eval_loss": 0.7479580044746399,
+      "eval_runtime": 15.5761,
+      "eval_samples_per_second": 1540.822,
+      "eval_steps_per_second": 4.815,
+      "step": 420
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 1e-06,
+      "loss": 0.7443,
+      "step": 440
+    },
+    {
+      "epoch": 0.18,
+      "eval_accuracy": 0.872,
+      "eval_loss": 0.7175572514533997,
+      "eval_runtime": 15.9105,
+      "eval_samples_per_second": 1508.436,
+      "eval_steps_per_second": 4.714,
+      "step": 440
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1e-06,
+      "loss": 0.7233,
+      "step": 460
+    },
+    {
+      "epoch": 0.19,
+      "eval_accuracy": 0.87075,
+      "eval_loss": 0.6889370679855347,
+      "eval_runtime": 16.1633,
+      "eval_samples_per_second": 1484.847,
+      "eval_steps_per_second": 4.64,
+      "step": 460
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 1e-06,
+      "loss": 0.6985,
+      "step": 480
+    },
+    {
+      "epoch": 0.2,
+      "eval_accuracy": 0.8758333333333334,
+      "eval_loss": 0.6611708402633667,
+      "eval_runtime": 15.5607,
+      "eval_samples_per_second": 1542.351,
+      "eval_steps_per_second": 4.82,
+      "step": 480
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1e-06,
+      "loss": 0.6754,
+      "step": 500
+    },
+    {
+      "epoch": 0.21,
+      "eval_accuracy": 0.8768333333333334,
+      "eval_loss": 0.635998547077179,
+      "eval_runtime": 15.8738,
+      "eval_samples_per_second": 1511.926,
+      "eval_steps_per_second": 4.725,
+      "step": 500
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 1e-06,
+      "loss": 0.6536,
+      "step": 520
+    },
+    {
+      "epoch": 0.22,
+      "eval_accuracy": 0.8788333333333334,
+      "eval_loss": 0.6131625175476074,
+      "eval_runtime": 15.5202,
+      "eval_samples_per_second": 1546.371,
+      "eval_steps_per_second": 4.832,
+      "step": 520
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1e-06,
+      "loss": 0.614,
+      "step": 540
+    },
+    {
+      "epoch": 0.23,
+      "eval_accuracy": 0.8811666666666667,
+      "eval_loss": 0.5912534594535828,
+      "eval_runtime": 16.1599,
+      "eval_samples_per_second": 1485.154,
+      "eval_steps_per_second": 4.641,
+      "step": 540
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1e-06,
+      "loss": 0.5963,
+      "step": 560
+    },
+    {
+      "epoch": 0.23,
+      "eval_accuracy": 0.8809583333333333,
+      "eval_loss": 0.5706831216812134,
+      "eval_runtime": 15.5035,
+      "eval_samples_per_second": 1548.034,
+      "eval_steps_per_second": 4.838,
+      "step": 560
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1e-06,
+      "loss": 0.5894,
+      "step": 580
+    },
+    {
+      "epoch": 0.24,
+      "eval_accuracy": 0.8815416666666667,
+      "eval_loss": 0.5526705980300903,
+      "eval_runtime": 15.515,
+      "eval_samples_per_second": 1546.894,
+      "eval_steps_per_second": 4.834,
+      "step": 580
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 1e-06,
+      "loss": 0.5504,
+      "step": 600
+    },
+    {
+      "epoch": 0.25,
+      "eval_accuracy": 0.8835416666666667,
+      "eval_loss": 0.5349943041801453,
+      "eval_runtime": 15.5986,
+      "eval_samples_per_second": 1538.598,
+      "eval_steps_per_second": 4.808,
+      "step": 600
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1e-06,
+      "loss": 0.5434,
+      "step": 620
+    },
+    {
+      "epoch": 0.26,
+      "eval_accuracy": 0.8850833333333333,
+      "eval_loss": 0.5199493169784546,
+      "eval_runtime": 15.5783,
+      "eval_samples_per_second": 1540.605,
+      "eval_steps_per_second": 4.814,
+      "step": 620
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1e-06,
+      "loss": 0.5325,
+      "step": 640
+    },
+    {
+      "epoch": 0.27,
+      "eval_accuracy": 0.8842083333333334,
+      "eval_loss": 0.5056445598602295,
+      "eval_runtime": 16.1601,
+      "eval_samples_per_second": 1485.138,
+      "eval_steps_per_second": 4.641,
+      "step": 640
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1e-06,
+      "loss": 0.4894,
+      "step": 660
+    },
+    {
+      "epoch": 0.28,
+      "eval_accuracy": 0.8850416666666666,
+      "eval_loss": 0.4912012815475464,
+      "eval_runtime": 15.5535,
+      "eval_samples_per_second": 1543.06,
+      "eval_steps_per_second": 4.822,
+      "step": 660
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1e-06,
+      "loss": 0.4955,
+      "step": 680
+    },
+    {
+      "epoch": 0.28,
+      "eval_accuracy": 0.8854166666666666,
+      "eval_loss": 0.4787778854370117,
+      "eval_runtime": 15.9742,
+      "eval_samples_per_second": 1502.42,
+      "eval_steps_per_second": 4.695,
+      "step": 680
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1e-06,
+      "loss": 0.5002,
+      "step": 700
+    },
+    {
+      "epoch": 0.29,
+      "eval_accuracy": 0.8854166666666666,
+      "eval_loss": 0.4684358835220337,
+      "eval_runtime": 15.4896,
+      "eval_samples_per_second": 1549.429,
+      "eval_steps_per_second": 4.842,
+      "step": 700
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 1e-06,
+      "loss": 0.4621,
+      "step": 720
+    },
+    {
+      "epoch": 0.3,
+      "eval_accuracy": 0.8866666666666667,
+      "eval_loss": 0.4581121802330017,
+      "eval_runtime": 15.7197,
+      "eval_samples_per_second": 1526.742,
+      "eval_steps_per_second": 4.771,
+      "step": 720
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1e-06,
+      "loss": 0.4873,
+      "step": 740
+    },
+    {
+      "epoch": 0.31,
+      "eval_accuracy": 0.8860833333333333,
+      "eval_loss": 0.4509838819503784,
+      "eval_runtime": 15.7293,
+      "eval_samples_per_second": 1525.818,
+      "eval_steps_per_second": 4.768,
+      "step": 740
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 1e-06,
+      "loss": 0.4575,
+      "step": 760
+    },
+    {
+      "epoch": 0.32,
+      "eval_accuracy": 0.88875,
+      "eval_loss": 0.4410632252693176,
+      "eval_runtime": 15.4276,
+      "eval_samples_per_second": 1555.649,
+      "eval_steps_per_second": 4.861,
+      "step": 760
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1e-06,
+      "loss": 0.4365,
+      "step": 780
+    },
+    {
+      "epoch": 0.33,
+      "eval_accuracy": 0.8889583333333333,
+      "eval_loss": 0.4337460398674011,
+      "eval_runtime": 15.9845,
+      "eval_samples_per_second": 1501.455,
+      "eval_steps_per_second": 4.692,
+      "step": 780
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1e-06,
+      "loss": 0.4306,
+      "step": 800
+    },
+    {
+      "epoch": 0.33,
+      "eval_accuracy": 0.8873333333333333,
+      "eval_loss": 0.42743387818336487,
+      "eval_runtime": 15.705,
+      "eval_samples_per_second": 1528.179,
+      "eval_steps_per_second": 4.776,
+      "step": 800
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 1e-06,
+      "loss": 0.4367,
+      "step": 820
+    },
+    {
+      "epoch": 0.34,
+      "eval_accuracy": 0.889625,
+      "eval_loss": 0.41997990012168884,
+      "eval_runtime": 15.2578,
+      "eval_samples_per_second": 1572.97,
+      "eval_steps_per_second": 4.916,
+      "step": 820
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1e-06,
+      "loss": 0.4643,
+      "step": 840
+    },
+    {
+      "epoch": 0.35,
+      "eval_accuracy": 0.8909583333333333,
+      "eval_loss": 0.41470062732696533,
+      "eval_runtime": 15.4309,
+      "eval_samples_per_second": 1555.323,
+      "eval_steps_per_second": 4.86,
+      "step": 840
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1e-06,
+      "loss": 0.4483,
+      "step": 860
+    },
+    {
+      "epoch": 0.36,
+      "eval_accuracy": 0.8922916666666667,
+      "eval_loss": 0.40827029943466187,
+      "eval_runtime": 15.9503,
+      "eval_samples_per_second": 1504.673,
+      "eval_steps_per_second": 4.702,
+      "step": 860
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 1e-06,
+      "loss": 0.3937,
+      "step": 880
+    },
+    {
+      "epoch": 0.37,
+      "eval_accuracy": 0.8928333333333334,
+      "eval_loss": 0.40155351161956787,
+      "eval_runtime": 15.6448,
+      "eval_samples_per_second": 1534.053,
+      "eval_steps_per_second": 4.794,
+      "step": 880
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1e-06,
+      "loss": 0.4074,
+      "step": 900
+    },
+    {
+      "epoch": 0.38,
+      "eval_accuracy": 0.89125,
+      "eval_loss": 0.39755621552467346,
+      "eval_runtime": 16.1245,
+      "eval_samples_per_second": 1488.416,
+      "eval_steps_per_second": 4.651,
+      "step": 900
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1e-06,
+      "loss": 0.4153,
+      "step": 920
+    },
+    {
+      "epoch": 0.38,
+      "eval_accuracy": 0.891125,
+      "eval_loss": 0.39380690455436707,
+      "eval_runtime": 15.6816,
+      "eval_samples_per_second": 1530.453,
+      "eval_steps_per_second": 4.783,
+      "step": 920
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 1e-06,
+      "loss": 0.3934,
+      "step": 940
+    },
+    {
+      "epoch": 0.39,
+      "eval_accuracy": 0.8914166666666666,
+      "eval_loss": 0.3885686695575714,
+      "eval_runtime": 15.0444,
+      "eval_samples_per_second": 1595.283,
+      "eval_steps_per_second": 4.985,
+      "step": 940
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1e-06,
+      "loss": 0.4486,
+      "step": 960
+    },
+    {
+      "epoch": 0.4,
+      "eval_accuracy": 0.8915,
+      "eval_loss": 0.3852112293243408,
+      "eval_runtime": 15.3141,
+      "eval_samples_per_second": 1567.186,
+      "eval_steps_per_second": 4.897,
+      "step": 960
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1e-06,
+      "loss": 0.3908,
+      "step": 980
+    },
+    {
+      "epoch": 0.41,
+      "eval_accuracy": 0.8928333333333334,
+      "eval_loss": 0.3800646662712097,
+      "eval_runtime": 15.4314,
+      "eval_samples_per_second": 1555.271,
+      "eval_steps_per_second": 4.86,
+      "step": 980
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1e-06,
+      "loss": 0.3888,
+      "step": 1000
+    },
+    {
+      "epoch": 0.42,
+      "eval_accuracy": 0.894375,
+      "eval_loss": 0.3761395514011383,
+      "eval_runtime": 15.8671,
+      "eval_samples_per_second": 1512.567,
+      "eval_steps_per_second": 4.727,
+      "step": 1000
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1e-06,
+      "loss": 0.3801,
+      "step": 1020
+    },
+    {
+      "epoch": 0.42,
+      "eval_accuracy": 0.8949583333333333,
+      "eval_loss": 0.3725646734237671,
+      "eval_runtime": 15.3688,
+      "eval_samples_per_second": 1561.603,
+      "eval_steps_per_second": 4.88,
+      "step": 1020
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1e-06,
+      "loss": 0.387,
+      "step": 1040
+    },
+    {
+      "epoch": 0.43,
+      "eval_accuracy": 0.894125,
+      "eval_loss": 0.3705978989601135,
+      "eval_runtime": 15.7377,
+      "eval_samples_per_second": 1525.005,
+      "eval_steps_per_second": 4.766,
+      "step": 1040
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 1e-06,
+      "loss": 0.4346,
+      "step": 1060
+    },
+    {
+      "epoch": 0.44,
+      "eval_accuracy": 0.8947916666666667,
+      "eval_loss": 0.367171049118042,
+      "eval_runtime": 14.9848,
+      "eval_samples_per_second": 1601.62,
+      "eval_steps_per_second": 5.005,
+      "step": 1060
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1e-06,
+      "loss": 0.3806,
+      "step": 1080
+    },
+    {
+      "epoch": 0.45,
+      "eval_accuracy": 0.8936666666666667,
+      "eval_loss": 0.3656676709651947,
+      "eval_runtime": 15.8929,
+      "eval_samples_per_second": 1510.112,
+      "eval_steps_per_second": 4.719,
+      "step": 1080
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 1e-06,
+      "loss": 0.3633,
+      "step": 1100
+    },
+    {
+      "epoch": 0.46,
+      "eval_accuracy": 0.8948333333333334,
+      "eval_loss": 0.3619757294654846,
+      "eval_runtime": 15.254,
+      "eval_samples_per_second": 1573.363,
+      "eval_steps_per_second": 4.917,
+      "step": 1100
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1e-06,
+      "loss": 0.3429,
+      "step": 1120
+    },
+    {
+      "epoch": 0.47,
+      "eval_accuracy": 0.8954583333333334,
+      "eval_loss": 0.35918018221855164,
+      "eval_runtime": 15.6719,
+      "eval_samples_per_second": 1531.4,
+      "eval_steps_per_second": 4.786,
+      "step": 1120
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1e-06,
+      "loss": 0.3681,
+      "step": 1140
+    },
+    {
+      "epoch": 0.47,
+      "eval_accuracy": 0.896625,
+      "eval_loss": 0.3563116788864136,
+      "eval_runtime": 16.068,
+      "eval_samples_per_second": 1493.649,
+      "eval_steps_per_second": 4.668,
+      "step": 1140
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1e-06,
+      "loss": 0.3624,
+      "step": 1160
+    },
+    {
+      "epoch": 0.48,
+      "eval_accuracy": 0.8964583333333334,
+      "eval_loss": 0.35381361842155457,
+      "eval_runtime": 15.992,
+      "eval_samples_per_second": 1500.748,
+      "eval_steps_per_second": 4.69,
+      "step": 1160
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 1e-06,
+      "loss": 0.3808,
+      "step": 1180
+    },
+    {
+      "epoch": 0.49,
+      "eval_accuracy": 0.895625,
+      "eval_loss": 0.35254761576652527,
+      "eval_runtime": 16.151,
+      "eval_samples_per_second": 1485.976,
+      "eval_steps_per_second": 4.644,
+      "step": 1180
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1e-06,
+      "loss": 0.3626,
+      "step": 1200
+    },
+    {
+      "epoch": 0.5,
+      "eval_accuracy": 0.897,
+      "eval_loss": 0.34913721680641174,
+      "eval_runtime": 15.769,
+      "eval_samples_per_second": 1521.97,
+      "eval_steps_per_second": 4.756,
+      "step": 1200
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 1e-06,
+      "loss": 0.3278,
+      "step": 1220
+    },
+    {
+      "epoch": 0.51,
+      "eval_accuracy": 0.897375,
+      "eval_loss": 0.3466154932975769,
+      "eval_runtime": 16.1468,
+      "eval_samples_per_second": 1486.363,
+      "eval_steps_per_second": 4.645,
+      "step": 1220
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 1e-06,
+      "loss": 0.3463,
+      "step": 1240
+    },
+    {
+      "epoch": 0.52,
+      "eval_accuracy": 0.8966666666666666,
+      "eval_loss": 0.34570518136024475,
+      "eval_runtime": 15.3513,
+      "eval_samples_per_second": 1563.389,
+      "eval_steps_per_second": 4.886,
+      "step": 1240
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1e-06,
+      "loss": 0.371,
+      "step": 1260
+    },
+    {
+      "epoch": 0.53,
+      "eval_accuracy": 0.8975416666666667,
+      "eval_loss": 0.3436819016933441,
+      "eval_runtime": 15.6461,
+      "eval_samples_per_second": 1533.926,
+      "eval_steps_per_second": 4.794,
+      "step": 1260
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 1e-06,
+      "loss": 0.326,
+      "step": 1280
+    },
+    {
+      "epoch": 0.53,
+      "eval_accuracy": 0.8983333333333333,
+      "eval_loss": 0.3413917124271393,
+      "eval_runtime": 15.3084,
+      "eval_samples_per_second": 1567.764,
+      "eval_steps_per_second": 4.899,
+      "step": 1280
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 1e-06,
+      "loss": 0.3639,
+      "step": 1300
+    },
+    {
+      "epoch": 0.54,
+      "eval_accuracy": 0.898,
+      "eval_loss": 0.33931395411491394,
+      "eval_runtime": 15.4706,
+      "eval_samples_per_second": 1551.329,
+      "eval_steps_per_second": 4.848,
+      "step": 1300
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 1e-06,
+      "loss": 0.3537,
+      "step": 1320
+    },
+    {
+      "epoch": 0.55,
+      "eval_accuracy": 0.8990416666666666,
+      "eval_loss": 0.33674728870391846,
+      "eval_runtime": 15.8671,
+      "eval_samples_per_second": 1512.563,
+      "eval_steps_per_second": 4.727,
+      "step": 1320
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 1e-06,
+      "loss": 0.3831,
+      "step": 1340
+    },
+    {
+      "epoch": 0.56,
+      "eval_accuracy": 0.898875,
+      "eval_loss": 0.3362838625907898,
+      "eval_runtime": 15.4851,
+      "eval_samples_per_second": 1549.872,
+      "eval_steps_per_second": 4.843,
+      "step": 1340
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1e-06,
+      "loss": 0.3559,
+      "step": 1360
+    },
+    {
+      "epoch": 0.57,
+      "eval_accuracy": 0.8994166666666666,
+      "eval_loss": 0.3338731527328491,
+      "eval_runtime": 15.8107,
+      "eval_samples_per_second": 1517.958,
+      "eval_steps_per_second": 4.744,
+      "step": 1360
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 1e-06,
+      "loss": 0.3706,
+      "step": 1380
+    },
+    {
+      "epoch": 0.57,
+      "eval_accuracy": 0.899375,
+      "eval_loss": 0.3323739767074585,
+      "eval_runtime": 15.3746,
+      "eval_samples_per_second": 1561.014,
+      "eval_steps_per_second": 4.878,
+      "step": 1380
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 1e-06,
+      "loss": 0.374,
+      "step": 1400
+    },
+    {
+      "epoch": 0.58,
+      "eval_accuracy": 0.8998333333333334,
+      "eval_loss": 0.33071157336235046,
+      "eval_runtime": 16.2242,
+      "eval_samples_per_second": 1479.273,
+      "eval_steps_per_second": 4.623,
+      "step": 1400
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 1e-06,
+      "loss": 0.3515,
+      "step": 1420
+    },
+    {
+      "epoch": 0.59,
+      "eval_accuracy": 0.8999583333333333,
+      "eval_loss": 0.3292597532272339,
+      "eval_runtime": 15.6689,
+      "eval_samples_per_second": 1531.696,
+      "eval_steps_per_second": 4.787,
+      "step": 1420
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 1e-06,
+      "loss": 0.3854,
+      "step": 1440
+    },
+    {
+      "epoch": 0.6,
+      "eval_accuracy": 0.8996666666666666,
+      "eval_loss": 0.32925406098365784,
+      "eval_runtime": 15.8799,
+      "eval_samples_per_second": 1511.341,
+      "eval_steps_per_second": 4.723,
+      "step": 1440
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 1e-06,
+      "loss": 0.3363,
+      "step": 1460
+    },
+    {
+      "epoch": 0.61,
+      "eval_accuracy": 0.8998333333333334,
+      "eval_loss": 0.327946275472641,
+      "eval_runtime": 15.54,
+      "eval_samples_per_second": 1544.404,
+      "eval_steps_per_second": 4.826,
+      "step": 1460
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1e-06,
+      "loss": 0.3199,
+      "step": 1480
+    },
+    {
+      "epoch": 0.62,
+      "eval_accuracy": 0.900625,
+      "eval_loss": 0.32561612129211426,
+      "eval_runtime": 15.692,
+      "eval_samples_per_second": 1529.44,
+      "eval_steps_per_second": 4.779,
+      "step": 1480
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 1e-06,
+      "loss": 0.356,
+      "step": 1500
+    },
+    {
+      "epoch": 0.62,
+      "eval_accuracy": 0.9007916666666667,
+      "eval_loss": 0.324897825717926,
+      "eval_runtime": 15.6415,
+      "eval_samples_per_second": 1534.381,
+      "eval_steps_per_second": 4.795,
+      "step": 1500
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 1e-06,
+      "loss": 0.3566,
+      "step": 1520
+    },
+    {
+      "epoch": 0.63,
+      "eval_accuracy": 0.9013333333333333,
+      "eval_loss": 0.32291579246520996,
+      "eval_runtime": 16.0404,
+      "eval_samples_per_second": 1496.219,
+      "eval_steps_per_second": 4.676,
+      "step": 1520
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1e-06,
+      "loss": 0.3419,
+      "step": 1540
+    },
+    {
+      "epoch": 0.64,
+      "eval_accuracy": 0.9010833333333333,
+      "eval_loss": 0.3221406042575836,
+      "eval_runtime": 15.5217,
+      "eval_samples_per_second": 1546.222,
+      "eval_steps_per_second": 4.832,
+      "step": 1540
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 1e-06,
+      "loss": 0.3453,
+      "step": 1560
+    },
+    {
+      "epoch": 0.65,
+      "eval_accuracy": 0.9005833333333333,
+      "eval_loss": 0.32188284397125244,
+      "eval_runtime": 15.8978,
+      "eval_samples_per_second": 1509.639,
+      "eval_steps_per_second": 4.718,
+      "step": 1560
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 1e-06,
+      "loss": 0.3132,
+      "step": 1580
+    },
+    {
+      "epoch": 0.66,
+      "eval_accuracy": 0.901875,
+      "eval_loss": 0.31994542479515076,
+      "eval_runtime": 15.5539,
+      "eval_samples_per_second": 1543.025,
+      "eval_steps_per_second": 4.822,
+      "step": 1580
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 1e-06,
+      "loss": 0.3457,
+      "step": 1600
+    },
+    {
+      "epoch": 0.67,
+      "eval_accuracy": 0.9012083333333333,
+      "eval_loss": 0.32011812925338745,
+      "eval_runtime": 15.7109,
+      "eval_samples_per_second": 1527.598,
+      "eval_steps_per_second": 4.774,
+      "step": 1600
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1e-06,
+      "loss": 0.3036,
+      "step": 1620
+    },
+    {
+      "epoch": 0.68,
+      "eval_accuracy": 0.9016666666666666,
+      "eval_loss": 0.31820276379585266,
+      "eval_runtime": 15.5575,
+      "eval_samples_per_second": 1542.661,
+      "eval_steps_per_second": 4.821,
+      "step": 1620
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 1e-06,
+      "loss": 0.3359,
+      "step": 1640
+    },
+    {
+      "epoch": 0.68,
+      "eval_accuracy": 0.902125,
+      "eval_loss": 0.316621333360672,
+      "eval_runtime": 16.0532,
+      "eval_samples_per_second": 1495.031,
+      "eval_steps_per_second": 4.672,
+      "step": 1640
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 1e-06,
+      "loss": 0.3291,
+      "step": 1660
+    },
+    {
+      "epoch": 0.69,
+      "eval_accuracy": 0.9020833333333333,
+      "eval_loss": 0.31562095880508423,
+      "eval_runtime": 15.7321,
+      "eval_samples_per_second": 1525.545,
+      "eval_steps_per_second": 4.767,
+      "step": 1660
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1e-06,
+      "loss": 0.3586,
+      "step": 1680
+    },
+    {
+      "epoch": 0.7,
+      "eval_accuracy": 0.9005833333333333,
+      "eval_loss": 0.3162485659122467,
+      "eval_runtime": 15.5682,
+      "eval_samples_per_second": 1541.608,
+      "eval_steps_per_second": 4.818,
+      "step": 1680
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 1e-06,
+      "loss": 0.3002,
+      "step": 1700
+    },
+    {
+      "epoch": 0.71,
+      "eval_accuracy": 0.9013333333333333,
+      "eval_loss": 0.31555384397506714,
+      "eval_runtime": 15.6458,
+      "eval_samples_per_second": 1533.959,
+      "eval_steps_per_second": 4.794,
+      "step": 1700
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1e-06,
+      "loss": 0.3743,
+      "step": 1720
+    },
+    {
+      "epoch": 0.72,
+      "eval_accuracy": 0.9024583333333334,
+      "eval_loss": 0.3135637044906616,
+      "eval_runtime": 15.9237,
+      "eval_samples_per_second": 1507.184,
+      "eval_steps_per_second": 4.71,
+      "step": 1720
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 1e-06,
+      "loss": 0.3506,
+      "step": 1740
+    },
+    {
+      "epoch": 0.72,
+      "eval_accuracy": 0.9029166666666667,
+      "eval_loss": 0.3118599057197571,
+      "eval_runtime": 15.6284,
+      "eval_samples_per_second": 1535.667,
+      "eval_steps_per_second": 4.799,
+      "step": 1740
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 1e-06,
+      "loss": 0.3328,
+      "step": 1760
+    },
+    {
+      "epoch": 0.73,
+      "eval_accuracy": 0.9022083333333333,
+      "eval_loss": 0.31298351287841797,
+      "eval_runtime": 15.7141,
+      "eval_samples_per_second": 1527.295,
+      "eval_steps_per_second": 4.773,
+      "step": 1760
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 1e-06,
+      "loss": 0.328,
+      "step": 1780
+    },
+    {
+      "epoch": 0.74,
+      "eval_accuracy": 0.9025416666666667,
+      "eval_loss": 0.31090086698532104,
+      "eval_runtime": 16.0505,
+      "eval_samples_per_second": 1495.277,
+      "eval_steps_per_second": 4.673,
+      "step": 1780
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 1e-06,
+      "loss": 0.3312,
+      "step": 1800
+    },
+    {
+      "epoch": 0.75,
+      "eval_accuracy": 0.9022083333333333,
+      "eval_loss": 0.31056439876556396,
+      "eval_runtime": 15.8194,
+      "eval_samples_per_second": 1517.124,
+      "eval_steps_per_second": 4.741,
+      "step": 1800
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 1e-06,
+      "loss": 0.3176,
+      "step": 1820
+    },
+    {
+      "epoch": 0.76,
+      "eval_accuracy": 0.9030833333333333,
+      "eval_loss": 0.3086493909358978,
+      "eval_runtime": 15.8105,
+      "eval_samples_per_second": 1517.979,
+      "eval_steps_per_second": 4.744,
+      "step": 1820
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 1e-06,
+      "loss": 0.326,
+      "step": 1840
+    },
+    {
+      "epoch": 0.77,
+      "eval_accuracy": 0.9035416666666667,
+      "eval_loss": 0.30771201848983765,
+      "eval_runtime": 15.6087,
+      "eval_samples_per_second": 1537.601,
+      "eval_steps_per_second": 4.805,
+      "step": 1840
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1e-06,
+      "loss": 0.3308,
+      "step": 1860
+    },
+    {
+      "epoch": 0.78,
+      "eval_accuracy": 0.9033333333333333,
+      "eval_loss": 0.30720219016075134,
+      "eval_runtime": 15.7737,
+      "eval_samples_per_second": 1521.525,
+      "eval_steps_per_second": 4.755,
+      "step": 1860
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2876,
+      "step": 1880
+    },
+    {
+      "epoch": 0.78,
+      "eval_accuracy": 0.90275,
+      "eval_loss": 0.30769696831703186,
+      "eval_runtime": 16.0101,
+      "eval_samples_per_second": 1499.051,
+      "eval_steps_per_second": 4.685,
+      "step": 1880
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 1e-06,
+      "loss": 0.3326,
+      "step": 1900
+    },
+    {
+      "epoch": 0.79,
+      "eval_accuracy": 0.9028333333333334,
+      "eval_loss": 0.3074987530708313,
+      "eval_runtime": 15.7279,
+      "eval_samples_per_second": 1525.953,
+      "eval_steps_per_second": 4.769,
+      "step": 1900
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1e-06,
+      "loss": 0.2999,
+      "step": 1920
+    },
+    {
+      "epoch": 0.8,
+      "eval_accuracy": 0.9040416666666666,
+      "eval_loss": 0.3049142062664032,
+      "eval_runtime": 15.7379,
+      "eval_samples_per_second": 1524.978,
+      "eval_steps_per_second": 4.766,
+      "step": 1920
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1e-06,
+      "loss": 0.3207,
+      "step": 1940
+    },
+    {
+      "epoch": 0.81,
+      "eval_accuracy": 0.9042083333333333,
+      "eval_loss": 0.30403196811676025,
+      "eval_runtime": 16.0526,
+      "eval_samples_per_second": 1495.086,
+      "eval_steps_per_second": 4.672,
+      "step": 1940
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1e-06,
+      "loss": 0.3126,
+      "step": 1960
+    },
+    {
+      "epoch": 0.82,
+      "eval_accuracy": 0.903625,
+      "eval_loss": 0.3041483461856842,
+      "eval_runtime": 15.4127,
+      "eval_samples_per_second": 1557.161,
+      "eval_steps_per_second": 4.866,
+      "step": 1960
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 1e-06,
+      "loss": 0.3785,
+      "step": 1980
+    },
+    {
+      "epoch": 0.82,
+      "eval_accuracy": 0.9036666666666666,
+      "eval_loss": 0.30376511812210083,
+      "eval_runtime": 15.6455,
+      "eval_samples_per_second": 1533.984,
+      "eval_steps_per_second": 4.794,
+      "step": 1980
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1e-06,
+      "loss": 0.3015,
+      "step": 2000
+    },
+    {
+      "epoch": 0.83,
+      "eval_accuracy": 0.9044166666666666,
+      "eval_loss": 0.30233460664749146,
+      "eval_runtime": 15.6366,
+      "eval_samples_per_second": 1534.859,
+      "eval_steps_per_second": 4.796,
+      "step": 2000
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 1e-06,
+      "loss": 0.348,
+      "step": 2020
+    },
+    {
+      "epoch": 0.84,
+      "eval_accuracy": 0.9045833333333333,
+      "eval_loss": 0.30240559577941895,
+      "eval_runtime": 15.8212,
+      "eval_samples_per_second": 1516.949,
+      "eval_steps_per_second": 4.74,
+      "step": 2020
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1e-06,
+      "loss": 0.2947,
+      "step": 2040
+    },
+    {
+      "epoch": 0.85,
+      "eval_accuracy": 0.904875,
+      "eval_loss": 0.30133891105651855,
+      "eval_runtime": 15.8234,
+      "eval_samples_per_second": 1516.737,
+      "eval_steps_per_second": 4.74,
+      "step": 2040
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 1e-06,
+      "loss": 0.3344,
+      "step": 2060
+    },
+    {
+      "epoch": 0.86,
+      "eval_accuracy": 0.9049583333333333,
+      "eval_loss": 0.30154794454574585,
+      "eval_runtime": 15.2879,
+      "eval_samples_per_second": 1569.873,
+      "eval_steps_per_second": 4.906,
+      "step": 2060
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 1e-06,
+      "loss": 0.3436,
+      "step": 2080
+    },
+    {
+      "epoch": 0.87,
+      "eval_accuracy": 0.9045833333333333,
+      "eval_loss": 0.29952293634414673,
+      "eval_runtime": 16.2949,
+      "eval_samples_per_second": 1472.856,
+      "eval_steps_per_second": 4.603,
+      "step": 2080
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1e-06,
+      "loss": 0.3179,
+      "step": 2100
+    },
+    {
+      "epoch": 0.88,
+      "eval_accuracy": 0.9036666666666666,
+      "eval_loss": 0.30100810527801514,
+      "eval_runtime": 15.803,
+      "eval_samples_per_second": 1518.698,
+      "eval_steps_per_second": 4.746,
+      "step": 2100
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1e-06,
+      "loss": 0.3045,
+      "step": 2120
+    },
+    {
+      "epoch": 0.88,
+      "eval_accuracy": 0.9050833333333334,
+      "eval_loss": 0.29866886138916016,
+      "eval_runtime": 15.5575,
+      "eval_samples_per_second": 1542.663,
+      "eval_steps_per_second": 4.821,
+      "step": 2120
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 1e-06,
+      "loss": 0.2797,
+      "step": 2140
+    },
+    {
+      "epoch": 0.89,
+      "eval_accuracy": 0.905125,
+      "eval_loss": 0.2977401614189148,
+      "eval_runtime": 15.9997,
+      "eval_samples_per_second": 1500.024,
+      "eval_steps_per_second": 4.688,
+      "step": 2140
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1e-06,
+      "loss": 0.3044,
+      "step": 2160
+    },
+    {
+      "epoch": 0.9,
+      "eval_accuracy": 0.9046666666666666,
+      "eval_loss": 0.29765772819519043,
+      "eval_runtime": 15.813,
+      "eval_samples_per_second": 1517.737,
+      "eval_steps_per_second": 4.743,
+      "step": 2160
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 1e-06,
+      "loss": 0.2948,
+      "step": 2180
+    },
+    {
+      "epoch": 0.91,
+      "eval_accuracy": 0.9055,
+      "eval_loss": 0.2965656518936157,
+      "eval_runtime": 15.4407,
+      "eval_samples_per_second": 1554.336,
+      "eval_steps_per_second": 4.857,
+      "step": 2180
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 1e-06,
+      "loss": 0.2803,
+      "step": 2200
+    },
+    {
+      "epoch": 0.92,
+      "eval_accuracy": 0.9052083333333333,
+      "eval_loss": 0.2959022521972656,
+      "eval_runtime": 15.9407,
+      "eval_samples_per_second": 1505.578,
+      "eval_steps_per_second": 4.705,
+      "step": 2200
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1e-06,
+      "loss": 0.3008,
+      "step": 2220
+    },
+    {
+      "epoch": 0.93,
+      "eval_accuracy": 0.9055833333333333,
+      "eval_loss": 0.2946693003177643,
+      "eval_runtime": 15.495,
+      "eval_samples_per_second": 1548.887,
+      "eval_steps_per_second": 4.84,
+      "step": 2220
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 1e-06,
+      "loss": 0.3163,
+      "step": 2240
+    },
+    {
+      "epoch": 0.93,
+      "eval_accuracy": 0.905625,
+      "eval_loss": 0.2954801321029663,
+      "eval_runtime": 15.9747,
+      "eval_samples_per_second": 1502.377,
+      "eval_steps_per_second": 4.695,
+      "step": 2240
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 1e-06,
+      "loss": 0.3094,
+      "step": 2260
+    },
+    {
+      "epoch": 0.94,
+      "eval_accuracy": 0.90525,
+      "eval_loss": 0.2948620617389679,
+      "eval_runtime": 15.5314,
+      "eval_samples_per_second": 1545.257,
+      "eval_steps_per_second": 4.829,
+      "step": 2260
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 1e-06,
+      "loss": 0.2932,
+      "step": 2280
+    },
+    {
+      "epoch": 0.95,
+      "eval_accuracy": 0.9058333333333334,
+      "eval_loss": 0.29374802112579346,
+      "eval_runtime": 15.9691,
+      "eval_samples_per_second": 1502.905,
+      "eval_steps_per_second": 4.697,
+      "step": 2280
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 1e-06,
+      "loss": 0.2963,
+      "step": 2300
+    },
+    {
+      "epoch": 0.96,
+      "eval_accuracy": 0.9058333333333334,
+      "eval_loss": 0.29317864775657654,
+      "eval_runtime": 15.765,
+      "eval_samples_per_second": 1522.362,
+      "eval_steps_per_second": 4.757,
+      "step": 2300
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1e-06,
+      "loss": 0.3027,
+      "step": 2320
+    },
+    {
+      "epoch": 0.97,
+      "eval_accuracy": 0.905375,
+      "eval_loss": 0.2939450442790985,
+      "eval_runtime": 15.8745,
+      "eval_samples_per_second": 1511.86,
+      "eval_steps_per_second": 4.725,
+      "step": 2320
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 1e-06,
+      "loss": 0.3017,
+      "step": 2340
+    },
+    {
+      "epoch": 0.97,
+      "eval_accuracy": 0.9062083333333333,
+      "eval_loss": 0.29151102900505066,
+      "eval_runtime": 15.7805,
+      "eval_samples_per_second": 1520.865,
+      "eval_steps_per_second": 4.753,
+      "step": 2340
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 1e-06,
+      "loss": 0.347,
+      "step": 2360
+    },
+    {
+      "epoch": 0.98,
+      "eval_accuracy": 0.9069583333333333,
+      "eval_loss": 0.2903897166252136,
+      "eval_runtime": 15.6602,
+      "eval_samples_per_second": 1532.552,
+      "eval_steps_per_second": 4.789,
+      "step": 2360
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 1e-06,
+      "loss": 0.311,
+      "step": 2380
+    },
+    {
+      "epoch": 0.99,
+      "eval_accuracy": 0.906375,
+      "eval_loss": 0.29053354263305664,
+      "eval_runtime": 15.4212,
+      "eval_samples_per_second": 1556.296,
+      "eval_steps_per_second": 4.863,
+      "step": 2380
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 1e-06,
+      "loss": 0.3007,
+      "step": 2400
+    },
+    {
+      "epoch": 1.0,
+      "eval_accuracy": 0.9072083333333333,
+      "eval_loss": 0.290608286857605,
+      "eval_runtime": 15.7304,
+      "eval_samples_per_second": 1525.709,
+      "eval_steps_per_second": 4.768,
+      "step": 2400
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1e-06,
+      "loss": 0.3577,
+      "step": 2420
+    },
+    {
+      "epoch": 1.01,
+      "eval_accuracy": 0.9072083333333333,
+      "eval_loss": 0.289432168006897,
+      "eval_runtime": 16.5863,
+      "eval_samples_per_second": 1446.973,
+      "eval_steps_per_second": 4.522,
+      "step": 2420
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2775,
+      "step": 2440
+    },
+    {
+      "epoch": 1.02,
+      "eval_accuracy": 0.906625,
+      "eval_loss": 0.2914583384990692,
+      "eval_runtime": 15.8866,
+      "eval_samples_per_second": 1510.711,
+      "eval_steps_per_second": 4.721,
+      "step": 2440
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1e-06,
+      "loss": 0.326,
+      "step": 2460
+    },
+    {
+      "epoch": 1.02,
+      "eval_accuracy": 0.907,
+      "eval_loss": 0.28882813453674316,
+      "eval_runtime": 15.4052,
+      "eval_samples_per_second": 1557.919,
+      "eval_steps_per_second": 4.868,
+      "step": 2460
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 1e-06,
+      "loss": 0.2958,
+      "step": 2480
+    },
+    {
+      "epoch": 1.03,
+      "eval_accuracy": 0.9076666666666666,
+      "eval_loss": 0.2885444760322571,
+      "eval_runtime": 15.4629,
+      "eval_samples_per_second": 1552.103,
+      "eval_steps_per_second": 4.85,
+      "step": 2480
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1e-06,
+      "loss": 0.2543,
+      "step": 2500
+    },
+    {
+      "epoch": 1.04,
+      "eval_accuracy": 0.90825,
+      "eval_loss": 0.28599992394447327,
+      "eval_runtime": 15.7097,
+      "eval_samples_per_second": 1527.717,
+      "eval_steps_per_second": 4.774,
+      "step": 2500
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1e-06,
+      "loss": 0.287,
+      "step": 2520
+    },
+    {
+      "epoch": 1.05,
+      "eval_accuracy": 0.908875,
+      "eval_loss": 0.285250186920166,
+      "eval_runtime": 15.7948,
+      "eval_samples_per_second": 1519.484,
+      "eval_steps_per_second": 4.748,
+      "step": 2520
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1e-06,
+      "loss": 0.3226,
+      "step": 2540
+    },
+    {
+      "epoch": 1.06,
+      "eval_accuracy": 0.908375,
+      "eval_loss": 0.28508585691452026,
+      "eval_runtime": 15.6253,
+      "eval_samples_per_second": 1535.966,
+      "eval_steps_per_second": 4.8,
+      "step": 2540
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1e-06,
+      "loss": 0.2912,
+      "step": 2560
+    },
+    {
+      "epoch": 1.07,
+      "eval_accuracy": 0.9077083333333333,
+      "eval_loss": 0.28835633397102356,
+      "eval_runtime": 15.302,
+      "eval_samples_per_second": 1568.422,
+      "eval_steps_per_second": 4.901,
+      "step": 2560
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1e-06,
+      "loss": 0.3344,
+      "step": 2580
+    },
+    {
+      "epoch": 1.07,
+      "eval_accuracy": 0.9082916666666667,
+      "eval_loss": 0.2841767966747284,
+      "eval_runtime": 15.7333,
+      "eval_samples_per_second": 1525.424,
+      "eval_steps_per_second": 4.767,
+      "step": 2580
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2883,
+      "step": 2600
+    },
+    {
+      "epoch": 1.08,
+      "eval_accuracy": 0.9075,
+      "eval_loss": 0.28568732738494873,
+      "eval_runtime": 15.9862,
+      "eval_samples_per_second": 1501.294,
+      "eval_steps_per_second": 4.692,
+      "step": 2600
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 1e-06,
+      "loss": 0.2958,
+      "step": 2620
+    },
+    {
+      "epoch": 1.09,
+      "eval_accuracy": 0.90875,
+      "eval_loss": 0.2833400368690491,
+      "eval_runtime": 15.9204,
+      "eval_samples_per_second": 1507.497,
+      "eval_steps_per_second": 4.711,
+      "step": 2620
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1e-06,
+      "loss": 0.2913,
+      "step": 2640
+    },
+    {
+      "epoch": 1.1,
+      "eval_accuracy": 0.9085,
+      "eval_loss": 0.2836286723613739,
+      "eval_runtime": 16.0314,
+      "eval_samples_per_second": 1497.058,
+      "eval_steps_per_second": 4.678,
+      "step": 2640
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1e-06,
+      "loss": 0.3007,
+      "step": 2660
+    },
+    {
+      "epoch": 1.11,
+      "eval_accuracy": 0.908625,
+      "eval_loss": 0.2823469638824463,
+      "eval_runtime": 15.5726,
+      "eval_samples_per_second": 1541.172,
+      "eval_steps_per_second": 4.816,
+      "step": 2660
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1e-06,
+      "loss": 0.3152,
+      "step": 2680
+    },
+    {
+      "epoch": 1.12,
+      "eval_accuracy": 0.9082083333333333,
+      "eval_loss": 0.2841881811618805,
+      "eval_runtime": 15.7571,
+      "eval_samples_per_second": 1523.126,
+      "eval_steps_per_second": 4.76,
+      "step": 2680
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2857,
+      "step": 2700
+    },
+    {
+      "epoch": 1.12,
+      "eval_accuracy": 0.9079583333333333,
+      "eval_loss": 0.28267306089401245,
+      "eval_runtime": 15.9307,
+      "eval_samples_per_second": 1506.523,
+      "eval_steps_per_second": 4.708,
+      "step": 2700
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1e-06,
+      "loss": 0.2821,
+      "step": 2720
+    },
+    {
+      "epoch": 1.13,
+      "eval_accuracy": 0.9085416666666667,
+      "eval_loss": 0.28154897689819336,
+      "eval_runtime": 15.5765,
+      "eval_samples_per_second": 1540.779,
+      "eval_steps_per_second": 4.815,
+      "step": 2720
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1e-06,
+      "loss": 0.2775,
+      "step": 2740
+    },
+    {
+      "epoch": 1.14,
+      "eval_accuracy": 0.9074166666666666,
+      "eval_loss": 0.2842314839363098,
+      "eval_runtime": 16.5328,
+      "eval_samples_per_second": 1451.656,
+      "eval_steps_per_second": 4.536,
+      "step": 2740
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 1e-06,
+      "loss": 0.2964,
+      "step": 2760
+    },
+    {
+      "epoch": 1.15,
+      "eval_accuracy": 0.9084166666666667,
+      "eval_loss": 0.2818900942802429,
+      "eval_runtime": 15.7457,
+      "eval_samples_per_second": 1524.229,
+      "eval_steps_per_second": 4.763,
+      "step": 2760
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1e-06,
+      "loss": 0.3245,
+      "step": 2780
+    },
+    {
+      "epoch": 1.16,
+      "eval_accuracy": 0.9088333333333334,
+      "eval_loss": 0.2803751826286316,
+      "eval_runtime": 15.6563,
+      "eval_samples_per_second": 1532.925,
+      "eval_steps_per_second": 4.79,
+      "step": 2780
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1e-06,
+      "loss": 0.2706,
+      "step": 2800
+    },
+    {
+      "epoch": 1.17,
+      "eval_accuracy": 0.909125,
+      "eval_loss": 0.2802983820438385,
+      "eval_runtime": 15.9707,
+      "eval_samples_per_second": 1502.752,
+      "eval_steps_per_second": 4.696,
+      "step": 2800
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1e-06,
+      "loss": 0.2674,
+      "step": 2820
+    },
+    {
+      "epoch": 1.18,
+      "eval_accuracy": 0.908875,
+      "eval_loss": 0.2793467342853546,
+      "eval_runtime": 15.9613,
+      "eval_samples_per_second": 1503.634,
+      "eval_steps_per_second": 4.699,
+      "step": 2820
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1e-06,
+      "loss": 0.3296,
+      "step": 2840
+    },
+    {
+      "epoch": 1.18,
+      "eval_accuracy": 0.9093333333333333,
+      "eval_loss": 0.2792114019393921,
+      "eval_runtime": 15.5813,
+      "eval_samples_per_second": 1540.308,
+      "eval_steps_per_second": 4.813,
+      "step": 2840
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1e-06,
+      "loss": 0.2993,
+      "step": 2860
+    },
+    {
+      "epoch": 1.19,
+      "eval_accuracy": 0.909125,
+      "eval_loss": 0.27838194370269775,
+      "eval_runtime": 15.9048,
+      "eval_samples_per_second": 1508.979,
+      "eval_steps_per_second": 4.716,
+      "step": 2860
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1e-06,
+      "loss": 0.3305,
+      "step": 2880
+    },
+    {
+      "epoch": 1.2,
+      "eval_accuracy": 0.9084583333333334,
+      "eval_loss": 0.2802034914493561,
+      "eval_runtime": 15.8179,
+      "eval_samples_per_second": 1517.271,
+      "eval_steps_per_second": 4.741,
+      "step": 2880
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 1e-06,
+      "loss": 0.2984,
+      "step": 2900
+    },
+    {
+      "epoch": 1.21,
+      "eval_accuracy": 0.9088333333333334,
+      "eval_loss": 0.2789742052555084,
+      "eval_runtime": 15.8715,
+      "eval_samples_per_second": 1512.14,
+      "eval_steps_per_second": 4.725,
+      "step": 2900
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1e-06,
+      "loss": 0.2748,
+      "step": 2920
+    },
+    {
+      "epoch": 1.22,
+      "eval_accuracy": 0.9088333333333334,
+      "eval_loss": 0.2786828875541687,
+      "eval_runtime": 15.744,
+      "eval_samples_per_second": 1524.387,
+      "eval_steps_per_second": 4.764,
+      "step": 2920
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1e-06,
+      "loss": 0.2708,
+      "step": 2940
+    },
+    {
+      "epoch": 1.23,
+      "eval_accuracy": 0.909125,
+      "eval_loss": 0.2787318527698517,
+      "eval_runtime": 15.8587,
+      "eval_samples_per_second": 1513.363,
+      "eval_steps_per_second": 4.729,
+      "step": 2940
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1e-06,
+      "loss": 0.3062,
+      "step": 2960
+    },
+    {
+      "epoch": 1.23,
+      "eval_accuracy": 0.910125,
+      "eval_loss": 0.2767701745033264,
+      "eval_runtime": 15.5088,
+      "eval_samples_per_second": 1547.508,
+      "eval_steps_per_second": 4.836,
+      "step": 2960
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1e-06,
+      "loss": 0.3039,
+      "step": 2980
+    },
+    {
+      "epoch": 1.24,
+      "eval_accuracy": 0.910375,
+      "eval_loss": 0.2769327163696289,
+      "eval_runtime": 15.7715,
+      "eval_samples_per_second": 1521.733,
+      "eval_steps_per_second": 4.755,
+      "step": 2980
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1e-06,
+      "loss": 0.2889,
+      "step": 3000
+    },
+    {
+      "epoch": 1.25,
+      "eval_accuracy": 0.9106666666666666,
+      "eval_loss": 0.2753864526748657,
+      "eval_runtime": 16.0825,
+      "eval_samples_per_second": 1492.309,
+      "eval_steps_per_second": 4.663,
+      "step": 3000
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1e-06,
+      "loss": 0.2964,
+      "step": 3020
+    },
+    {
+      "epoch": 1.26,
+      "eval_accuracy": 0.9102916666666667,
+      "eval_loss": 0.2752026319503784,
+      "eval_runtime": 15.8059,
+      "eval_samples_per_second": 1518.419,
+      "eval_steps_per_second": 4.745,
+      "step": 3020
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1e-06,
+      "loss": 0.2951,
+      "step": 3040
+    },
+    {
+      "epoch": 1.27,
+      "eval_accuracy": 0.9099583333333333,
+      "eval_loss": 0.27485111355781555,
+      "eval_runtime": 15.7664,
+      "eval_samples_per_second": 1522.226,
+      "eval_steps_per_second": 4.757,
+      "step": 3040
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1e-06,
+      "loss": 0.2879,
+      "step": 3060
+    },
+    {
+      "epoch": 1.27,
+      "eval_accuracy": 0.91075,
+      "eval_loss": 0.27414020895957947,
+      "eval_runtime": 15.6142,
+      "eval_samples_per_second": 1537.064,
+      "eval_steps_per_second": 4.803,
+      "step": 3060
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1e-06,
+      "loss": 0.2795,
+      "step": 3080
+    },
+    {
+      "epoch": 1.28,
+      "eval_accuracy": 0.9106666666666666,
+      "eval_loss": 0.2753457725048065,
+      "eval_runtime": 15.8605,
+      "eval_samples_per_second": 1513.198,
+      "eval_steps_per_second": 4.729,
+      "step": 3080
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1e-06,
+      "loss": 0.2794,
+      "step": 3100
+    },
+    {
+      "epoch": 1.29,
+      "eval_accuracy": 0.9100416666666666,
+      "eval_loss": 0.27380895614624023,
+      "eval_runtime": 16.3184,
+      "eval_samples_per_second": 1470.733,
+      "eval_steps_per_second": 4.596,
+      "step": 3100
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1e-06,
+      "loss": 0.258,
+      "step": 3120
+    },
+    {
+      "epoch": 1.3,
+      "eval_accuracy": 0.910625,
+      "eval_loss": 0.27472469210624695,
+      "eval_runtime": 15.8282,
+      "eval_samples_per_second": 1516.281,
+      "eval_steps_per_second": 4.738,
+      "step": 3120
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1e-06,
+      "loss": 0.2551,
+      "step": 3140
+    },
+    {
+      "epoch": 1.31,
+      "eval_accuracy": 0.91,
+      "eval_loss": 0.274568110704422,
+      "eval_runtime": 15.7686,
+      "eval_samples_per_second": 1522.008,
+      "eval_steps_per_second": 4.756,
+      "step": 3140
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1e-06,
+      "loss": 0.3086,
+      "step": 3160
+    },
+    {
+      "epoch": 1.32,
+      "eval_accuracy": 0.9097083333333333,
+      "eval_loss": 0.27524352073669434,
+      "eval_runtime": 16.163,
+      "eval_samples_per_second": 1484.869,
+      "eval_steps_per_second": 4.64,
+      "step": 3160
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1e-06,
+      "loss": 0.2812,
+      "step": 3180
+    },
+    {
+      "epoch": 1.32,
+      "eval_accuracy": 0.9117083333333333,
+      "eval_loss": 0.27365365624427795,
+      "eval_runtime": 15.5225,
+      "eval_samples_per_second": 1546.138,
+      "eval_steps_per_second": 4.832,
+      "step": 3180
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1e-06,
+      "loss": 0.3016,
+      "step": 3200
+    },
+    {
+      "epoch": 1.33,
+      "eval_accuracy": 0.911,
+      "eval_loss": 0.271597683429718,
+      "eval_runtime": 15.8761,
+      "eval_samples_per_second": 1511.709,
+      "eval_steps_per_second": 4.724,
+      "step": 3200
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1e-06,
+      "loss": 0.2913,
+      "step": 3220
+    },
+    {
+      "epoch": 1.34,
+      "eval_accuracy": 0.9109583333333333,
+      "eval_loss": 0.27143821120262146,
+      "eval_runtime": 16.2761,
+      "eval_samples_per_second": 1474.556,
+      "eval_steps_per_second": 4.608,
+      "step": 3220
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1e-06,
+      "loss": 0.2634,
+      "step": 3240
+    },
+    {
+      "epoch": 1.35,
+      "eval_accuracy": 0.909,
+      "eval_loss": 0.2747085988521576,
+      "eval_runtime": 16.3191,
+      "eval_samples_per_second": 1470.669,
+      "eval_steps_per_second": 4.596,
+      "step": 3240
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1e-06,
+      "loss": 0.2845,
+      "step": 3260
+    },
+    {
+      "epoch": 1.36,
+      "eval_accuracy": 0.9109166666666667,
+      "eval_loss": 0.2722584903240204,
+      "eval_runtime": 15.7837,
+      "eval_samples_per_second": 1520.554,
+      "eval_steps_per_second": 4.752,
+      "step": 3260
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1e-06,
+      "loss": 0.2864,
+      "step": 3280
+    },
+    {
+      "epoch": 1.37,
+      "eval_accuracy": 0.9112083333333333,
+      "eval_loss": 0.2707350552082062,
+      "eval_runtime": 15.9926,
+      "eval_samples_per_second": 1500.697,
+      "eval_steps_per_second": 4.69,
+      "step": 3280
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1e-06,
+      "loss": 0.2794,
+      "step": 3300
+    },
+    {
+      "epoch": 1.38,
+      "eval_accuracy": 0.9111666666666667,
+      "eval_loss": 0.2700752317905426,
+      "eval_runtime": 15.6416,
+      "eval_samples_per_second": 1534.367,
+      "eval_steps_per_second": 4.795,
+      "step": 3300
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 1e-06,
+      "loss": 0.3216,
+      "step": 3320
+    },
+    {
+      "epoch": 1.38,
+      "eval_accuracy": 0.9105,
+      "eval_loss": 0.27095627784729004,
+      "eval_runtime": 16.2168,
+      "eval_samples_per_second": 1479.95,
+      "eval_steps_per_second": 4.625,
+      "step": 3320
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1e-06,
+      "loss": 0.2738,
+      "step": 3340
+    },
+    {
+      "epoch": 1.39,
+      "eval_accuracy": 0.9102083333333333,
+      "eval_loss": 0.2725074291229248,
+      "eval_runtime": 15.7189,
+      "eval_samples_per_second": 1526.821,
+      "eval_steps_per_second": 4.771,
+      "step": 3340
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1e-06,
+      "loss": 0.301,
+      "step": 3360
+    },
+    {
+      "epoch": 1.4,
+      "eval_accuracy": 0.9102083333333333,
+      "eval_loss": 0.27188044786453247,
+      "eval_runtime": 15.3899,
+      "eval_samples_per_second": 1559.463,
+      "eval_steps_per_second": 4.873,
+      "step": 3360
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1e-06,
+      "loss": 0.3144,
+      "step": 3380
+    },
+    {
+      "epoch": 1.41,
+      "eval_accuracy": 0.911125,
+      "eval_loss": 0.269314169883728,
+      "eval_runtime": 16.1663,
+      "eval_samples_per_second": 1484.568,
+      "eval_steps_per_second": 4.639,
+      "step": 3380
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2914,
+      "step": 3400
+    },
+    {
+      "epoch": 1.42,
+      "eval_accuracy": 0.9120416666666666,
+      "eval_loss": 0.26871222257614136,
+      "eval_runtime": 15.7364,
+      "eval_samples_per_second": 1525.13,
+      "eval_steps_per_second": 4.766,
+      "step": 3400
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1e-06,
+      "loss": 0.252,
+      "step": 3420
+    },
+    {
+      "epoch": 1.43,
+      "eval_accuracy": 0.9114166666666667,
+      "eval_loss": 0.26832154393196106,
+      "eval_runtime": 16.1928,
+      "eval_samples_per_second": 1482.143,
+      "eval_steps_per_second": 4.632,
+      "step": 3420
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1e-06,
+      "loss": 0.2616,
+      "step": 3440
+    },
+    {
+      "epoch": 1.43,
+      "eval_accuracy": 0.9119583333333333,
+      "eval_loss": 0.2678380608558655,
+      "eval_runtime": 15.9974,
+      "eval_samples_per_second": 1500.246,
+      "eval_steps_per_second": 4.688,
+      "step": 3440
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 1e-06,
+      "loss": 0.247,
+      "step": 3460
+    },
+    {
+      "epoch": 1.44,
+      "eval_accuracy": 0.9126666666666666,
+      "eval_loss": 0.2679081857204437,
+      "eval_runtime": 15.5041,
+      "eval_samples_per_second": 1547.976,
+      "eval_steps_per_second": 4.837,
+      "step": 3460
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1e-06,
+      "loss": 0.279,
+      "step": 3480
+    },
+    {
+      "epoch": 1.45,
+      "eval_accuracy": 0.912,
+      "eval_loss": 0.2675539553165436,
+      "eval_runtime": 16.0139,
+      "eval_samples_per_second": 1498.702,
+      "eval_steps_per_second": 4.683,
+      "step": 3480
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1e-06,
+      "loss": 0.2823,
+      "step": 3500
+    },
+    {
+      "epoch": 1.46,
+      "eval_accuracy": 0.9124166666666667,
+      "eval_loss": 0.26708924770355225,
+      "eval_runtime": 15.8242,
+      "eval_samples_per_second": 1516.667,
+      "eval_steps_per_second": 4.74,
+      "step": 3500
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1e-06,
+      "loss": 0.2769,
+      "step": 3520
+    },
+    {
+      "epoch": 1.47,
+      "eval_accuracy": 0.9125,
+      "eval_loss": 0.26735562086105347,
+      "eval_runtime": 16.0609,
+      "eval_samples_per_second": 1494.311,
+      "eval_steps_per_second": 4.67,
+      "step": 3520
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1e-06,
+      "loss": 0.253,
+      "step": 3540
+    },
+    {
+      "epoch": 1.48,
+      "eval_accuracy": 0.9122916666666666,
+      "eval_loss": 0.2679881453514099,
+      "eval_runtime": 15.8153,
+      "eval_samples_per_second": 1517.514,
+      "eval_steps_per_second": 4.742,
+      "step": 3540
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2398,
+      "step": 3560
+    },
+    {
+      "epoch": 1.48,
+      "eval_accuracy": 0.9125,
+      "eval_loss": 0.26599201560020447,
+      "eval_runtime": 15.854,
+      "eval_samples_per_second": 1513.812,
+      "eval_steps_per_second": 4.731,
+      "step": 3560
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1e-06,
+      "loss": 0.2524,
+      "step": 3580
+    },
+    {
+      "epoch": 1.49,
+      "eval_accuracy": 0.912875,
+      "eval_loss": 0.2660870850086212,
+      "eval_runtime": 15.72,
+      "eval_samples_per_second": 1526.713,
+      "eval_steps_per_second": 4.771,
+      "step": 3580
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 1e-06,
+      "loss": 0.2794,
+      "step": 3600
+    },
+    {
+      "epoch": 1.5,
+      "eval_accuracy": 0.9122916666666666,
+      "eval_loss": 0.2667754590511322,
+      "eval_runtime": 15.9732,
+      "eval_samples_per_second": 1502.513,
+      "eval_steps_per_second": 4.695,
+      "step": 3600
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1e-06,
+      "loss": 0.2378,
+      "step": 3620
+    },
+    {
+      "epoch": 1.51,
+      "eval_accuracy": 0.912375,
+      "eval_loss": 0.26728901267051697,
+      "eval_runtime": 15.3559,
+      "eval_samples_per_second": 1562.918,
+      "eval_steps_per_second": 4.884,
+      "step": 3620
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2309,
+      "step": 3640
+    },
+    {
+      "epoch": 1.52,
+      "eval_accuracy": 0.91175,
+      "eval_loss": 0.2677074074745178,
+      "eval_runtime": 15.7934,
+      "eval_samples_per_second": 1519.623,
+      "eval_steps_per_second": 4.749,
+      "step": 3640
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2414,
+      "step": 3660
+    },
+    {
+      "epoch": 1.52,
+      "eval_accuracy": 0.9127083333333333,
+      "eval_loss": 0.2664356529712677,
+      "eval_runtime": 16.3142,
+      "eval_samples_per_second": 1471.115,
+      "eval_steps_per_second": 4.597,
+      "step": 3660
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1e-06,
+      "loss": 0.2698,
+      "step": 3680
+    },
+    {
+      "epoch": 1.53,
+      "eval_accuracy": 0.9130416666666666,
+      "eval_loss": 0.26516926288604736,
+      "eval_runtime": 15.5498,
+      "eval_samples_per_second": 1543.428,
+      "eval_steps_per_second": 4.823,
+      "step": 3680
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1e-06,
+      "loss": 0.2674,
+      "step": 3700
+    },
+    {
+      "epoch": 1.54,
+      "eval_accuracy": 0.912875,
+      "eval_loss": 0.26480814814567566,
+      "eval_runtime": 15.9433,
+      "eval_samples_per_second": 1505.331,
+      "eval_steps_per_second": 4.704,
+      "step": 3700
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 1e-06,
+      "loss": 0.3004,
+      "step": 3720
+    },
+    {
+      "epoch": 1.55,
+      "eval_accuracy": 0.9120833333333334,
+      "eval_loss": 0.26566416025161743,
+      "eval_runtime": 16.4814,
+      "eval_samples_per_second": 1456.188,
+      "eval_steps_per_second": 4.551,
+      "step": 3720
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 1e-06,
+      "loss": 0.2621,
+      "step": 3740
+    },
+    {
+      "epoch": 1.56,
+      "eval_accuracy": 0.9130833333333334,
+      "eval_loss": 0.2644825577735901,
+      "eval_runtime": 15.8572,
+      "eval_samples_per_second": 1513.507,
+      "eval_steps_per_second": 4.73,
+      "step": 3740
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 1e-06,
+      "loss": 0.2691,
+      "step": 3760
+    },
+    {
+      "epoch": 1.57,
+      "eval_accuracy": 0.9137916666666667,
+      "eval_loss": 0.26423123478889465,
+      "eval_runtime": 15.9709,
+      "eval_samples_per_second": 1502.733,
+      "eval_steps_per_second": 4.696,
+      "step": 3760
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 1e-06,
+      "loss": 0.2768,
+      "step": 3780
+    },
+    {
+      "epoch": 1.57,
+      "eval_accuracy": 0.9114583333333334,
+      "eval_loss": 0.2679901421070099,
+      "eval_runtime": 15.6044,
+      "eval_samples_per_second": 1538.026,
+      "eval_steps_per_second": 4.806,
+      "step": 3780
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2768,
+      "step": 3800
+    },
+    {
+      "epoch": 1.58,
+      "eval_accuracy": 0.9123333333333333,
+      "eval_loss": 0.2639557421207428,
+      "eval_runtime": 16.2935,
+      "eval_samples_per_second": 1472.978,
+      "eval_steps_per_second": 4.603,
+      "step": 3800
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 1e-06,
+      "loss": 0.2308,
+      "step": 3820
+    },
+    {
+      "epoch": 1.59,
+      "eval_accuracy": 0.9118333333333334,
+      "eval_loss": 0.2661728262901306,
+      "eval_runtime": 15.4841,
+      "eval_samples_per_second": 1549.978,
+      "eval_steps_per_second": 4.844,
+      "step": 3820
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 1e-06,
+      "loss": 0.2591,
+      "step": 3840
+    },
+    {
+      "epoch": 1.6,
+      "eval_accuracy": 0.9131666666666667,
+      "eval_loss": 0.2664397954940796,
+      "eval_runtime": 15.6675,
+      "eval_samples_per_second": 1531.834,
+      "eval_steps_per_second": 4.787,
+      "step": 3840
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 1e-06,
+      "loss": 0.2496,
+      "step": 3860
+    },
+    {
+      "epoch": 1.61,
+      "eval_accuracy": 0.9125,
+      "eval_loss": 0.2657550573348999,
+      "eval_runtime": 15.6288,
+      "eval_samples_per_second": 1535.631,
+      "eval_steps_per_second": 4.799,
+      "step": 3860
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 1e-06,
+      "loss": 0.2609,
+      "step": 3880
+    },
+    {
+      "epoch": 1.62,
+      "eval_accuracy": 0.91225,
+      "eval_loss": 0.26461654901504517,
+      "eval_runtime": 15.6184,
+      "eval_samples_per_second": 1536.644,
+      "eval_steps_per_second": 4.802,
+      "step": 3880
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 1e-06,
+      "loss": 0.2491,
+      "step": 3900
+    },
+    {
+      "epoch": 1.62,
+      "eval_accuracy": 0.9132916666666666,
+      "eval_loss": 0.2625100314617157,
+      "eval_runtime": 15.868,
+      "eval_samples_per_second": 1512.482,
+      "eval_steps_per_second": 4.727,
+      "step": 3900
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 1e-06,
+      "loss": 0.2609,
+      "step": 3920
+    },
+    {
+      "epoch": 1.63,
+      "eval_accuracy": 0.91375,
+      "eval_loss": 0.2650201916694641,
+      "eval_runtime": 15.7536,
+      "eval_samples_per_second": 1523.463,
+      "eval_steps_per_second": 4.761,
+      "step": 3920
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 1e-06,
+      "loss": 0.225,
+      "step": 3940
+    },
+    {
+      "epoch": 1.64,
+      "eval_accuracy": 0.9124166666666667,
+      "eval_loss": 0.262962281703949,
+      "eval_runtime": 15.7571,
+      "eval_samples_per_second": 1523.123,
+      "eval_steps_per_second": 4.76,
+      "step": 3940
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1e-06,
+      "loss": 0.3099,
+      "step": 3960
+    },
+    {
+      "epoch": 1.65,
+      "eval_accuracy": 0.9139583333333333,
+      "eval_loss": 0.26255694031715393,
+      "eval_runtime": 15.7275,
+      "eval_samples_per_second": 1525.985,
+      "eval_steps_per_second": 4.769,
+      "step": 3960
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1e-06,
+      "loss": 0.2627,
+      "step": 3980
+    },
+    {
+      "epoch": 1.66,
+      "eval_accuracy": 0.91425,
+      "eval_loss": 0.26082727313041687,
+      "eval_runtime": 15.785,
+      "eval_samples_per_second": 1520.427,
+      "eval_steps_per_second": 4.751,
+      "step": 3980
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 1e-06,
+      "loss": 0.2831,
+      "step": 4000
+    },
+    {
+      "epoch": 1.67,
+      "eval_accuracy": 0.9128333333333334,
+      "eval_loss": 0.2629248797893524,
+      "eval_runtime": 15.9672,
+      "eval_samples_per_second": 1503.078,
+      "eval_steps_per_second": 4.697,
+      "step": 4000
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 1e-06,
+      "loss": 0.2806,
+      "step": 4020
+    },
+    {
+      "epoch": 1.68,
+      "eval_accuracy": 0.9131666666666667,
+      "eval_loss": 0.2608546316623688,
+      "eval_runtime": 15.7614,
+      "eval_samples_per_second": 1522.706,
+      "eval_steps_per_second": 4.758,
+      "step": 4020
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 1e-06,
+      "loss": 0.3095,
+      "step": 4040
+    },
+    {
+      "epoch": 1.68,
+      "eval_accuracy": 0.914125,
+      "eval_loss": 0.263700008392334,
+      "eval_runtime": 15.7288,
+      "eval_samples_per_second": 1525.864,
+      "eval_steps_per_second": 4.768,
+      "step": 4040
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 1e-06,
+      "loss": 0.2637,
+      "step": 4060
+    },
+    {
+      "epoch": 1.69,
+      "eval_accuracy": 0.9152083333333333,
+      "eval_loss": 0.26146170496940613,
+      "eval_runtime": 15.6613,
+      "eval_samples_per_second": 1532.442,
+      "eval_steps_per_second": 4.789,
+      "step": 4060
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 1e-06,
+      "loss": 0.3105,
+      "step": 4080
+    },
+    {
+      "epoch": 1.7,
+      "eval_accuracy": 0.9137083333333333,
+      "eval_loss": 0.25999248027801514,
+      "eval_runtime": 15.6145,
+      "eval_samples_per_second": 1537.029,
+      "eval_steps_per_second": 4.803,
+      "step": 4080
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 1e-06,
+      "loss": 0.2665,
+      "step": 4100
+    },
+    {
+      "epoch": 1.71,
+      "eval_accuracy": 0.912875,
+      "eval_loss": 0.2612546980381012,
+      "eval_runtime": 16.3137,
+      "eval_samples_per_second": 1471.158,
+      "eval_steps_per_second": 4.597,
+      "step": 4100
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 1e-06,
+      "loss": 0.2582,
+      "step": 4120
+    },
+    {
+      "epoch": 1.72,
+      "eval_accuracy": 0.9130416666666666,
+      "eval_loss": 0.2623673975467682,
+      "eval_runtime": 15.824,
+      "eval_samples_per_second": 1516.686,
+      "eval_steps_per_second": 4.74,
+      "step": 4120
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 1e-06,
+      "loss": 0.2425,
+      "step": 4140
+    },
+    {
+      "epoch": 1.73,
+      "eval_accuracy": 0.91425,
+      "eval_loss": 0.25996851921081543,
+      "eval_runtime": 16.2345,
+      "eval_samples_per_second": 1478.333,
+      "eval_steps_per_second": 4.62,
+      "step": 4140
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 1e-06,
+      "loss": 0.2431,
+      "step": 4160
+    },
+    {
+      "epoch": 1.73,
+      "eval_accuracy": 0.914,
+      "eval_loss": 0.25984445214271545,
+      "eval_runtime": 15.7995,
+      "eval_samples_per_second": 1519.04,
+      "eval_steps_per_second": 4.747,
+      "step": 4160
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 1e-06,
+      "loss": 0.2423,
+      "step": 4180
+    },
+    {
+      "epoch": 1.74,
+      "eval_accuracy": 0.9142916666666666,
+      "eval_loss": 0.2597646713256836,
+      "eval_runtime": 15.5687,
+      "eval_samples_per_second": 1541.552,
+      "eval_steps_per_second": 4.817,
+      "step": 4180
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1e-06,
+      "loss": 0.2594,
+      "step": 4200
+    },
+    {
+      "epoch": 1.75,
+      "eval_accuracy": 0.9134583333333334,
+      "eval_loss": 0.26066410541534424,
+      "eval_runtime": 15.9904,
+      "eval_samples_per_second": 1500.899,
+      "eval_steps_per_second": 4.69,
+      "step": 4200
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 1e-06,
+      "loss": 0.2647,
+      "step": 4220
+    },
+    {
+      "epoch": 1.76,
+      "eval_accuracy": 0.914,
+      "eval_loss": 0.25917357206344604,
+      "eval_runtime": 15.5439,
+      "eval_samples_per_second": 1544.015,
+      "eval_steps_per_second": 4.825,
+      "step": 4220
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 1e-06,
+      "loss": 0.282,
+      "step": 4240
+    },
+    {
+      "epoch": 1.77,
+      "eval_accuracy": 0.9147083333333333,
+      "eval_loss": 0.258999764919281,
+      "eval_runtime": 15.9706,
+      "eval_samples_per_second": 1502.761,
+      "eval_steps_per_second": 4.696,
+      "step": 4240
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 1e-06,
+      "loss": 0.246,
+      "step": 4260
+    },
+    {
+      "epoch": 1.77,
+      "eval_accuracy": 0.9139166666666667,
+      "eval_loss": 0.25847962498664856,
+      "eval_runtime": 15.9552,
+      "eval_samples_per_second": 1504.21,
+      "eval_steps_per_second": 4.701,
+      "step": 4260
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2558,
+      "step": 4280
+    },
+    {
+      "epoch": 1.78,
+      "eval_accuracy": 0.9138333333333334,
+      "eval_loss": 0.25929775834083557,
+      "eval_runtime": 15.5972,
+      "eval_samples_per_second": 1538.742,
+      "eval_steps_per_second": 4.809,
+      "step": 4280
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 1e-06,
+      "loss": 0.2249,
+      "step": 4300
+    },
+    {
+      "epoch": 1.79,
+      "eval_accuracy": 0.9144166666666667,
+      "eval_loss": 0.258432537317276,
+      "eval_runtime": 15.9462,
+      "eval_samples_per_second": 1505.062,
+      "eval_steps_per_second": 4.703,
+      "step": 4300
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 1e-06,
+      "loss": 0.2619,
+      "step": 4320
+    },
+    {
+      "epoch": 1.8,
+      "eval_accuracy": 0.9135416666666667,
+      "eval_loss": 0.25964200496673584,
+      "eval_runtime": 15.7728,
+      "eval_samples_per_second": 1521.607,
+      "eval_steps_per_second": 4.755,
+      "step": 4320
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 1e-06,
+      "loss": 0.2789,
+      "step": 4340
+    },
+    {
+      "epoch": 1.81,
+      "eval_accuracy": 0.9133333333333333,
+      "eval_loss": 0.26067379117012024,
+      "eval_runtime": 16.1036,
+      "eval_samples_per_second": 1490.346,
+      "eval_steps_per_second": 4.657,
+      "step": 4340
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 1e-06,
+      "loss": 0.2714,
+      "step": 4360
+    },
+    {
+      "epoch": 1.82,
+      "eval_accuracy": 0.9144583333333334,
+      "eval_loss": 0.25738218426704407,
+      "eval_runtime": 15.9298,
+      "eval_samples_per_second": 1506.611,
+      "eval_steps_per_second": 4.708,
+      "step": 4360
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 1e-06,
+      "loss": 0.2662,
+      "step": 4380
+    },
+    {
+      "epoch": 1.82,
+      "eval_accuracy": 0.9156666666666666,
+      "eval_loss": 0.2561679184436798,
+      "eval_runtime": 16.0834,
+      "eval_samples_per_second": 1492.218,
+      "eval_steps_per_second": 4.663,
+      "step": 4380
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 1e-06,
+      "loss": 0.2773,
+      "step": 4400
+    },
+    {
+      "epoch": 1.83,
+      "eval_accuracy": 0.914,
+      "eval_loss": 0.2571386694908142,
+      "eval_runtime": 15.7636,
+      "eval_samples_per_second": 1522.493,
+      "eval_steps_per_second": 4.758,
+      "step": 4400
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 1e-06,
+      "loss": 0.2478,
+      "step": 4420
+    },
+    {
+      "epoch": 1.84,
+      "eval_accuracy": 0.9152916666666666,
+      "eval_loss": 0.25946521759033203,
+      "eval_runtime": 15.8816,
+      "eval_samples_per_second": 1511.187,
+      "eval_steps_per_second": 4.722,
+      "step": 4420
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 1e-06,
+      "loss": 0.2517,
+      "step": 4440
+    },
+    {
+      "epoch": 1.85,
+      "eval_accuracy": 0.9144583333333334,
+      "eval_loss": 0.2570332884788513,
+      "eval_runtime": 15.9553,
+      "eval_samples_per_second": 1504.206,
+      "eval_steps_per_second": 4.701,
+      "step": 4440
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 1e-06,
+      "loss": 0.2539,
+      "step": 4460
+    },
+    {
+      "epoch": 1.86,
+      "eval_accuracy": 0.9137916666666667,
+      "eval_loss": 0.25786763429641724,
+      "eval_runtime": 15.6911,
+      "eval_samples_per_second": 1529.534,
+      "eval_steps_per_second": 4.78,
+      "step": 4460
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 1e-06,
+      "loss": 0.2635,
+      "step": 4480
+    },
+    {
+      "epoch": 1.87,
+      "eval_accuracy": 0.9154166666666667,
+      "eval_loss": 0.25599122047424316,
+      "eval_runtime": 15.95,
+      "eval_samples_per_second": 1504.705,
+      "eval_steps_per_second": 4.702,
+      "step": 4480
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1e-06,
+      "loss": 0.3007,
+      "step": 4500
+    },
+    {
+      "epoch": 1.88,
+      "eval_accuracy": 0.914625,
+      "eval_loss": 0.25718793272972107,
+      "eval_runtime": 16.0115,
+      "eval_samples_per_second": 1498.923,
+      "eval_steps_per_second": 4.684,
+      "step": 4500
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1e-06,
+      "loss": 0.2865,
+      "step": 4520
+    },
+    {
+      "epoch": 1.88,
+      "eval_accuracy": 0.915125,
+      "eval_loss": 0.25561970472335815,
+      "eval_runtime": 15.2788,
+      "eval_samples_per_second": 1570.809,
+      "eval_steps_per_second": 4.909,
+      "step": 4520
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 1e-06,
+      "loss": 0.2234,
+      "step": 4540
+    },
+    {
+      "epoch": 1.89,
+      "eval_accuracy": 0.9139583333333333,
+      "eval_loss": 0.2578865885734558,
+      "eval_runtime": 15.8725,
+      "eval_samples_per_second": 1512.049,
+      "eval_steps_per_second": 4.725,
+      "step": 4540
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 1e-06,
+      "loss": 0.2864,
+      "step": 4560
+    },
+    {
+      "epoch": 1.9,
+      "eval_accuracy": 0.9142083333333333,
+      "eval_loss": 0.25844380259513855,
+      "eval_runtime": 15.9686,
+      "eval_samples_per_second": 1502.947,
+      "eval_steps_per_second": 4.697,
+      "step": 4560
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 1e-06,
+      "loss": 0.229,
+      "step": 4580
+    },
+    {
+      "epoch": 1.91,
+      "eval_accuracy": 0.9151666666666667,
+      "eval_loss": 0.2548525631427765,
+      "eval_runtime": 15.8099,
+      "eval_samples_per_second": 1518.04,
+      "eval_steps_per_second": 4.744,
+      "step": 4580
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 1e-06,
+      "loss": 0.2584,
+      "step": 4600
+    },
+    {
+      "epoch": 1.92,
+      "eval_accuracy": 0.9155,
+      "eval_loss": 0.25401976704597473,
+      "eval_runtime": 15.9608,
+      "eval_samples_per_second": 1503.688,
+      "eval_steps_per_second": 4.699,
+      "step": 4600
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 1e-06,
+      "loss": 0.3175,
+      "step": 4620
+    },
+    {
+      "epoch": 1.93,
+      "eval_accuracy": 0.9151666666666667,
+      "eval_loss": 0.2564151883125305,
+      "eval_runtime": 16.163,
+      "eval_samples_per_second": 1484.877,
+      "eval_steps_per_second": 4.64,
+      "step": 4620
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 1e-06,
+      "loss": 0.3066,
+      "step": 4640
+    },
+    {
+      "epoch": 1.93,
+      "eval_accuracy": 0.9160416666666666,
+      "eval_loss": 0.2530251443386078,
+      "eval_runtime": 15.8188,
+      "eval_samples_per_second": 1517.184,
+      "eval_steps_per_second": 4.741,
+      "step": 4640
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 1e-06,
+      "loss": 0.2746,
+      "step": 4660
+    },
+    {
+      "epoch": 1.94,
+      "eval_accuracy": 0.9144583333333334,
+      "eval_loss": 0.25700142979621887,
+      "eval_runtime": 15.9454,
+      "eval_samples_per_second": 1505.134,
+      "eval_steps_per_second": 4.704,
+      "step": 4660
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 1e-06,
+      "loss": 0.2608,
+      "step": 4680
+    },
+    {
+      "epoch": 1.95,
+      "eval_accuracy": 0.915625,
+      "eval_loss": 0.25347229838371277,
+      "eval_runtime": 15.8024,
+      "eval_samples_per_second": 1518.76,
+      "eval_steps_per_second": 4.746,
+      "step": 4680
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 1e-06,
+      "loss": 0.233,
+      "step": 4700
+    },
+    {
+      "epoch": 1.96,
+      "eval_accuracy": 0.9157083333333333,
+      "eval_loss": 0.2529941201210022,
+      "eval_runtime": 16.3246,
+      "eval_samples_per_second": 1470.174,
+      "eval_steps_per_second": 4.594,
+      "step": 4700
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 1e-06,
+      "loss": 0.2556,
+      "step": 4720
+    },
+    {
+      "epoch": 1.97,
+      "eval_accuracy": 0.9156666666666666,
+      "eval_loss": 0.25271856784820557,
+      "eval_runtime": 15.8607,
+      "eval_samples_per_second": 1513.176,
+      "eval_steps_per_second": 4.729,
+      "step": 4720
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1e-06,
+      "loss": 0.253,
+      "step": 4740
+    },
+    {
+      "epoch": 1.98,
+      "eval_accuracy": 0.9163333333333333,
+      "eval_loss": 0.251432865858078,
+      "eval_runtime": 16.1859,
+      "eval_samples_per_second": 1482.768,
+      "eval_steps_per_second": 4.634,
+      "step": 4740
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1e-06,
+      "loss": 0.212,
+      "step": 4760
+    },
+    {
+      "epoch": 1.98,
+      "eval_accuracy": 0.9162916666666666,
+      "eval_loss": 0.25113439559936523,
+      "eval_runtime": 15.5374,
+      "eval_samples_per_second": 1544.657,
+      "eval_steps_per_second": 4.827,
+      "step": 4760
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1e-06,
+      "loss": 0.2573,
+      "step": 4780
+    },
+    {
+      "epoch": 1.99,
+      "eval_accuracy": 0.9160833333333334,
+      "eval_loss": 0.25215280055999756,
+      "eval_runtime": 16.3238,
+      "eval_samples_per_second": 1470.242,
+      "eval_steps_per_second": 4.595,
+      "step": 4780
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 1e-06,
+      "loss": 0.2533,
+      "step": 4800
+    },
+    {
+      "epoch": 2.0,
+      "eval_accuracy": 0.9161666666666667,
+      "eval_loss": 0.25234469771385193,
+      "eval_runtime": 15.7714,
+      "eval_samples_per_second": 1521.744,
+      "eval_steps_per_second": 4.755,
+      "step": 4800
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 1e-06,
+      "loss": 0.2638,
+      "step": 4820
+    },
+    {
+      "epoch": 2.01,
+      "eval_accuracy": 0.9162916666666666,
+      "eval_loss": 0.25334304571151733,
+      "eval_runtime": 16.0308,
+      "eval_samples_per_second": 1497.122,
+      "eval_steps_per_second": 4.679,
+      "step": 4820
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2763,
+      "step": 4840
+    },
+    {
+      "epoch": 2.02,
+      "eval_accuracy": 0.9144166666666667,
+      "eval_loss": 0.25832322239875793,
+      "eval_runtime": 15.8766,
+      "eval_samples_per_second": 1511.661,
+      "eval_steps_per_second": 4.724,
+      "step": 4840
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2503,
+      "step": 4860
+    },
+    {
+      "epoch": 2.02,
+      "eval_accuracy": 0.9145416666666667,
+      "eval_loss": 0.2566870450973511,
+      "eval_runtime": 15.4218,
+      "eval_samples_per_second": 1556.237,
+      "eval_steps_per_second": 4.863,
+      "step": 4860
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 1e-06,
+      "loss": 0.2439,
+      "step": 4880
+    },
+    {
+      "epoch": 2.03,
+      "eval_accuracy": 0.916,
+      "eval_loss": 0.2518457770347595,
+      "eval_runtime": 16.4312,
+      "eval_samples_per_second": 1460.639,
+      "eval_steps_per_second": 4.564,
+      "step": 4880
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 1e-06,
+      "loss": 0.2539,
+      "step": 4900
+    },
+    {
+      "epoch": 2.04,
+      "eval_accuracy": 0.9162083333333333,
+      "eval_loss": 0.25036895275115967,
+      "eval_runtime": 15.7834,
+      "eval_samples_per_second": 1520.583,
+      "eval_steps_per_second": 4.752,
+      "step": 4900
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 1e-06,
+      "loss": 0.2508,
+      "step": 4920
+    },
+    {
+      "epoch": 2.05,
+      "eval_accuracy": 0.9161666666666667,
+      "eval_loss": 0.25022852420806885,
+      "eval_runtime": 15.2801,
+      "eval_samples_per_second": 1570.674,
+      "eval_steps_per_second": 4.908,
+      "step": 4920
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 1e-06,
+      "loss": 0.2793,
+      "step": 4940
+    },
+    {
+      "epoch": 2.06,
+      "eval_accuracy": 0.917125,
+      "eval_loss": 0.25100791454315186,
+      "eval_runtime": 15.9757,
+      "eval_samples_per_second": 1502.281,
+      "eval_steps_per_second": 4.695,
+      "step": 4940
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 1e-06,
+      "loss": 0.2369,
+      "step": 4960
+    },
+    {
+      "epoch": 2.07,
+      "eval_accuracy": 0.9169166666666667,
+      "eval_loss": 0.25172725319862366,
+      "eval_runtime": 15.9075,
+      "eval_samples_per_second": 1508.718,
+      "eval_steps_per_second": 4.715,
+      "step": 4960
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2337,
+      "step": 4980
+    },
+    {
+      "epoch": 2.08,
+      "eval_accuracy": 0.9175833333333333,
+      "eval_loss": 0.24899640679359436,
+      "eval_runtime": 15.735,
+      "eval_samples_per_second": 1525.264,
+      "eval_steps_per_second": 4.766,
+      "step": 4980
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2683,
+      "step": 5000
+    },
+    {
+      "epoch": 2.08,
+      "eval_accuracy": 0.9167916666666667,
+      "eval_loss": 0.24926243722438812,
+      "eval_runtime": 16.2679,
+      "eval_samples_per_second": 1475.302,
+      "eval_steps_per_second": 4.61,
+      "step": 5000
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 1e-06,
+      "loss": 0.2288,
+      "step": 5020
+    },
+    {
+      "epoch": 2.09,
+      "eval_accuracy": 0.9171666666666667,
+      "eval_loss": 0.24971893429756165,
+      "eval_runtime": 15.7076,
+      "eval_samples_per_second": 1527.924,
+      "eval_steps_per_second": 4.775,
+      "step": 5020
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 1e-06,
+      "loss": 0.209,
+      "step": 5040
+    },
+    {
+      "epoch": 2.1,
+      "eval_accuracy": 0.9169583333333333,
+      "eval_loss": 0.24874989688396454,
+      "eval_runtime": 15.9175,
+      "eval_samples_per_second": 1507.777,
+      "eval_steps_per_second": 4.712,
+      "step": 5040
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 1e-06,
+      "loss": 0.2272,
+      "step": 5060
+    },
+    {
+      "epoch": 2.11,
+      "eval_accuracy": 0.9155416666666667,
+      "eval_loss": 0.2508015036582947,
+      "eval_runtime": 15.4995,
+      "eval_samples_per_second": 1548.44,
+      "eval_steps_per_second": 4.839,
+      "step": 5060
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2537,
+      "step": 5080
+    },
+    {
+      "epoch": 2.12,
+      "eval_accuracy": 0.9165833333333333,
+      "eval_loss": 0.250166654586792,
+      "eval_runtime": 16.1014,
+      "eval_samples_per_second": 1490.551,
+      "eval_steps_per_second": 4.658,
+      "step": 5080
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 1e-06,
+      "loss": 0.3171,
+      "step": 5100
+    },
+    {
+      "epoch": 2.12,
+      "eval_accuracy": 0.9159166666666667,
+      "eval_loss": 0.25174057483673096,
+      "eval_runtime": 15.868,
+      "eval_samples_per_second": 1512.48,
+      "eval_steps_per_second": 4.727,
+      "step": 5100
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 1e-06,
+      "loss": 0.1955,
+      "step": 5120
+    },
+    {
+      "epoch": 2.13,
+      "eval_accuracy": 0.9165833333333333,
+      "eval_loss": 0.2481849193572998,
+      "eval_runtime": 15.8739,
+      "eval_samples_per_second": 1511.915,
+      "eval_steps_per_second": 4.725,
+      "step": 5120
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 1e-06,
+      "loss": 0.2618,
+      "step": 5140
+    },
+    {
+      "epoch": 2.14,
+      "eval_accuracy": 0.9178333333333333,
+      "eval_loss": 0.24656133353710175,
+      "eval_runtime": 16.2181,
+      "eval_samples_per_second": 1479.83,
+      "eval_steps_per_second": 4.624,
+      "step": 5140
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 1e-06,
+      "loss": 0.2334,
+      "step": 5160
+    },
+    {
+      "epoch": 2.15,
+      "eval_accuracy": 0.9174583333333334,
+      "eval_loss": 0.24714942276477814,
+      "eval_runtime": 15.6119,
+      "eval_samples_per_second": 1537.293,
+      "eval_steps_per_second": 4.804,
+      "step": 5160
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 1e-06,
+      "loss": 0.2253,
+      "step": 5180
+    },
+    {
+      "epoch": 2.16,
+      "eval_accuracy": 0.917125,
+      "eval_loss": 0.2473263293504715,
+      "eval_runtime": 16.4346,
+      "eval_samples_per_second": 1460.331,
+      "eval_steps_per_second": 4.564,
+      "step": 5180
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 1e-06,
+      "loss": 0.2735,
+      "step": 5200
+    },
+    {
+      "epoch": 2.17,
+      "eval_accuracy": 0.9169583333333333,
+      "eval_loss": 0.24832595884799957,
+      "eval_runtime": 15.9352,
+      "eval_samples_per_second": 1506.103,
+      "eval_steps_per_second": 4.707,
+      "step": 5200
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 1e-06,
+      "loss": 0.2119,
+      "step": 5220
+    },
+    {
+      "epoch": 2.17,
+      "eval_accuracy": 0.9175,
+      "eval_loss": 0.24637369811534882,
+      "eval_runtime": 15.8502,
+      "eval_samples_per_second": 1514.172,
+      "eval_steps_per_second": 4.732,
+      "step": 5220
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 1e-06,
+      "loss": 0.2338,
+      "step": 5240
+    },
+    {
+      "epoch": 2.18,
+      "eval_accuracy": 0.9175416666666667,
+      "eval_loss": 0.24698366224765778,
+      "eval_runtime": 15.7545,
+      "eval_samples_per_second": 1523.378,
+      "eval_steps_per_second": 4.761,
+      "step": 5240
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 1e-06,
+      "loss": 0.2233,
+      "step": 5260
+    },
+    {
+      "epoch": 2.19,
+      "eval_accuracy": 0.9155,
+      "eval_loss": 0.24797461926937103,
+      "eval_runtime": 15.7338,
+      "eval_samples_per_second": 1525.377,
+      "eval_steps_per_second": 4.767,
+      "step": 5260
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 1e-06,
+      "loss": 0.243,
+      "step": 5280
+    },
+    {
+      "epoch": 2.2,
+      "eval_accuracy": 0.916375,
+      "eval_loss": 0.24534855782985687,
+      "eval_runtime": 15.9549,
+      "eval_samples_per_second": 1504.244,
+      "eval_steps_per_second": 4.701,
+      "step": 5280
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 1e-06,
+      "loss": 0.2476,
+      "step": 5300
+    },
+    {
+      "epoch": 2.21,
+      "eval_accuracy": 0.917125,
+      "eval_loss": 0.24429555237293243,
+      "eval_runtime": 16.2563,
+      "eval_samples_per_second": 1476.355,
+      "eval_steps_per_second": 4.614,
+      "step": 5300
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 1e-06,
+      "loss": 0.2481,
+      "step": 5320
+    },
+    {
+      "epoch": 2.22,
+      "eval_accuracy": 0.9174166666666667,
+      "eval_loss": 0.24369333684444427,
+      "eval_runtime": 15.6541,
+      "eval_samples_per_second": 1533.143,
+      "eval_steps_per_second": 4.791,
+      "step": 5320
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 1e-06,
+      "loss": 0.2505,
+      "step": 5340
+    },
+    {
+      "epoch": 2.23,
+      "eval_accuracy": 0.9179583333333333,
+      "eval_loss": 0.24328631162643433,
+      "eval_runtime": 15.9128,
+      "eval_samples_per_second": 1508.223,
+      "eval_steps_per_second": 4.713,
+      "step": 5340
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 1e-06,
+      "loss": 0.2591,
+      "step": 5360
+    },
+    {
+      "epoch": 2.23,
+      "eval_accuracy": 0.9164583333333334,
+      "eval_loss": 0.24514025449752808,
+      "eval_runtime": 16.021,
+      "eval_samples_per_second": 1498.036,
+      "eval_steps_per_second": 4.681,
+      "step": 5360
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 1e-06,
+      "loss": 0.2342,
+      "step": 5380
+    },
+    {
+      "epoch": 2.24,
+      "eval_accuracy": 0.9154166666666667,
+      "eval_loss": 0.2472737431526184,
+      "eval_runtime": 15.688,
+      "eval_samples_per_second": 1529.832,
+      "eval_steps_per_second": 4.781,
+      "step": 5380
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 1e-06,
+      "loss": 0.2405,
+      "step": 5400
+    },
+    {
+      "epoch": 2.25,
+      "eval_accuracy": 0.9179166666666667,
+      "eval_loss": 0.24237360060214996,
+      "eval_runtime": 16.4849,
+      "eval_samples_per_second": 1455.882,
+      "eval_steps_per_second": 4.55,
+      "step": 5400
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 1e-06,
+      "loss": 0.2803,
+      "step": 5420
+    },
+    {
+      "epoch": 2.26,
+      "eval_accuracy": 0.9185416666666667,
+      "eval_loss": 0.2412734031677246,
+      "eval_runtime": 15.7027,
+      "eval_samples_per_second": 1528.403,
+      "eval_steps_per_second": 4.776,
+      "step": 5420
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 1e-06,
+      "loss": 0.2816,
+      "step": 5440
+    },
+    {
+      "epoch": 2.27,
+      "eval_accuracy": 0.9178333333333333,
+      "eval_loss": 0.2425076812505722,
+      "eval_runtime": 15.8848,
+      "eval_samples_per_second": 1510.874,
+      "eval_steps_per_second": 4.721,
+      "step": 5440
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 1e-06,
+      "loss": 0.2489,
+      "step": 5460
+    },
+    {
+      "epoch": 2.27,
+      "eval_accuracy": 0.9172916666666666,
+      "eval_loss": 0.24504542350769043,
+      "eval_runtime": 16.3712,
+      "eval_samples_per_second": 1465.986,
+      "eval_steps_per_second": 4.581,
+      "step": 5460
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 1e-06,
+      "loss": 0.2346,
+      "step": 5480
+    },
+    {
+      "epoch": 2.28,
+      "eval_accuracy": 0.9180833333333334,
+      "eval_loss": 0.2423253357410431,
+      "eval_runtime": 15.5334,
+      "eval_samples_per_second": 1545.062,
+      "eval_steps_per_second": 4.828,
+      "step": 5480
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 1e-06,
+      "loss": 0.251,
+      "step": 5500
+    },
+    {
+      "epoch": 2.29,
+      "eval_accuracy": 0.9185,
+      "eval_loss": 0.24269212782382965,
+      "eval_runtime": 16.0617,
+      "eval_samples_per_second": 1494.234,
+      "eval_steps_per_second": 4.669,
+      "step": 5500
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 1e-06,
+      "loss": 0.2574,
+      "step": 5520
+    },
+    {
+      "epoch": 2.3,
+      "eval_accuracy": 0.9187083333333333,
+      "eval_loss": 0.2401323914527893,
+      "eval_runtime": 15.7783,
+      "eval_samples_per_second": 1521.078,
+      "eval_steps_per_second": 4.753,
+      "step": 5520
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 1e-06,
+      "loss": 0.2227,
+      "step": 5540
+    },
+    {
+      "epoch": 2.31,
+      "eval_accuracy": 0.9179583333333333,
+      "eval_loss": 0.24057930707931519,
+      "eval_runtime": 16.3047,
+      "eval_samples_per_second": 1471.971,
+      "eval_steps_per_second": 4.6,
+      "step": 5540
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 1e-06,
+      "loss": 0.2362,
+      "step": 5560
+    },
+    {
+      "epoch": 2.32,
+      "eval_accuracy": 0.9185833333333333,
+      "eval_loss": 0.23938481509685516,
+      "eval_runtime": 15.6702,
+      "eval_samples_per_second": 1531.572,
+      "eval_steps_per_second": 4.786,
+      "step": 5560
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 1e-06,
+      "loss": 0.2346,
+      "step": 5580
+    },
+    {
+      "epoch": 2.33,
+      "eval_accuracy": 0.918625,
+      "eval_loss": 0.23980508744716644,
+      "eval_runtime": 15.999,
+      "eval_samples_per_second": 1500.095,
+      "eval_steps_per_second": 4.688,
+      "step": 5580
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 1e-06,
+      "loss": 0.2308,
+      "step": 5600
+    },
+    {
+      "epoch": 2.33,
+      "eval_accuracy": 0.9172083333333333,
+      "eval_loss": 0.24078369140625,
+      "eval_runtime": 16.2118,
+      "eval_samples_per_second": 1480.402,
+      "eval_steps_per_second": 4.626,
+      "step": 5600
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 1e-06,
+      "loss": 0.2668,
+      "step": 5620
+    },
+    {
+      "epoch": 2.34,
+      "eval_accuracy": 0.9185833333333333,
+      "eval_loss": 0.23922114074230194,
+      "eval_runtime": 15.9955,
+      "eval_samples_per_second": 1500.419,
+      "eval_steps_per_second": 4.689,
+      "step": 5620
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 1e-06,
+      "loss": 0.2363,
+      "step": 5640
+    },
+    {
+      "epoch": 2.35,
+      "eval_accuracy": 0.9182083333333333,
+      "eval_loss": 0.23886892199516296,
+      "eval_runtime": 15.6875,
+      "eval_samples_per_second": 1529.876,
+      "eval_steps_per_second": 4.781,
+      "step": 5640
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 1e-06,
+      "loss": 0.2152,
+      "step": 5660
+    },
+    {
+      "epoch": 2.36,
+      "eval_accuracy": 0.918125,
+      "eval_loss": 0.23896987736225128,
+      "eval_runtime": 15.4793,
+      "eval_samples_per_second": 1550.456,
+      "eval_steps_per_second": 4.845,
+      "step": 5660
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 1e-06,
+      "loss": 0.2759,
+      "step": 5680
+    },
+    {
+      "epoch": 2.37,
+      "eval_accuracy": 0.9181666666666667,
+      "eval_loss": 0.23786494135856628,
+      "eval_runtime": 15.5978,
+      "eval_samples_per_second": 1538.675,
+      "eval_steps_per_second": 4.808,
+      "step": 5680
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 1e-06,
+      "loss": 0.2274,
+      "step": 5700
+    },
+    {
+      "epoch": 2.38,
+      "eval_accuracy": 0.9188333333333333,
+      "eval_loss": 0.23672978579998016,
+      "eval_runtime": 15.6992,
+      "eval_samples_per_second": 1528.744,
+      "eval_steps_per_second": 4.777,
+      "step": 5700
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 1e-06,
+      "loss": 0.239,
+      "step": 5720
+    },
+    {
+      "epoch": 2.38,
+      "eval_accuracy": 0.9190833333333334,
+      "eval_loss": 0.23753681778907776,
+      "eval_runtime": 15.6792,
+      "eval_samples_per_second": 1530.692,
+      "eval_steps_per_second": 4.783,
+      "step": 5720
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 1e-06,
+      "loss": 0.2195,
+      "step": 5740
+    },
+    {
+      "epoch": 2.39,
+      "eval_accuracy": 0.9187083333333333,
+      "eval_loss": 0.2378937304019928,
+      "eval_runtime": 16.0105,
+      "eval_samples_per_second": 1499.013,
+      "eval_steps_per_second": 4.684,
+      "step": 5740
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 1e-06,
+      "loss": 0.2454,
+      "step": 5760
+    },
+    {
+      "epoch": 2.4,
+      "eval_accuracy": 0.9187916666666667,
+      "eval_loss": 0.23726117610931396,
+      "eval_runtime": 15.4532,
+      "eval_samples_per_second": 1553.077,
+      "eval_steps_per_second": 4.853,
+      "step": 5760
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1e-06,
+      "loss": 0.2376,
+      "step": 5780
+    },
+    {
+      "epoch": 2.41,
+      "eval_accuracy": 0.9192916666666666,
+      "eval_loss": 0.23683802783489227,
+      "eval_runtime": 16.4115,
+      "eval_samples_per_second": 1462.385,
+      "eval_steps_per_second": 4.57,
+      "step": 5780
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1e-06,
+      "loss": 0.253,
+      "step": 5800
+    },
+    {
+      "epoch": 2.42,
+      "eval_accuracy": 0.9193333333333333,
+      "eval_loss": 0.23695851862430573,
+      "eval_runtime": 15.7863,
+      "eval_samples_per_second": 1520.307,
+      "eval_steps_per_second": 4.751,
+      "step": 5800
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2365,
+      "step": 5820
+    },
+    {
+      "epoch": 2.42,
+      "eval_accuracy": 0.919375,
+      "eval_loss": 0.23560389876365662,
+      "eval_runtime": 16.216,
+      "eval_samples_per_second": 1480.022,
+      "eval_steps_per_second": 4.625,
+      "step": 5820
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 1e-06,
+      "loss": 0.2025,
+      "step": 5840
+    },
+    {
+      "epoch": 2.43,
+      "eval_accuracy": 0.9199583333333333,
+      "eval_loss": 0.2355504035949707,
+      "eval_runtime": 15.4916,
+      "eval_samples_per_second": 1549.222,
+      "eval_steps_per_second": 4.841,
+      "step": 5840
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 1e-06,
+      "loss": 0.2115,
+      "step": 5860
+    },
+    {
+      "epoch": 2.44,
+      "eval_accuracy": 0.9187083333333333,
+      "eval_loss": 0.23711217939853668,
+      "eval_runtime": 15.6181,
+      "eval_samples_per_second": 1536.68,
+      "eval_steps_per_second": 4.802,
+      "step": 5860
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1e-06,
+      "loss": 0.191,
+      "step": 5880
+    },
+    {
+      "epoch": 2.45,
+      "eval_accuracy": 0.9195,
+      "eval_loss": 0.2373773753643036,
+      "eval_runtime": 15.8036,
+      "eval_samples_per_second": 1518.638,
+      "eval_steps_per_second": 4.746,
+      "step": 5880
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1e-06,
+      "loss": 0.2228,
+      "step": 5900
+    },
+    {
+      "epoch": 2.46,
+      "eval_accuracy": 0.9200416666666666,
+      "eval_loss": 0.23531056940555573,
+      "eval_runtime": 15.9661,
+      "eval_samples_per_second": 1503.188,
+      "eval_steps_per_second": 4.697,
+      "step": 5900
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 1e-06,
+      "loss": 0.2151,
+      "step": 5920
+    },
+    {
+      "epoch": 2.47,
+      "eval_accuracy": 0.919375,
+      "eval_loss": 0.23577865958213806,
+      "eval_runtime": 15.3031,
+      "eval_samples_per_second": 1568.307,
+      "eval_steps_per_second": 4.901,
+      "step": 5920
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2351,
+      "step": 5940
+    },
+    {
+      "epoch": 2.48,
+      "eval_accuracy": 0.9205,
+      "eval_loss": 0.23470845818519592,
+      "eval_runtime": 15.6649,
+      "eval_samples_per_second": 1532.088,
+      "eval_steps_per_second": 4.788,
+      "step": 5940
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2535,
+      "step": 5960
+    },
+    {
+      "epoch": 2.48,
+      "eval_accuracy": 0.9204166666666667,
+      "eval_loss": 0.23469573259353638,
+      "eval_runtime": 15.8743,
+      "eval_samples_per_second": 1511.875,
+      "eval_steps_per_second": 4.725,
+      "step": 5960
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1e-06,
+      "loss": 0.2646,
+      "step": 5980
+    },
+    {
+      "epoch": 2.49,
+      "eval_accuracy": 0.9199166666666667,
+      "eval_loss": 0.23572835326194763,
+      "eval_runtime": 15.9499,
+      "eval_samples_per_second": 1504.715,
+      "eval_steps_per_second": 4.702,
+      "step": 5980
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 1e-06,
+      "loss": 0.2495,
+      "step": 6000
+    },
+    {
+      "epoch": 2.5,
+      "eval_accuracy": 0.9185,
+      "eval_loss": 0.237389475107193,
+      "eval_runtime": 15.8761,
+      "eval_samples_per_second": 1511.71,
+      "eval_steps_per_second": 4.724,
+      "step": 6000
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1e-06,
+      "loss": 0.2383,
+      "step": 6020
+    },
+    {
+      "epoch": 2.51,
+      "eval_accuracy": 0.91875,
+      "eval_loss": 0.23722399771213531,
+      "eval_runtime": 16.1535,
+      "eval_samples_per_second": 1485.747,
+      "eval_steps_per_second": 4.643,
+      "step": 6020
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2103,
+      "step": 6040
+    },
+    {
+      "epoch": 2.52,
+      "eval_accuracy": 0.9197916666666667,
+      "eval_loss": 0.23570208251476288,
+      "eval_runtime": 15.6602,
+      "eval_samples_per_second": 1532.549,
+      "eval_steps_per_second": 4.789,
+      "step": 6040
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2667,
+      "step": 6060
+    },
+    {
+      "epoch": 2.52,
+      "eval_accuracy": 0.9205,
+      "eval_loss": 0.2345227301120758,
+      "eval_runtime": 15.9993,
+      "eval_samples_per_second": 1500.065,
+      "eval_steps_per_second": 4.688,
+      "step": 6060
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1e-06,
+      "loss": 0.2229,
+      "step": 6080
+    },
+    {
+      "epoch": 2.53,
+      "eval_accuracy": 0.9203333333333333,
+      "eval_loss": 0.23675419390201569,
+      "eval_runtime": 16.0513,
+      "eval_samples_per_second": 1495.205,
+      "eval_steps_per_second": 4.673,
+      "step": 6080
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1e-06,
+      "loss": 0.2794,
+      "step": 6100
+    },
+    {
+      "epoch": 2.54,
+      "eval_accuracy": 0.9181666666666667,
+      "eval_loss": 0.2398298680782318,
+      "eval_runtime": 15.8057,
+      "eval_samples_per_second": 1518.443,
+      "eval_steps_per_second": 4.745,
+      "step": 6100
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1e-06,
+      "loss": 0.2191,
+      "step": 6120
+    },
+    {
+      "epoch": 2.55,
+      "eval_accuracy": 0.919125,
+      "eval_loss": 0.23680013418197632,
+      "eval_runtime": 16.0686,
+      "eval_samples_per_second": 1493.6,
+      "eval_steps_per_second": 4.668,
+      "step": 6120
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 1e-06,
+      "loss": 0.263,
+      "step": 6140
+    },
+    {
+      "epoch": 2.56,
+      "eval_accuracy": 0.9210833333333334,
+      "eval_loss": 0.23318885266780853,
+      "eval_runtime": 15.8077,
+      "eval_samples_per_second": 1518.25,
+      "eval_steps_per_second": 4.745,
+      "step": 6140
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1e-06,
+      "loss": 0.2008,
+      "step": 6160
+    },
+    {
+      "epoch": 2.57,
+      "eval_accuracy": 0.9214166666666667,
+      "eval_loss": 0.2328587919473648,
+      "eval_runtime": 15.7426,
+      "eval_samples_per_second": 1524.526,
+      "eval_steps_per_second": 4.764,
+      "step": 6160
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2226,
+      "step": 6180
+    },
+    {
+      "epoch": 2.58,
+      "eval_accuracy": 0.9210416666666666,
+      "eval_loss": 0.23366689682006836,
+      "eval_runtime": 15.6914,
+      "eval_samples_per_second": 1529.497,
+      "eval_steps_per_second": 4.78,
+      "step": 6180
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2261,
+      "step": 6200
+    },
+    {
+      "epoch": 2.58,
+      "eval_accuracy": 0.9207083333333334,
+      "eval_loss": 0.23435989022254944,
+      "eval_runtime": 15.7413,
+      "eval_samples_per_second": 1524.655,
+      "eval_steps_per_second": 4.765,
+      "step": 6200
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 1e-06,
+      "loss": 0.2481,
+      "step": 6220
+    },
+    {
+      "epoch": 2.59,
+      "eval_accuracy": 0.9221666666666667,
+      "eval_loss": 0.23215261101722717,
+      "eval_runtime": 15.7855,
+      "eval_samples_per_second": 1520.387,
+      "eval_steps_per_second": 4.751,
+      "step": 6220
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 1e-06,
+      "loss": 0.2235,
+      "step": 6240
+    },
+    {
+      "epoch": 2.6,
+      "eval_accuracy": 0.9214583333333334,
+      "eval_loss": 0.23404672741889954,
+      "eval_runtime": 16.0667,
+      "eval_samples_per_second": 1493.777,
+      "eval_steps_per_second": 4.668,
+      "step": 6240
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 1e-06,
+      "loss": 0.2821,
+      "step": 6260
+    },
+    {
+      "epoch": 2.61,
+      "eval_accuracy": 0.9206666666666666,
+      "eval_loss": 0.2340671271085739,
+      "eval_runtime": 16.2416,
+      "eval_samples_per_second": 1477.689,
+      "eval_steps_per_second": 4.618,
+      "step": 6260
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1988,
+      "step": 6280
+    },
+    {
+      "epoch": 2.62,
+      "eval_accuracy": 0.9215,
+      "eval_loss": 0.2360154539346695,
+      "eval_runtime": 15.5902,
+      "eval_samples_per_second": 1539.425,
+      "eval_steps_per_second": 4.811,
+      "step": 6280
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 1e-06,
+      "loss": 0.2407,
+      "step": 6300
+    },
+    {
+      "epoch": 2.62,
+      "eval_accuracy": 0.9207083333333334,
+      "eval_loss": 0.23405057191848755,
+      "eval_runtime": 16.1083,
+      "eval_samples_per_second": 1489.915,
+      "eval_steps_per_second": 4.656,
+      "step": 6300
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 1e-06,
+      "loss": 0.2138,
+      "step": 6320
+    },
+    {
+      "epoch": 2.63,
+      "eval_accuracy": 0.9222916666666666,
+      "eval_loss": 0.23155492544174194,
+      "eval_runtime": 15.9219,
+      "eval_samples_per_second": 1507.362,
+      "eval_steps_per_second": 4.711,
+      "step": 6320
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 1e-06,
+      "loss": 0.2171,
+      "step": 6340
+    },
+    {
+      "epoch": 2.64,
+      "eval_accuracy": 0.9215833333333333,
+      "eval_loss": 0.23096613585948944,
+      "eval_runtime": 15.2563,
+      "eval_samples_per_second": 1573.117,
+      "eval_steps_per_second": 4.916,
+      "step": 6340
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 1e-06,
+      "loss": 0.2302,
+      "step": 6360
+    },
+    {
+      "epoch": 2.65,
+      "eval_accuracy": 0.9212083333333333,
+      "eval_loss": 0.23060353100299835,
+      "eval_runtime": 16.0532,
+      "eval_samples_per_second": 1495.028,
+      "eval_steps_per_second": 4.672,
+      "step": 6360
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 1e-06,
+      "loss": 0.2394,
+      "step": 6380
+    },
+    {
+      "epoch": 2.66,
+      "eval_accuracy": 0.9209166666666667,
+      "eval_loss": 0.23088908195495605,
+      "eval_runtime": 15.9256,
+      "eval_samples_per_second": 1507.003,
+      "eval_steps_per_second": 4.709,
+      "step": 6380
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 1e-06,
+      "loss": 0.2321,
+      "step": 6400
+    },
+    {
+      "epoch": 2.67,
+      "eval_accuracy": 0.9219166666666667,
+      "eval_loss": 0.22929558157920837,
+      "eval_runtime": 15.7096,
+      "eval_samples_per_second": 1527.727,
+      "eval_steps_per_second": 4.774,
+      "step": 6400
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 1e-06,
+      "loss": 0.2011,
+      "step": 6420
+    },
+    {
+      "epoch": 2.67,
+      "eval_accuracy": 0.92225,
+      "eval_loss": 0.2290237993001938,
+      "eval_runtime": 15.5552,
+      "eval_samples_per_second": 1542.895,
+      "eval_steps_per_second": 4.822,
+      "step": 6420
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 1e-06,
+      "loss": 0.2235,
+      "step": 6440
+    },
+    {
+      "epoch": 2.68,
+      "eval_accuracy": 0.9219583333333333,
+      "eval_loss": 0.23050343990325928,
+      "eval_runtime": 16.0981,
+      "eval_samples_per_second": 1490.855,
+      "eval_steps_per_second": 4.659,
+      "step": 6440
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 1e-06,
+      "loss": 0.2638,
+      "step": 6460
+    },
+    {
+      "epoch": 2.69,
+      "eval_accuracy": 0.9218333333333333,
+      "eval_loss": 0.23081812262535095,
+      "eval_runtime": 16.0097,
+      "eval_samples_per_second": 1499.096,
+      "eval_steps_per_second": 4.685,
+      "step": 6460
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 1e-06,
+      "loss": 0.2767,
+      "step": 6480
+    },
+    {
+      "epoch": 2.7,
+      "eval_accuracy": 0.92225,
+      "eval_loss": 0.23322133719921112,
+      "eval_runtime": 16.0242,
+      "eval_samples_per_second": 1497.736,
+      "eval_steps_per_second": 4.68,
+      "step": 6480
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 1e-06,
+      "loss": 0.2332,
+      "step": 6500
+    },
+    {
+      "epoch": 2.71,
+      "eval_accuracy": 0.9226666666666666,
+      "eval_loss": 0.22899757325649261,
+      "eval_runtime": 16.6235,
+      "eval_samples_per_second": 1443.741,
+      "eval_steps_per_second": 4.512,
+      "step": 6500
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 1e-06,
+      "loss": 0.2104,
+      "step": 6520
+    },
+    {
+      "epoch": 2.72,
+      "eval_accuracy": 0.9220833333333334,
+      "eval_loss": 0.22903680801391602,
+      "eval_runtime": 15.8598,
+      "eval_samples_per_second": 1513.257,
+      "eval_steps_per_second": 4.729,
+      "step": 6520
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 1e-06,
+      "loss": 0.2148,
+      "step": 6540
+    },
+    {
+      "epoch": 2.73,
+      "eval_accuracy": 0.9219166666666667,
+      "eval_loss": 0.2291133552789688,
+      "eval_runtime": 16.0042,
+      "eval_samples_per_second": 1499.604,
+      "eval_steps_per_second": 4.686,
+      "step": 6540
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 1e-06,
+      "loss": 0.2708,
+      "step": 6560
+    },
+    {
+      "epoch": 2.73,
+      "eval_accuracy": 0.922,
+      "eval_loss": 0.22891123592853546,
+      "eval_runtime": 15.8338,
+      "eval_samples_per_second": 1515.741,
+      "eval_steps_per_second": 4.737,
+      "step": 6560
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 1e-06,
+      "loss": 0.2155,
+      "step": 6580
+    },
+    {
+      "epoch": 2.74,
+      "eval_accuracy": 0.9213333333333333,
+      "eval_loss": 0.2305193543434143,
+      "eval_runtime": 15.7359,
+      "eval_samples_per_second": 1525.179,
+      "eval_steps_per_second": 4.766,
+      "step": 6580
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 1e-06,
+      "loss": 0.2149,
+      "step": 6600
+    },
+    {
+      "epoch": 2.75,
+      "eval_accuracy": 0.921375,
+      "eval_loss": 0.22942766547203064,
+      "eval_runtime": 15.983,
+      "eval_samples_per_second": 1501.595,
+      "eval_steps_per_second": 4.692,
+      "step": 6600
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 1e-06,
+      "loss": 0.2396,
+      "step": 6620
+    },
+    {
+      "epoch": 2.76,
+      "eval_accuracy": 0.9215,
+      "eval_loss": 0.23061256110668182,
+      "eval_runtime": 15.913,
+      "eval_samples_per_second": 1508.203,
+      "eval_steps_per_second": 4.713,
+      "step": 6620
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 1e-06,
+      "loss": 0.2572,
+      "step": 6640
+    },
+    {
+      "epoch": 2.77,
+      "eval_accuracy": 0.9215,
+      "eval_loss": 0.23231017589569092,
+      "eval_runtime": 15.4396,
+      "eval_samples_per_second": 1554.44,
+      "eval_steps_per_second": 4.858,
+      "step": 6640
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 1e-06,
+      "loss": 0.2536,
+      "step": 6660
+    },
+    {
+      "epoch": 2.77,
+      "eval_accuracy": 0.9224166666666667,
+      "eval_loss": 0.2280372679233551,
+      "eval_runtime": 15.4345,
+      "eval_samples_per_second": 1554.958,
+      "eval_steps_per_second": 4.859,
+      "step": 6660
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1892,
+      "step": 6680
+    },
+    {
+      "epoch": 2.78,
+      "eval_accuracy": 0.9220416666666666,
+      "eval_loss": 0.22818611562252045,
+      "eval_runtime": 16.0452,
+      "eval_samples_per_second": 1495.777,
+      "eval_steps_per_second": 4.674,
+      "step": 6680
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 1e-06,
+      "loss": 0.2689,
+      "step": 6700
+    },
+    {
+      "epoch": 2.79,
+      "eval_accuracy": 0.92125,
+      "eval_loss": 0.229068323969841,
+      "eval_runtime": 15.6442,
+      "eval_samples_per_second": 1534.111,
+      "eval_steps_per_second": 4.794,
+      "step": 6700
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 1e-06,
+      "loss": 0.3018,
+      "step": 6720
+    },
+    {
+      "epoch": 2.8,
+      "eval_accuracy": 0.9225,
+      "eval_loss": 0.22712692618370056,
+      "eval_runtime": 15.8562,
+      "eval_samples_per_second": 1513.605,
+      "eval_steps_per_second": 4.73,
+      "step": 6720
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1971,
+      "step": 6740
+    },
+    {
+      "epoch": 2.81,
+      "eval_accuracy": 0.9229583333333333,
+      "eval_loss": 0.22696258127689362,
+      "eval_runtime": 15.7829,
+      "eval_samples_per_second": 1520.631,
+      "eval_steps_per_second": 4.752,
+      "step": 6740
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 1e-06,
+      "loss": 0.2276,
+      "step": 6760
+    },
+    {
+      "epoch": 2.82,
+      "eval_accuracy": 0.923375,
+      "eval_loss": 0.22678209841251373,
+      "eval_runtime": 15.81,
+      "eval_samples_per_second": 1518.027,
+      "eval_steps_per_second": 4.744,
+      "step": 6760
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 1e-06,
+      "loss": 0.2141,
+      "step": 6780
+    },
+    {
+      "epoch": 2.83,
+      "eval_accuracy": 0.923,
+      "eval_loss": 0.2268705815076828,
+      "eval_runtime": 16.2609,
+      "eval_samples_per_second": 1475.93,
+      "eval_steps_per_second": 4.612,
+      "step": 6780
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 1e-06,
+      "loss": 0.2376,
+      "step": 6800
+    },
+    {
+      "epoch": 2.83,
+      "eval_accuracy": 0.9234583333333334,
+      "eval_loss": 0.22707809507846832,
+      "eval_runtime": 15.6779,
+      "eval_samples_per_second": 1530.813,
+      "eval_steps_per_second": 4.784,
+      "step": 6800
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 1e-06,
+      "loss": 0.2237,
+      "step": 6820
+    },
+    {
+      "epoch": 2.84,
+      "eval_accuracy": 0.9207916666666667,
+      "eval_loss": 0.231236070394516,
+      "eval_runtime": 16.1718,
+      "eval_samples_per_second": 1484.064,
+      "eval_steps_per_second": 4.638,
+      "step": 6820
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 1e-06,
+      "loss": 0.2114,
+      "step": 6840
+    },
+    {
+      "epoch": 2.85,
+      "eval_accuracy": 0.9222083333333333,
+      "eval_loss": 0.2280959039926529,
+      "eval_runtime": 15.646,
+      "eval_samples_per_second": 1533.937,
+      "eval_steps_per_second": 4.794,
+      "step": 6840
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 1e-06,
+      "loss": 0.2037,
+      "step": 6860
+    },
+    {
+      "epoch": 2.86,
+      "eval_accuracy": 0.9230833333333334,
+      "eval_loss": 0.22740183770656586,
+      "eval_runtime": 16.2073,
+      "eval_samples_per_second": 1480.812,
+      "eval_steps_per_second": 4.628,
+      "step": 6860
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 1e-06,
+      "loss": 0.2412,
+      "step": 6880
+    },
+    {
+      "epoch": 2.87,
+      "eval_accuracy": 0.9233333333333333,
+      "eval_loss": 0.22586235404014587,
+      "eval_runtime": 15.8595,
+      "eval_samples_per_second": 1513.292,
+      "eval_steps_per_second": 4.729,
+      "step": 6880
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 1e-06,
+      "loss": 0.2105,
+      "step": 6900
+    },
+    {
+      "epoch": 2.88,
+      "eval_accuracy": 0.9239583333333333,
+      "eval_loss": 0.22604452073574066,
+      "eval_runtime": 15.5265,
+      "eval_samples_per_second": 1545.746,
+      "eval_steps_per_second": 4.83,
+      "step": 6900
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 1e-06,
+      "loss": 0.2209,
+      "step": 6920
+    },
+    {
+      "epoch": 2.88,
+      "eval_accuracy": 0.922625,
+      "eval_loss": 0.22909581661224365,
+      "eval_runtime": 16.0771,
+      "eval_samples_per_second": 1492.802,
+      "eval_steps_per_second": 4.665,
+      "step": 6920
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 1e-06,
+      "loss": 0.2323,
+      "step": 6940
+    },
+    {
+      "epoch": 2.89,
+      "eval_accuracy": 0.9240833333333334,
+      "eval_loss": 0.2258378565311432,
+      "eval_runtime": 15.5648,
+      "eval_samples_per_second": 1541.936,
+      "eval_steps_per_second": 4.819,
+      "step": 6940
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 1e-06,
+      "loss": 0.2416,
+      "step": 6960
+    },
+    {
+      "epoch": 2.9,
+      "eval_accuracy": 0.9242916666666666,
+      "eval_loss": 0.22519205510616302,
+      "eval_runtime": 16.1579,
+      "eval_samples_per_second": 1485.345,
+      "eval_steps_per_second": 4.642,
+      "step": 6960
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 1e-06,
+      "loss": 0.2369,
+      "step": 6980
+    },
+    {
+      "epoch": 2.91,
+      "eval_accuracy": 0.9219166666666667,
+      "eval_loss": 0.22782516479492188,
+      "eval_runtime": 15.7414,
+      "eval_samples_per_second": 1524.641,
+      "eval_steps_per_second": 4.765,
+      "step": 6980
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 1e-06,
+      "loss": 0.2218,
+      "step": 7000
+    },
+    {
+      "epoch": 2.92,
+      "eval_accuracy": 0.91975,
+      "eval_loss": 0.23249071836471558,
+      "eval_runtime": 16.1495,
+      "eval_samples_per_second": 1486.113,
+      "eval_steps_per_second": 4.644,
+      "step": 7000
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 1e-06,
+      "loss": 0.2479,
+      "step": 7020
+    },
+    {
+      "epoch": 2.92,
+      "eval_accuracy": 0.922375,
+      "eval_loss": 0.2269277721643448,
+      "eval_runtime": 15.707,
+      "eval_samples_per_second": 1527.979,
+      "eval_steps_per_second": 4.775,
+      "step": 7020
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 1e-06,
+      "loss": 0.2174,
+      "step": 7040
+    },
+    {
+      "epoch": 2.93,
+      "eval_accuracy": 0.9234583333333334,
+      "eval_loss": 0.22529129683971405,
+      "eval_runtime": 15.7426,
+      "eval_samples_per_second": 1524.524,
+      "eval_steps_per_second": 4.764,
+      "step": 7040
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 1e-06,
+      "loss": 0.2243,
+      "step": 7060
+    },
+    {
+      "epoch": 2.94,
+      "eval_accuracy": 0.9230416666666666,
+      "eval_loss": 0.22688570618629456,
+      "eval_runtime": 16.3366,
+      "eval_samples_per_second": 1469.094,
+      "eval_steps_per_second": 4.591,
+      "step": 7060
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 1e-06,
+      "loss": 0.2822,
+      "step": 7080
+    },
+    {
+      "epoch": 2.95,
+      "eval_accuracy": 0.9227916666666667,
+      "eval_loss": 0.23044590651988983,
+      "eval_runtime": 15.8096,
+      "eval_samples_per_second": 1518.063,
+      "eval_steps_per_second": 4.744,
+      "step": 7080
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 1e-06,
+      "loss": 0.2161,
+      "step": 7100
+    },
+    {
+      "epoch": 2.96,
+      "eval_accuracy": 0.9220833333333334,
+      "eval_loss": 0.22724518179893494,
+      "eval_runtime": 16.2528,
+      "eval_samples_per_second": 1476.67,
+      "eval_steps_per_second": 4.615,
+      "step": 7100
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 1e-06,
+      "loss": 0.238,
+      "step": 7120
+    },
+    {
+      "epoch": 2.97,
+      "eval_accuracy": 0.9245,
+      "eval_loss": 0.22448720037937164,
+      "eval_runtime": 15.5514,
+      "eval_samples_per_second": 1543.268,
+      "eval_steps_per_second": 4.823,
+      "step": 7120
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 1e-06,
+      "loss": 0.238,
+      "step": 7140
+    },
+    {
+      "epoch": 2.98,
+      "eval_accuracy": 0.924125,
+      "eval_loss": 0.22591613233089447,
+      "eval_runtime": 15.8997,
+      "eval_samples_per_second": 1509.461,
+      "eval_steps_per_second": 4.717,
+      "step": 7140
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1969,
+      "step": 7160
+    },
+    {
+      "epoch": 2.98,
+      "eval_accuracy": 0.924125,
+      "eval_loss": 0.22452926635742188,
+      "eval_runtime": 15.5234,
+      "eval_samples_per_second": 1546.054,
+      "eval_steps_per_second": 4.831,
+      "step": 7160
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1897,
+      "step": 7180
+    },
+    {
+      "epoch": 2.99,
+      "eval_accuracy": 0.923625,
+      "eval_loss": 0.22510305047035217,
+      "eval_runtime": 15.9816,
+      "eval_samples_per_second": 1501.726,
+      "eval_steps_per_second": 4.693,
+      "step": 7180
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 1e-06,
+      "loss": 0.2168,
+      "step": 7200
+    },
+    {
+      "epoch": 3.0,
+      "eval_accuracy": 0.9232916666666666,
+      "eval_loss": 0.22502471506595612,
+      "eval_runtime": 15.6986,
+      "eval_samples_per_second": 1528.802,
+      "eval_steps_per_second": 4.778,
+      "step": 7200
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 1e-06,
+      "loss": 0.251,
+      "step": 7220
+    },
+    {
+      "epoch": 3.01,
+      "eval_accuracy": 0.9230833333333334,
+      "eval_loss": 0.22653472423553467,
+      "eval_runtime": 16.1799,
+      "eval_samples_per_second": 1483.32,
+      "eval_steps_per_second": 4.635,
+      "step": 7220
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2315,
+      "step": 7240
+    },
+    {
+      "epoch": 3.02,
+      "eval_accuracy": 0.923125,
+      "eval_loss": 0.2254115790128708,
+      "eval_runtime": 15.9152,
+      "eval_samples_per_second": 1507.997,
+      "eval_steps_per_second": 4.712,
+      "step": 7240
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2055,
+      "step": 7260
+    },
+    {
+      "epoch": 3.02,
+      "eval_accuracy": 0.9245416666666667,
+      "eval_loss": 0.22353042662143707,
+      "eval_runtime": 15.5398,
+      "eval_samples_per_second": 1544.419,
+      "eval_steps_per_second": 4.826,
+      "step": 7260
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 1e-06,
+      "loss": 0.2277,
+      "step": 7280
+    },
+    {
+      "epoch": 3.03,
+      "eval_accuracy": 0.9244583333333334,
+      "eval_loss": 0.2269069403409958,
+      "eval_runtime": 16.1706,
+      "eval_samples_per_second": 1484.176,
+      "eval_steps_per_second": 4.638,
+      "step": 7280
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 1e-06,
+      "loss": 0.2379,
+      "step": 7300
+    },
+    {
+      "epoch": 3.04,
+      "eval_accuracy": 0.9248333333333333,
+      "eval_loss": 0.22591404616832733,
+      "eval_runtime": 16.0606,
+      "eval_samples_per_second": 1494.339,
+      "eval_steps_per_second": 4.67,
+      "step": 7300
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 1e-06,
+      "loss": 0.1889,
+      "step": 7320
+    },
+    {
+      "epoch": 3.05,
+      "eval_accuracy": 0.9246666666666666,
+      "eval_loss": 0.22398823499679565,
+      "eval_runtime": 15.816,
+      "eval_samples_per_second": 1517.45,
+      "eval_steps_per_second": 4.742,
+      "step": 7320
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 1e-06,
+      "loss": 0.2386,
+      "step": 7340
+    },
+    {
+      "epoch": 3.06,
+      "eval_accuracy": 0.9240416666666667,
+      "eval_loss": 0.22369486093521118,
+      "eval_runtime": 16.3833,
+      "eval_samples_per_second": 1464.908,
+      "eval_steps_per_second": 4.578,
+      "step": 7340
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 1e-06,
+      "loss": 0.2135,
+      "step": 7360
+    },
+    {
+      "epoch": 3.07,
+      "eval_accuracy": 0.9237083333333334,
+      "eval_loss": 0.22461163997650146,
+      "eval_runtime": 15.8713,
+      "eval_samples_per_second": 1512.165,
+      "eval_steps_per_second": 4.726,
+      "step": 7360
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2411,
+      "step": 7380
+    },
+    {
+      "epoch": 3.08,
+      "eval_accuracy": 0.923875,
+      "eval_loss": 0.2259892225265503,
+      "eval_runtime": 16.3476,
+      "eval_samples_per_second": 1468.103,
+      "eval_steps_per_second": 4.588,
+      "step": 7380
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2206,
+      "step": 7400
+    },
+    {
+      "epoch": 3.08,
+      "eval_accuracy": 0.9233333333333333,
+      "eval_loss": 0.22540873289108276,
+      "eval_runtime": 15.7971,
+      "eval_samples_per_second": 1519.265,
+      "eval_steps_per_second": 4.748,
+      "step": 7400
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 1e-06,
+      "loss": 0.221,
+      "step": 7420
+    },
+    {
+      "epoch": 3.09,
+      "eval_accuracy": 0.9237083333333334,
+      "eval_loss": 0.2240631878376007,
+      "eval_runtime": 15.554,
+      "eval_samples_per_second": 1543.013,
+      "eval_steps_per_second": 4.822,
+      "step": 7420
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 1e-06,
+      "loss": 0.2189,
+      "step": 7440
+    },
+    {
+      "epoch": 3.1,
+      "eval_accuracy": 0.9245,
+      "eval_loss": 0.2240666151046753,
+      "eval_runtime": 15.7074,
+      "eval_samples_per_second": 1527.938,
+      "eval_steps_per_second": 4.775,
+      "step": 7440
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 1e-06,
+      "loss": 0.2222,
+      "step": 7460
+    },
+    {
+      "epoch": 3.11,
+      "eval_accuracy": 0.9242916666666666,
+      "eval_loss": 0.22323697805404663,
+      "eval_runtime": 15.5541,
+      "eval_samples_per_second": 1543.003,
+      "eval_steps_per_second": 4.822,
+      "step": 7460
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2227,
+      "step": 7480
+    },
+    {
+      "epoch": 3.12,
+      "eval_accuracy": 0.92575,
+      "eval_loss": 0.221858412027359,
+      "eval_runtime": 16.1116,
+      "eval_samples_per_second": 1489.607,
+      "eval_steps_per_second": 4.655,
+      "step": 7480
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2375,
+      "step": 7500
+    },
+    {
+      "epoch": 3.12,
+      "eval_accuracy": 0.9249166666666667,
+      "eval_loss": 0.2222757488489151,
+      "eval_runtime": 15.423,
+      "eval_samples_per_second": 1556.117,
+      "eval_steps_per_second": 4.863,
+      "step": 7500
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 1e-06,
+      "loss": 0.2124,
+      "step": 7520
+    },
+    {
+      "epoch": 3.13,
+      "eval_accuracy": 0.9242083333333333,
+      "eval_loss": 0.2229667603969574,
+      "eval_runtime": 16.1122,
+      "eval_samples_per_second": 1489.556,
+      "eval_steps_per_second": 4.655,
+      "step": 7520
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 1e-06,
+      "loss": 0.2383,
+      "step": 7540
+    },
+    {
+      "epoch": 3.14,
+      "eval_accuracy": 0.9243333333333333,
+      "eval_loss": 0.2248881459236145,
+      "eval_runtime": 15.7471,
+      "eval_samples_per_second": 1524.086,
+      "eval_steps_per_second": 4.763,
+      "step": 7540
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 1e-06,
+      "loss": 0.2735,
+      "step": 7560
+    },
+    {
+      "epoch": 3.15,
+      "eval_accuracy": 0.9245833333333333,
+      "eval_loss": 0.22355443239212036,
+      "eval_runtime": 16.0099,
+      "eval_samples_per_second": 1499.071,
+      "eval_steps_per_second": 4.685,
+      "step": 7560
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 1e-06,
+      "loss": 0.195,
+      "step": 7580
+    },
+    {
+      "epoch": 3.16,
+      "eval_accuracy": 0.9247916666666667,
+      "eval_loss": 0.2215178906917572,
+      "eval_runtime": 16.1768,
+      "eval_samples_per_second": 1483.608,
+      "eval_steps_per_second": 4.636,
+      "step": 7580
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 1e-06,
+      "loss": 0.208,
+      "step": 7600
+    },
+    {
+      "epoch": 3.17,
+      "eval_accuracy": 0.9236666666666666,
+      "eval_loss": 0.22357864677906036,
+      "eval_runtime": 15.8058,
+      "eval_samples_per_second": 1518.434,
+      "eval_steps_per_second": 4.745,
+      "step": 7600
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 1e-06,
+      "loss": 0.2019,
+      "step": 7620
+    },
+    {
+      "epoch": 3.17,
+      "eval_accuracy": 0.9232916666666666,
+      "eval_loss": 0.2229277789592743,
+      "eval_runtime": 16.1663,
+      "eval_samples_per_second": 1484.567,
+      "eval_steps_per_second": 4.639,
+      "step": 7620
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 1e-06,
+      "loss": 0.242,
+      "step": 7640
+    },
+    {
+      "epoch": 3.18,
+      "eval_accuracy": 0.925125,
+      "eval_loss": 0.22199295461177826,
+      "eval_runtime": 15.6259,
+      "eval_samples_per_second": 1535.909,
+      "eval_steps_per_second": 4.8,
+      "step": 7640
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 1e-06,
+      "loss": 0.2209,
+      "step": 7660
+    },
+    {
+      "epoch": 3.19,
+      "eval_accuracy": 0.924375,
+      "eval_loss": 0.22275349497795105,
+      "eval_runtime": 15.853,
+      "eval_samples_per_second": 1513.909,
+      "eval_steps_per_second": 4.731,
+      "step": 7660
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 1e-06,
+      "loss": 0.263,
+      "step": 7680
+    },
+    {
+      "epoch": 3.2,
+      "eval_accuracy": 0.9236666666666666,
+      "eval_loss": 0.22205360233783722,
+      "eval_runtime": 15.6683,
+      "eval_samples_per_second": 1531.757,
+      "eval_steps_per_second": 4.787,
+      "step": 7680
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 1e-06,
+      "loss": 0.1923,
+      "step": 7700
+    },
+    {
+      "epoch": 3.21,
+      "eval_accuracy": 0.9255,
+      "eval_loss": 0.22050580382347107,
+      "eval_runtime": 16.26,
+      "eval_samples_per_second": 1476.012,
+      "eval_steps_per_second": 4.613,
+      "step": 7700
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 1e-06,
+      "loss": 0.2203,
+      "step": 7720
+    },
+    {
+      "epoch": 3.22,
+      "eval_accuracy": 0.9250833333333334,
+      "eval_loss": 0.22197723388671875,
+      "eval_runtime": 16.2074,
+      "eval_samples_per_second": 1480.801,
+      "eval_steps_per_second": 4.628,
+      "step": 7720
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 1e-06,
+      "loss": 0.2166,
+      "step": 7740
+    },
+    {
+      "epoch": 3.23,
+      "eval_accuracy": 0.9254583333333334,
+      "eval_loss": 0.2208959013223648,
+      "eval_runtime": 16.0355,
+      "eval_samples_per_second": 1496.675,
+      "eval_steps_per_second": 4.677,
+      "step": 7740
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 1e-06,
+      "loss": 0.2545,
+      "step": 7760
+    },
+    {
+      "epoch": 3.23,
+      "eval_accuracy": 0.9252083333333333,
+      "eval_loss": 0.22131887078285217,
+      "eval_runtime": 15.9945,
+      "eval_samples_per_second": 1500.516,
+      "eval_steps_per_second": 4.689,
+      "step": 7760
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 1e-06,
+      "loss": 0.1604,
+      "step": 7780
+    },
+    {
+      "epoch": 3.24,
+      "eval_accuracy": 0.92425,
+      "eval_loss": 0.22284860908985138,
+      "eval_runtime": 16.005,
+      "eval_samples_per_second": 1499.53,
+      "eval_steps_per_second": 4.686,
+      "step": 7780
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 1e-06,
+      "loss": 0.2,
+      "step": 7800
+    },
+    {
+      "epoch": 3.25,
+      "eval_accuracy": 0.923875,
+      "eval_loss": 0.22088997066020966,
+      "eval_runtime": 15.4149,
+      "eval_samples_per_second": 1556.938,
+      "eval_steps_per_second": 4.865,
+      "step": 7800
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 1e-06,
+      "loss": 0.2373,
+      "step": 7820
+    },
+    {
+      "epoch": 3.26,
+      "eval_accuracy": 0.9215833333333333,
+      "eval_loss": 0.2259444147348404,
+      "eval_runtime": 16.167,
+      "eval_samples_per_second": 1484.503,
+      "eval_steps_per_second": 4.639,
+      "step": 7820
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 1e-06,
+      "loss": 0.217,
+      "step": 7840
+    },
+    {
+      "epoch": 3.27,
+      "eval_accuracy": 0.9252083333333333,
+      "eval_loss": 0.22143015265464783,
+      "eval_runtime": 15.727,
+      "eval_samples_per_second": 1526.036,
+      "eval_steps_per_second": 4.769,
+      "step": 7840
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 1e-06,
+      "loss": 0.2172,
+      "step": 7860
+    },
+    {
+      "epoch": 3.27,
+      "eval_accuracy": 0.9234583333333334,
+      "eval_loss": 0.22263799607753754,
+      "eval_runtime": 16.2672,
+      "eval_samples_per_second": 1475.362,
+      "eval_steps_per_second": 4.611,
+      "step": 7860
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1959,
+      "step": 7880
+    },
+    {
+      "epoch": 3.28,
+      "eval_accuracy": 0.9235833333333333,
+      "eval_loss": 0.22203776240348816,
+      "eval_runtime": 15.4597,
+      "eval_samples_per_second": 1552.421,
+      "eval_steps_per_second": 4.851,
+      "step": 7880
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 1e-06,
+      "loss": 0.1781,
+      "step": 7900
+    },
+    {
+      "epoch": 3.29,
+      "eval_accuracy": 0.9245,
+      "eval_loss": 0.22064486145973206,
+      "eval_runtime": 15.8085,
+      "eval_samples_per_second": 1518.167,
+      "eval_steps_per_second": 4.744,
+      "step": 7900
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 1e-06,
+      "loss": 0.2024,
+      "step": 7920
+    },
+    {
+      "epoch": 3.3,
+      "eval_accuracy": 0.9228333333333333,
+      "eval_loss": 0.22356781363487244,
+      "eval_runtime": 16.1166,
+      "eval_samples_per_second": 1489.15,
+      "eval_steps_per_second": 4.654,
+      "step": 7920
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 1e-06,
+      "loss": 0.196,
+      "step": 7940
+    },
+    {
+      "epoch": 3.31,
+      "eval_accuracy": 0.923375,
+      "eval_loss": 0.22183743119239807,
+      "eval_runtime": 15.9128,
+      "eval_samples_per_second": 1508.22,
+      "eval_steps_per_second": 4.713,
+      "step": 7940
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 1e-06,
+      "loss": 0.2462,
+      "step": 7960
+    },
+    {
+      "epoch": 3.32,
+      "eval_accuracy": 0.925875,
+      "eval_loss": 0.21925699710845947,
+      "eval_runtime": 15.9525,
+      "eval_samples_per_second": 1504.469,
+      "eval_steps_per_second": 4.701,
+      "step": 7960
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 1e-06,
+      "loss": 0.205,
+      "step": 7980
+    },
+    {
+      "epoch": 3.33,
+      "eval_accuracy": 0.9255833333333333,
+      "eval_loss": 0.21881194412708282,
+      "eval_runtime": 15.9116,
+      "eval_samples_per_second": 1508.338,
+      "eval_steps_per_second": 4.714,
+      "step": 7980
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1793,
+      "step": 8000
+    },
+    {
+      "epoch": 3.33,
+      "eval_accuracy": 0.9254583333333334,
+      "eval_loss": 0.21929273009300232,
+      "eval_runtime": 16.6645,
+      "eval_samples_per_second": 1440.183,
+      "eval_steps_per_second": 4.501,
+      "step": 8000
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 1e-06,
+      "loss": 0.2551,
+      "step": 8020
+    },
+    {
+      "epoch": 3.34,
+      "eval_accuracy": 0.9253333333333333,
+      "eval_loss": 0.21935345232486725,
+      "eval_runtime": 15.8842,
+      "eval_samples_per_second": 1510.936,
+      "eval_steps_per_second": 4.722,
+      "step": 8020
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 1e-06,
+      "loss": 0.2471,
+      "step": 8040
+    },
+    {
+      "epoch": 3.35,
+      "eval_accuracy": 0.9250416666666667,
+      "eval_loss": 0.220913827419281,
+      "eval_runtime": 15.7346,
+      "eval_samples_per_second": 1525.304,
+      "eval_steps_per_second": 4.767,
+      "step": 8040
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 1e-06,
+      "loss": 0.1765,
+      "step": 8060
+    },
+    {
+      "epoch": 3.36,
+      "eval_accuracy": 0.9254166666666667,
+      "eval_loss": 0.22260655462741852,
+      "eval_runtime": 15.7321,
+      "eval_samples_per_second": 1525.546,
+      "eval_steps_per_second": 4.767,
+      "step": 8060
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 1e-06,
+      "loss": 0.161,
+      "step": 8080
+    },
+    {
+      "epoch": 3.37,
+      "eval_accuracy": 0.9257083333333334,
+      "eval_loss": 0.2232980579137802,
+      "eval_runtime": 15.5673,
+      "eval_samples_per_second": 1541.698,
+      "eval_steps_per_second": 4.818,
+      "step": 8080
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 1e-06,
+      "loss": 0.2243,
+      "step": 8100
+    },
+    {
+      "epoch": 3.38,
+      "eval_accuracy": 0.9247083333333334,
+      "eval_loss": 0.22126658260822296,
+      "eval_runtime": 16.3031,
+      "eval_samples_per_second": 1472.111,
+      "eval_steps_per_second": 4.6,
+      "step": 8100
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 1e-06,
+      "loss": 0.2044,
+      "step": 8120
+    },
+    {
+      "epoch": 3.38,
+      "eval_accuracy": 0.92525,
+      "eval_loss": 0.21990346908569336,
+      "eval_runtime": 15.9234,
+      "eval_samples_per_second": 1507.215,
+      "eval_steps_per_second": 4.71,
+      "step": 8120
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 1e-06,
+      "loss": 0.2213,
+      "step": 8140
+    },
+    {
+      "epoch": 3.39,
+      "eval_accuracy": 0.9261666666666667,
+      "eval_loss": 0.21853962540626526,
+      "eval_runtime": 15.7101,
+      "eval_samples_per_second": 1527.683,
+      "eval_steps_per_second": 4.774,
+      "step": 8140
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 1e-06,
+      "loss": 0.23,
+      "step": 8160
+    },
+    {
+      "epoch": 3.4,
+      "eval_accuracy": 0.9267916666666667,
+      "eval_loss": 0.21748250722885132,
+      "eval_runtime": 15.4099,
+      "eval_samples_per_second": 1557.437,
+      "eval_steps_per_second": 4.867,
+      "step": 8160
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 1e-06,
+      "loss": 0.2377,
+      "step": 8180
+    },
+    {
+      "epoch": 3.41,
+      "eval_accuracy": 0.9244166666666667,
+      "eval_loss": 0.21937525272369385,
+      "eval_runtime": 15.8123,
+      "eval_samples_per_second": 1517.802,
+      "eval_steps_per_second": 4.743,
+      "step": 8180
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2349,
+      "step": 8200
+    },
+    {
+      "epoch": 3.42,
+      "eval_accuracy": 0.9235833333333333,
+      "eval_loss": 0.21988487243652344,
+      "eval_runtime": 15.604,
+      "eval_samples_per_second": 1538.067,
+      "eval_steps_per_second": 4.806,
+      "step": 8200
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2217,
+      "step": 8220
+    },
+    {
+      "epoch": 3.42,
+      "eval_accuracy": 0.9260416666666667,
+      "eval_loss": 0.21746017038822174,
+      "eval_runtime": 15.8834,
+      "eval_samples_per_second": 1511.008,
+      "eval_steps_per_second": 4.722,
+      "step": 8220
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 1e-06,
+      "loss": 0.2005,
+      "step": 8240
+    },
+    {
+      "epoch": 3.43,
+      "eval_accuracy": 0.9264166666666667,
+      "eval_loss": 0.21770620346069336,
+      "eval_runtime": 16.0248,
+      "eval_samples_per_second": 1497.677,
+      "eval_steps_per_second": 4.68,
+      "step": 8240
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 1e-06,
+      "loss": 0.2193,
+      "step": 8260
+    },
+    {
+      "epoch": 3.44,
+      "eval_accuracy": 0.9264583333333334,
+      "eval_loss": 0.21887263655662537,
+      "eval_runtime": 15.827,
+      "eval_samples_per_second": 1516.398,
+      "eval_steps_per_second": 4.739,
+      "step": 8260
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 1e-06,
+      "loss": 0.1551,
+      "step": 8280
+    },
+    {
+      "epoch": 3.45,
+      "eval_accuracy": 0.92625,
+      "eval_loss": 0.22045257687568665,
+      "eval_runtime": 15.6985,
+      "eval_samples_per_second": 1528.808,
+      "eval_steps_per_second": 4.778,
+      "step": 8280
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 1e-06,
+      "loss": 0.2399,
+      "step": 8300
+    },
+    {
+      "epoch": 3.46,
+      "eval_accuracy": 0.9254583333333334,
+      "eval_loss": 0.21784846484661102,
+      "eval_runtime": 15.9829,
+      "eval_samples_per_second": 1501.6,
+      "eval_steps_per_second": 4.693,
+      "step": 8300
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 1e-06,
+      "loss": 0.2308,
+      "step": 8320
+    },
+    {
+      "epoch": 3.47,
+      "eval_accuracy": 0.9246666666666666,
+      "eval_loss": 0.21774353086948395,
+      "eval_runtime": 15.8498,
+      "eval_samples_per_second": 1514.211,
+      "eval_steps_per_second": 4.732,
+      "step": 8320
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 1e-06,
+      "loss": 0.1661,
+      "step": 8340
+    },
+    {
+      "epoch": 3.48,
+      "eval_accuracy": 0.924625,
+      "eval_loss": 0.21844609081745148,
+      "eval_runtime": 16.0502,
+      "eval_samples_per_second": 1495.308,
+      "eval_steps_per_second": 4.673,
+      "step": 8340
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2014,
+      "step": 8360
+    },
+    {
+      "epoch": 3.48,
+      "eval_accuracy": 0.9242916666666666,
+      "eval_loss": 0.21898412704467773,
+      "eval_runtime": 15.9262,
+      "eval_samples_per_second": 1506.951,
+      "eval_steps_per_second": 4.709,
+      "step": 8360
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 1e-06,
+      "loss": 0.2667,
+      "step": 8380
+    },
+    {
+      "epoch": 3.49,
+      "eval_accuracy": 0.9242083333333333,
+      "eval_loss": 0.21949926018714905,
+      "eval_runtime": 15.9646,
+      "eval_samples_per_second": 1503.329,
+      "eval_steps_per_second": 4.698,
+      "step": 8380
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 1e-06,
+      "loss": 0.205,
+      "step": 8400
+    },
+    {
+      "epoch": 3.5,
+      "eval_accuracy": 0.9252916666666666,
+      "eval_loss": 0.21843333542346954,
+      "eval_runtime": 16.4182,
+      "eval_samples_per_second": 1461.79,
+      "eval_steps_per_second": 4.568,
+      "step": 8400
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 1e-06,
+      "loss": 0.1982,
+      "step": 8420
+    },
+    {
+      "epoch": 3.51,
+      "eval_accuracy": 0.9259166666666667,
+      "eval_loss": 0.21639133989810944,
+      "eval_runtime": 15.6132,
+      "eval_samples_per_second": 1537.157,
+      "eval_steps_per_second": 4.804,
+      "step": 8420
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2511,
+      "step": 8440
+    },
+    {
+      "epoch": 3.52,
+      "eval_accuracy": 0.926125,
+      "eval_loss": 0.21578435599803925,
+      "eval_runtime": 15.8122,
+      "eval_samples_per_second": 1517.813,
+      "eval_steps_per_second": 4.743,
+      "step": 8440
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2627,
+      "step": 8460
+    },
+    {
+      "epoch": 3.52,
+      "eval_accuracy": 0.9263333333333333,
+      "eval_loss": 0.2152308076620102,
+      "eval_runtime": 16.1172,
+      "eval_samples_per_second": 1489.089,
+      "eval_steps_per_second": 4.653,
+      "step": 8460
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1905,
+      "step": 8480
+    },
+    {
+      "epoch": 3.53,
+      "eval_accuracy": 0.9267083333333334,
+      "eval_loss": 0.215366929769516,
+      "eval_runtime": 16.1037,
+      "eval_samples_per_second": 1490.344,
+      "eval_steps_per_second": 4.657,
+      "step": 8480
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 1e-06,
+      "loss": 0.2349,
+      "step": 8500
+    },
+    {
+      "epoch": 3.54,
+      "eval_accuracy": 0.9254166666666667,
+      "eval_loss": 0.216691255569458,
+      "eval_runtime": 15.7789,
+      "eval_samples_per_second": 1521.022,
+      "eval_steps_per_second": 4.753,
+      "step": 8500
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 1e-06,
+      "loss": 0.1732,
+      "step": 8520
+    },
+    {
+      "epoch": 3.55,
+      "eval_accuracy": 0.9255833333333333,
+      "eval_loss": 0.2171410173177719,
+      "eval_runtime": 15.9111,
+      "eval_samples_per_second": 1508.379,
+      "eval_steps_per_second": 4.714,
+      "step": 8520
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 1e-06,
+      "loss": 0.2152,
+      "step": 8540
+    },
+    {
+      "epoch": 3.56,
+      "eval_accuracy": 0.9257916666666667,
+      "eval_loss": 0.21639755368232727,
+      "eval_runtime": 16.6264,
+      "eval_samples_per_second": 1443.488,
+      "eval_steps_per_second": 4.511,
+      "step": 8540
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 1e-06,
+      "loss": 0.2487,
+      "step": 8560
+    },
+    {
+      "epoch": 3.57,
+      "eval_accuracy": 0.92575,
+      "eval_loss": 0.21681177616119385,
+      "eval_runtime": 15.9061,
+      "eval_samples_per_second": 1508.852,
+      "eval_steps_per_second": 4.715,
+      "step": 8560
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2411,
+      "step": 8580
+    },
+    {
+      "epoch": 3.58,
+      "eval_accuracy": 0.9255833333333333,
+      "eval_loss": 0.2175518274307251,
+      "eval_runtime": 16.2026,
+      "eval_samples_per_second": 1481.242,
+      "eval_steps_per_second": 4.629,
+      "step": 8580
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2113,
+      "step": 8600
+    },
+    {
+      "epoch": 3.58,
+      "eval_accuracy": 0.9257916666666667,
+      "eval_loss": 0.21555760502815247,
+      "eval_runtime": 15.883,
+      "eval_samples_per_second": 1511.052,
+      "eval_steps_per_second": 4.722,
+      "step": 8600
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 1e-06,
+      "loss": 0.2065,
+      "step": 8620
+    },
+    {
+      "epoch": 3.59,
+      "eval_accuracy": 0.9265833333333333,
+      "eval_loss": 0.21459566056728363,
+      "eval_runtime": 15.8987,
+      "eval_samples_per_second": 1509.554,
+      "eval_steps_per_second": 4.717,
+      "step": 8620
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 1e-06,
+      "loss": 0.2317,
+      "step": 8640
+    },
+    {
+      "epoch": 3.6,
+      "eval_accuracy": 0.926,
+      "eval_loss": 0.21562263369560242,
+      "eval_runtime": 15.7909,
+      "eval_samples_per_second": 1519.863,
+      "eval_steps_per_second": 4.75,
+      "step": 8640
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 1e-06,
+      "loss": 0.2217,
+      "step": 8660
+    },
+    {
+      "epoch": 3.61,
+      "eval_accuracy": 0.9255416666666667,
+      "eval_loss": 0.21699927747249603,
+      "eval_runtime": 15.7983,
+      "eval_samples_per_second": 1519.146,
+      "eval_steps_per_second": 4.747,
+      "step": 8660
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 1e-06,
+      "loss": 0.2296,
+      "step": 8680
+    },
+    {
+      "epoch": 3.62,
+      "eval_accuracy": 0.9252916666666666,
+      "eval_loss": 0.21904636919498444,
+      "eval_runtime": 15.8426,
+      "eval_samples_per_second": 1514.899,
+      "eval_steps_per_second": 4.734,
+      "step": 8680
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 1e-06,
+      "loss": 0.2143,
+      "step": 8700
+    },
+    {
+      "epoch": 3.62,
+      "eval_accuracy": 0.9249166666666667,
+      "eval_loss": 0.21984702348709106,
+      "eval_runtime": 15.7532,
+      "eval_samples_per_second": 1523.499,
+      "eval_steps_per_second": 4.761,
+      "step": 8700
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 1e-06,
+      "loss": 0.215,
+      "step": 8720
+    },
+    {
+      "epoch": 3.63,
+      "eval_accuracy": 0.9260416666666667,
+      "eval_loss": 0.21465341746807098,
+      "eval_runtime": 15.8302,
+      "eval_samples_per_second": 1516.094,
+      "eval_steps_per_second": 4.738,
+      "step": 8720
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 1e-06,
+      "loss": 0.2109,
+      "step": 8740
+    },
+    {
+      "epoch": 3.64,
+      "eval_accuracy": 0.9271666666666667,
+      "eval_loss": 0.21410001814365387,
+      "eval_runtime": 15.7941,
+      "eval_samples_per_second": 1519.554,
+      "eval_steps_per_second": 4.749,
+      "step": 8740
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 1e-06,
+      "loss": 0.1908,
+      "step": 8760
+    },
+    {
+      "epoch": 3.65,
+      "eval_accuracy": 0.9242083333333333,
+      "eval_loss": 0.21745717525482178,
+      "eval_runtime": 16.1017,
+      "eval_samples_per_second": 1490.524,
+      "eval_steps_per_second": 4.658,
+      "step": 8760
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 1e-06,
+      "loss": 0.1899,
+      "step": 8780
+    },
+    {
+      "epoch": 3.66,
+      "eval_accuracy": 0.9242916666666666,
+      "eval_loss": 0.21800926327705383,
+      "eval_runtime": 15.5229,
+      "eval_samples_per_second": 1546.1,
+      "eval_steps_per_second": 4.832,
+      "step": 8780
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1681,
+      "step": 8800
+    },
+    {
+      "epoch": 3.67,
+      "eval_accuracy": 0.925625,
+      "eval_loss": 0.2183951884508133,
+      "eval_runtime": 15.9633,
+      "eval_samples_per_second": 1503.444,
+      "eval_steps_per_second": 4.698,
+      "step": 8800
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 1e-06,
+      "loss": 0.2134,
+      "step": 8820
+    },
+    {
+      "epoch": 3.67,
+      "eval_accuracy": 0.9258333333333333,
+      "eval_loss": 0.21758043766021729,
+      "eval_runtime": 15.6658,
+      "eval_samples_per_second": 1532.001,
+      "eval_steps_per_second": 4.788,
+      "step": 8820
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1788,
+      "step": 8840
+    },
+    {
+      "epoch": 3.68,
+      "eval_accuracy": 0.9257916666666667,
+      "eval_loss": 0.2156967669725418,
+      "eval_runtime": 16.1604,
+      "eval_samples_per_second": 1485.116,
+      "eval_steps_per_second": 4.641,
+      "step": 8840
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1963,
+      "step": 8860
+    },
+    {
+      "epoch": 3.69,
+      "eval_accuracy": 0.9265833333333333,
+      "eval_loss": 0.21555079519748688,
+      "eval_runtime": 15.9521,
+      "eval_samples_per_second": 1504.506,
+      "eval_steps_per_second": 4.702,
+      "step": 8860
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1962,
+      "step": 8880
+    },
+    {
+      "epoch": 3.7,
+      "eval_accuracy": 0.9252083333333333,
+      "eval_loss": 0.2161911576986313,
+      "eval_runtime": 16.0835,
+      "eval_samples_per_second": 1492.21,
+      "eval_steps_per_second": 4.663,
+      "step": 8880
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1587,
+      "step": 8900
+    },
+    {
+      "epoch": 3.71,
+      "eval_accuracy": 0.9257916666666667,
+      "eval_loss": 0.21644917130470276,
+      "eval_runtime": 16.0872,
+      "eval_samples_per_second": 1491.867,
+      "eval_steps_per_second": 4.662,
+      "step": 8900
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1926,
+      "step": 8920
+    },
+    {
+      "epoch": 3.72,
+      "eval_accuracy": 0.924875,
+      "eval_loss": 0.21802347898483276,
+      "eval_runtime": 15.7349,
+      "eval_samples_per_second": 1525.267,
+      "eval_steps_per_second": 4.766,
+      "step": 8920
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 1e-06,
+      "loss": 0.2268,
+      "step": 8940
+    },
+    {
+      "epoch": 3.73,
+      "eval_accuracy": 0.9250416666666667,
+      "eval_loss": 0.21704040467739105,
+      "eval_runtime": 16.0405,
+      "eval_samples_per_second": 1496.209,
+      "eval_steps_per_second": 4.676,
+      "step": 8940
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 1e-06,
+      "loss": 0.191,
+      "step": 8960
+    },
+    {
+      "epoch": 3.73,
+      "eval_accuracy": 0.9265416666666667,
+      "eval_loss": 0.21474313735961914,
+      "eval_runtime": 16.1261,
+      "eval_samples_per_second": 1488.27,
+      "eval_steps_per_second": 4.651,
+      "step": 8960
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 1e-06,
+      "loss": 0.242,
+      "step": 8980
+    },
+    {
+      "epoch": 3.74,
+      "eval_accuracy": 0.92675,
+      "eval_loss": 0.21450480818748474,
+      "eval_runtime": 15.8452,
+      "eval_samples_per_second": 1514.651,
+      "eval_steps_per_second": 4.733,
+      "step": 8980
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 1e-06,
+      "loss": 0.2096,
+      "step": 9000
+    },
+    {
+      "epoch": 3.75,
+      "eval_accuracy": 0.925625,
+      "eval_loss": 0.2159881889820099,
+      "eval_runtime": 15.8929,
+      "eval_samples_per_second": 1510.111,
+      "eval_steps_per_second": 4.719,
+      "step": 9000
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 1e-06,
+      "loss": 0.1713,
+      "step": 9020
+    },
+    {
+      "epoch": 3.76,
+      "eval_accuracy": 0.9254583333333334,
+      "eval_loss": 0.2180025428533554,
+      "eval_runtime": 16.0033,
+      "eval_samples_per_second": 1499.686,
+      "eval_steps_per_second": 4.687,
+      "step": 9020
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 1e-06,
+      "loss": 0.2437,
+      "step": 9040
+    },
+    {
+      "epoch": 3.77,
+      "eval_accuracy": 0.9272083333333333,
+      "eval_loss": 0.21439102292060852,
+      "eval_runtime": 15.8661,
+      "eval_samples_per_second": 1512.663,
+      "eval_steps_per_second": 4.727,
+      "step": 9040
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 1e-06,
+      "loss": 0.2058,
+      "step": 9060
+    },
+    {
+      "epoch": 3.77,
+      "eval_accuracy": 0.9265416666666667,
+      "eval_loss": 0.21493494510650635,
+      "eval_runtime": 16.417,
+      "eval_samples_per_second": 1461.901,
+      "eval_steps_per_second": 4.568,
+      "step": 9060
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2107,
+      "step": 9080
+    },
+    {
+      "epoch": 3.78,
+      "eval_accuracy": 0.9262083333333333,
+      "eval_loss": 0.21356363594532013,
+      "eval_runtime": 15.9759,
+      "eval_samples_per_second": 1502.265,
+      "eval_steps_per_second": 4.695,
+      "step": 9080
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 1e-06,
+      "loss": 0.2274,
+      "step": 9100
+    },
+    {
+      "epoch": 3.79,
+      "eval_accuracy": 0.9263333333333333,
+      "eval_loss": 0.21457020938396454,
+      "eval_runtime": 16.1086,
+      "eval_samples_per_second": 1489.887,
+      "eval_steps_per_second": 4.656,
+      "step": 9100
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 1e-06,
+      "loss": 0.1802,
+      "step": 9120
+    },
+    {
+      "epoch": 3.8,
+      "eval_accuracy": 0.9258333333333333,
+      "eval_loss": 0.21796223521232605,
+      "eval_runtime": 16.0162,
+      "eval_samples_per_second": 1498.485,
+      "eval_steps_per_second": 4.683,
+      "step": 9120
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1901,
+      "step": 9140
+    },
+    {
+      "epoch": 3.81,
+      "eval_accuracy": 0.9270416666666667,
+      "eval_loss": 0.21377238631248474,
+      "eval_runtime": 16.0799,
+      "eval_samples_per_second": 1492.542,
+      "eval_steps_per_second": 4.664,
+      "step": 9140
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 1e-06,
+      "loss": 0.1868,
+      "step": 9160
+    },
+    {
+      "epoch": 3.82,
+      "eval_accuracy": 0.9264583333333334,
+      "eval_loss": 0.214362233877182,
+      "eval_runtime": 15.7967,
+      "eval_samples_per_second": 1519.305,
+      "eval_steps_per_second": 4.748,
+      "step": 9160
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 1e-06,
+      "loss": 0.159,
+      "step": 9180
+    },
+    {
+      "epoch": 3.83,
+      "eval_accuracy": 0.927875,
+      "eval_loss": 0.21380971372127533,
+      "eval_runtime": 15.6723,
+      "eval_samples_per_second": 1531.362,
+      "eval_steps_per_second": 4.786,
+      "step": 9180
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 1e-06,
+      "loss": 0.2448,
+      "step": 9200
+    },
+    {
+      "epoch": 3.83,
+      "eval_accuracy": 0.9277083333333334,
+      "eval_loss": 0.2141939252614975,
+      "eval_runtime": 16.2577,
+      "eval_samples_per_second": 1476.225,
+      "eval_steps_per_second": 4.613,
+      "step": 9200
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 1e-06,
+      "loss": 0.2412,
+      "step": 9220
+    },
+    {
+      "epoch": 3.84,
+      "eval_accuracy": 0.9275416666666667,
+      "eval_loss": 0.2140486091375351,
+      "eval_runtime": 15.679,
+      "eval_samples_per_second": 1530.709,
+      "eval_steps_per_second": 4.783,
+      "step": 9220
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 1e-06,
+      "loss": 0.2457,
+      "step": 9240
+    },
+    {
+      "epoch": 3.85,
+      "eval_accuracy": 0.925375,
+      "eval_loss": 0.21611127257347107,
+      "eval_runtime": 16.288,
+      "eval_samples_per_second": 1473.473,
+      "eval_steps_per_second": 4.605,
+      "step": 9240
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1918,
+      "step": 9260
+    },
+    {
+      "epoch": 3.86,
+      "eval_accuracy": 0.92625,
+      "eval_loss": 0.21398600935935974,
+      "eval_runtime": 15.9302,
+      "eval_samples_per_second": 1506.573,
+      "eval_steps_per_second": 4.708,
+      "step": 9260
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 1e-06,
+      "loss": 0.2424,
+      "step": 9280
+    },
+    {
+      "epoch": 3.87,
+      "eval_accuracy": 0.92875,
+      "eval_loss": 0.21145953238010406,
+      "eval_runtime": 16.2556,
+      "eval_samples_per_second": 1476.419,
+      "eval_steps_per_second": 4.614,
+      "step": 9280
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 1e-06,
+      "loss": 0.2059,
+      "step": 9300
+    },
+    {
+      "epoch": 3.88,
+      "eval_accuracy": 0.9278333333333333,
+      "eval_loss": 0.21115827560424805,
+      "eval_runtime": 16.0018,
+      "eval_samples_per_second": 1499.836,
+      "eval_steps_per_second": 4.687,
+      "step": 9300
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 1e-06,
+      "loss": 0.2415,
+      "step": 9320
+    },
+    {
+      "epoch": 3.88,
+      "eval_accuracy": 0.9265,
+      "eval_loss": 0.212614968419075,
+      "eval_runtime": 15.9163,
+      "eval_samples_per_second": 1507.888,
+      "eval_steps_per_second": 4.712,
+      "step": 9320
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 1e-06,
+      "loss": 0.2312,
+      "step": 9340
+    },
+    {
+      "epoch": 3.89,
+      "eval_accuracy": 0.9279166666666666,
+      "eval_loss": 0.21092422306537628,
+      "eval_runtime": 15.7459,
+      "eval_samples_per_second": 1524.21,
+      "eval_steps_per_second": 4.763,
+      "step": 9340
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 1e-06,
+      "loss": 0.2002,
+      "step": 9360
+    },
+    {
+      "epoch": 3.9,
+      "eval_accuracy": 0.927,
+      "eval_loss": 0.21157173812389374,
+      "eval_runtime": 15.674,
+      "eval_samples_per_second": 1531.197,
+      "eval_steps_per_second": 4.785,
+      "step": 9360
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 1e-06,
+      "loss": 0.2061,
+      "step": 9380
+    },
+    {
+      "epoch": 3.91,
+      "eval_accuracy": 0.9255416666666667,
+      "eval_loss": 0.21427378058433533,
+      "eval_runtime": 16.16,
+      "eval_samples_per_second": 1485.144,
+      "eval_steps_per_second": 4.641,
+      "step": 9380
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1892,
+      "step": 9400
+    },
+    {
+      "epoch": 3.92,
+      "eval_accuracy": 0.9269583333333333,
+      "eval_loss": 0.21478785574436188,
+      "eval_runtime": 15.666,
+      "eval_samples_per_second": 1531.978,
+      "eval_steps_per_second": 4.787,
+      "step": 9400
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 1e-06,
+      "loss": 0.242,
+      "step": 9420
+    },
+    {
+      "epoch": 3.92,
+      "eval_accuracy": 0.9268333333333333,
+      "eval_loss": 0.21459507942199707,
+      "eval_runtime": 16.0867,
+      "eval_samples_per_second": 1491.912,
+      "eval_steps_per_second": 4.662,
+      "step": 9420
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 1e-06,
+      "loss": 0.2151,
+      "step": 9440
+    },
+    {
+      "epoch": 3.93,
+      "eval_accuracy": 0.9257916666666667,
+      "eval_loss": 0.21316887438297272,
+      "eval_runtime": 16.0112,
+      "eval_samples_per_second": 1498.948,
+      "eval_steps_per_second": 4.684,
+      "step": 9440
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 1e-06,
+      "loss": 0.2057,
+      "step": 9460
+    },
+    {
+      "epoch": 3.94,
+      "eval_accuracy": 0.927125,
+      "eval_loss": 0.21309146285057068,
+      "eval_runtime": 15.8401,
+      "eval_samples_per_second": 1515.139,
+      "eval_steps_per_second": 4.735,
+      "step": 9460
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 1e-06,
+      "loss": 0.2165,
+      "step": 9480
+    },
+    {
+      "epoch": 3.95,
+      "eval_accuracy": 0.9276666666666666,
+      "eval_loss": 0.21158146858215332,
+      "eval_runtime": 16.0289,
+      "eval_samples_per_second": 1497.291,
+      "eval_steps_per_second": 4.679,
+      "step": 9480
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 1e-06,
+      "loss": 0.1845,
+      "step": 9500
+    },
+    {
+      "epoch": 3.96,
+      "eval_accuracy": 0.9277083333333334,
+      "eval_loss": 0.21126095950603485,
+      "eval_runtime": 15.8162,
+      "eval_samples_per_second": 1517.433,
+      "eval_steps_per_second": 4.742,
+      "step": 9500
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1787,
+      "step": 9520
+    },
+    {
+      "epoch": 3.97,
+      "eval_accuracy": 0.927,
+      "eval_loss": 0.21392786502838135,
+      "eval_runtime": 16.2866,
+      "eval_samples_per_second": 1473.607,
+      "eval_steps_per_second": 4.605,
+      "step": 9520
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1947,
+      "step": 9540
+    },
+    {
+      "epoch": 3.98,
+      "eval_accuracy": 0.927875,
+      "eval_loss": 0.2135414332151413,
+      "eval_runtime": 15.7812,
+      "eval_samples_per_second": 1520.796,
+      "eval_steps_per_second": 4.752,
+      "step": 9540
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1802,
+      "step": 9560
+    },
+    {
+      "epoch": 3.98,
+      "eval_accuracy": 0.9288333333333333,
+      "eval_loss": 0.2114471048116684,
+      "eval_runtime": 15.9961,
+      "eval_samples_per_second": 1500.37,
+      "eval_steps_per_second": 4.689,
+      "step": 9560
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1865,
+      "step": 9580
+    },
+    {
+      "epoch": 3.99,
+      "eval_accuracy": 0.9280833333333334,
+      "eval_loss": 0.21309266984462738,
+      "eval_runtime": 16.0813,
+      "eval_samples_per_second": 1492.415,
+      "eval_steps_per_second": 4.664,
+      "step": 9580
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 1e-06,
+      "loss": 0.2346,
+      "step": 9600
+    },
+    {
+      "epoch": 4.0,
+      "eval_accuracy": 0.9282083333333333,
+      "eval_loss": 0.21224650740623474,
+      "eval_runtime": 15.806,
+      "eval_samples_per_second": 1518.415,
+      "eval_steps_per_second": 4.745,
+      "step": 9600
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 1e-06,
+      "loss": 0.1973,
+      "step": 9620
+    },
+    {
+      "epoch": 4.01,
+      "eval_accuracy": 0.9283333333333333,
+      "eval_loss": 0.2113197296857834,
+      "eval_runtime": 16.0449,
+      "eval_samples_per_second": 1495.802,
+      "eval_steps_per_second": 4.674,
+      "step": 9620
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 1e-06,
+      "loss": 0.1873,
+      "step": 9640
+    },
+    {
+      "epoch": 4.02,
+      "eval_accuracy": 0.928,
+      "eval_loss": 0.21195828914642334,
+      "eval_runtime": 16.0572,
+      "eval_samples_per_second": 1494.657,
+      "eval_steps_per_second": 4.671,
+      "step": 9640
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1883,
+      "step": 9660
+    },
+    {
+      "epoch": 4.03,
+      "eval_accuracy": 0.9275416666666667,
+      "eval_loss": 0.21404027938842773,
+      "eval_runtime": 16.1167,
+      "eval_samples_per_second": 1489.14,
+      "eval_steps_per_second": 4.654,
+      "step": 9660
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 1e-06,
+      "loss": 0.2386,
+      "step": 9680
+    },
+    {
+      "epoch": 4.03,
+      "eval_accuracy": 0.9255,
+      "eval_loss": 0.21560032665729523,
+      "eval_runtime": 15.8482,
+      "eval_samples_per_second": 1514.369,
+      "eval_steps_per_second": 4.732,
+      "step": 9680
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 1e-06,
+      "loss": 0.1594,
+      "step": 9700
+    },
+    {
+      "epoch": 4.04,
+      "eval_accuracy": 0.9266666666666666,
+      "eval_loss": 0.2141159623861313,
+      "eval_runtime": 15.8534,
+      "eval_samples_per_second": 1513.868,
+      "eval_steps_per_second": 4.731,
+      "step": 9700
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 1e-06,
+      "loss": 0.1805,
+      "step": 9720
+    },
+    {
+      "epoch": 4.05,
+      "eval_accuracy": 0.927875,
+      "eval_loss": 0.21176785230636597,
+      "eval_runtime": 15.5441,
+      "eval_samples_per_second": 1543.994,
+      "eval_steps_per_second": 4.825,
+      "step": 9720
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 1e-06,
+      "loss": 0.2121,
+      "step": 9740
+    },
+    {
+      "epoch": 4.06,
+      "eval_accuracy": 0.927,
+      "eval_loss": 0.21185992658138275,
+      "eval_runtime": 16.1338,
+      "eval_samples_per_second": 1487.557,
+      "eval_steps_per_second": 4.649,
+      "step": 9740
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1764,
+      "step": 9760
+    },
+    {
+      "epoch": 4.07,
+      "eval_accuracy": 0.926,
+      "eval_loss": 0.21480970084667206,
+      "eval_runtime": 15.7121,
+      "eval_samples_per_second": 1527.483,
+      "eval_steps_per_second": 4.773,
+      "step": 9760
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2067,
+      "step": 9780
+    },
+    {
+      "epoch": 4.08,
+      "eval_accuracy": 0.9278333333333333,
+      "eval_loss": 0.21278834342956543,
+      "eval_runtime": 15.7391,
+      "eval_samples_per_second": 1524.869,
+      "eval_steps_per_second": 4.765,
+      "step": 9780
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2219,
+      "step": 9800
+    },
+    {
+      "epoch": 4.08,
+      "eval_accuracy": 0.9279166666666666,
+      "eval_loss": 0.21168003976345062,
+      "eval_runtime": 15.591,
+      "eval_samples_per_second": 1539.347,
+      "eval_steps_per_second": 4.81,
+      "step": 9800
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 1e-06,
+      "loss": 0.1931,
+      "step": 9820
+    },
+    {
+      "epoch": 4.09,
+      "eval_accuracy": 0.9288333333333333,
+      "eval_loss": 0.2115764170885086,
+      "eval_runtime": 16.0212,
+      "eval_samples_per_second": 1498.017,
+      "eval_steps_per_second": 4.681,
+      "step": 9820
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 1e-06,
+      "loss": 0.198,
+      "step": 9840
+    },
+    {
+      "epoch": 4.1,
+      "eval_accuracy": 0.9270833333333334,
+      "eval_loss": 0.21085327863693237,
+      "eval_runtime": 16.2758,
+      "eval_samples_per_second": 1474.58,
+      "eval_steps_per_second": 4.608,
+      "step": 9840
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 1e-06,
+      "loss": 0.1664,
+      "step": 9860
+    },
+    {
+      "epoch": 4.11,
+      "eval_accuracy": 0.9275833333333333,
+      "eval_loss": 0.21043312549591064,
+      "eval_runtime": 15.7593,
+      "eval_samples_per_second": 1522.91,
+      "eval_steps_per_second": 4.759,
+      "step": 9860
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1998,
+      "step": 9880
+    },
+    {
+      "epoch": 4.12,
+      "eval_accuracy": 0.9276666666666666,
+      "eval_loss": 0.21023598313331604,
+      "eval_runtime": 15.8663,
+      "eval_samples_per_second": 1512.639,
+      "eval_steps_per_second": 4.727,
+      "step": 9880
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2338,
+      "step": 9900
+    },
+    {
+      "epoch": 4.12,
+      "eval_accuracy": 0.9267916666666667,
+      "eval_loss": 0.21208404004573822,
+      "eval_runtime": 15.987,
+      "eval_samples_per_second": 1501.219,
+      "eval_steps_per_second": 4.691,
+      "step": 9900
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 1e-06,
+      "loss": 0.1807,
+      "step": 9920
+    },
+    {
+      "epoch": 4.13,
+      "eval_accuracy": 0.9267083333333334,
+      "eval_loss": 0.21226836740970612,
+      "eval_runtime": 15.4893,
+      "eval_samples_per_second": 1549.453,
+      "eval_steps_per_second": 4.842,
+      "step": 9920
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 1e-06,
+      "loss": 0.2055,
+      "step": 9940
+    },
+    {
+      "epoch": 4.14,
+      "eval_accuracy": 0.92775,
+      "eval_loss": 0.21014344692230225,
+      "eval_runtime": 16.5702,
+      "eval_samples_per_second": 1448.387,
+      "eval_steps_per_second": 4.526,
+      "step": 9940
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 1e-06,
+      "loss": 0.2094,
+      "step": 9960
+    },
+    {
+      "epoch": 4.15,
+      "eval_accuracy": 0.9275,
+      "eval_loss": 0.21054843068122864,
+      "eval_runtime": 15.7855,
+      "eval_samples_per_second": 1520.387,
+      "eval_steps_per_second": 4.751,
+      "step": 9960
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 1e-06,
+      "loss": 0.1675,
+      "step": 9980
+    },
+    {
+      "epoch": 4.16,
+      "eval_accuracy": 0.9266666666666666,
+      "eval_loss": 0.21459108591079712,
+      "eval_runtime": 15.728,
+      "eval_samples_per_second": 1525.936,
+      "eval_steps_per_second": 4.769,
+      "step": 9980
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 1e-06,
+      "loss": 0.2043,
+      "step": 10000
+    },
+    {
+      "epoch": 4.17,
+      "eval_accuracy": 0.9267916666666667,
+      "eval_loss": 0.2115868628025055,
+      "eval_runtime": 16.2887,
+      "eval_samples_per_second": 1473.412,
+      "eval_steps_per_second": 4.604,
+      "step": 10000
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 1e-06,
+      "loss": 0.1625,
+      "step": 10020
+    },
+    {
+      "epoch": 4.17,
+      "eval_accuracy": 0.927625,
+      "eval_loss": 0.21186378598213196,
+      "eval_runtime": 15.6418,
+      "eval_samples_per_second": 1534.346,
+      "eval_steps_per_second": 4.795,
+      "step": 10020
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1761,
+      "step": 10040
+    },
+    {
+      "epoch": 4.18,
+      "eval_accuracy": 0.92725,
+      "eval_loss": 0.2122122347354889,
+      "eval_runtime": 16.1634,
+      "eval_samples_per_second": 1484.835,
+      "eval_steps_per_second": 4.64,
+      "step": 10040
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 1e-06,
+      "loss": 0.1787,
+      "step": 10060
+    },
+    {
+      "epoch": 4.19,
+      "eval_accuracy": 0.9278333333333333,
+      "eval_loss": 0.21212342381477356,
+      "eval_runtime": 15.87,
+      "eval_samples_per_second": 1512.288,
+      "eval_steps_per_second": 4.726,
+      "step": 10060
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 1e-06,
+      "loss": 0.1773,
+      "step": 10080
+    },
+    {
+      "epoch": 4.2,
+      "eval_accuracy": 0.927375,
+      "eval_loss": 0.21214234828948975,
+      "eval_runtime": 15.6627,
+      "eval_samples_per_second": 1532.306,
+      "eval_steps_per_second": 4.788,
+      "step": 10080
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 1e-06,
+      "loss": 0.2317,
+      "step": 10100
+    },
+    {
+      "epoch": 4.21,
+      "eval_accuracy": 0.9284166666666667,
+      "eval_loss": 0.21087035536766052,
+      "eval_runtime": 15.9309,
+      "eval_samples_per_second": 1506.51,
+      "eval_steps_per_second": 4.708,
+      "step": 10100
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1909,
+      "step": 10120
+    },
+    {
+      "epoch": 4.22,
+      "eval_accuracy": 0.9279583333333333,
+      "eval_loss": 0.2117665410041809,
+      "eval_runtime": 15.8849,
+      "eval_samples_per_second": 1510.864,
+      "eval_steps_per_second": 4.721,
+      "step": 10120
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 1e-06,
+      "loss": 0.2192,
+      "step": 10140
+    },
+    {
+      "epoch": 4.22,
+      "eval_accuracy": 0.9291666666666667,
+      "eval_loss": 0.20962686836719513,
+      "eval_runtime": 16.1628,
+      "eval_samples_per_second": 1484.887,
+      "eval_steps_per_second": 4.64,
+      "step": 10140
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 1e-06,
+      "loss": 0.1977,
+      "step": 10160
+    },
+    {
+      "epoch": 4.23,
+      "eval_accuracy": 0.9277083333333334,
+      "eval_loss": 0.21077552437782288,
+      "eval_runtime": 15.8089,
+      "eval_samples_per_second": 1518.136,
+      "eval_steps_per_second": 4.744,
+      "step": 10160
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 1e-06,
+      "loss": 0.2138,
+      "step": 10180
+    },
+    {
+      "epoch": 4.24,
+      "eval_accuracy": 0.9267083333333334,
+      "eval_loss": 0.2138536125421524,
+      "eval_runtime": 15.891,
+      "eval_samples_per_second": 1510.292,
+      "eval_steps_per_second": 4.72,
+      "step": 10180
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 1e-06,
+      "loss": 0.1753,
+      "step": 10200
+    },
+    {
+      "epoch": 4.25,
+      "eval_accuracy": 0.92675,
+      "eval_loss": 0.21307241916656494,
+      "eval_runtime": 16.0289,
+      "eval_samples_per_second": 1497.293,
+      "eval_steps_per_second": 4.679,
+      "step": 10200
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 1e-06,
+      "loss": 0.2094,
+      "step": 10220
+    },
+    {
+      "epoch": 4.26,
+      "eval_accuracy": 0.9283333333333333,
+      "eval_loss": 0.21122166514396667,
+      "eval_runtime": 16.1206,
+      "eval_samples_per_second": 1488.781,
+      "eval_steps_per_second": 4.652,
+      "step": 10220
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 1e-06,
+      "loss": 0.204,
+      "step": 10240
+    },
+    {
+      "epoch": 4.27,
+      "eval_accuracy": 0.92825,
+      "eval_loss": 0.20935103297233582,
+      "eval_runtime": 15.822,
+      "eval_samples_per_second": 1516.879,
+      "eval_steps_per_second": 4.74,
+      "step": 10240
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 1e-06,
+      "loss": 0.2196,
+      "step": 10260
+    },
+    {
+      "epoch": 4.28,
+      "eval_accuracy": 0.927,
+      "eval_loss": 0.2119808942079544,
+      "eval_runtime": 16.0008,
+      "eval_samples_per_second": 1499.92,
+      "eval_steps_per_second": 4.687,
+      "step": 10260
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 1e-06,
+      "loss": 0.2122,
+      "step": 10280
+    },
+    {
+      "epoch": 4.28,
+      "eval_accuracy": 0.9271666666666667,
+      "eval_loss": 0.21251654624938965,
+      "eval_runtime": 16.2621,
+      "eval_samples_per_second": 1475.821,
+      "eval_steps_per_second": 4.612,
+      "step": 10280
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 1e-06,
+      "loss": 0.1534,
+      "step": 10300
+    },
+    {
+      "epoch": 4.29,
+      "eval_accuracy": 0.9284166666666667,
+      "eval_loss": 0.2099744975566864,
+      "eval_runtime": 15.4991,
+      "eval_samples_per_second": 1548.474,
+      "eval_steps_per_second": 4.839,
+      "step": 10300
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 1e-06,
+      "loss": 0.1917,
+      "step": 10320
+    },
+    {
+      "epoch": 4.3,
+      "eval_accuracy": 0.9281666666666667,
+      "eval_loss": 0.2101805955171585,
+      "eval_runtime": 15.541,
+      "eval_samples_per_second": 1544.303,
+      "eval_steps_per_second": 4.826,
+      "step": 10320
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 1e-06,
+      "loss": 0.201,
+      "step": 10340
+    },
+    {
+      "epoch": 4.31,
+      "eval_accuracy": 0.928125,
+      "eval_loss": 0.20908385515213013,
+      "eval_runtime": 15.6138,
+      "eval_samples_per_second": 1537.104,
+      "eval_steps_per_second": 4.803,
+      "step": 10340
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1736,
+      "step": 10360
+    },
+    {
+      "epoch": 4.32,
+      "eval_accuracy": 0.9292083333333333,
+      "eval_loss": 0.20927385985851288,
+      "eval_runtime": 15.7173,
+      "eval_samples_per_second": 1526.975,
+      "eval_steps_per_second": 4.772,
+      "step": 10360
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1948,
+      "step": 10380
+    },
+    {
+      "epoch": 4.33,
+      "eval_accuracy": 0.928625,
+      "eval_loss": 0.2104508876800537,
+      "eval_runtime": 15.7589,
+      "eval_samples_per_second": 1522.952,
+      "eval_steps_per_second": 4.759,
+      "step": 10380
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1967,
+      "step": 10400
+    },
+    {
+      "epoch": 4.33,
+      "eval_accuracy": 0.9270833333333334,
+      "eval_loss": 0.21193169057369232,
+      "eval_runtime": 15.5786,
+      "eval_samples_per_second": 1540.575,
+      "eval_steps_per_second": 4.814,
+      "step": 10400
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 1e-06,
+      "loss": 0.1722,
+      "step": 10420
+    },
+    {
+      "epoch": 4.34,
+      "eval_accuracy": 0.9289583333333333,
+      "eval_loss": 0.20838379859924316,
+      "eval_runtime": 16.4064,
+      "eval_samples_per_second": 1462.84,
+      "eval_steps_per_second": 4.571,
+      "step": 10420
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 1e-06,
+      "loss": 0.1855,
+      "step": 10440
+    },
+    {
+      "epoch": 4.35,
+      "eval_accuracy": 0.928375,
+      "eval_loss": 0.20829389989376068,
+      "eval_runtime": 15.8035,
+      "eval_samples_per_second": 1518.649,
+      "eval_steps_per_second": 4.746,
+      "step": 10440
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 1e-06,
+      "loss": 0.2067,
+      "step": 10460
+    },
+    {
+      "epoch": 4.36,
+      "eval_accuracy": 0.92925,
+      "eval_loss": 0.20744042098522186,
+      "eval_runtime": 15.8104,
+      "eval_samples_per_second": 1517.993,
+      "eval_steps_per_second": 4.744,
+      "step": 10460
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 1e-06,
+      "loss": 0.1925,
+      "step": 10480
+    },
+    {
+      "epoch": 4.37,
+      "eval_accuracy": 0.9292916666666666,
+      "eval_loss": 0.20747624337673187,
+      "eval_runtime": 16.0704,
+      "eval_samples_per_second": 1493.428,
+      "eval_steps_per_second": 4.667,
+      "step": 10480
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 1e-06,
+      "loss": 0.2027,
+      "step": 10500
+    },
+    {
+      "epoch": 4.38,
+      "eval_accuracy": 0.92775,
+      "eval_loss": 0.2088412493467331,
+      "eval_runtime": 15.8843,
+      "eval_samples_per_second": 1510.928,
+      "eval_steps_per_second": 4.722,
+      "step": 10500
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1763,
+      "step": 10520
+    },
+    {
+      "epoch": 4.38,
+      "eval_accuracy": 0.928125,
+      "eval_loss": 0.20948942005634308,
+      "eval_runtime": 15.8584,
+      "eval_samples_per_second": 1513.39,
+      "eval_steps_per_second": 4.729,
+      "step": 10520
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 1e-06,
+      "loss": 0.2019,
+      "step": 10540
+    },
+    {
+      "epoch": 4.39,
+      "eval_accuracy": 0.9283333333333333,
+      "eval_loss": 0.20972661674022675,
+      "eval_runtime": 16.0436,
+      "eval_samples_per_second": 1495.92,
+      "eval_steps_per_second": 4.675,
+      "step": 10540
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 1e-06,
+      "loss": 0.2411,
+      "step": 10560
+    },
+    {
+      "epoch": 4.4,
+      "eval_accuracy": 0.9285,
+      "eval_loss": 0.21000780165195465,
+      "eval_runtime": 15.8037,
+      "eval_samples_per_second": 1518.635,
+      "eval_steps_per_second": 4.746,
+      "step": 10560
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 1e-06,
+      "loss": 0.1896,
+      "step": 10580
+    },
+    {
+      "epoch": 4.41,
+      "eval_accuracy": 0.9274583333333334,
+      "eval_loss": 0.20897161960601807,
+      "eval_runtime": 15.8212,
+      "eval_samples_per_second": 1516.948,
+      "eval_steps_per_second": 4.74,
+      "step": 10580
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2153,
+      "step": 10600
+    },
+    {
+      "epoch": 4.42,
+      "eval_accuracy": 0.9294166666666667,
+      "eval_loss": 0.2074136584997177,
+      "eval_runtime": 16.2219,
+      "eval_samples_per_second": 1479.478,
+      "eval_steps_per_second": 4.623,
+      "step": 10600
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2157,
+      "step": 10620
+    },
+    {
+      "epoch": 4.42,
+      "eval_accuracy": 0.9291666666666667,
+      "eval_loss": 0.20682406425476074,
+      "eval_runtime": 15.4922,
+      "eval_samples_per_second": 1549.169,
+      "eval_steps_per_second": 4.841,
+      "step": 10620
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1676,
+      "step": 10640
+    },
+    {
+      "epoch": 4.43,
+      "eval_accuracy": 0.9277916666666667,
+      "eval_loss": 0.20833227038383484,
+      "eval_runtime": 16.1598,
+      "eval_samples_per_second": 1485.169,
+      "eval_steps_per_second": 4.641,
+      "step": 10640
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 1e-06,
+      "loss": 0.2071,
+      "step": 10660
+    },
+    {
+      "epoch": 4.44,
+      "eval_accuracy": 0.9270833333333334,
+      "eval_loss": 0.20793978869915009,
+      "eval_runtime": 15.9207,
+      "eval_samples_per_second": 1507.475,
+      "eval_steps_per_second": 4.711,
+      "step": 10660
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 1e-06,
+      "loss": 0.2311,
+      "step": 10680
+    },
+    {
+      "epoch": 4.45,
+      "eval_accuracy": 0.9275,
+      "eval_loss": 0.2090083658695221,
+      "eval_runtime": 15.8466,
+      "eval_samples_per_second": 1514.52,
+      "eval_steps_per_second": 4.733,
+      "step": 10680
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 1e-06,
+      "loss": 0.1938,
+      "step": 10700
+    },
+    {
+      "epoch": 4.46,
+      "eval_accuracy": 0.9301666666666667,
+      "eval_loss": 0.20576812326908112,
+      "eval_runtime": 16.5431,
+      "eval_samples_per_second": 1450.759,
+      "eval_steps_per_second": 4.534,
+      "step": 10700
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 1e-06,
+      "loss": 0.2202,
+      "step": 10720
+    },
+    {
+      "epoch": 4.47,
+      "eval_accuracy": 0.9292083333333333,
+      "eval_loss": 0.20580460131168365,
+      "eval_runtime": 15.9051,
+      "eval_samples_per_second": 1508.949,
+      "eval_steps_per_second": 4.715,
+      "step": 10720
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1872,
+      "step": 10740
+    },
+    {
+      "epoch": 4.47,
+      "eval_accuracy": 0.9278333333333333,
+      "eval_loss": 0.20827758312225342,
+      "eval_runtime": 15.8978,
+      "eval_samples_per_second": 1509.642,
+      "eval_steps_per_second": 4.718,
+      "step": 10740
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2034,
+      "step": 10760
+    },
+    {
+      "epoch": 4.48,
+      "eval_accuracy": 0.927875,
+      "eval_loss": 0.2091016322374344,
+      "eval_runtime": 15.7146,
+      "eval_samples_per_second": 1527.242,
+      "eval_steps_per_second": 4.773,
+      "step": 10760
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 1e-06,
+      "loss": 0.2204,
+      "step": 10780
+    },
+    {
+      "epoch": 4.49,
+      "eval_accuracy": 0.9276666666666666,
+      "eval_loss": 0.20921491086483002,
+      "eval_runtime": 16.115,
+      "eval_samples_per_second": 1489.295,
+      "eval_steps_per_second": 4.654,
+      "step": 10780
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 1e-06,
+      "loss": 0.216,
+      "step": 10800
+    },
+    {
+      "epoch": 4.5,
+      "eval_accuracy": 0.927625,
+      "eval_loss": 0.20790547132492065,
+      "eval_runtime": 15.9443,
+      "eval_samples_per_second": 1505.242,
+      "eval_steps_per_second": 4.704,
+      "step": 10800
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 1e-06,
+      "loss": 0.1824,
+      "step": 10820
+    },
+    {
+      "epoch": 4.51,
+      "eval_accuracy": 0.9280833333333334,
+      "eval_loss": 0.20656706392765045,
+      "eval_runtime": 16.1164,
+      "eval_samples_per_second": 1489.165,
+      "eval_steps_per_second": 4.654,
+      "step": 10820
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 1e-06,
+      "loss": 0.2098,
+      "step": 10840
+    },
+    {
+      "epoch": 4.52,
+      "eval_accuracy": 0.9296666666666666,
+      "eval_loss": 0.20553721487522125,
+      "eval_runtime": 15.8252,
+      "eval_samples_per_second": 1516.573,
+      "eval_steps_per_second": 4.739,
+      "step": 10840
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 1e-06,
+      "loss": 0.2258,
+      "step": 10860
+    },
+    {
+      "epoch": 4.53,
+      "eval_accuracy": 0.9294583333333334,
+      "eval_loss": 0.2055697739124298,
+      "eval_runtime": 15.8329,
+      "eval_samples_per_second": 1515.829,
+      "eval_steps_per_second": 4.737,
+      "step": 10860
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 1e-06,
+      "loss": 0.2086,
+      "step": 10880
+    },
+    {
+      "epoch": 4.53,
+      "eval_accuracy": 0.9282083333333333,
+      "eval_loss": 0.20601527392864227,
+      "eval_runtime": 16.046,
+      "eval_samples_per_second": 1495.703,
+      "eval_steps_per_second": 4.674,
+      "step": 10880
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 1e-06,
+      "loss": 0.1725,
+      "step": 10900
+    },
+    {
+      "epoch": 4.54,
+      "eval_accuracy": 0.9290416666666667,
+      "eval_loss": 0.20634058117866516,
+      "eval_runtime": 15.8771,
+      "eval_samples_per_second": 1511.608,
+      "eval_steps_per_second": 4.724,
+      "step": 10900
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 1e-06,
+      "loss": 0.1899,
+      "step": 10920
+    },
+    {
+      "epoch": 4.55,
+      "eval_accuracy": 0.9292916666666666,
+      "eval_loss": 0.20672395825386047,
+      "eval_runtime": 16.1064,
+      "eval_samples_per_second": 1490.089,
+      "eval_steps_per_second": 4.657,
+      "step": 10920
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 1e-06,
+      "loss": 0.1959,
+      "step": 10940
+    },
+    {
+      "epoch": 4.56,
+      "eval_accuracy": 0.9266666666666666,
+      "eval_loss": 0.21057891845703125,
+      "eval_runtime": 15.9754,
+      "eval_samples_per_second": 1502.309,
+      "eval_steps_per_second": 4.695,
+      "step": 10940
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 1e-06,
+      "loss": 0.2029,
+      "step": 10960
+    },
+    {
+      "epoch": 4.57,
+      "eval_accuracy": 0.928875,
+      "eval_loss": 0.20633479952812195,
+      "eval_runtime": 15.7725,
+      "eval_samples_per_second": 1521.637,
+      "eval_steps_per_second": 4.755,
+      "step": 10960
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 1e-06,
+      "loss": 0.1843,
+      "step": 10980
+    },
+    {
+      "epoch": 4.58,
+      "eval_accuracy": 0.9300833333333334,
+      "eval_loss": 0.2045080065727234,
+      "eval_runtime": 16.6018,
+      "eval_samples_per_second": 1445.625,
+      "eval_steps_per_second": 4.518,
+      "step": 10980
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2249,
+      "step": 11000
+    },
+    {
+      "epoch": 4.58,
+      "eval_accuracy": 0.93025,
+      "eval_loss": 0.20456919074058533,
+      "eval_runtime": 15.8868,
+      "eval_samples_per_second": 1510.688,
+      "eval_steps_per_second": 4.721,
+      "step": 11000
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 1e-06,
+      "loss": 0.1764,
+      "step": 11020
+    },
+    {
+      "epoch": 4.59,
+      "eval_accuracy": 0.92925,
+      "eval_loss": 0.20611906051635742,
+      "eval_runtime": 16.0015,
+      "eval_samples_per_second": 1499.856,
+      "eval_steps_per_second": 4.687,
+      "step": 11020
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 1e-06,
+      "loss": 0.2269,
+      "step": 11040
+    },
+    {
+      "epoch": 4.6,
+      "eval_accuracy": 0.9290416666666667,
+      "eval_loss": 0.20687995851039886,
+      "eval_runtime": 15.9876,
+      "eval_samples_per_second": 1501.163,
+      "eval_steps_per_second": 4.691,
+      "step": 11040
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 1e-06,
+      "loss": 0.234,
+      "step": 11060
+    },
+    {
+      "epoch": 4.61,
+      "eval_accuracy": 0.92825,
+      "eval_loss": 0.20855723321437836,
+      "eval_runtime": 15.799,
+      "eval_samples_per_second": 1519.081,
+      "eval_steps_per_second": 4.747,
+      "step": 11060
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1925,
+      "step": 11080
+    },
+    {
+      "epoch": 4.62,
+      "eval_accuracy": 0.9294166666666667,
+      "eval_loss": 0.2057270109653473,
+      "eval_runtime": 15.765,
+      "eval_samples_per_second": 1522.358,
+      "eval_steps_per_second": 4.757,
+      "step": 11080
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1949,
+      "step": 11100
+    },
+    {
+      "epoch": 4.62,
+      "eval_accuracy": 0.9285,
+      "eval_loss": 0.206070676445961,
+      "eval_runtime": 16.1103,
+      "eval_samples_per_second": 1489.726,
+      "eval_steps_per_second": 4.655,
+      "step": 11100
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 1e-06,
+      "loss": 0.1928,
+      "step": 11120
+    },
+    {
+      "epoch": 4.63,
+      "eval_accuracy": 0.9290416666666667,
+      "eval_loss": 0.20591707527637482,
+      "eval_runtime": 15.8556,
+      "eval_samples_per_second": 1513.659,
+      "eval_steps_per_second": 4.73,
+      "step": 11120
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 1e-06,
+      "loss": 0.1744,
+      "step": 11140
+    },
+    {
+      "epoch": 4.64,
+      "eval_accuracy": 0.9286666666666666,
+      "eval_loss": 0.20713801681995392,
+      "eval_runtime": 16.1571,
+      "eval_samples_per_second": 1485.413,
+      "eval_steps_per_second": 4.642,
+      "step": 11140
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 1e-06,
+      "loss": 0.2161,
+      "step": 11160
+    },
+    {
+      "epoch": 4.65,
+      "eval_accuracy": 0.92925,
+      "eval_loss": 0.20511361956596375,
+      "eval_runtime": 16.4581,
+      "eval_samples_per_second": 1458.25,
+      "eval_steps_per_second": 4.557,
+      "step": 11160
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 1e-06,
+      "loss": 0.168,
+      "step": 11180
+    },
+    {
+      "epoch": 4.66,
+      "eval_accuracy": 0.92925,
+      "eval_loss": 0.2065957635641098,
+      "eval_runtime": 15.7867,
+      "eval_samples_per_second": 1520.267,
+      "eval_steps_per_second": 4.751,
+      "step": 11180
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1739,
+      "step": 11200
+    },
+    {
+      "epoch": 4.67,
+      "eval_accuracy": 0.9285833333333333,
+      "eval_loss": 0.20561246573925018,
+      "eval_runtime": 15.6163,
+      "eval_samples_per_second": 1536.854,
+      "eval_steps_per_second": 4.803,
+      "step": 11200
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1816,
+      "step": 11220
+    },
+    {
+      "epoch": 4.67,
+      "eval_accuracy": 0.9287916666666667,
+      "eval_loss": 0.20476531982421875,
+      "eval_runtime": 16.2877,
+      "eval_samples_per_second": 1473.506,
+      "eval_steps_per_second": 4.605,
+      "step": 11220
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1931,
+      "step": 11240
+    },
+    {
+      "epoch": 4.68,
+      "eval_accuracy": 0.9285416666666667,
+      "eval_loss": 0.20711849629878998,
+      "eval_runtime": 15.8294,
+      "eval_samples_per_second": 1516.17,
+      "eval_steps_per_second": 4.738,
+      "step": 11240
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1945,
+      "step": 11260
+    },
+    {
+      "epoch": 4.69,
+      "eval_accuracy": 0.9281666666666667,
+      "eval_loss": 0.20717017352581024,
+      "eval_runtime": 16.0348,
+      "eval_samples_per_second": 1496.744,
+      "eval_steps_per_second": 4.677,
+      "step": 11260
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1929,
+      "step": 11280
+    },
+    {
+      "epoch": 4.7,
+      "eval_accuracy": 0.928875,
+      "eval_loss": 0.20755107700824738,
+      "eval_runtime": 16.3373,
+      "eval_samples_per_second": 1469.027,
+      "eval_steps_per_second": 4.591,
+      "step": 11280
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1903,
+      "step": 11300
+    },
+    {
+      "epoch": 4.71,
+      "eval_accuracy": 0.929875,
+      "eval_loss": 0.20404241979122162,
+      "eval_runtime": 16.0151,
+      "eval_samples_per_second": 1498.586,
+      "eval_steps_per_second": 4.683,
+      "step": 11300
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 1e-06,
+      "loss": 0.2051,
+      "step": 11320
+    },
+    {
+      "epoch": 4.72,
+      "eval_accuracy": 0.9290416666666667,
+      "eval_loss": 0.20393158495426178,
+      "eval_runtime": 16.464,
+      "eval_samples_per_second": 1457.729,
+      "eval_steps_per_second": 4.555,
+      "step": 11320
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1614,
+      "step": 11340
+    },
+    {
+      "epoch": 4.72,
+      "eval_accuracy": 0.929125,
+      "eval_loss": 0.20461545884609222,
+      "eval_runtime": 15.8151,
+      "eval_samples_per_second": 1517.534,
+      "eval_steps_per_second": 4.742,
+      "step": 11340
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 1e-06,
+      "loss": 0.1968,
+      "step": 11360
+    },
+    {
+      "epoch": 4.73,
+      "eval_accuracy": 0.9283333333333333,
+      "eval_loss": 0.20622652769088745,
+      "eval_runtime": 15.9434,
+      "eval_samples_per_second": 1505.328,
+      "eval_steps_per_second": 4.704,
+      "step": 11360
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 1e-06,
+      "loss": 0.2091,
+      "step": 11380
+    },
+    {
+      "epoch": 4.74,
+      "eval_accuracy": 0.926375,
+      "eval_loss": 0.2111745923757553,
+      "eval_runtime": 15.9338,
+      "eval_samples_per_second": 1506.236,
+      "eval_steps_per_second": 4.707,
+      "step": 11380
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 1e-06,
+      "loss": 0.2072,
+      "step": 11400
+    },
+    {
+      "epoch": 4.75,
+      "eval_accuracy": 0.9277916666666667,
+      "eval_loss": 0.2086385041475296,
+      "eval_runtime": 15.7955,
+      "eval_samples_per_second": 1519.425,
+      "eval_steps_per_second": 4.748,
+      "step": 11400
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 1e-06,
+      "loss": 0.2298,
+      "step": 11420
+    },
+    {
+      "epoch": 4.76,
+      "eval_accuracy": 0.9287083333333334,
+      "eval_loss": 0.2052876353263855,
+      "eval_runtime": 16.0941,
+      "eval_samples_per_second": 1491.226,
+      "eval_steps_per_second": 4.66,
+      "step": 11420
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 1e-06,
+      "loss": 0.1874,
+      "step": 11440
+    },
+    {
+      "epoch": 4.77,
+      "eval_accuracy": 0.9292916666666666,
+      "eval_loss": 0.20625039935112,
+      "eval_runtime": 15.9003,
+      "eval_samples_per_second": 1509.408,
+      "eval_steps_per_second": 4.717,
+      "step": 11440
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1838,
+      "step": 11460
+    },
+    {
+      "epoch": 4.78,
+      "eval_accuracy": 0.928,
+      "eval_loss": 0.20656456053256989,
+      "eval_runtime": 16.1911,
+      "eval_samples_per_second": 1482.299,
+      "eval_steps_per_second": 4.632,
+      "step": 11460
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2015,
+      "step": 11480
+    },
+    {
+      "epoch": 4.78,
+      "eval_accuracy": 0.929625,
+      "eval_loss": 0.20533688366413116,
+      "eval_runtime": 15.5948,
+      "eval_samples_per_second": 1538.974,
+      "eval_steps_per_second": 4.809,
+      "step": 11480
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 1e-06,
+      "loss": 0.1822,
+      "step": 11500
+    },
+    {
+      "epoch": 4.79,
+      "eval_accuracy": 0.9284583333333334,
+      "eval_loss": 0.20839504897594452,
+      "eval_runtime": 15.9851,
+      "eval_samples_per_second": 1501.396,
+      "eval_steps_per_second": 4.692,
+      "step": 11500
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 1e-06,
+      "loss": 0.2209,
+      "step": 11520
+    },
+    {
+      "epoch": 4.8,
+      "eval_accuracy": 0.9295,
+      "eval_loss": 0.20546448230743408,
+      "eval_runtime": 15.8868,
+      "eval_samples_per_second": 1510.688,
+      "eval_steps_per_second": 4.721,
+      "step": 11520
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1918,
+      "step": 11540
+    },
+    {
+      "epoch": 4.81,
+      "eval_accuracy": 0.929625,
+      "eval_loss": 0.20511560142040253,
+      "eval_runtime": 16.5324,
+      "eval_samples_per_second": 1451.696,
+      "eval_steps_per_second": 4.537,
+      "step": 11540
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 1e-06,
+      "loss": 0.2252,
+      "step": 11560
+    },
+    {
+      "epoch": 4.82,
+      "eval_accuracy": 0.9294166666666667,
+      "eval_loss": 0.2052285224199295,
+      "eval_runtime": 15.5246,
+      "eval_samples_per_second": 1545.931,
+      "eval_steps_per_second": 4.831,
+      "step": 11560
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1929,
+      "step": 11580
+    },
+    {
+      "epoch": 4.83,
+      "eval_accuracy": 0.9300833333333334,
+      "eval_loss": 0.20385289192199707,
+      "eval_runtime": 15.4266,
+      "eval_samples_per_second": 1555.758,
+      "eval_steps_per_second": 4.862,
+      "step": 11580
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1889,
+      "step": 11600
+    },
+    {
+      "epoch": 4.83,
+      "eval_accuracy": 0.9298333333333333,
+      "eval_loss": 0.2047443836927414,
+      "eval_runtime": 15.9168,
+      "eval_samples_per_second": 1507.841,
+      "eval_steps_per_second": 4.712,
+      "step": 11600
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 1e-06,
+      "loss": 0.1812,
+      "step": 11620
+    },
+    {
+      "epoch": 4.84,
+      "eval_accuracy": 0.929,
+      "eval_loss": 0.20585575699806213,
+      "eval_runtime": 15.6349,
+      "eval_samples_per_second": 1535.024,
+      "eval_steps_per_second": 4.797,
+      "step": 11620
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 1e-06,
+      "loss": 0.2473,
+      "step": 11640
+    },
+    {
+      "epoch": 4.85,
+      "eval_accuracy": 0.9304166666666667,
+      "eval_loss": 0.2033381462097168,
+      "eval_runtime": 16.0063,
+      "eval_samples_per_second": 1499.413,
+      "eval_steps_per_second": 4.686,
+      "step": 11640
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1757,
+      "step": 11660
+    },
+    {
+      "epoch": 4.86,
+      "eval_accuracy": 0.9302083333333333,
+      "eval_loss": 0.20283745229244232,
+      "eval_runtime": 15.8227,
+      "eval_samples_per_second": 1516.806,
+      "eval_steps_per_second": 4.74,
+      "step": 11660
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 1e-06,
+      "loss": 0.2138,
+      "step": 11680
+    },
+    {
+      "epoch": 4.87,
+      "eval_accuracy": 0.9295833333333333,
+      "eval_loss": 0.20379288494586945,
+      "eval_runtime": 15.9957,
+      "eval_samples_per_second": 1500.406,
+      "eval_steps_per_second": 4.689,
+      "step": 11680
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 1e-06,
+      "loss": 0.2594,
+      "step": 11700
+    },
+    {
+      "epoch": 4.88,
+      "eval_accuracy": 0.9294166666666667,
+      "eval_loss": 0.20390905439853668,
+      "eval_runtime": 15.8805,
+      "eval_samples_per_second": 1511.292,
+      "eval_steps_per_second": 4.723,
+      "step": 11700
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 1e-06,
+      "loss": 0.183,
+      "step": 11720
+    },
+    {
+      "epoch": 4.88,
+      "eval_accuracy": 0.9290416666666667,
+      "eval_loss": 0.2046515792608261,
+      "eval_runtime": 16.3804,
+      "eval_samples_per_second": 1465.166,
+      "eval_steps_per_second": 4.579,
+      "step": 11720
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 1e-06,
+      "loss": 0.1479,
+      "step": 11740
+    },
+    {
+      "epoch": 4.89,
+      "eval_accuracy": 0.9282083333333333,
+      "eval_loss": 0.20830100774765015,
+      "eval_runtime": 15.9731,
+      "eval_samples_per_second": 1502.525,
+      "eval_steps_per_second": 4.695,
+      "step": 11740
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 1e-06,
+      "loss": 0.1684,
+      "step": 11760
+    },
+    {
+      "epoch": 4.9,
+      "eval_accuracy": 0.9296666666666666,
+      "eval_loss": 0.20390328764915466,
+      "eval_runtime": 15.9815,
+      "eval_samples_per_second": 1501.737,
+      "eval_steps_per_second": 4.693,
+      "step": 11760
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 1e-06,
+      "loss": 0.1802,
+      "step": 11780
+    },
+    {
+      "epoch": 4.91,
+      "eval_accuracy": 0.928125,
+      "eval_loss": 0.2074248492717743,
+      "eval_runtime": 16.0187,
+      "eval_samples_per_second": 1498.245,
+      "eval_steps_per_second": 4.682,
+      "step": 11780
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1707,
+      "step": 11800
+    },
+    {
+      "epoch": 4.92,
+      "eval_accuracy": 0.92875,
+      "eval_loss": 0.20657067000865936,
+      "eval_runtime": 15.9205,
+      "eval_samples_per_second": 1507.488,
+      "eval_steps_per_second": 4.711,
+      "step": 11800
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 1e-06,
+      "loss": 0.2031,
+      "step": 11820
+    },
+    {
+      "epoch": 4.92,
+      "eval_accuracy": 0.928375,
+      "eval_loss": 0.21004259586334229,
+      "eval_runtime": 15.5995,
+      "eval_samples_per_second": 1538.507,
+      "eval_steps_per_second": 4.808,
+      "step": 11820
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 1e-06,
+      "loss": 0.2169,
+      "step": 11840
+    },
+    {
+      "epoch": 4.93,
+      "eval_accuracy": 0.9290416666666667,
+      "eval_loss": 0.20725244283676147,
+      "eval_runtime": 15.6197,
+      "eval_samples_per_second": 1536.517,
+      "eval_steps_per_second": 4.802,
+      "step": 11840
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 1e-06,
+      "loss": 0.1941,
+      "step": 11860
+    },
+    {
+      "epoch": 4.94,
+      "eval_accuracy": 0.930125,
+      "eval_loss": 0.2036478966474533,
+      "eval_runtime": 16.2424,
+      "eval_samples_per_second": 1477.617,
+      "eval_steps_per_second": 4.618,
+      "step": 11860
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 1e-06,
+      "loss": 0.2131,
+      "step": 11880
+    },
+    {
+      "epoch": 4.95,
+      "eval_accuracy": 0.9295,
+      "eval_loss": 0.2052011936903,
+      "eval_runtime": 15.8259,
+      "eval_samples_per_second": 1516.501,
+      "eval_steps_per_second": 4.739,
+      "step": 11880
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 1e-06,
+      "loss": 0.1867,
+      "step": 11900
+    },
+    {
+      "epoch": 4.96,
+      "eval_accuracy": 0.9289583333333333,
+      "eval_loss": 0.20610161125659943,
+      "eval_runtime": 15.9156,
+      "eval_samples_per_second": 1507.958,
+      "eval_steps_per_second": 4.712,
+      "step": 11900
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 1e-06,
+      "loss": 0.176,
+      "step": 11920
+    },
+    {
+      "epoch": 4.97,
+      "eval_accuracy": 0.9302916666666666,
+      "eval_loss": 0.20278537273406982,
+      "eval_runtime": 16.3933,
+      "eval_samples_per_second": 1464.01,
+      "eval_steps_per_second": 4.575,
+      "step": 11920
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1932,
+      "step": 11940
+    },
+    {
+      "epoch": 4.97,
+      "eval_accuracy": 0.9305833333333333,
+      "eval_loss": 0.2031938135623932,
+      "eval_runtime": 15.7386,
+      "eval_samples_per_second": 1524.912,
+      "eval_steps_per_second": 4.765,
+      "step": 11940
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 1e-06,
+      "loss": 0.2253,
+      "step": 11960
+    },
+    {
+      "epoch": 4.98,
+      "eval_accuracy": 0.930125,
+      "eval_loss": 0.20466017723083496,
+      "eval_runtime": 15.9542,
+      "eval_samples_per_second": 1504.308,
+      "eval_steps_per_second": 4.701,
+      "step": 11960
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 1e-06,
+      "loss": 0.2001,
+      "step": 11980
+    },
+    {
+      "epoch": 4.99,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.20326323807239532,
+      "eval_runtime": 16.2585,
+      "eval_samples_per_second": 1476.147,
+      "eval_steps_per_second": 4.613,
+      "step": 11980
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 1e-06,
+      "loss": 0.2234,
+      "step": 12000
+    },
+    {
+      "epoch": 5.0,
+      "eval_accuracy": 0.9305,
+      "eval_loss": 0.20482754707336426,
+      "eval_runtime": 15.5729,
+      "eval_samples_per_second": 1541.134,
+      "eval_steps_per_second": 4.816,
+      "step": 12000
+    },
+    {
+      "epoch": 5.01,
+      "learning_rate": 1e-06,
+      "loss": 0.2004,
+      "step": 12020
+    },
+    {
+      "epoch": 5.01,
+      "eval_accuracy": 0.9305416666666667,
+      "eval_loss": 0.20364505052566528,
+      "eval_runtime": 16.9313,
+      "eval_samples_per_second": 1417.493,
+      "eval_steps_per_second": 4.43,
+      "step": 12020
+    },
+    {
+      "epoch": 5.02,
+      "learning_rate": 1e-06,
+      "loss": 0.1859,
+      "step": 12040
+    },
+    {
+      "epoch": 5.02,
+      "eval_accuracy": 0.9309583333333333,
+      "eval_loss": 0.2025202065706253,
+      "eval_runtime": 16.472,
+      "eval_samples_per_second": 1457.018,
+      "eval_steps_per_second": 4.553,
+      "step": 12040
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 1e-06,
+      "loss": 0.19,
+      "step": 12060
+    },
+    {
+      "epoch": 5.03,
+      "eval_accuracy": 0.931375,
+      "eval_loss": 0.20227618515491486,
+      "eval_runtime": 15.727,
+      "eval_samples_per_second": 1526.038,
+      "eval_steps_per_second": 4.769,
+      "step": 12060
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1645,
+      "step": 12080
+    },
+    {
+      "epoch": 5.03,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.20448292791843414,
+      "eval_runtime": 16.1326,
+      "eval_samples_per_second": 1487.669,
+      "eval_steps_per_second": 4.649,
+      "step": 12080
+    },
+    {
+      "epoch": 5.04,
+      "learning_rate": 1e-06,
+      "loss": 0.2003,
+      "step": 12100
+    },
+    {
+      "epoch": 5.04,
+      "eval_accuracy": 0.931875,
+      "eval_loss": 0.20208178460597992,
+      "eval_runtime": 16.0837,
+      "eval_samples_per_second": 1492.198,
+      "eval_steps_per_second": 4.663,
+      "step": 12100
+    },
+    {
+      "epoch": 5.05,
+      "learning_rate": 1e-06,
+      "loss": 0.1788,
+      "step": 12120
+    },
+    {
+      "epoch": 5.05,
+      "eval_accuracy": 0.9320833333333334,
+      "eval_loss": 0.2024490088224411,
+      "eval_runtime": 15.8984,
+      "eval_samples_per_second": 1509.589,
+      "eval_steps_per_second": 4.717,
+      "step": 12120
+    },
+    {
+      "epoch": 5.06,
+      "learning_rate": 1e-06,
+      "loss": 0.1697,
+      "step": 12140
+    },
+    {
+      "epoch": 5.06,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.202731654047966,
+      "eval_runtime": 15.9929,
+      "eval_samples_per_second": 1500.664,
+      "eval_steps_per_second": 4.69,
+      "step": 12140
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1824,
+      "step": 12160
+    },
+    {
+      "epoch": 5.07,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.2048906683921814,
+      "eval_runtime": 15.963,
+      "eval_samples_per_second": 1503.477,
+      "eval_steps_per_second": 4.698,
+      "step": 12160
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1773,
+      "step": 12180
+    },
+    {
+      "epoch": 5.08,
+      "eval_accuracy": 0.9302916666666666,
+      "eval_loss": 0.20466017723083496,
+      "eval_runtime": 16.2544,
+      "eval_samples_per_second": 1476.528,
+      "eval_steps_per_second": 4.614,
+      "step": 12180
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 1e-06,
+      "loss": 0.2324,
+      "step": 12200
+    },
+    {
+      "epoch": 5.08,
+      "eval_accuracy": 0.9304166666666667,
+      "eval_loss": 0.20540772378444672,
+      "eval_runtime": 16.2048,
+      "eval_samples_per_second": 1481.039,
+      "eval_steps_per_second": 4.628,
+      "step": 12200
+    },
+    {
+      "epoch": 5.09,
+      "learning_rate": 1e-06,
+      "loss": 0.1513,
+      "step": 12220
+    },
+    {
+      "epoch": 5.09,
+      "eval_accuracy": 0.9298333333333333,
+      "eval_loss": 0.20598597824573517,
+      "eval_runtime": 15.5756,
+      "eval_samples_per_second": 1540.872,
+      "eval_steps_per_second": 4.815,
+      "step": 12220
+    },
+    {
+      "epoch": 5.1,
+      "learning_rate": 1e-06,
+      "loss": 0.1975,
+      "step": 12240
+    },
+    {
+      "epoch": 5.1,
+      "eval_accuracy": 0.9305833333333333,
+      "eval_loss": 0.20424893498420715,
+      "eval_runtime": 15.9543,
+      "eval_samples_per_second": 1504.299,
+      "eval_steps_per_second": 4.701,
+      "step": 12240
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 1e-06,
+      "loss": 0.2149,
+      "step": 12260
+    },
+    {
+      "epoch": 5.11,
+      "eval_accuracy": 0.9297916666666667,
+      "eval_loss": 0.205413356423378,
+      "eval_runtime": 15.6474,
+      "eval_samples_per_second": 1533.804,
+      "eval_steps_per_second": 4.793,
+      "step": 12260
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1685,
+      "step": 12280
+    },
+    {
+      "epoch": 5.12,
+      "eval_accuracy": 0.9296666666666666,
+      "eval_loss": 0.20603446662425995,
+      "eval_runtime": 16.0146,
+      "eval_samples_per_second": 1498.636,
+      "eval_steps_per_second": 4.683,
+      "step": 12280
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2097,
+      "step": 12300
+    },
+    {
+      "epoch": 5.12,
+      "eval_accuracy": 0.93125,
+      "eval_loss": 0.20300358533859253,
+      "eval_runtime": 15.8596,
+      "eval_samples_per_second": 1513.277,
+      "eval_steps_per_second": 4.729,
+      "step": 12300
+    },
+    {
+      "epoch": 5.13,
+      "learning_rate": 1e-06,
+      "loss": 0.2318,
+      "step": 12320
+    },
+    {
+      "epoch": 5.13,
+      "eval_accuracy": 0.931375,
+      "eval_loss": 0.20188292860984802,
+      "eval_runtime": 15.9958,
+      "eval_samples_per_second": 1500.398,
+      "eval_steps_per_second": 4.689,
+      "step": 12320
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 1e-06,
+      "loss": 0.1662,
+      "step": 12340
+    },
+    {
+      "epoch": 5.14,
+      "eval_accuracy": 0.9291666666666667,
+      "eval_loss": 0.20399628579616547,
+      "eval_runtime": 15.942,
+      "eval_samples_per_second": 1505.459,
+      "eval_steps_per_second": 4.705,
+      "step": 12340
+    },
+    {
+      "epoch": 5.15,
+      "learning_rate": 1e-06,
+      "loss": 0.1792,
+      "step": 12360
+    },
+    {
+      "epoch": 5.15,
+      "eval_accuracy": 0.92925,
+      "eval_loss": 0.205734983086586,
+      "eval_runtime": 16.3128,
+      "eval_samples_per_second": 1471.239,
+      "eval_steps_per_second": 4.598,
+      "step": 12360
+    },
+    {
+      "epoch": 5.16,
+      "learning_rate": 1e-06,
+      "loss": 0.1762,
+      "step": 12380
+    },
+    {
+      "epoch": 5.16,
+      "eval_accuracy": 0.9287916666666667,
+      "eval_loss": 0.2092132568359375,
+      "eval_runtime": 15.9137,
+      "eval_samples_per_second": 1508.133,
+      "eval_steps_per_second": 4.713,
+      "step": 12380
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 1e-06,
+      "loss": 0.1693,
+      "step": 12400
+    },
+    {
+      "epoch": 5.17,
+      "eval_accuracy": 0.929375,
+      "eval_loss": 0.20484225451946259,
+      "eval_runtime": 16.0154,
+      "eval_samples_per_second": 1498.562,
+      "eval_steps_per_second": 4.683,
+      "step": 12400
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 1e-06,
+      "loss": 0.1659,
+      "step": 12420
+    },
+    {
+      "epoch": 5.17,
+      "eval_accuracy": 0.9309166666666666,
+      "eval_loss": 0.2023383527994156,
+      "eval_runtime": 16.3394,
+      "eval_samples_per_second": 1468.841,
+      "eval_steps_per_second": 4.59,
+      "step": 12420
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1561,
+      "step": 12440
+    },
+    {
+      "epoch": 5.18,
+      "eval_accuracy": 0.930875,
+      "eval_loss": 0.20242756605148315,
+      "eval_runtime": 15.5429,
+      "eval_samples_per_second": 1544.111,
+      "eval_steps_per_second": 4.825,
+      "step": 12440
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 1e-06,
+      "loss": 0.1826,
+      "step": 12460
+    },
+    {
+      "epoch": 5.19,
+      "eval_accuracy": 0.9317916666666667,
+      "eval_loss": 0.2020563930273056,
+      "eval_runtime": 16.0095,
+      "eval_samples_per_second": 1499.108,
+      "eval_steps_per_second": 4.685,
+      "step": 12460
+    },
+    {
+      "epoch": 5.2,
+      "learning_rate": 1e-06,
+      "loss": 0.1544,
+      "step": 12480
+    },
+    {
+      "epoch": 5.2,
+      "eval_accuracy": 0.93075,
+      "eval_loss": 0.2041223645210266,
+      "eval_runtime": 15.9404,
+      "eval_samples_per_second": 1505.609,
+      "eval_steps_per_second": 4.705,
+      "step": 12480
+    },
+    {
+      "epoch": 5.21,
+      "learning_rate": 1e-06,
+      "loss": 0.1836,
+      "step": 12500
+    },
+    {
+      "epoch": 5.21,
+      "eval_accuracy": 0.9310833333333334,
+      "eval_loss": 0.20302866399288177,
+      "eval_runtime": 15.672,
+      "eval_samples_per_second": 1531.396,
+      "eval_steps_per_second": 4.786,
+      "step": 12500
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1792,
+      "step": 12520
+    },
+    {
+      "epoch": 5.22,
+      "eval_accuracy": 0.93125,
+      "eval_loss": 0.20235809683799744,
+      "eval_runtime": 16.0568,
+      "eval_samples_per_second": 1494.695,
+      "eval_steps_per_second": 4.671,
+      "step": 12520
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1666,
+      "step": 12540
+    },
+    {
+      "epoch": 5.22,
+      "eval_accuracy": 0.931875,
+      "eval_loss": 0.2016390711069107,
+      "eval_runtime": 15.811,
+      "eval_samples_per_second": 1517.93,
+      "eval_steps_per_second": 4.744,
+      "step": 12540
+    },
+    {
+      "epoch": 5.23,
+      "learning_rate": 1e-06,
+      "loss": 0.1714,
+      "step": 12560
+    },
+    {
+      "epoch": 5.23,
+      "eval_accuracy": 0.9312916666666666,
+      "eval_loss": 0.20161676406860352,
+      "eval_runtime": 16.4914,
+      "eval_samples_per_second": 1455.303,
+      "eval_steps_per_second": 4.548,
+      "step": 12560
+    },
+    {
+      "epoch": 5.24,
+      "learning_rate": 1e-06,
+      "loss": 0.2023,
+      "step": 12580
+    },
+    {
+      "epoch": 5.24,
+      "eval_accuracy": 0.930125,
+      "eval_loss": 0.20450520515441895,
+      "eval_runtime": 16.0345,
+      "eval_samples_per_second": 1496.773,
+      "eval_steps_per_second": 4.677,
+      "step": 12580
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 1e-06,
+      "loss": 0.1996,
+      "step": 12600
+    },
+    {
+      "epoch": 5.25,
+      "eval_accuracy": 0.9287916666666667,
+      "eval_loss": 0.20813672244548798,
+      "eval_runtime": 15.8781,
+      "eval_samples_per_second": 1511.512,
+      "eval_steps_per_second": 4.723,
+      "step": 12600
+    },
+    {
+      "epoch": 5.26,
+      "learning_rate": 1e-06,
+      "loss": 0.2113,
+      "step": 12620
+    },
+    {
+      "epoch": 5.26,
+      "eval_accuracy": 0.931125,
+      "eval_loss": 0.20202632248401642,
+      "eval_runtime": 15.9398,
+      "eval_samples_per_second": 1505.669,
+      "eval_steps_per_second": 4.705,
+      "step": 12620
+    },
+    {
+      "epoch": 5.27,
+      "learning_rate": 1e-06,
+      "loss": 0.1826,
+      "step": 12640
+    },
+    {
+      "epoch": 5.27,
+      "eval_accuracy": 0.9315416666666667,
+      "eval_loss": 0.2010980248451233,
+      "eval_runtime": 15.6531,
+      "eval_samples_per_second": 1533.238,
+      "eval_steps_per_second": 4.791,
+      "step": 12640
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 1e-06,
+      "loss": 0.2069,
+      "step": 12660
+    },
+    {
+      "epoch": 5.28,
+      "eval_accuracy": 0.9294583333333334,
+      "eval_loss": 0.20446471869945526,
+      "eval_runtime": 15.9931,
+      "eval_samples_per_second": 1500.646,
+      "eval_steps_per_second": 4.69,
+      "step": 12660
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1621,
+      "step": 12680
+    },
+    {
+      "epoch": 5.28,
+      "eval_accuracy": 0.9308333333333333,
+      "eval_loss": 0.20204661786556244,
+      "eval_runtime": 16.4396,
+      "eval_samples_per_second": 1459.894,
+      "eval_steps_per_second": 4.562,
+      "step": 12680
+    },
+    {
+      "epoch": 5.29,
+      "learning_rate": 1e-06,
+      "loss": 0.16,
+      "step": 12700
+    },
+    {
+      "epoch": 5.29,
+      "eval_accuracy": 0.93075,
+      "eval_loss": 0.20316839218139648,
+      "eval_runtime": 15.9576,
+      "eval_samples_per_second": 1503.99,
+      "eval_steps_per_second": 4.7,
+      "step": 12700
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 1e-06,
+      "loss": 0.165,
+      "step": 12720
+    },
+    {
+      "epoch": 5.3,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.20250557363033295,
+      "eval_runtime": 15.8087,
+      "eval_samples_per_second": 1518.148,
+      "eval_steps_per_second": 4.744,
+      "step": 12720
+    },
+    {
+      "epoch": 5.31,
+      "learning_rate": 1e-06,
+      "loss": 0.1894,
+      "step": 12740
+    },
+    {
+      "epoch": 5.31,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.20032641291618347,
+      "eval_runtime": 15.4644,
+      "eval_samples_per_second": 1551.949,
+      "eval_steps_per_second": 4.85,
+      "step": 12740
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1633,
+      "step": 12760
+    },
+    {
+      "epoch": 5.32,
+      "eval_accuracy": 0.9314166666666667,
+      "eval_loss": 0.20086656510829926,
+      "eval_runtime": 16.2759,
+      "eval_samples_per_second": 1474.571,
+      "eval_steps_per_second": 4.608,
+      "step": 12760
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 1e-06,
+      "loss": 0.2289,
+      "step": 12780
+    },
+    {
+      "epoch": 5.33,
+      "eval_accuracy": 0.932625,
+      "eval_loss": 0.20015648007392883,
+      "eval_runtime": 16.3054,
+      "eval_samples_per_second": 1471.905,
+      "eval_steps_per_second": 4.6,
+      "step": 12780
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1571,
+      "step": 12800
+    },
+    {
+      "epoch": 5.33,
+      "eval_accuracy": 0.932,
+      "eval_loss": 0.20201164484024048,
+      "eval_runtime": 16.2407,
+      "eval_samples_per_second": 1477.773,
+      "eval_steps_per_second": 4.618,
+      "step": 12800
+    },
+    {
+      "epoch": 5.34,
+      "learning_rate": 1e-06,
+      "loss": 0.2073,
+      "step": 12820
+    },
+    {
+      "epoch": 5.34,
+      "eval_accuracy": 0.9295416666666667,
+      "eval_loss": 0.2069837599992752,
+      "eval_runtime": 17.0384,
+      "eval_samples_per_second": 1408.581,
+      "eval_steps_per_second": 4.402,
+      "step": 12820
+    },
+    {
+      "epoch": 5.35,
+      "learning_rate": 1e-06,
+      "loss": 0.1605,
+      "step": 12840
+    },
+    {
+      "epoch": 5.35,
+      "eval_accuracy": 0.93075,
+      "eval_loss": 0.20616813004016876,
+      "eval_runtime": 16.8287,
+      "eval_samples_per_second": 1426.138,
+      "eval_steps_per_second": 4.457,
+      "step": 12840
+    },
+    {
+      "epoch": 5.36,
+      "learning_rate": 1e-06,
+      "loss": 0.1879,
+      "step": 12860
+    },
+    {
+      "epoch": 5.36,
+      "eval_accuracy": 0.9306666666666666,
+      "eval_loss": 0.20293764770030975,
+      "eval_runtime": 17.5436,
+      "eval_samples_per_second": 1368.02,
+      "eval_steps_per_second": 4.275,
+      "step": 12860
+    },
+    {
+      "epoch": 5.37,
+      "learning_rate": 1e-06,
+      "loss": 0.2104,
+      "step": 12880
+    },
+    {
+      "epoch": 5.37,
+      "eval_accuracy": 0.9311666666666667,
+      "eval_loss": 0.20171169936656952,
+      "eval_runtime": 16.6339,
+      "eval_samples_per_second": 1442.838,
+      "eval_steps_per_second": 4.509,
+      "step": 12880
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 1e-06,
+      "loss": 0.2058,
+      "step": 12900
+    },
+    {
+      "epoch": 5.38,
+      "eval_accuracy": 0.9315416666666667,
+      "eval_loss": 0.20113909244537354,
+      "eval_runtime": 16.0022,
+      "eval_samples_per_second": 1499.798,
+      "eval_steps_per_second": 4.687,
+      "step": 12900
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1624,
+      "step": 12920
+    },
+    {
+      "epoch": 5.38,
+      "eval_accuracy": 0.9311666666666667,
+      "eval_loss": 0.20383241772651672,
+      "eval_runtime": 15.8073,
+      "eval_samples_per_second": 1518.289,
+      "eval_steps_per_second": 4.745,
+      "step": 12920
+    },
+    {
+      "epoch": 5.39,
+      "learning_rate": 1e-06,
+      "loss": 0.1765,
+      "step": 12940
+    },
+    {
+      "epoch": 5.39,
+      "eval_accuracy": 0.9310416666666667,
+      "eval_loss": 0.20215220749378204,
+      "eval_runtime": 16.3386,
+      "eval_samples_per_second": 1468.917,
+      "eval_steps_per_second": 4.59,
+      "step": 12940
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 1e-06,
+      "loss": 0.187,
+      "step": 12960
+    },
+    {
+      "epoch": 5.4,
+      "eval_accuracy": 0.931375,
+      "eval_loss": 0.20157550275325775,
+      "eval_runtime": 16.0967,
+      "eval_samples_per_second": 1490.991,
+      "eval_steps_per_second": 4.659,
+      "step": 12960
+    },
+    {
+      "epoch": 5.41,
+      "learning_rate": 1e-06,
+      "loss": 0.1565,
+      "step": 12980
+    },
+    {
+      "epoch": 5.41,
+      "eval_accuracy": 0.9317083333333334,
+      "eval_loss": 0.20059089362621307,
+      "eval_runtime": 15.7591,
+      "eval_samples_per_second": 1522.928,
+      "eval_steps_per_second": 4.759,
+      "step": 12980
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 1e-06,
+      "loss": 0.1901,
+      "step": 13000
+    },
+    {
+      "epoch": 5.42,
+      "eval_accuracy": 0.9322916666666666,
+      "eval_loss": 0.20068366825580597,
+      "eval_runtime": 15.8354,
+      "eval_samples_per_second": 1515.591,
+      "eval_steps_per_second": 4.736,
+      "step": 13000
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2473,
+      "step": 13020
+    },
+    {
+      "epoch": 5.42,
+      "eval_accuracy": 0.9309583333333333,
+      "eval_loss": 0.20135918259620667,
+      "eval_runtime": 15.7908,
+      "eval_samples_per_second": 1519.875,
+      "eval_steps_per_second": 4.75,
+      "step": 13020
+    },
+    {
+      "epoch": 5.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1836,
+      "step": 13040
+    },
+    {
+      "epoch": 5.43,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.20122550427913666,
+      "eval_runtime": 17.2214,
+      "eval_samples_per_second": 1393.613,
+      "eval_steps_per_second": 4.355,
+      "step": 13040
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 1e-06,
+      "loss": 0.1759,
+      "step": 13060
+    },
+    {
+      "epoch": 5.44,
+      "eval_accuracy": 0.9310416666666667,
+      "eval_loss": 0.20236973464488983,
+      "eval_runtime": 17.2846,
+      "eval_samples_per_second": 1388.518,
+      "eval_steps_per_second": 4.339,
+      "step": 13060
+    },
+    {
+      "epoch": 5.45,
+      "learning_rate": 1e-06,
+      "loss": 0.2385,
+      "step": 13080
+    },
+    {
+      "epoch": 5.45,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.202514186501503,
+      "eval_runtime": 17.5809,
+      "eval_samples_per_second": 1365.119,
+      "eval_steps_per_second": 4.266,
+      "step": 13080
+    },
+    {
+      "epoch": 5.46,
+      "learning_rate": 1e-06,
+      "loss": 0.1586,
+      "step": 13100
+    },
+    {
+      "epoch": 5.46,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.20214223861694336,
+      "eval_runtime": 17.3368,
+      "eval_samples_per_second": 1384.339,
+      "eval_steps_per_second": 4.326,
+      "step": 13100
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1734,
+      "step": 13120
+    },
+    {
+      "epoch": 5.47,
+      "eval_accuracy": 0.9310833333333334,
+      "eval_loss": 0.20263217389583588,
+      "eval_runtime": 15.8849,
+      "eval_samples_per_second": 1510.867,
+      "eval_steps_per_second": 4.721,
+      "step": 13120
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1888,
+      "step": 13140
+    },
+    {
+      "epoch": 5.47,
+      "eval_accuracy": 0.9295833333333333,
+      "eval_loss": 0.20405107736587524,
+      "eval_runtime": 15.9922,
+      "eval_samples_per_second": 1500.728,
+      "eval_steps_per_second": 4.69,
+      "step": 13140
+    },
+    {
+      "epoch": 5.48,
+      "learning_rate": 1e-06,
+      "loss": 0.1676,
+      "step": 13160
+    },
+    {
+      "epoch": 5.48,
+      "eval_accuracy": 0.9306666666666666,
+      "eval_loss": 0.20299804210662842,
+      "eval_runtime": 15.9348,
+      "eval_samples_per_second": 1506.14,
+      "eval_steps_per_second": 4.707,
+      "step": 13160
+    },
+    {
+      "epoch": 5.49,
+      "learning_rate": 1e-06,
+      "loss": 0.2462,
+      "step": 13180
+    },
+    {
+      "epoch": 5.49,
+      "eval_accuracy": 0.9300833333333334,
+      "eval_loss": 0.20278708636760712,
+      "eval_runtime": 16.0737,
+      "eval_samples_per_second": 1493.119,
+      "eval_steps_per_second": 4.666,
+      "step": 13180
+    },
+    {
+      "epoch": 5.5,
+      "learning_rate": 1e-06,
+      "loss": 0.1751,
+      "step": 13200
+    },
+    {
+      "epoch": 5.5,
+      "eval_accuracy": 0.93125,
+      "eval_loss": 0.20159900188446045,
+      "eval_runtime": 15.4631,
+      "eval_samples_per_second": 1552.08,
+      "eval_steps_per_second": 4.85,
+      "step": 13200
+    },
+    {
+      "epoch": 5.51,
+      "learning_rate": 1e-06,
+      "loss": 0.2113,
+      "step": 13220
+    },
+    {
+      "epoch": 5.51,
+      "eval_accuracy": 0.929125,
+      "eval_loss": 0.20587190985679626,
+      "eval_runtime": 15.9217,
+      "eval_samples_per_second": 1507.375,
+      "eval_steps_per_second": 4.711,
+      "step": 13220
+    },
+    {
+      "epoch": 5.52,
+      "learning_rate": 1e-06,
+      "loss": 0.1884,
+      "step": 13240
+    },
+    {
+      "epoch": 5.52,
+      "eval_accuracy": 0.92875,
+      "eval_loss": 0.20635303854942322,
+      "eval_runtime": 15.8747,
+      "eval_samples_per_second": 1511.836,
+      "eval_steps_per_second": 4.724,
+      "step": 13240
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1545,
+      "step": 13260
+    },
+    {
+      "epoch": 5.53,
+      "eval_accuracy": 0.9289166666666666,
+      "eval_loss": 0.20559687912464142,
+      "eval_runtime": 15.9786,
+      "eval_samples_per_second": 1502.007,
+      "eval_steps_per_second": 4.694,
+      "step": 13260
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 1e-06,
+      "loss": 0.209,
+      "step": 13280
+    },
+    {
+      "epoch": 5.53,
+      "eval_accuracy": 0.9307916666666667,
+      "eval_loss": 0.20248527824878693,
+      "eval_runtime": 15.7154,
+      "eval_samples_per_second": 1527.166,
+      "eval_steps_per_second": 4.772,
+      "step": 13280
+    },
+    {
+      "epoch": 5.54,
+      "learning_rate": 1e-06,
+      "loss": 0.1843,
+      "step": 13300
+    },
+    {
+      "epoch": 5.54,
+      "eval_accuracy": 0.9307916666666667,
+      "eval_loss": 0.20135876536369324,
+      "eval_runtime": 16.3985,
+      "eval_samples_per_second": 1463.552,
+      "eval_steps_per_second": 4.574,
+      "step": 13300
+    },
+    {
+      "epoch": 5.55,
+      "learning_rate": 1e-06,
+      "loss": 0.2096,
+      "step": 13320
+    },
+    {
+      "epoch": 5.55,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.20034638047218323,
+      "eval_runtime": 15.8661,
+      "eval_samples_per_second": 1512.659,
+      "eval_steps_per_second": 4.727,
+      "step": 13320
+    },
+    {
+      "epoch": 5.56,
+      "learning_rate": 1e-06,
+      "loss": 0.1821,
+      "step": 13340
+    },
+    {
+      "epoch": 5.56,
+      "eval_accuracy": 0.9300416666666667,
+      "eval_loss": 0.20381678640842438,
+      "eval_runtime": 16.2244,
+      "eval_samples_per_second": 1479.257,
+      "eval_steps_per_second": 4.623,
+      "step": 13340
+    },
+    {
+      "epoch": 5.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1898,
+      "step": 13360
+    },
+    {
+      "epoch": 5.57,
+      "eval_accuracy": 0.9309166666666666,
+      "eval_loss": 0.20166015625,
+      "eval_runtime": 15.944,
+      "eval_samples_per_second": 1505.264,
+      "eval_steps_per_second": 4.704,
+      "step": 13360
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 1e-06,
+      "loss": 0.2068,
+      "step": 13380
+    },
+    {
+      "epoch": 5.58,
+      "eval_accuracy": 0.9309583333333333,
+      "eval_loss": 0.2012653946876526,
+      "eval_runtime": 16.2049,
+      "eval_samples_per_second": 1481.035,
+      "eval_steps_per_second": 4.628,
+      "step": 13380
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 1e-06,
+      "loss": 0.1822,
+      "step": 13400
+    },
+    {
+      "epoch": 5.58,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.20043647289276123,
+      "eval_runtime": 15.8256,
+      "eval_samples_per_second": 1516.532,
+      "eval_steps_per_second": 4.739,
+      "step": 13400
+    },
+    {
+      "epoch": 5.59,
+      "learning_rate": 1e-06,
+      "loss": 0.1974,
+      "step": 13420
+    },
+    {
+      "epoch": 5.59,
+      "eval_accuracy": 0.9320416666666667,
+      "eval_loss": 0.19964149594306946,
+      "eval_runtime": 15.6329,
+      "eval_samples_per_second": 1535.22,
+      "eval_steps_per_second": 4.798,
+      "step": 13420
+    },
+    {
+      "epoch": 5.6,
+      "learning_rate": 1e-06,
+      "loss": 0.1634,
+      "step": 13440
+    },
+    {
+      "epoch": 5.6,
+      "eval_accuracy": 0.93,
+      "eval_loss": 0.2027622014284134,
+      "eval_runtime": 15.6182,
+      "eval_samples_per_second": 1536.672,
+      "eval_steps_per_second": 4.802,
+      "step": 13440
+    },
+    {
+      "epoch": 5.61,
+      "learning_rate": 1e-06,
+      "loss": 0.2075,
+      "step": 13460
+    },
+    {
+      "epoch": 5.61,
+      "eval_accuracy": 0.92925,
+      "eval_loss": 0.2044484168291092,
+      "eval_runtime": 15.6933,
+      "eval_samples_per_second": 1529.315,
+      "eval_steps_per_second": 4.779,
+      "step": 13460
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1775,
+      "step": 13480
+    },
+    {
+      "epoch": 5.62,
+      "eval_accuracy": 0.9289583333333333,
+      "eval_loss": 0.20532788336277008,
+      "eval_runtime": 15.7257,
+      "eval_samples_per_second": 1526.161,
+      "eval_steps_per_second": 4.769,
+      "step": 13480
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1648,
+      "step": 13500
+    },
+    {
+      "epoch": 5.62,
+      "eval_accuracy": 0.9272083333333333,
+      "eval_loss": 0.2097957581281662,
+      "eval_runtime": 15.4877,
+      "eval_samples_per_second": 1549.617,
+      "eval_steps_per_second": 4.843,
+      "step": 13500
+    },
+    {
+      "epoch": 5.63,
+      "learning_rate": 1e-06,
+      "loss": 0.1702,
+      "step": 13520
+    },
+    {
+      "epoch": 5.63,
+      "eval_accuracy": 0.9284166666666667,
+      "eval_loss": 0.2071061134338379,
+      "eval_runtime": 15.7223,
+      "eval_samples_per_second": 1526.495,
+      "eval_steps_per_second": 4.77,
+      "step": 13520
+    },
+    {
+      "epoch": 5.64,
+      "learning_rate": 1e-06,
+      "loss": 0.1701,
+      "step": 13540
+    },
+    {
+      "epoch": 5.64,
+      "eval_accuracy": 0.9302083333333333,
+      "eval_loss": 0.205330029129982,
+      "eval_runtime": 16.1803,
+      "eval_samples_per_second": 1483.284,
+      "eval_steps_per_second": 4.635,
+      "step": 13540
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 1e-06,
+      "loss": 0.2158,
+      "step": 13560
+    },
+    {
+      "epoch": 5.65,
+      "eval_accuracy": 0.929375,
+      "eval_loss": 0.20697665214538574,
+      "eval_runtime": 15.6178,
+      "eval_samples_per_second": 1536.706,
+      "eval_steps_per_second": 4.802,
+      "step": 13560
+    },
+    {
+      "epoch": 5.66,
+      "learning_rate": 1e-06,
+      "loss": 0.1622,
+      "step": 13580
+    },
+    {
+      "epoch": 5.66,
+      "eval_accuracy": 0.9305,
+      "eval_loss": 0.2030411958694458,
+      "eval_runtime": 15.9497,
+      "eval_samples_per_second": 1504.728,
+      "eval_steps_per_second": 4.702,
+      "step": 13580
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1764,
+      "step": 13600
+    },
+    {
+      "epoch": 5.67,
+      "eval_accuracy": 0.9311666666666667,
+      "eval_loss": 0.2010955512523651,
+      "eval_runtime": 15.6626,
+      "eval_samples_per_second": 1532.316,
+      "eval_steps_per_second": 4.788,
+      "step": 13600
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 1e-06,
+      "loss": 0.2191,
+      "step": 13620
+    },
+    {
+      "epoch": 5.67,
+      "eval_accuracy": 0.93125,
+      "eval_loss": 0.1993076354265213,
+      "eval_runtime": 16.3922,
+      "eval_samples_per_second": 1464.11,
+      "eval_steps_per_second": 4.575,
+      "step": 13620
+    },
+    {
+      "epoch": 5.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1962,
+      "step": 13640
+    },
+    {
+      "epoch": 5.68,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.19903969764709473,
+      "eval_runtime": 16.0762,
+      "eval_samples_per_second": 1492.89,
+      "eval_steps_per_second": 4.665,
+      "step": 13640
+    },
+    {
+      "epoch": 5.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1954,
+      "step": 13660
+    },
+    {
+      "epoch": 5.69,
+      "eval_accuracy": 0.9315833333333333,
+      "eval_loss": 0.20040030777454376,
+      "eval_runtime": 16.0741,
+      "eval_samples_per_second": 1493.088,
+      "eval_steps_per_second": 4.666,
+      "step": 13660
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1903,
+      "step": 13680
+    },
+    {
+      "epoch": 5.7,
+      "eval_accuracy": 0.9295,
+      "eval_loss": 0.20218642055988312,
+      "eval_runtime": 17.6769,
+      "eval_samples_per_second": 1357.707,
+      "eval_steps_per_second": 4.243,
+      "step": 13680
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1867,
+      "step": 13700
+    },
+    {
+      "epoch": 5.71,
+      "eval_accuracy": 0.930375,
+      "eval_loss": 0.20176221430301666,
+      "eval_runtime": 16.5988,
+      "eval_samples_per_second": 1445.891,
+      "eval_steps_per_second": 4.518,
+      "step": 13700
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1991,
+      "step": 13720
+    },
+    {
+      "epoch": 5.72,
+      "eval_accuracy": 0.93025,
+      "eval_loss": 0.20237615704536438,
+      "eval_runtime": 16.9965,
+      "eval_samples_per_second": 1412.055,
+      "eval_steps_per_second": 4.413,
+      "step": 13720
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1646,
+      "step": 13740
+    },
+    {
+      "epoch": 5.72,
+      "eval_accuracy": 0.9304583333333334,
+      "eval_loss": 0.20343907177448273,
+      "eval_runtime": 17.7322,
+      "eval_samples_per_second": 1353.468,
+      "eval_steps_per_second": 4.23,
+      "step": 13740
+    },
+    {
+      "epoch": 5.73,
+      "learning_rate": 1e-06,
+      "loss": 0.2051,
+      "step": 13760
+    },
+    {
+      "epoch": 5.73,
+      "eval_accuracy": 0.9300416666666667,
+      "eval_loss": 0.2030051052570343,
+      "eval_runtime": 17.541,
+      "eval_samples_per_second": 1368.227,
+      "eval_steps_per_second": 4.276,
+      "step": 13760
+    },
+    {
+      "epoch": 5.74,
+      "learning_rate": 1e-06,
+      "loss": 0.1693,
+      "step": 13780
+    },
+    {
+      "epoch": 5.74,
+      "eval_accuracy": 0.931,
+      "eval_loss": 0.1998627781867981,
+      "eval_runtime": 16.6335,
+      "eval_samples_per_second": 1442.869,
+      "eval_steps_per_second": 4.509,
+      "step": 13780
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 1e-06,
+      "loss": 0.1663,
+      "step": 13800
+    },
+    {
+      "epoch": 5.75,
+      "eval_accuracy": 0.9325,
+      "eval_loss": 0.1989893913269043,
+      "eval_runtime": 17.515,
+      "eval_samples_per_second": 1370.253,
+      "eval_steps_per_second": 4.282,
+      "step": 13800
+    },
+    {
+      "epoch": 5.76,
+      "learning_rate": 1e-06,
+      "loss": 0.1452,
+      "step": 13820
+    },
+    {
+      "epoch": 5.76,
+      "eval_accuracy": 0.9319583333333333,
+      "eval_loss": 0.1993263214826584,
+      "eval_runtime": 17.7237,
+      "eval_samples_per_second": 1354.116,
+      "eval_steps_per_second": 4.232,
+      "step": 13820
+    },
+    {
+      "epoch": 5.77,
+      "learning_rate": 1e-06,
+      "loss": 0.1535,
+      "step": 13840
+    },
+    {
+      "epoch": 5.77,
+      "eval_accuracy": 0.932125,
+      "eval_loss": 0.19914114475250244,
+      "eval_runtime": 17.0395,
+      "eval_samples_per_second": 1408.496,
+      "eval_steps_per_second": 4.402,
+      "step": 13840
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2072,
+      "step": 13860
+    },
+    {
+      "epoch": 5.78,
+      "eval_accuracy": 0.9300833333333334,
+      "eval_loss": 0.20136070251464844,
+      "eval_runtime": 17.6045,
+      "eval_samples_per_second": 1363.289,
+      "eval_steps_per_second": 4.26,
+      "step": 13860
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2069,
+      "step": 13880
+    },
+    {
+      "epoch": 5.78,
+      "eval_accuracy": 0.9298333333333333,
+      "eval_loss": 0.2024083137512207,
+      "eval_runtime": 17.3144,
+      "eval_samples_per_second": 1386.128,
+      "eval_steps_per_second": 4.332,
+      "step": 13880
+    },
+    {
+      "epoch": 5.79,
+      "learning_rate": 1e-06,
+      "loss": 0.2298,
+      "step": 13900
+    },
+    {
+      "epoch": 5.79,
+      "eval_accuracy": 0.93225,
+      "eval_loss": 0.19839055836200714,
+      "eval_runtime": 18.3284,
+      "eval_samples_per_second": 1309.44,
+      "eval_steps_per_second": 4.092,
+      "step": 13900
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 1e-06,
+      "loss": 0.201,
+      "step": 13920
+    },
+    {
+      "epoch": 5.8,
+      "eval_accuracy": 0.932125,
+      "eval_loss": 0.1982399821281433,
+      "eval_runtime": 18.1938,
+      "eval_samples_per_second": 1319.131,
+      "eval_steps_per_second": 4.122,
+      "step": 13920
+    },
+    {
+      "epoch": 5.81,
+      "learning_rate": 1e-06,
+      "loss": 0.184,
+      "step": 13940
+    },
+    {
+      "epoch": 5.81,
+      "eval_accuracy": 0.9317083333333334,
+      "eval_loss": 0.1989286094903946,
+      "eval_runtime": 18.406,
+      "eval_samples_per_second": 1303.923,
+      "eval_steps_per_second": 4.075,
+      "step": 13940
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 1e-06,
+      "loss": 0.1942,
+      "step": 13960
+    },
+    {
+      "epoch": 5.82,
+      "eval_accuracy": 0.9319166666666666,
+      "eval_loss": 0.19913487136363983,
+      "eval_runtime": 18.2526,
+      "eval_samples_per_second": 1314.879,
+      "eval_steps_per_second": 4.109,
+      "step": 13960
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 1e-06,
+      "loss": 0.2085,
+      "step": 13980
+    },
+    {
+      "epoch": 5.83,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.19867363572120667,
+      "eval_runtime": 17.554,
+      "eval_samples_per_second": 1367.21,
+      "eval_steps_per_second": 4.273,
+      "step": 13980
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 1e-06,
+      "loss": 0.177,
+      "step": 14000
+    },
+    {
+      "epoch": 5.83,
+      "eval_accuracy": 0.9326666666666666,
+      "eval_loss": 0.19808466732501984,
+      "eval_runtime": 16.6018,
+      "eval_samples_per_second": 1445.627,
+      "eval_steps_per_second": 4.518,
+      "step": 14000
+    },
+    {
+      "epoch": 5.84,
+      "learning_rate": 1e-06,
+      "loss": 0.1746,
+      "step": 14020
+    },
+    {
+      "epoch": 5.84,
+      "eval_accuracy": 0.9325416666666667,
+      "eval_loss": 0.19782015681266785,
+      "eval_runtime": 17.9546,
+      "eval_samples_per_second": 1336.706,
+      "eval_steps_per_second": 4.177,
+      "step": 14020
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 1e-06,
+      "loss": 0.1731,
+      "step": 14040
+    },
+    {
+      "epoch": 5.85,
+      "eval_accuracy": 0.9310416666666667,
+      "eval_loss": 0.20012834668159485,
+      "eval_runtime": 17.9142,
+      "eval_samples_per_second": 1339.719,
+      "eval_steps_per_second": 4.187,
+      "step": 14040
+    },
+    {
+      "epoch": 5.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1888,
+      "step": 14060
+    },
+    {
+      "epoch": 5.86,
+      "eval_accuracy": 0.93075,
+      "eval_loss": 0.20140434801578522,
+      "eval_runtime": 18.4741,
+      "eval_samples_per_second": 1299.116,
+      "eval_steps_per_second": 4.06,
+      "step": 14060
+    },
+    {
+      "epoch": 5.87,
+      "learning_rate": 1e-06,
+      "loss": 0.1708,
+      "step": 14080
+    },
+    {
+      "epoch": 5.87,
+      "eval_accuracy": 0.9315416666666667,
+      "eval_loss": 0.19907809793949127,
+      "eval_runtime": 18.0588,
+      "eval_samples_per_second": 1328.991,
+      "eval_steps_per_second": 4.153,
+      "step": 14080
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1927,
+      "step": 14100
+    },
+    {
+      "epoch": 5.88,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.2014717161655426,
+      "eval_runtime": 17.5458,
+      "eval_samples_per_second": 1367.846,
+      "eval_steps_per_second": 4.275,
+      "step": 14100
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1756,
+      "step": 14120
+    },
+    {
+      "epoch": 5.88,
+      "eval_accuracy": 0.9304583333333334,
+      "eval_loss": 0.20116497576236725,
+      "eval_runtime": 16.0925,
+      "eval_samples_per_second": 1491.379,
+      "eval_steps_per_second": 4.661,
+      "step": 14120
+    },
+    {
+      "epoch": 5.89,
+      "learning_rate": 1e-06,
+      "loss": 0.1829,
+      "step": 14140
+    },
+    {
+      "epoch": 5.89,
+      "eval_accuracy": 0.9317083333333334,
+      "eval_loss": 0.2000524401664734,
+      "eval_runtime": 16.8659,
+      "eval_samples_per_second": 1422.991,
+      "eval_steps_per_second": 4.447,
+      "step": 14140
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 1e-06,
+      "loss": 0.1348,
+      "step": 14160
+    },
+    {
+      "epoch": 5.9,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.19840501248836517,
+      "eval_runtime": 16.5412,
+      "eval_samples_per_second": 1450.921,
+      "eval_steps_per_second": 4.534,
+      "step": 14160
+    },
+    {
+      "epoch": 5.91,
+      "learning_rate": 1e-06,
+      "loss": 0.1773,
+      "step": 14180
+    },
+    {
+      "epoch": 5.91,
+      "eval_accuracy": 0.9315416666666667,
+      "eval_loss": 0.20062309503555298,
+      "eval_runtime": 16.2967,
+      "eval_samples_per_second": 1472.691,
+      "eval_steps_per_second": 4.602,
+      "step": 14180
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1639,
+      "step": 14200
+    },
+    {
+      "epoch": 5.92,
+      "eval_accuracy": 0.9306666666666666,
+      "eval_loss": 0.2011023461818695,
+      "eval_runtime": 16.2819,
+      "eval_samples_per_second": 1474.03,
+      "eval_steps_per_second": 4.606,
+      "step": 14200
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 1e-06,
+      "loss": 0.2275,
+      "step": 14220
+    },
+    {
+      "epoch": 5.92,
+      "eval_accuracy": 0.9318333333333333,
+      "eval_loss": 0.19922685623168945,
+      "eval_runtime": 17.1405,
+      "eval_samples_per_second": 1400.191,
+      "eval_steps_per_second": 4.376,
+      "step": 14220
+    },
+    {
+      "epoch": 5.93,
+      "learning_rate": 1e-06,
+      "loss": 0.1934,
+      "step": 14240
+    },
+    {
+      "epoch": 5.93,
+      "eval_accuracy": 0.932125,
+      "eval_loss": 0.19886505603790283,
+      "eval_runtime": 16.0782,
+      "eval_samples_per_second": 1492.703,
+      "eval_steps_per_second": 4.665,
+      "step": 14240
+    },
+    {
+      "epoch": 5.94,
+      "learning_rate": 1e-06,
+      "loss": 0.1847,
+      "step": 14260
+    },
+    {
+      "epoch": 5.94,
+      "eval_accuracy": 0.9330833333333334,
+      "eval_loss": 0.19734573364257812,
+      "eval_runtime": 17.2888,
+      "eval_samples_per_second": 1388.183,
+      "eval_steps_per_second": 4.338,
+      "step": 14260
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 1e-06,
+      "loss": 0.1572,
+      "step": 14280
+    },
+    {
+      "epoch": 5.95,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.1978907287120819,
+      "eval_runtime": 19.318,
+      "eval_samples_per_second": 1242.362,
+      "eval_steps_per_second": 3.882,
+      "step": 14280
+    },
+    {
+      "epoch": 5.96,
+      "learning_rate": 1e-06,
+      "loss": 0.2191,
+      "step": 14300
+    },
+    {
+      "epoch": 5.96,
+      "eval_accuracy": 0.9322083333333333,
+      "eval_loss": 0.19908125698566437,
+      "eval_runtime": 17.7053,
+      "eval_samples_per_second": 1355.53,
+      "eval_steps_per_second": 4.236,
+      "step": 14300
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 1e-06,
+      "loss": 0.2022,
+      "step": 14320
+    },
+    {
+      "epoch": 5.97,
+      "eval_accuracy": 0.9321666666666667,
+      "eval_loss": 0.19820664823055267,
+      "eval_runtime": 17.7778,
+      "eval_samples_per_second": 1350.001,
+      "eval_steps_per_second": 4.219,
+      "step": 14320
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1575,
+      "step": 14340
+    },
+    {
+      "epoch": 5.97,
+      "eval_accuracy": 0.9321666666666667,
+      "eval_loss": 0.19856464862823486,
+      "eval_runtime": 17.5559,
+      "eval_samples_per_second": 1367.059,
+      "eval_steps_per_second": 4.272,
+      "step": 14340
+    },
+    {
+      "epoch": 5.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1826,
+      "step": 14360
+    },
+    {
+      "epoch": 5.98,
+      "eval_accuracy": 0.9322916666666666,
+      "eval_loss": 0.19818070530891418,
+      "eval_runtime": 17.7144,
+      "eval_samples_per_second": 1354.827,
+      "eval_steps_per_second": 4.234,
+      "step": 14360
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1855,
+      "step": 14380
+    },
+    {
+      "epoch": 5.99,
+      "eval_accuracy": 0.9326666666666666,
+      "eval_loss": 0.19788843393325806,
+      "eval_runtime": 17.5812,
+      "eval_samples_per_second": 1365.094,
+      "eval_steps_per_second": 4.266,
+      "step": 14380
+    },
+    {
+      "epoch": 6.0,
+      "learning_rate": 1e-06,
+      "loss": 0.2027,
+      "step": 14400
+    },
+    {
+      "epoch": 6.0,
+      "eval_accuracy": 0.9319583333333333,
+      "eval_loss": 0.19935990869998932,
+      "eval_runtime": 17.5236,
+      "eval_samples_per_second": 1369.581,
+      "eval_steps_per_second": 4.28,
+      "step": 14400
+    },
+    {
+      "epoch": 6.01,
+      "learning_rate": 1e-06,
+      "loss": 0.1585,
+      "step": 14420
+    },
+    {
+      "epoch": 6.01,
+      "eval_accuracy": 0.9324583333333333,
+      "eval_loss": 0.20092272758483887,
+      "eval_runtime": 17.5852,
+      "eval_samples_per_second": 1364.781,
+      "eval_steps_per_second": 4.265,
+      "step": 14420
+    },
+    {
+      "epoch": 6.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2092,
+      "step": 14440
+    },
+    {
+      "epoch": 6.02,
+      "eval_accuracy": 0.9339583333333333,
+      "eval_loss": 0.1970815807580948,
+      "eval_runtime": 17.1214,
+      "eval_samples_per_second": 1401.757,
+      "eval_steps_per_second": 4.38,
+      "step": 14440
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1773,
+      "step": 14460
+    },
+    {
+      "epoch": 6.03,
+      "eval_accuracy": 0.9334583333333333,
+      "eval_loss": 0.19676099717617035,
+      "eval_runtime": 16.7515,
+      "eval_samples_per_second": 1432.707,
+      "eval_steps_per_second": 4.477,
+      "step": 14460
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 1e-06,
+      "loss": 0.186,
+      "step": 14480
+    },
+    {
+      "epoch": 6.03,
+      "eval_accuracy": 0.9323333333333333,
+      "eval_loss": 0.19818897545337677,
+      "eval_runtime": 18.0357,
+      "eval_samples_per_second": 1330.693,
+      "eval_steps_per_second": 4.158,
+      "step": 14480
+    },
+    {
+      "epoch": 6.04,
+      "learning_rate": 1e-06,
+      "loss": 0.1759,
+      "step": 14500
+    },
+    {
+      "epoch": 6.04,
+      "eval_accuracy": 0.9315,
+      "eval_loss": 0.19834274053573608,
+      "eval_runtime": 16.9189,
+      "eval_samples_per_second": 1418.535,
+      "eval_steps_per_second": 4.433,
+      "step": 14500
+    },
+    {
+      "epoch": 6.05,
+      "learning_rate": 1e-06,
+      "loss": 0.168,
+      "step": 14520
+    },
+    {
+      "epoch": 6.05,
+      "eval_accuracy": 0.9335416666666667,
+      "eval_loss": 0.19659017026424408,
+      "eval_runtime": 16.8237,
+      "eval_samples_per_second": 1426.555,
+      "eval_steps_per_second": 4.458,
+      "step": 14520
+    },
+    {
+      "epoch": 6.06,
+      "learning_rate": 1e-06,
+      "loss": 0.168,
+      "step": 14540
+    },
+    {
+      "epoch": 6.06,
+      "eval_accuracy": 0.9322083333333333,
+      "eval_loss": 0.1991969645023346,
+      "eval_runtime": 17.2701,
+      "eval_samples_per_second": 1389.683,
+      "eval_steps_per_second": 4.343,
+      "step": 14540
+    },
+    {
+      "epoch": 6.07,
+      "learning_rate": 1e-06,
+      "loss": 0.2085,
+      "step": 14560
+    },
+    {
+      "epoch": 6.07,
+      "eval_accuracy": 0.9324583333333333,
+      "eval_loss": 0.19824150204658508,
+      "eval_runtime": 17.4283,
+      "eval_samples_per_second": 1377.067,
+      "eval_steps_per_second": 4.303,
+      "step": 14560
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1866,
+      "step": 14580
+    },
+    {
+      "epoch": 6.08,
+      "eval_accuracy": 0.93325,
+      "eval_loss": 0.19639739394187927,
+      "eval_runtime": 17.7717,
+      "eval_samples_per_second": 1350.465,
+      "eval_steps_per_second": 4.22,
+      "step": 14580
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1772,
+      "step": 14600
+    },
+    {
+      "epoch": 6.08,
+      "eval_accuracy": 0.93325,
+      "eval_loss": 0.19745591282844543,
+      "eval_runtime": 16.2574,
+      "eval_samples_per_second": 1476.246,
+      "eval_steps_per_second": 4.613,
+      "step": 14600
+    },
+    {
+      "epoch": 6.09,
+      "learning_rate": 1e-06,
+      "loss": 0.1772,
+      "step": 14620
+    },
+    {
+      "epoch": 6.09,
+      "eval_accuracy": 0.933375,
+      "eval_loss": 0.1968727856874466,
+      "eval_runtime": 16.8404,
+      "eval_samples_per_second": 1425.142,
+      "eval_steps_per_second": 4.454,
+      "step": 14620
+    },
+    {
+      "epoch": 6.1,
+      "learning_rate": 1e-06,
+      "loss": 0.1442,
+      "step": 14640
+    },
+    {
+      "epoch": 6.1,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.19765588641166687,
+      "eval_runtime": 16.5146,
+      "eval_samples_per_second": 1453.262,
+      "eval_steps_per_second": 4.541,
+      "step": 14640
+    },
+    {
+      "epoch": 6.11,
+      "learning_rate": 1e-06,
+      "loss": 0.1531,
+      "step": 14660
+    },
+    {
+      "epoch": 6.11,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.19997546076774597,
+      "eval_runtime": 16.1908,
+      "eval_samples_per_second": 1482.324,
+      "eval_steps_per_second": 4.632,
+      "step": 14660
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2187,
+      "step": 14680
+    },
+    {
+      "epoch": 6.12,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.20033469796180725,
+      "eval_runtime": 16.76,
+      "eval_samples_per_second": 1431.985,
+      "eval_steps_per_second": 4.475,
+      "step": 14680
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1788,
+      "step": 14700
+    },
+    {
+      "epoch": 6.12,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.2029111534357071,
+      "eval_runtime": 16.3811,
+      "eval_samples_per_second": 1465.1,
+      "eval_steps_per_second": 4.578,
+      "step": 14700
+    },
+    {
+      "epoch": 6.13,
+      "learning_rate": 1e-06,
+      "loss": 0.2018,
+      "step": 14720
+    },
+    {
+      "epoch": 6.13,
+      "eval_accuracy": 0.9309583333333333,
+      "eval_loss": 0.20074012875556946,
+      "eval_runtime": 16.4407,
+      "eval_samples_per_second": 1459.794,
+      "eval_steps_per_second": 4.562,
+      "step": 14720
+    },
+    {
+      "epoch": 6.14,
+      "learning_rate": 1e-06,
+      "loss": 0.1636,
+      "step": 14740
+    },
+    {
+      "epoch": 6.14,
+      "eval_accuracy": 0.9325833333333333,
+      "eval_loss": 0.1991911381483078,
+      "eval_runtime": 16.8493,
+      "eval_samples_per_second": 1424.389,
+      "eval_steps_per_second": 4.451,
+      "step": 14740
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 1e-06,
+      "loss": 0.1622,
+      "step": 14760
+    },
+    {
+      "epoch": 6.15,
+      "eval_accuracy": 0.9322916666666666,
+      "eval_loss": 0.19850997626781464,
+      "eval_runtime": 16.3552,
+      "eval_samples_per_second": 1467.424,
+      "eval_steps_per_second": 4.586,
+      "step": 14760
+    },
+    {
+      "epoch": 6.16,
+      "learning_rate": 1e-06,
+      "loss": 0.1947,
+      "step": 14780
+    },
+    {
+      "epoch": 6.16,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.19700346887111664,
+      "eval_runtime": 16.8167,
+      "eval_samples_per_second": 1427.151,
+      "eval_steps_per_second": 4.46,
+      "step": 14780
+    },
+    {
+      "epoch": 6.17,
+      "learning_rate": 1e-06,
+      "loss": 0.198,
+      "step": 14800
+    },
+    {
+      "epoch": 6.17,
+      "eval_accuracy": 0.933875,
+      "eval_loss": 0.19635510444641113,
+      "eval_runtime": 17.4336,
+      "eval_samples_per_second": 1376.652,
+      "eval_steps_per_second": 4.302,
+      "step": 14800
+    },
+    {
+      "epoch": 6.17,
+      "learning_rate": 1e-06,
+      "loss": 0.2261,
+      "step": 14820
+    },
+    {
+      "epoch": 6.17,
+      "eval_accuracy": 0.9331666666666667,
+      "eval_loss": 0.19697488844394684,
+      "eval_runtime": 16.4872,
+      "eval_samples_per_second": 1455.677,
+      "eval_steps_per_second": 4.549,
+      "step": 14820
+    },
+    {
+      "epoch": 6.18,
+      "learning_rate": 1e-06,
+      "loss": 0.2095,
+      "step": 14840
+    },
+    {
+      "epoch": 6.18,
+      "eval_accuracy": 0.9303333333333333,
+      "eval_loss": 0.2015983760356903,
+      "eval_runtime": 16.9074,
+      "eval_samples_per_second": 1419.494,
+      "eval_steps_per_second": 4.436,
+      "step": 14840
+    },
+    {
+      "epoch": 6.19,
+      "learning_rate": 1e-06,
+      "loss": 0.1589,
+      "step": 14860
+    },
+    {
+      "epoch": 6.19,
+      "eval_accuracy": 0.9330833333333334,
+      "eval_loss": 0.19708770513534546,
+      "eval_runtime": 16.2449,
+      "eval_samples_per_second": 1477.385,
+      "eval_steps_per_second": 4.617,
+      "step": 14860
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 1e-06,
+      "loss": 0.191,
+      "step": 14880
+    },
+    {
+      "epoch": 6.2,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.19845397770404816,
+      "eval_runtime": 17.192,
+      "eval_samples_per_second": 1395.999,
+      "eval_steps_per_second": 4.362,
+      "step": 14880
+    },
+    {
+      "epoch": 6.21,
+      "learning_rate": 1e-06,
+      "loss": 0.1484,
+      "step": 14900
+    },
+    {
+      "epoch": 6.21,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.19584734737873077,
+      "eval_runtime": 17.8255,
+      "eval_samples_per_second": 1346.386,
+      "eval_steps_per_second": 4.207,
+      "step": 14900
+    },
+    {
+      "epoch": 6.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1791,
+      "step": 14920
+    },
+    {
+      "epoch": 6.22,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.19735968112945557,
+      "eval_runtime": 17.6186,
+      "eval_samples_per_second": 1362.195,
+      "eval_steps_per_second": 4.257,
+      "step": 14920
+    },
+    {
+      "epoch": 6.22,
+      "learning_rate": 1e-06,
+      "loss": 0.2077,
+      "step": 14940
+    },
+    {
+      "epoch": 6.22,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.19711482524871826,
+      "eval_runtime": 17.4097,
+      "eval_samples_per_second": 1378.545,
+      "eval_steps_per_second": 4.308,
+      "step": 14940
+    },
+    {
+      "epoch": 6.23,
+      "learning_rate": 1e-06,
+      "loss": 0.1918,
+      "step": 14960
+    },
+    {
+      "epoch": 6.23,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.19635051488876343,
+      "eval_runtime": 16.6079,
+      "eval_samples_per_second": 1445.094,
+      "eval_steps_per_second": 4.516,
+      "step": 14960
+    },
+    {
+      "epoch": 6.24,
+      "learning_rate": 1e-06,
+      "loss": 0.2291,
+      "step": 14980
+    },
+    {
+      "epoch": 6.24,
+      "eval_accuracy": 0.9330833333333334,
+      "eval_loss": 0.19559039175510406,
+      "eval_runtime": 16.5817,
+      "eval_samples_per_second": 1447.382,
+      "eval_steps_per_second": 4.523,
+      "step": 14980
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 1e-06,
+      "loss": 0.1597,
+      "step": 15000
+    },
+    {
+      "epoch": 6.25,
+      "eval_accuracy": 0.9320833333333334,
+      "eval_loss": 0.1965901404619217,
+      "eval_runtime": 16.2198,
+      "eval_samples_per_second": 1479.673,
+      "eval_steps_per_second": 4.624,
+      "step": 15000
+    },
+    {
+      "epoch": 6.26,
+      "learning_rate": 1e-06,
+      "loss": 0.1844,
+      "step": 15020
+    },
+    {
+      "epoch": 6.26,
+      "eval_accuracy": 0.9323333333333333,
+      "eval_loss": 0.19741342961788177,
+      "eval_runtime": 16.1794,
+      "eval_samples_per_second": 1483.371,
+      "eval_steps_per_second": 4.636,
+      "step": 15020
+    },
+    {
+      "epoch": 6.27,
+      "learning_rate": 1e-06,
+      "loss": 0.1653,
+      "step": 15040
+    },
+    {
+      "epoch": 6.27,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.1963772028684616,
+      "eval_runtime": 16.9569,
+      "eval_samples_per_second": 1415.356,
+      "eval_steps_per_second": 4.423,
+      "step": 15040
+    },
+    {
+      "epoch": 6.28,
+      "learning_rate": 1e-06,
+      "loss": 0.158,
+      "step": 15060
+    },
+    {
+      "epoch": 6.28,
+      "eval_accuracy": 0.9309583333333333,
+      "eval_loss": 0.20028692483901978,
+      "eval_runtime": 16.1902,
+      "eval_samples_per_second": 1482.38,
+      "eval_steps_per_second": 4.632,
+      "step": 15060
+    },
+    {
+      "epoch": 6.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1602,
+      "step": 15080
+    },
+    {
+      "epoch": 6.28,
+      "eval_accuracy": 0.932625,
+      "eval_loss": 0.19667039811611176,
+      "eval_runtime": 16.4266,
+      "eval_samples_per_second": 1461.041,
+      "eval_steps_per_second": 4.566,
+      "step": 15080
+    },
+    {
+      "epoch": 6.29,
+      "learning_rate": 1e-06,
+      "loss": 0.1656,
+      "step": 15100
+    },
+    {
+      "epoch": 6.29,
+      "eval_accuracy": 0.9329166666666666,
+      "eval_loss": 0.19664861261844635,
+      "eval_runtime": 17.1479,
+      "eval_samples_per_second": 1399.589,
+      "eval_steps_per_second": 4.374,
+      "step": 15100
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 1e-06,
+      "loss": 0.1691,
+      "step": 15120
+    },
+    {
+      "epoch": 6.3,
+      "eval_accuracy": 0.9322916666666666,
+      "eval_loss": 0.19764386117458344,
+      "eval_runtime": 17.5239,
+      "eval_samples_per_second": 1369.555,
+      "eval_steps_per_second": 4.28,
+      "step": 15120
+    },
+    {
+      "epoch": 6.31,
+      "learning_rate": 1e-06,
+      "loss": 0.1598,
+      "step": 15140
+    },
+    {
+      "epoch": 6.31,
+      "eval_accuracy": 0.9333333333333333,
+      "eval_loss": 0.1968608945608139,
+      "eval_runtime": 17.6137,
+      "eval_samples_per_second": 1362.576,
+      "eval_steps_per_second": 4.258,
+      "step": 15140
+    },
+    {
+      "epoch": 6.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1768,
+      "step": 15160
+    },
+    {
+      "epoch": 6.32,
+      "eval_accuracy": 0.9334583333333333,
+      "eval_loss": 0.19633811712265015,
+      "eval_runtime": 16.2937,
+      "eval_samples_per_second": 1472.96,
+      "eval_steps_per_second": 4.603,
+      "step": 15160
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1653,
+      "step": 15180
+    },
+    {
+      "epoch": 6.33,
+      "eval_accuracy": 0.9334166666666667,
+      "eval_loss": 0.1963397115468979,
+      "eval_runtime": 16.377,
+      "eval_samples_per_second": 1465.471,
+      "eval_steps_per_second": 4.58,
+      "step": 15180
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1822,
+      "step": 15200
+    },
+    {
+      "epoch": 6.33,
+      "eval_accuracy": 0.9312916666666666,
+      "eval_loss": 0.20008063316345215,
+      "eval_runtime": 16.4032,
+      "eval_samples_per_second": 1463.126,
+      "eval_steps_per_second": 4.572,
+      "step": 15200
+    },
+    {
+      "epoch": 6.34,
+      "learning_rate": 1e-06,
+      "loss": 0.1895,
+      "step": 15220
+    },
+    {
+      "epoch": 6.34,
+      "eval_accuracy": 0.9325416666666667,
+      "eval_loss": 0.19740186631679535,
+      "eval_runtime": 17.3348,
+      "eval_samples_per_second": 1384.502,
+      "eval_steps_per_second": 4.327,
+      "step": 15220
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 1e-06,
+      "loss": 0.1622,
+      "step": 15240
+    },
+    {
+      "epoch": 6.35,
+      "eval_accuracy": 0.9315,
+      "eval_loss": 0.19918110966682434,
+      "eval_runtime": 16.9401,
+      "eval_samples_per_second": 1416.753,
+      "eval_steps_per_second": 4.427,
+      "step": 15240
+    },
+    {
+      "epoch": 6.36,
+      "learning_rate": 1e-06,
+      "loss": 0.1982,
+      "step": 15260
+    },
+    {
+      "epoch": 6.36,
+      "eval_accuracy": 0.93225,
+      "eval_loss": 0.19843068718910217,
+      "eval_runtime": 17.3631,
+      "eval_samples_per_second": 1382.241,
+      "eval_steps_per_second": 4.32,
+      "step": 15260
+    },
+    {
+      "epoch": 6.37,
+      "learning_rate": 1e-06,
+      "loss": 0.172,
+      "step": 15280
+    },
+    {
+      "epoch": 6.37,
+      "eval_accuracy": 0.9322916666666666,
+      "eval_loss": 0.19916561245918274,
+      "eval_runtime": 17.0127,
+      "eval_samples_per_second": 1410.711,
+      "eval_steps_per_second": 4.408,
+      "step": 15280
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1573,
+      "step": 15300
+    },
+    {
+      "epoch": 6.38,
+      "eval_accuracy": 0.9306666666666666,
+      "eval_loss": 0.20143656432628632,
+      "eval_runtime": 17.4933,
+      "eval_samples_per_second": 1371.953,
+      "eval_steps_per_second": 4.287,
+      "step": 15300
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 1e-06,
+      "loss": 0.158,
+      "step": 15320
+    },
+    {
+      "epoch": 6.38,
+      "eval_accuracy": 0.9324583333333333,
+      "eval_loss": 0.19875669479370117,
+      "eval_runtime": 16.9838,
+      "eval_samples_per_second": 1413.115,
+      "eval_steps_per_second": 4.416,
+      "step": 15320
+    },
+    {
+      "epoch": 6.39,
+      "learning_rate": 1e-06,
+      "loss": 0.1359,
+      "step": 15340
+    },
+    {
+      "epoch": 6.39,
+      "eval_accuracy": 0.9317916666666667,
+      "eval_loss": 0.19885526597499847,
+      "eval_runtime": 16.3343,
+      "eval_samples_per_second": 1469.303,
+      "eval_steps_per_second": 4.592,
+      "step": 15340
+    },
+    {
+      "epoch": 6.4,
+      "learning_rate": 1e-06,
+      "loss": 0.2088,
+      "step": 15360
+    },
+    {
+      "epoch": 6.4,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.19839046895503998,
+      "eval_runtime": 16.2551,
+      "eval_samples_per_second": 1476.461,
+      "eval_steps_per_second": 4.614,
+      "step": 15360
+    },
+    {
+      "epoch": 6.41,
+      "learning_rate": 1e-06,
+      "loss": 0.1952,
+      "step": 15380
+    },
+    {
+      "epoch": 6.41,
+      "eval_accuracy": 0.9315833333333333,
+      "eval_loss": 0.1983661949634552,
+      "eval_runtime": 16.7153,
+      "eval_samples_per_second": 1435.806,
+      "eval_steps_per_second": 4.487,
+      "step": 15380
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 1e-06,
+      "loss": 0.1516,
+      "step": 15400
+    },
+    {
+      "epoch": 6.42,
+      "eval_accuracy": 0.9327083333333334,
+      "eval_loss": 0.1965571790933609,
+      "eval_runtime": 16.5815,
+      "eval_samples_per_second": 1447.4,
+      "eval_steps_per_second": 4.523,
+      "step": 15400
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 1e-06,
+      "loss": 0.2063,
+      "step": 15420
+    },
+    {
+      "epoch": 6.42,
+      "eval_accuracy": 0.9333333333333333,
+      "eval_loss": 0.19558171927928925,
+      "eval_runtime": 17.0847,
+      "eval_samples_per_second": 1404.766,
+      "eval_steps_per_second": 4.39,
+      "step": 15420
+    },
+    {
+      "epoch": 6.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1766,
+      "step": 15440
+    },
+    {
+      "epoch": 6.43,
+      "eval_accuracy": 0.932875,
+      "eval_loss": 0.19576992094516754,
+      "eval_runtime": 17.3112,
+      "eval_samples_per_second": 1386.388,
+      "eval_steps_per_second": 4.332,
+      "step": 15440
+    },
+    {
+      "epoch": 6.44,
+      "learning_rate": 1e-06,
+      "loss": 0.1711,
+      "step": 15460
+    },
+    {
+      "epoch": 6.44,
+      "eval_accuracy": 0.93275,
+      "eval_loss": 0.1965101659297943,
+      "eval_runtime": 16.7762,
+      "eval_samples_per_second": 1430.595,
+      "eval_steps_per_second": 4.471,
+      "step": 15460
+    },
+    {
+      "epoch": 6.45,
+      "learning_rate": 1e-06,
+      "loss": 0.1621,
+      "step": 15480
+    },
+    {
+      "epoch": 6.45,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.19722963869571686,
+      "eval_runtime": 16.5923,
+      "eval_samples_per_second": 1446.456,
+      "eval_steps_per_second": 4.52,
+      "step": 15480
+    },
+    {
+      "epoch": 6.46,
+      "learning_rate": 1e-06,
+      "loss": 0.1854,
+      "step": 15500
+    },
+    {
+      "epoch": 6.46,
+      "eval_accuracy": 0.9300833333333334,
+      "eval_loss": 0.20203134417533875,
+      "eval_runtime": 16.701,
+      "eval_samples_per_second": 1437.04,
+      "eval_steps_per_second": 4.491,
+      "step": 15500
+    },
+    {
+      "epoch": 6.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1731,
+      "step": 15520
+    },
+    {
+      "epoch": 6.47,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.19797883927822113,
+      "eval_runtime": 17.3415,
+      "eval_samples_per_second": 1383.962,
+      "eval_steps_per_second": 4.325,
+      "step": 15520
+    },
+    {
+      "epoch": 6.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1948,
+      "step": 15540
+    },
+    {
+      "epoch": 6.47,
+      "eval_accuracy": 0.9307916666666667,
+      "eval_loss": 0.20005568861961365,
+      "eval_runtime": 17.5386,
+      "eval_samples_per_second": 1368.409,
+      "eval_steps_per_second": 4.276,
+      "step": 15540
+    },
+    {
+      "epoch": 6.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2325,
+      "step": 15560
+    },
+    {
+      "epoch": 6.48,
+      "eval_accuracy": 0.93075,
+      "eval_loss": 0.1986798346042633,
+      "eval_runtime": 17.0336,
+      "eval_samples_per_second": 1408.979,
+      "eval_steps_per_second": 4.403,
+      "step": 15560
+    },
+    {
+      "epoch": 6.49,
+      "learning_rate": 1e-06,
+      "loss": 0.1913,
+      "step": 15580
+    },
+    {
+      "epoch": 6.49,
+      "eval_accuracy": 0.9315833333333333,
+      "eval_loss": 0.19669051468372345,
+      "eval_runtime": 16.2579,
+      "eval_samples_per_second": 1476.206,
+      "eval_steps_per_second": 4.613,
+      "step": 15580
+    },
+    {
+      "epoch": 6.5,
+      "learning_rate": 1e-06,
+      "loss": 0.1755,
+      "step": 15600
+    },
+    {
+      "epoch": 6.5,
+      "eval_accuracy": 0.9321666666666667,
+      "eval_loss": 0.1981978416442871,
+      "eval_runtime": 16.3857,
+      "eval_samples_per_second": 1464.688,
+      "eval_steps_per_second": 4.577,
+      "step": 15600
+    },
+    {
+      "epoch": 6.51,
+      "learning_rate": 1e-06,
+      "loss": 0.1944,
+      "step": 15620
+    },
+    {
+      "epoch": 6.51,
+      "eval_accuracy": 0.9316666666666666,
+      "eval_loss": 0.20202693343162537,
+      "eval_runtime": 16.8881,
+      "eval_samples_per_second": 1421.117,
+      "eval_steps_per_second": 4.441,
+      "step": 15620
+    },
+    {
+      "epoch": 6.52,
+      "learning_rate": 1e-06,
+      "loss": 0.1487,
+      "step": 15640
+    },
+    {
+      "epoch": 6.52,
+      "eval_accuracy": 0.9341666666666667,
+      "eval_loss": 0.19385740160942078,
+      "eval_runtime": 16.5171,
+      "eval_samples_per_second": 1453.042,
+      "eval_steps_per_second": 4.541,
+      "step": 15640
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1854,
+      "step": 15660
+    },
+    {
+      "epoch": 6.53,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.19531066715717316,
+      "eval_runtime": 17.796,
+      "eval_samples_per_second": 1348.616,
+      "eval_steps_per_second": 4.214,
+      "step": 15660
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1331,
+      "step": 15680
+    },
+    {
+      "epoch": 6.53,
+      "eval_accuracy": 0.933125,
+      "eval_loss": 0.1954515278339386,
+      "eval_runtime": 17.1458,
+      "eval_samples_per_second": 1399.762,
+      "eval_steps_per_second": 4.374,
+      "step": 15680
+    },
+    {
+      "epoch": 6.54,
+      "learning_rate": 1e-06,
+      "loss": 0.2017,
+      "step": 15700
+    },
+    {
+      "epoch": 6.54,
+      "eval_accuracy": 0.93275,
+      "eval_loss": 0.19526571035385132,
+      "eval_runtime": 17.0288,
+      "eval_samples_per_second": 1409.381,
+      "eval_steps_per_second": 4.404,
+      "step": 15700
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 1e-06,
+      "loss": 0.1507,
+      "step": 15720
+    },
+    {
+      "epoch": 6.55,
+      "eval_accuracy": 0.9327083333333334,
+      "eval_loss": 0.19576336443424225,
+      "eval_runtime": 16.53,
+      "eval_samples_per_second": 1451.904,
+      "eval_steps_per_second": 4.537,
+      "step": 15720
+    },
+    {
+      "epoch": 6.56,
+      "learning_rate": 1e-06,
+      "loss": 0.1459,
+      "step": 15740
+    },
+    {
+      "epoch": 6.56,
+      "eval_accuracy": 0.9308333333333333,
+      "eval_loss": 0.19955700635910034,
+      "eval_runtime": 16.5253,
+      "eval_samples_per_second": 1452.316,
+      "eval_steps_per_second": 4.538,
+      "step": 15740
+    },
+    {
+      "epoch": 6.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1585,
+      "step": 15760
+    },
+    {
+      "epoch": 6.57,
+      "eval_accuracy": 0.932625,
+      "eval_loss": 0.1977219432592392,
+      "eval_runtime": 16.8034,
+      "eval_samples_per_second": 1428.285,
+      "eval_steps_per_second": 4.463,
+      "step": 15760
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 1e-06,
+      "loss": 0.1814,
+      "step": 15780
+    },
+    {
+      "epoch": 6.58,
+      "eval_accuracy": 0.93375,
+      "eval_loss": 0.19515223801136017,
+      "eval_runtime": 17.0293,
+      "eval_samples_per_second": 1409.337,
+      "eval_steps_per_second": 4.404,
+      "step": 15780
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 1e-06,
+      "loss": 0.1383,
+      "step": 15800
+    },
+    {
+      "epoch": 6.58,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.19543632864952087,
+      "eval_runtime": 17.0436,
+      "eval_samples_per_second": 1408.152,
+      "eval_steps_per_second": 4.4,
+      "step": 15800
+    },
+    {
+      "epoch": 6.59,
+      "learning_rate": 1e-06,
+      "loss": 0.1981,
+      "step": 15820
+    },
+    {
+      "epoch": 6.59,
+      "eval_accuracy": 0.9323333333333333,
+      "eval_loss": 0.19736984372138977,
+      "eval_runtime": 16.4411,
+      "eval_samples_per_second": 1459.753,
+      "eval_steps_per_second": 4.562,
+      "step": 15820
+    },
+    {
+      "epoch": 6.6,
+      "learning_rate": 1e-06,
+      "loss": 0.1715,
+      "step": 15840
+    },
+    {
+      "epoch": 6.6,
+      "eval_accuracy": 0.93225,
+      "eval_loss": 0.1977250874042511,
+      "eval_runtime": 16.9027,
+      "eval_samples_per_second": 1419.887,
+      "eval_steps_per_second": 4.437,
+      "step": 15840
+    },
+    {
+      "epoch": 6.61,
+      "learning_rate": 1e-06,
+      "loss": 0.1392,
+      "step": 15860
+    },
+    {
+      "epoch": 6.61,
+      "eval_accuracy": 0.9314166666666667,
+      "eval_loss": 0.19858193397521973,
+      "eval_runtime": 16.2175,
+      "eval_samples_per_second": 1479.884,
+      "eval_steps_per_second": 4.625,
+      "step": 15860
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1587,
+      "step": 15880
+    },
+    {
+      "epoch": 6.62,
+      "eval_accuracy": 0.93225,
+      "eval_loss": 0.19824790954589844,
+      "eval_runtime": 16.4062,
+      "eval_samples_per_second": 1462.863,
+      "eval_steps_per_second": 4.571,
+      "step": 15880
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1697,
+      "step": 15900
+    },
+    {
+      "epoch": 6.62,
+      "eval_accuracy": 0.9299583333333333,
+      "eval_loss": 0.2014349400997162,
+      "eval_runtime": 16.3856,
+      "eval_samples_per_second": 1464.702,
+      "eval_steps_per_second": 4.577,
+      "step": 15900
+    },
+    {
+      "epoch": 6.63,
+      "learning_rate": 1e-06,
+      "loss": 0.1861,
+      "step": 15920
+    },
+    {
+      "epoch": 6.63,
+      "eval_accuracy": 0.9314166666666667,
+      "eval_loss": 0.1988927125930786,
+      "eval_runtime": 16.3893,
+      "eval_samples_per_second": 1464.366,
+      "eval_steps_per_second": 4.576,
+      "step": 15920
+    },
+    {
+      "epoch": 6.64,
+      "learning_rate": 1e-06,
+      "loss": 0.174,
+      "step": 15940
+    },
+    {
+      "epoch": 6.64,
+      "eval_accuracy": 0.931375,
+      "eval_loss": 0.1998647302389145,
+      "eval_runtime": 16.7731,
+      "eval_samples_per_second": 1430.861,
+      "eval_steps_per_second": 4.471,
+      "step": 15940
+    },
+    {
+      "epoch": 6.65,
+      "learning_rate": 1e-06,
+      "loss": 0.1641,
+      "step": 15960
+    },
+    {
+      "epoch": 6.65,
+      "eval_accuracy": 0.9314166666666667,
+      "eval_loss": 0.20076820254325867,
+      "eval_runtime": 16.2481,
+      "eval_samples_per_second": 1477.098,
+      "eval_steps_per_second": 4.616,
+      "step": 15960
+    },
+    {
+      "epoch": 6.66,
+      "learning_rate": 1e-06,
+      "loss": 0.1408,
+      "step": 15980
+    },
+    {
+      "epoch": 6.66,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.19549201428890228,
+      "eval_runtime": 16.2222,
+      "eval_samples_per_second": 1479.458,
+      "eval_steps_per_second": 4.623,
+      "step": 15980
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1878,
+      "step": 16000
+    },
+    {
+      "epoch": 6.67,
+      "eval_accuracy": 0.9340833333333334,
+      "eval_loss": 0.1959511786699295,
+      "eval_runtime": 16.3546,
+      "eval_samples_per_second": 1467.474,
+      "eval_steps_per_second": 4.586,
+      "step": 16000
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1425,
+      "step": 16020
+    },
+    {
+      "epoch": 6.67,
+      "eval_accuracy": 0.9337916666666667,
+      "eval_loss": 0.1953056901693344,
+      "eval_runtime": 16.458,
+      "eval_samples_per_second": 1458.255,
+      "eval_steps_per_second": 4.557,
+      "step": 16020
+    },
+    {
+      "epoch": 6.68,
+      "learning_rate": 1e-06,
+      "loss": 0.184,
+      "step": 16040
+    },
+    {
+      "epoch": 6.68,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.19988182187080383,
+      "eval_runtime": 17.0453,
+      "eval_samples_per_second": 1408.016,
+      "eval_steps_per_second": 4.4,
+      "step": 16040
+    },
+    {
+      "epoch": 6.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1609,
+      "step": 16060
+    },
+    {
+      "epoch": 6.69,
+      "eval_accuracy": 0.9342916666666666,
+      "eval_loss": 0.19486786425113678,
+      "eval_runtime": 16.3548,
+      "eval_samples_per_second": 1467.457,
+      "eval_steps_per_second": 4.586,
+      "step": 16060
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1757,
+      "step": 16080
+    },
+    {
+      "epoch": 6.7,
+      "eval_accuracy": 0.93375,
+      "eval_loss": 0.195814847946167,
+      "eval_runtime": 16.6686,
+      "eval_samples_per_second": 1439.835,
+      "eval_steps_per_second": 4.499,
+      "step": 16080
+    },
+    {
+      "epoch": 6.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1657,
+      "step": 16100
+    },
+    {
+      "epoch": 6.71,
+      "eval_accuracy": 0.9322083333333333,
+      "eval_loss": 0.19810351729393005,
+      "eval_runtime": 17.5937,
+      "eval_samples_per_second": 1364.125,
+      "eval_steps_per_second": 4.263,
+      "step": 16100
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1496,
+      "step": 16120
+    },
+    {
+      "epoch": 6.72,
+      "eval_accuracy": 0.933875,
+      "eval_loss": 0.19454562664031982,
+      "eval_runtime": 17.5068,
+      "eval_samples_per_second": 1370.9,
+      "eval_steps_per_second": 4.284,
+      "step": 16120
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 1e-06,
+      "loss": 0.2053,
+      "step": 16140
+    },
+    {
+      "epoch": 6.72,
+      "eval_accuracy": 0.9334166666666667,
+      "eval_loss": 0.19530512392520905,
+      "eval_runtime": 17.6899,
+      "eval_samples_per_second": 1356.703,
+      "eval_steps_per_second": 4.24,
+      "step": 16140
+    },
+    {
+      "epoch": 6.73,
+      "learning_rate": 1e-06,
+      "loss": 0.1905,
+      "step": 16160
+    },
+    {
+      "epoch": 6.73,
+      "eval_accuracy": 0.933375,
+      "eval_loss": 0.19494038820266724,
+      "eval_runtime": 17.39,
+      "eval_samples_per_second": 1380.102,
+      "eval_steps_per_second": 4.313,
+      "step": 16160
+    },
+    {
+      "epoch": 6.74,
+      "learning_rate": 1e-06,
+      "loss": 0.1515,
+      "step": 16180
+    },
+    {
+      "epoch": 6.74,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.19500760734081268,
+      "eval_runtime": 17.8113,
+      "eval_samples_per_second": 1347.457,
+      "eval_steps_per_second": 4.211,
+      "step": 16180
+    },
+    {
+      "epoch": 6.75,
+      "learning_rate": 1e-06,
+      "loss": 0.1705,
+      "step": 16200
+    },
+    {
+      "epoch": 6.75,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.19622166454792023,
+      "eval_runtime": 17.3835,
+      "eval_samples_per_second": 1380.617,
+      "eval_steps_per_second": 4.314,
+      "step": 16200
+    },
+    {
+      "epoch": 6.76,
+      "learning_rate": 1e-06,
+      "loss": 0.1756,
+      "step": 16220
+    },
+    {
+      "epoch": 6.76,
+      "eval_accuracy": 0.9332916666666666,
+      "eval_loss": 0.1943114697933197,
+      "eval_runtime": 17.2208,
+      "eval_samples_per_second": 1393.664,
+      "eval_steps_per_second": 4.355,
+      "step": 16220
+    },
+    {
+      "epoch": 6.77,
+      "learning_rate": 1e-06,
+      "loss": 0.1729,
+      "step": 16240
+    },
+    {
+      "epoch": 6.77,
+      "eval_accuracy": 0.9343333333333333,
+      "eval_loss": 0.19441558420658112,
+      "eval_runtime": 17.5483,
+      "eval_samples_per_second": 1367.653,
+      "eval_steps_per_second": 4.274,
+      "step": 16240
+    },
+    {
+      "epoch": 6.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1526,
+      "step": 16260
+    },
+    {
+      "epoch": 6.78,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.1942491978406906,
+      "eval_runtime": 16.3394,
+      "eval_samples_per_second": 1468.842,
+      "eval_steps_per_second": 4.59,
+      "step": 16260
+    },
+    {
+      "epoch": 6.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1684,
+      "step": 16280
+    },
+    {
+      "epoch": 6.78,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.1948528289794922,
+      "eval_runtime": 16.0087,
+      "eval_samples_per_second": 1499.183,
+      "eval_steps_per_second": 4.685,
+      "step": 16280
+    },
+    {
+      "epoch": 6.79,
+      "learning_rate": 1e-06,
+      "loss": 0.1756,
+      "step": 16300
+    },
+    {
+      "epoch": 6.79,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.19495390355587006,
+      "eval_runtime": 16.737,
+      "eval_samples_per_second": 1433.945,
+      "eval_steps_per_second": 4.481,
+      "step": 16300
+    },
+    {
+      "epoch": 6.8,
+      "learning_rate": 1e-06,
+      "loss": 0.1706,
+      "step": 16320
+    },
+    {
+      "epoch": 6.8,
+      "eval_accuracy": 0.9318333333333333,
+      "eval_loss": 0.19855649769306183,
+      "eval_runtime": 16.285,
+      "eval_samples_per_second": 1473.752,
+      "eval_steps_per_second": 4.605,
+      "step": 16320
+    },
+    {
+      "epoch": 6.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1655,
+      "step": 16340
+    },
+    {
+      "epoch": 6.81,
+      "eval_accuracy": 0.9319166666666666,
+      "eval_loss": 0.19920918345451355,
+      "eval_runtime": 17.3955,
+      "eval_samples_per_second": 1379.671,
+      "eval_steps_per_second": 4.311,
+      "step": 16340
+    },
+    {
+      "epoch": 6.82,
+      "learning_rate": 1e-06,
+      "loss": 0.1858,
+      "step": 16360
+    },
+    {
+      "epoch": 6.82,
+      "eval_accuracy": 0.9325,
+      "eval_loss": 0.19792011380195618,
+      "eval_runtime": 17.319,
+      "eval_samples_per_second": 1385.762,
+      "eval_steps_per_second": 4.331,
+      "step": 16360
+    },
+    {
+      "epoch": 6.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1926,
+      "step": 16380
+    },
+    {
+      "epoch": 6.83,
+      "eval_accuracy": 0.9332083333333333,
+      "eval_loss": 0.19599467515945435,
+      "eval_runtime": 16.5773,
+      "eval_samples_per_second": 1447.759,
+      "eval_steps_per_second": 4.524,
+      "step": 16380
+    },
+    {
+      "epoch": 6.83,
+      "learning_rate": 1e-06,
+      "loss": 0.162,
+      "step": 16400
+    },
+    {
+      "epoch": 6.83,
+      "eval_accuracy": 0.9325,
+      "eval_loss": 0.1975654661655426,
+      "eval_runtime": 16.0954,
+      "eval_samples_per_second": 1491.106,
+      "eval_steps_per_second": 4.66,
+      "step": 16400
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 1e-06,
+      "loss": 0.2168,
+      "step": 16420
+    },
+    {
+      "epoch": 6.84,
+      "eval_accuracy": 0.9322083333333333,
+      "eval_loss": 0.19847214221954346,
+      "eval_runtime": 16.3923,
+      "eval_samples_per_second": 1464.098,
+      "eval_steps_per_second": 4.575,
+      "step": 16420
+    },
+    {
+      "epoch": 6.85,
+      "learning_rate": 1e-06,
+      "loss": 0.173,
+      "step": 16440
+    },
+    {
+      "epoch": 6.85,
+      "eval_accuracy": 0.9312916666666666,
+      "eval_loss": 0.2017899602651596,
+      "eval_runtime": 16.2391,
+      "eval_samples_per_second": 1477.915,
+      "eval_steps_per_second": 4.618,
+      "step": 16440
+    },
+    {
+      "epoch": 6.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1891,
+      "step": 16460
+    },
+    {
+      "epoch": 6.86,
+      "eval_accuracy": 0.93225,
+      "eval_loss": 0.19827592372894287,
+      "eval_runtime": 16.3796,
+      "eval_samples_per_second": 1465.236,
+      "eval_steps_per_second": 4.579,
+      "step": 16460
+    },
+    {
+      "epoch": 6.87,
+      "learning_rate": 1e-06,
+      "loss": 0.1619,
+      "step": 16480
+    },
+    {
+      "epoch": 6.87,
+      "eval_accuracy": 0.9332916666666666,
+      "eval_loss": 0.19625312089920044,
+      "eval_runtime": 17.7432,
+      "eval_samples_per_second": 1352.634,
+      "eval_steps_per_second": 4.227,
+      "step": 16480
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1884,
+      "step": 16500
+    },
+    {
+      "epoch": 6.88,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.19771744310855865,
+      "eval_runtime": 16.6178,
+      "eval_samples_per_second": 1444.233,
+      "eval_steps_per_second": 4.513,
+      "step": 16500
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1735,
+      "step": 16520
+    },
+    {
+      "epoch": 6.88,
+      "eval_accuracy": 0.9327916666666667,
+      "eval_loss": 0.19762367010116577,
+      "eval_runtime": 16.9624,
+      "eval_samples_per_second": 1414.896,
+      "eval_steps_per_second": 4.422,
+      "step": 16520
+    },
+    {
+      "epoch": 6.89,
+      "learning_rate": 1e-06,
+      "loss": 0.1949,
+      "step": 16540
+    },
+    {
+      "epoch": 6.89,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.19725541770458221,
+      "eval_runtime": 16.0777,
+      "eval_samples_per_second": 1492.755,
+      "eval_steps_per_second": 4.665,
+      "step": 16540
+    },
+    {
+      "epoch": 6.9,
+      "learning_rate": 1e-06,
+      "loss": 0.2021,
+      "step": 16560
+    },
+    {
+      "epoch": 6.9,
+      "eval_accuracy": 0.9340833333333334,
+      "eval_loss": 0.1951657384634018,
+      "eval_runtime": 16.8714,
+      "eval_samples_per_second": 1422.525,
+      "eval_steps_per_second": 4.445,
+      "step": 16560
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 1e-06,
+      "loss": 0.1742,
+      "step": 16580
+    },
+    {
+      "epoch": 6.91,
+      "eval_accuracy": 0.9345833333333333,
+      "eval_loss": 0.19460086524486542,
+      "eval_runtime": 16.336,
+      "eval_samples_per_second": 1469.149,
+      "eval_steps_per_second": 4.591,
+      "step": 16580
+    },
+    {
+      "epoch": 6.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1935,
+      "step": 16600
+    },
+    {
+      "epoch": 6.92,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.19422036409378052,
+      "eval_runtime": 16.245,
+      "eval_samples_per_second": 1477.38,
+      "eval_steps_per_second": 4.617,
+      "step": 16600
+    },
+    {
+      "epoch": 6.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1709,
+      "step": 16620
+    },
+    {
+      "epoch": 6.92,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.1949995458126068,
+      "eval_runtime": 16.3728,
+      "eval_samples_per_second": 1465.845,
+      "eval_steps_per_second": 4.581,
+      "step": 16620
+    },
+    {
+      "epoch": 6.93,
+      "learning_rate": 1e-06,
+      "loss": 0.1675,
+      "step": 16640
+    },
+    {
+      "epoch": 6.93,
+      "eval_accuracy": 0.9334166666666667,
+      "eval_loss": 0.19600391387939453,
+      "eval_runtime": 16.421,
+      "eval_samples_per_second": 1461.543,
+      "eval_steps_per_second": 4.567,
+      "step": 16640
+    },
+    {
+      "epoch": 6.94,
+      "learning_rate": 1e-06,
+      "loss": 0.1648,
+      "step": 16660
+    },
+    {
+      "epoch": 6.94,
+      "eval_accuracy": 0.9329583333333333,
+      "eval_loss": 0.19674494862556458,
+      "eval_runtime": 16.3324,
+      "eval_samples_per_second": 1469.47,
+      "eval_steps_per_second": 4.592,
+      "step": 16660
+    },
+    {
+      "epoch": 6.95,
+      "learning_rate": 1e-06,
+      "loss": 0.1757,
+      "step": 16680
+    },
+    {
+      "epoch": 6.95,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.1960502564907074,
+      "eval_runtime": 17.1637,
+      "eval_samples_per_second": 1398.302,
+      "eval_steps_per_second": 4.37,
+      "step": 16680
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 1e-06,
+      "loss": 0.1743,
+      "step": 16700
+    },
+    {
+      "epoch": 6.96,
+      "eval_accuracy": 0.9329583333333333,
+      "eval_loss": 0.19483575224876404,
+      "eval_runtime": 15.8961,
+      "eval_samples_per_second": 1509.806,
+      "eval_steps_per_second": 4.718,
+      "step": 16700
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 1e-06,
+      "loss": 0.17,
+      "step": 16720
+    },
+    {
+      "epoch": 6.97,
+      "eval_accuracy": 0.9332083333333333,
+      "eval_loss": 0.19447939097881317,
+      "eval_runtime": 16.8287,
+      "eval_samples_per_second": 1426.132,
+      "eval_steps_per_second": 4.457,
+      "step": 16720
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1625,
+      "step": 16740
+    },
+    {
+      "epoch": 6.97,
+      "eval_accuracy": 0.933875,
+      "eval_loss": 0.1952148824930191,
+      "eval_runtime": 16.0124,
+      "eval_samples_per_second": 1498.839,
+      "eval_steps_per_second": 4.684,
+      "step": 16740
+    },
+    {
+      "epoch": 6.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1802,
+      "step": 16760
+    },
+    {
+      "epoch": 6.98,
+      "eval_accuracy": 0.9338333333333333,
+      "eval_loss": 0.19577091932296753,
+      "eval_runtime": 16.0326,
+      "eval_samples_per_second": 1496.951,
+      "eval_steps_per_second": 4.678,
+      "step": 16760
+    },
+    {
+      "epoch": 6.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1855,
+      "step": 16780
+    },
+    {
+      "epoch": 6.99,
+      "eval_accuracy": 0.932625,
+      "eval_loss": 0.19726844131946564,
+      "eval_runtime": 17.465,
+      "eval_samples_per_second": 1374.176,
+      "eval_steps_per_second": 4.294,
+      "step": 16780
+    },
+    {
+      "epoch": 7.0,
+      "learning_rate": 1e-06,
+      "loss": 0.1623,
+      "step": 16800
+    },
+    {
+      "epoch": 7.0,
+      "eval_accuracy": 0.9333333333333333,
+      "eval_loss": 0.19629396498203278,
+      "eval_runtime": 17.5957,
+      "eval_samples_per_second": 1363.968,
+      "eval_steps_per_second": 4.262,
+      "step": 16800
+    },
+    {
+      "epoch": 7.01,
+      "learning_rate": 1e-06,
+      "loss": 0.1521,
+      "step": 16820
+    },
+    {
+      "epoch": 7.01,
+      "eval_accuracy": 0.931625,
+      "eval_loss": 0.1997893750667572,
+      "eval_runtime": 17.3287,
+      "eval_samples_per_second": 1384.986,
+      "eval_steps_per_second": 4.328,
+      "step": 16820
+    },
+    {
+      "epoch": 7.02,
+      "learning_rate": 1e-06,
+      "loss": 0.2071,
+      "step": 16840
+    },
+    {
+      "epoch": 7.02,
+      "eval_accuracy": 0.9344166666666667,
+      "eval_loss": 0.1939399242401123,
+      "eval_runtime": 16.6723,
+      "eval_samples_per_second": 1439.518,
+      "eval_steps_per_second": 4.498,
+      "step": 16840
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1669,
+      "step": 16860
+    },
+    {
+      "epoch": 7.03,
+      "eval_accuracy": 0.9325833333333333,
+      "eval_loss": 0.1989278346300125,
+      "eval_runtime": 16.9163,
+      "eval_samples_per_second": 1418.746,
+      "eval_steps_per_second": 4.434,
+      "step": 16860
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1695,
+      "step": 16880
+    },
+    {
+      "epoch": 7.03,
+      "eval_accuracy": 0.9313333333333333,
+      "eval_loss": 0.20084674656391144,
+      "eval_runtime": 17.4712,
+      "eval_samples_per_second": 1373.69,
+      "eval_steps_per_second": 4.293,
+      "step": 16880
+    },
+    {
+      "epoch": 7.04,
+      "learning_rate": 1e-06,
+      "loss": 0.2092,
+      "step": 16900
+    },
+    {
+      "epoch": 7.04,
+      "eval_accuracy": 0.9324166666666667,
+      "eval_loss": 0.19925980269908905,
+      "eval_runtime": 17.5432,
+      "eval_samples_per_second": 1368.052,
+      "eval_steps_per_second": 4.275,
+      "step": 16900
+    },
+    {
+      "epoch": 7.05,
+      "learning_rate": 1e-06,
+      "loss": 0.1859,
+      "step": 16920
+    },
+    {
+      "epoch": 7.05,
+      "eval_accuracy": 0.9331666666666667,
+      "eval_loss": 0.19550496339797974,
+      "eval_runtime": 17.387,
+      "eval_samples_per_second": 1380.338,
+      "eval_steps_per_second": 4.314,
+      "step": 16920
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 1e-06,
+      "loss": 0.1407,
+      "step": 16940
+    },
+    {
+      "epoch": 7.06,
+      "eval_accuracy": 0.9350833333333334,
+      "eval_loss": 0.19279736280441284,
+      "eval_runtime": 17.6476,
+      "eval_samples_per_second": 1359.958,
+      "eval_steps_per_second": 4.25,
+      "step": 16940
+    },
+    {
+      "epoch": 7.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1564,
+      "step": 16960
+    },
+    {
+      "epoch": 7.07,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.19368121027946472,
+      "eval_runtime": 17.8587,
+      "eval_samples_per_second": 1343.884,
+      "eval_steps_per_second": 4.2,
+      "step": 16960
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1514,
+      "step": 16980
+    },
+    {
+      "epoch": 7.08,
+      "eval_accuracy": 0.9342083333333333,
+      "eval_loss": 0.19475901126861572,
+      "eval_runtime": 17.1608,
+      "eval_samples_per_second": 1398.536,
+      "eval_steps_per_second": 4.37,
+      "step": 16980
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1425,
+      "step": 17000
+    },
+    {
+      "epoch": 7.08,
+      "eval_accuracy": 0.932125,
+      "eval_loss": 0.19865868985652924,
+      "eval_runtime": 17.0882,
+      "eval_samples_per_second": 1404.476,
+      "eval_steps_per_second": 4.389,
+      "step": 17000
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 1e-06,
+      "loss": 0.1849,
+      "step": 17020
+    },
+    {
+      "epoch": 7.09,
+      "eval_accuracy": 0.9346666666666666,
+      "eval_loss": 0.1954139769077301,
+      "eval_runtime": 17.452,
+      "eval_samples_per_second": 1375.197,
+      "eval_steps_per_second": 4.297,
+      "step": 17020
+    },
+    {
+      "epoch": 7.1,
+      "learning_rate": 1e-06,
+      "loss": 0.1662,
+      "step": 17040
+    },
+    {
+      "epoch": 7.1,
+      "eval_accuracy": 0.934875,
+      "eval_loss": 0.19401773810386658,
+      "eval_runtime": 16.9959,
+      "eval_samples_per_second": 1412.107,
+      "eval_steps_per_second": 4.413,
+      "step": 17040
+    },
+    {
+      "epoch": 7.11,
+      "learning_rate": 1e-06,
+      "loss": 0.1887,
+      "step": 17060
+    },
+    {
+      "epoch": 7.11,
+      "eval_accuracy": 0.9336666666666666,
+      "eval_loss": 0.1946951448917389,
+      "eval_runtime": 16.4323,
+      "eval_samples_per_second": 1460.542,
+      "eval_steps_per_second": 4.564,
+      "step": 17060
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1704,
+      "step": 17080
+    },
+    {
+      "epoch": 7.12,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.1954944133758545,
+      "eval_runtime": 17.7938,
+      "eval_samples_per_second": 1348.781,
+      "eval_steps_per_second": 4.215,
+      "step": 17080
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 1e-06,
+      "loss": 0.2087,
+      "step": 17100
+    },
+    {
+      "epoch": 7.12,
+      "eval_accuracy": 0.9334166666666667,
+      "eval_loss": 0.19571976363658905,
+      "eval_runtime": 16.3621,
+      "eval_samples_per_second": 1466.803,
+      "eval_steps_per_second": 4.584,
+      "step": 17100
+    },
+    {
+      "epoch": 7.13,
+      "learning_rate": 1e-06,
+      "loss": 0.1576,
+      "step": 17120
+    },
+    {
+      "epoch": 7.13,
+      "eval_accuracy": 0.934875,
+      "eval_loss": 0.19232991337776184,
+      "eval_runtime": 17.0875,
+      "eval_samples_per_second": 1404.537,
+      "eval_steps_per_second": 4.389,
+      "step": 17120
+    },
+    {
+      "epoch": 7.14,
+      "learning_rate": 1e-06,
+      "loss": 0.1837,
+      "step": 17140
+    },
+    {
+      "epoch": 7.14,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.19416970014572144,
+      "eval_runtime": 17.2421,
+      "eval_samples_per_second": 1391.941,
+      "eval_steps_per_second": 4.35,
+      "step": 17140
+    },
+    {
+      "epoch": 7.15,
+      "learning_rate": 1e-06,
+      "loss": 0.1771,
+      "step": 17160
+    },
+    {
+      "epoch": 7.15,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.19288769364356995,
+      "eval_runtime": 15.9944,
+      "eval_samples_per_second": 1500.527,
+      "eval_steps_per_second": 4.689,
+      "step": 17160
+    },
+    {
+      "epoch": 7.16,
+      "learning_rate": 1e-06,
+      "loss": 0.1661,
+      "step": 17180
+    },
+    {
+      "epoch": 7.16,
+      "eval_accuracy": 0.9346666666666666,
+      "eval_loss": 0.19377556443214417,
+      "eval_runtime": 16.1242,
+      "eval_samples_per_second": 1488.45,
+      "eval_steps_per_second": 4.651,
+      "step": 17180
+    },
+    {
+      "epoch": 7.17,
+      "learning_rate": 1e-06,
+      "loss": 0.1839,
+      "step": 17200
+    },
+    {
+      "epoch": 7.17,
+      "eval_accuracy": 0.9345833333333333,
+      "eval_loss": 0.19333358108997345,
+      "eval_runtime": 16.31,
+      "eval_samples_per_second": 1471.488,
+      "eval_steps_per_second": 4.598,
+      "step": 17200
+    },
+    {
+      "epoch": 7.17,
+      "learning_rate": 1e-06,
+      "loss": 0.172,
+      "step": 17220
+    },
+    {
+      "epoch": 7.17,
+      "eval_accuracy": 0.9345416666666667,
+      "eval_loss": 0.19157427549362183,
+      "eval_runtime": 16.7874,
+      "eval_samples_per_second": 1429.64,
+      "eval_steps_per_second": 4.468,
+      "step": 17220
+    },
+    {
+      "epoch": 7.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1563,
+      "step": 17240
+    },
+    {
+      "epoch": 7.18,
+      "eval_accuracy": 0.9334583333333333,
+      "eval_loss": 0.19396959245204926,
+      "eval_runtime": 16.9407,
+      "eval_samples_per_second": 1416.708,
+      "eval_steps_per_second": 4.427,
+      "step": 17240
+    },
+    {
+      "epoch": 7.19,
+      "learning_rate": 1e-06,
+      "loss": 0.1835,
+      "step": 17260
+    },
+    {
+      "epoch": 7.19,
+      "eval_accuracy": 0.9312916666666666,
+      "eval_loss": 0.20071102678775787,
+      "eval_runtime": 17.2109,
+      "eval_samples_per_second": 1394.462,
+      "eval_steps_per_second": 4.358,
+      "step": 17260
+    },
+    {
+      "epoch": 7.2,
+      "learning_rate": 1e-06,
+      "loss": 0.1794,
+      "step": 17280
+    },
+    {
+      "epoch": 7.2,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.1961183249950409,
+      "eval_runtime": 16.6977,
+      "eval_samples_per_second": 1437.327,
+      "eval_steps_per_second": 4.492,
+      "step": 17280
+    },
+    {
+      "epoch": 7.21,
+      "learning_rate": 1e-06,
+      "loss": 0.2048,
+      "step": 17300
+    },
+    {
+      "epoch": 7.21,
+      "eval_accuracy": 0.934125,
+      "eval_loss": 0.1937197893857956,
+      "eval_runtime": 17.0498,
+      "eval_samples_per_second": 1407.643,
+      "eval_steps_per_second": 4.399,
+      "step": 17300
+    },
+    {
+      "epoch": 7.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1609,
+      "step": 17320
+    },
+    {
+      "epoch": 7.22,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.1952856183052063,
+      "eval_runtime": 16.2943,
+      "eval_samples_per_second": 1472.908,
+      "eval_steps_per_second": 4.603,
+      "step": 17320
+    },
+    {
+      "epoch": 7.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1588,
+      "step": 17340
+    },
+    {
+      "epoch": 7.22,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.19826608896255493,
+      "eval_runtime": 16.8162,
+      "eval_samples_per_second": 1427.195,
+      "eval_steps_per_second": 4.46,
+      "step": 17340
+    },
+    {
+      "epoch": 7.23,
+      "learning_rate": 1e-06,
+      "loss": 0.1796,
+      "step": 17360
+    },
+    {
+      "epoch": 7.23,
+      "eval_accuracy": 0.932375,
+      "eval_loss": 0.1968182921409607,
+      "eval_runtime": 17.6501,
+      "eval_samples_per_second": 1359.765,
+      "eval_steps_per_second": 4.249,
+      "step": 17360
+    },
+    {
+      "epoch": 7.24,
+      "learning_rate": 1e-06,
+      "loss": 0.1834,
+      "step": 17380
+    },
+    {
+      "epoch": 7.24,
+      "eval_accuracy": 0.932625,
+      "eval_loss": 0.19632968306541443,
+      "eval_runtime": 17.2336,
+      "eval_samples_per_second": 1392.628,
+      "eval_steps_per_second": 4.352,
+      "step": 17380
+    },
+    {
+      "epoch": 7.25,
+      "learning_rate": 1e-06,
+      "loss": 0.1564,
+      "step": 17400
+    },
+    {
+      "epoch": 7.25,
+      "eval_accuracy": 0.9337916666666667,
+      "eval_loss": 0.19396378099918365,
+      "eval_runtime": 17.7707,
+      "eval_samples_per_second": 1350.539,
+      "eval_steps_per_second": 4.22,
+      "step": 17400
+    },
+    {
+      "epoch": 7.26,
+      "learning_rate": 1e-06,
+      "loss": 0.1513,
+      "step": 17420
+    },
+    {
+      "epoch": 7.26,
+      "eval_accuracy": 0.932875,
+      "eval_loss": 0.19559015333652496,
+      "eval_runtime": 17.1532,
+      "eval_samples_per_second": 1399.156,
+      "eval_steps_per_second": 4.372,
+      "step": 17420
+    },
+    {
+      "epoch": 7.27,
+      "learning_rate": 1e-06,
+      "loss": 0.1568,
+      "step": 17440
+    },
+    {
+      "epoch": 7.27,
+      "eval_accuracy": 0.935125,
+      "eval_loss": 0.19202855229377747,
+      "eval_runtime": 17.1051,
+      "eval_samples_per_second": 1403.089,
+      "eval_steps_per_second": 4.385,
+      "step": 17440
+    },
+    {
+      "epoch": 7.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1748,
+      "step": 17460
+    },
+    {
+      "epoch": 7.28,
+      "eval_accuracy": 0.9349583333333333,
+      "eval_loss": 0.19286368787288666,
+      "eval_runtime": 15.5136,
+      "eval_samples_per_second": 1547.025,
+      "eval_steps_per_second": 4.834,
+      "step": 17460
+    },
+    {
+      "epoch": 7.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1578,
+      "step": 17480
+    },
+    {
+      "epoch": 7.28,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.194309800863266,
+      "eval_runtime": 16.1632,
+      "eval_samples_per_second": 1484.852,
+      "eval_steps_per_second": 4.64,
+      "step": 17480
+    },
+    {
+      "epoch": 7.29,
+      "learning_rate": 1e-06,
+      "loss": 0.1321,
+      "step": 17500
+    },
+    {
+      "epoch": 7.29,
+      "eval_accuracy": 0.9351666666666667,
+      "eval_loss": 0.19201384484767914,
+      "eval_runtime": 16.5178,
+      "eval_samples_per_second": 1452.974,
+      "eval_steps_per_second": 4.541,
+      "step": 17500
+    },
+    {
+      "epoch": 7.3,
+      "learning_rate": 1e-06,
+      "loss": 0.1963,
+      "step": 17520
+    },
+    {
+      "epoch": 7.3,
+      "eval_accuracy": 0.9339166666666666,
+      "eval_loss": 0.19566002488136292,
+      "eval_runtime": 16.995,
+      "eval_samples_per_second": 1412.184,
+      "eval_steps_per_second": 4.413,
+      "step": 17520
+    },
+    {
+      "epoch": 7.31,
+      "learning_rate": 1e-06,
+      "loss": 0.1927,
+      "step": 17540
+    },
+    {
+      "epoch": 7.31,
+      "eval_accuracy": 0.9330416666666667,
+      "eval_loss": 0.19617126882076263,
+      "eval_runtime": 17.1067,
+      "eval_samples_per_second": 1402.962,
+      "eval_steps_per_second": 4.384,
+      "step": 17540
+    },
+    {
+      "epoch": 7.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1658,
+      "step": 17560
+    },
+    {
+      "epoch": 7.32,
+      "eval_accuracy": 0.9332916666666666,
+      "eval_loss": 0.195390984416008,
+      "eval_runtime": 17.0629,
+      "eval_samples_per_second": 1406.565,
+      "eval_steps_per_second": 4.396,
+      "step": 17560
+    },
+    {
+      "epoch": 7.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1452,
+      "step": 17580
+    },
+    {
+      "epoch": 7.33,
+      "eval_accuracy": 0.9345833333333333,
+      "eval_loss": 0.19409912824630737,
+      "eval_runtime": 16.3617,
+      "eval_samples_per_second": 1466.84,
+      "eval_steps_per_second": 4.584,
+      "step": 17580
+    },
+    {
+      "epoch": 7.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1992,
+      "step": 17600
+    },
+    {
+      "epoch": 7.33,
+      "eval_accuracy": 0.934125,
+      "eval_loss": 0.1932896226644516,
+      "eval_runtime": 16.2659,
+      "eval_samples_per_second": 1475.476,
+      "eval_steps_per_second": 4.611,
+      "step": 17600
+    },
+    {
+      "epoch": 7.34,
+      "learning_rate": 1e-06,
+      "loss": 0.1824,
+      "step": 17620
+    },
+    {
+      "epoch": 7.34,
+      "eval_accuracy": 0.934625,
+      "eval_loss": 0.1922486424446106,
+      "eval_runtime": 16.3428,
+      "eval_samples_per_second": 1468.54,
+      "eval_steps_per_second": 4.589,
+      "step": 17620
+    },
+    {
+      "epoch": 7.35,
+      "learning_rate": 1e-06,
+      "loss": 0.1388,
+      "step": 17640
+    },
+    {
+      "epoch": 7.35,
+      "eval_accuracy": 0.9345,
+      "eval_loss": 0.19244614243507385,
+      "eval_runtime": 17.0407,
+      "eval_samples_per_second": 1408.391,
+      "eval_steps_per_second": 4.401,
+      "step": 17640
+    },
+    {
+      "epoch": 7.36,
+      "learning_rate": 1e-06,
+      "loss": 0.1732,
+      "step": 17660
+    },
+    {
+      "epoch": 7.36,
+      "eval_accuracy": 0.9343333333333333,
+      "eval_loss": 0.19296354055404663,
+      "eval_runtime": 16.3329,
+      "eval_samples_per_second": 1469.426,
+      "eval_steps_per_second": 4.592,
+      "step": 17660
+    },
+    {
+      "epoch": 7.37,
+      "learning_rate": 1e-06,
+      "loss": 0.1824,
+      "step": 17680
+    },
+    {
+      "epoch": 7.37,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.19450555741786957,
+      "eval_runtime": 16.422,
+      "eval_samples_per_second": 1461.453,
+      "eval_steps_per_second": 4.567,
+      "step": 17680
+    },
+    {
+      "epoch": 7.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1715,
+      "step": 17700
+    },
+    {
+      "epoch": 7.38,
+      "eval_accuracy": 0.9337916666666667,
+      "eval_loss": 0.19444973766803741,
+      "eval_runtime": 17.29,
+      "eval_samples_per_second": 1388.083,
+      "eval_steps_per_second": 4.338,
+      "step": 17700
+    },
+    {
+      "epoch": 7.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1228,
+      "step": 17720
+    },
+    {
+      "epoch": 7.38,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.1941564828157425,
+      "eval_runtime": 17.8125,
+      "eval_samples_per_second": 1347.37,
+      "eval_steps_per_second": 4.211,
+      "step": 17720
+    },
+    {
+      "epoch": 7.39,
+      "learning_rate": 1e-06,
+      "loss": 0.1787,
+      "step": 17740
+    },
+    {
+      "epoch": 7.39,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.19363898038864136,
+      "eval_runtime": 16.6565,
+      "eval_samples_per_second": 1440.875,
+      "eval_steps_per_second": 4.503,
+      "step": 17740
+    },
+    {
+      "epoch": 7.4,
+      "learning_rate": 1e-06,
+      "loss": 0.1422,
+      "step": 17760
+    },
+    {
+      "epoch": 7.4,
+      "eval_accuracy": 0.9340833333333334,
+      "eval_loss": 0.19510914385318756,
+      "eval_runtime": 16.7029,
+      "eval_samples_per_second": 1436.875,
+      "eval_steps_per_second": 4.49,
+      "step": 17760
+    },
+    {
+      "epoch": 7.41,
+      "learning_rate": 1e-06,
+      "loss": 0.1541,
+      "step": 17780
+    },
+    {
+      "epoch": 7.41,
+      "eval_accuracy": 0.934,
+      "eval_loss": 0.19492153823375702,
+      "eval_runtime": 16.117,
+      "eval_samples_per_second": 1489.115,
+      "eval_steps_per_second": 4.653,
+      "step": 17780
+    },
+    {
+      "epoch": 7.42,
+      "learning_rate": 1e-06,
+      "loss": 0.188,
+      "step": 17800
+    },
+    {
+      "epoch": 7.42,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.1961117386817932,
+      "eval_runtime": 16.3107,
+      "eval_samples_per_second": 1471.429,
+      "eval_steps_per_second": 4.598,
+      "step": 17800
+    },
+    {
+      "epoch": 7.42,
+      "learning_rate": 1e-06,
+      "loss": 0.1591,
+      "step": 17820
+    },
+    {
+      "epoch": 7.42,
+      "eval_accuracy": 0.93475,
+      "eval_loss": 0.19403943419456482,
+      "eval_runtime": 16.2708,
+      "eval_samples_per_second": 1475.037,
+      "eval_steps_per_second": 4.609,
+      "step": 17820
+    },
+    {
+      "epoch": 7.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1526,
+      "step": 17840
+    },
+    {
+      "epoch": 7.43,
+      "eval_accuracy": 0.933,
+      "eval_loss": 0.19669285416603088,
+      "eval_runtime": 16.4196,
+      "eval_samples_per_second": 1461.668,
+      "eval_steps_per_second": 4.568,
+      "step": 17840
+    },
+    {
+      "epoch": 7.44,
+      "learning_rate": 1e-06,
+      "loss": 0.1988,
+      "step": 17860
+    },
+    {
+      "epoch": 7.44,
+      "eval_accuracy": 0.933,
+      "eval_loss": 0.19575349986553192,
+      "eval_runtime": 17.4894,
+      "eval_samples_per_second": 1372.256,
+      "eval_steps_per_second": 4.288,
+      "step": 17860
+    },
+    {
+      "epoch": 7.45,
+      "learning_rate": 1e-06,
+      "loss": 0.1471,
+      "step": 17880
+    },
+    {
+      "epoch": 7.45,
+      "eval_accuracy": 0.9339166666666666,
+      "eval_loss": 0.19493769109249115,
+      "eval_runtime": 17.3622,
+      "eval_samples_per_second": 1382.312,
+      "eval_steps_per_second": 4.32,
+      "step": 17880
+    },
+    {
+      "epoch": 7.46,
+      "learning_rate": 1e-06,
+      "loss": 0.1631,
+      "step": 17900
+    },
+    {
+      "epoch": 7.46,
+      "eval_accuracy": 0.9326666666666666,
+      "eval_loss": 0.1954115778207779,
+      "eval_runtime": 16.3935,
+      "eval_samples_per_second": 1463.997,
+      "eval_steps_per_second": 4.575,
+      "step": 17900
+    },
+    {
+      "epoch": 7.47,
+      "learning_rate": 1e-06,
+      "loss": 0.2076,
+      "step": 17920
+    },
+    {
+      "epoch": 7.47,
+      "eval_accuracy": 0.9314583333333334,
+      "eval_loss": 0.20004239678382874,
+      "eval_runtime": 16.8491,
+      "eval_samples_per_second": 1424.406,
+      "eval_steps_per_second": 4.451,
+      "step": 17920
+    },
+    {
+      "epoch": 7.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1541,
+      "step": 17940
+    },
+    {
+      "epoch": 7.47,
+      "eval_accuracy": 0.9330833333333334,
+      "eval_loss": 0.1946871429681778,
+      "eval_runtime": 16.9255,
+      "eval_samples_per_second": 1417.978,
+      "eval_steps_per_second": 4.431,
+      "step": 17940
+    },
+    {
+      "epoch": 7.48,
+      "learning_rate": 1e-06,
+      "loss": 0.2057,
+      "step": 17960
+    },
+    {
+      "epoch": 7.48,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.194530189037323,
+      "eval_runtime": 17.2774,
+      "eval_samples_per_second": 1389.094,
+      "eval_steps_per_second": 4.341,
+      "step": 17960
+    },
+    {
+      "epoch": 7.49,
+      "learning_rate": 1e-06,
+      "loss": 0.1721,
+      "step": 17980
+    },
+    {
+      "epoch": 7.49,
+      "eval_accuracy": 0.934,
+      "eval_loss": 0.19205187261104584,
+      "eval_runtime": 16.4737,
+      "eval_samples_per_second": 1456.864,
+      "eval_steps_per_second": 4.553,
+      "step": 17980
+    },
+    {
+      "epoch": 7.5,
+      "learning_rate": 1e-06,
+      "loss": 0.1528,
+      "step": 18000
+    },
+    {
+      "epoch": 7.5,
+      "eval_accuracy": 0.934375,
+      "eval_loss": 0.19169993698596954,
+      "eval_runtime": 16.6059,
+      "eval_samples_per_second": 1445.27,
+      "eval_steps_per_second": 4.516,
+      "step": 18000
+    },
+    {
+      "epoch": 7.51,
+      "learning_rate": 1e-06,
+      "loss": 0.1758,
+      "step": 18020
+    },
+    {
+      "epoch": 7.51,
+      "eval_accuracy": 0.9336666666666666,
+      "eval_loss": 0.19537770748138428,
+      "eval_runtime": 16.2403,
+      "eval_samples_per_second": 1477.804,
+      "eval_steps_per_second": 4.618,
+      "step": 18020
+    },
+    {
+      "epoch": 7.52,
+      "learning_rate": 1e-06,
+      "loss": 0.1757,
+      "step": 18040
+    },
+    {
+      "epoch": 7.52,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.19484874606132507,
+      "eval_runtime": 16.9236,
+      "eval_samples_per_second": 1418.134,
+      "eval_steps_per_second": 4.432,
+      "step": 18040
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1499,
+      "step": 18060
+    },
+    {
+      "epoch": 7.53,
+      "eval_accuracy": 0.9348333333333333,
+      "eval_loss": 0.19193783402442932,
+      "eval_runtime": 16.6031,
+      "eval_samples_per_second": 1445.509,
+      "eval_steps_per_second": 4.517,
+      "step": 18060
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 1e-06,
+      "loss": 0.2056,
+      "step": 18080
+    },
+    {
+      "epoch": 7.53,
+      "eval_accuracy": 0.9350416666666667,
+      "eval_loss": 0.1917405128479004,
+      "eval_runtime": 17.4615,
+      "eval_samples_per_second": 1374.451,
+      "eval_steps_per_second": 4.295,
+      "step": 18080
+    },
+    {
+      "epoch": 7.54,
+      "learning_rate": 1e-06,
+      "loss": 0.1646,
+      "step": 18100
+    },
+    {
+      "epoch": 7.54,
+      "eval_accuracy": 0.9340833333333334,
+      "eval_loss": 0.1928911805152893,
+      "eval_runtime": 17.1226,
+      "eval_samples_per_second": 1401.658,
+      "eval_steps_per_second": 4.38,
+      "step": 18100
+    },
+    {
+      "epoch": 7.55,
+      "learning_rate": 1e-06,
+      "loss": 0.1312,
+      "step": 18120
+    },
+    {
+      "epoch": 7.55,
+      "eval_accuracy": 0.9336666666666666,
+      "eval_loss": 0.19554804265499115,
+      "eval_runtime": 16.8658,
+      "eval_samples_per_second": 1422.998,
+      "eval_steps_per_second": 4.447,
+      "step": 18120
+    },
+    {
+      "epoch": 7.56,
+      "learning_rate": 1e-06,
+      "loss": 0.1759,
+      "step": 18140
+    },
+    {
+      "epoch": 7.56,
+      "eval_accuracy": 0.9339166666666666,
+      "eval_loss": 0.1938326209783554,
+      "eval_runtime": 16.5184,
+      "eval_samples_per_second": 1452.925,
+      "eval_steps_per_second": 4.54,
+      "step": 18140
+    },
+    {
+      "epoch": 7.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1907,
+      "step": 18160
+    },
+    {
+      "epoch": 7.57,
+      "eval_accuracy": 0.934125,
+      "eval_loss": 0.19356830418109894,
+      "eval_runtime": 16.2934,
+      "eval_samples_per_second": 1472.985,
+      "eval_steps_per_second": 4.603,
+      "step": 18160
+    },
+    {
+      "epoch": 7.58,
+      "learning_rate": 1e-06,
+      "loss": 0.156,
+      "step": 18180
+    },
+    {
+      "epoch": 7.58,
+      "eval_accuracy": 0.932625,
+      "eval_loss": 0.19519713521003723,
+      "eval_runtime": 16.5186,
+      "eval_samples_per_second": 1452.905,
+      "eval_steps_per_second": 4.54,
+      "step": 18180
+    },
+    {
+      "epoch": 7.58,
+      "learning_rate": 1e-06,
+      "loss": 0.1656,
+      "step": 18200
+    },
+    {
+      "epoch": 7.58,
+      "eval_accuracy": 0.934,
+      "eval_loss": 0.19366022944450378,
+      "eval_runtime": 17.0576,
+      "eval_samples_per_second": 1406.998,
+      "eval_steps_per_second": 4.397,
+      "step": 18200
+    },
+    {
+      "epoch": 7.59,
+      "learning_rate": 1e-06,
+      "loss": 0.1321,
+      "step": 18220
+    },
+    {
+      "epoch": 7.59,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.19374357163906097,
+      "eval_runtime": 16.6623,
+      "eval_samples_per_second": 1440.375,
+      "eval_steps_per_second": 4.501,
+      "step": 18220
+    },
+    {
+      "epoch": 7.6,
+      "learning_rate": 1e-06,
+      "loss": 0.1367,
+      "step": 18240
+    },
+    {
+      "epoch": 7.6,
+      "eval_accuracy": 0.934875,
+      "eval_loss": 0.1920449137687683,
+      "eval_runtime": 17.6693,
+      "eval_samples_per_second": 1358.287,
+      "eval_steps_per_second": 4.245,
+      "step": 18240
+    },
+    {
+      "epoch": 7.61,
+      "learning_rate": 1e-06,
+      "loss": 0.1935,
+      "step": 18260
+    },
+    {
+      "epoch": 7.61,
+      "eval_accuracy": 0.9331666666666667,
+      "eval_loss": 0.19506241381168365,
+      "eval_runtime": 16.8545,
+      "eval_samples_per_second": 1423.948,
+      "eval_steps_per_second": 4.45,
+      "step": 18260
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1625,
+      "step": 18280
+    },
+    {
+      "epoch": 7.62,
+      "eval_accuracy": 0.9312083333333333,
+      "eval_loss": 0.20022885501384735,
+      "eval_runtime": 15.6569,
+      "eval_samples_per_second": 1532.875,
+      "eval_steps_per_second": 4.79,
+      "step": 18280
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1959,
+      "step": 18300
+    },
+    {
+      "epoch": 7.62,
+      "eval_accuracy": 0.932,
+      "eval_loss": 0.19843071699142456,
+      "eval_runtime": 17.5133,
+      "eval_samples_per_second": 1370.39,
+      "eval_steps_per_second": 4.282,
+      "step": 18300
+    },
+    {
+      "epoch": 7.63,
+      "learning_rate": 1e-06,
+      "loss": 0.1523,
+      "step": 18320
+    },
+    {
+      "epoch": 7.63,
+      "eval_accuracy": 0.9332916666666666,
+      "eval_loss": 0.19627678394317627,
+      "eval_runtime": 17.5177,
+      "eval_samples_per_second": 1370.041,
+      "eval_steps_per_second": 4.281,
+      "step": 18320
+    },
+    {
+      "epoch": 7.64,
+      "learning_rate": 1e-06,
+      "loss": 0.1248,
+      "step": 18340
+    },
+    {
+      "epoch": 7.64,
+      "eval_accuracy": 0.9340416666666667,
+      "eval_loss": 0.1947159618139267,
+      "eval_runtime": 17.1988,
+      "eval_samples_per_second": 1395.448,
+      "eval_steps_per_second": 4.361,
+      "step": 18340
+    },
+    {
+      "epoch": 7.65,
+      "learning_rate": 1e-06,
+      "loss": 0.1575,
+      "step": 18360
+    },
+    {
+      "epoch": 7.65,
+      "eval_accuracy": 0.9345416666666667,
+      "eval_loss": 0.1939171701669693,
+      "eval_runtime": 16.9923,
+      "eval_samples_per_second": 1412.401,
+      "eval_steps_per_second": 4.414,
+      "step": 18360
+    },
+    {
+      "epoch": 7.66,
+      "learning_rate": 1e-06,
+      "loss": 0.145,
+      "step": 18380
+    },
+    {
+      "epoch": 7.66,
+      "eval_accuracy": 0.9348333333333333,
+      "eval_loss": 0.19376207888126373,
+      "eval_runtime": 16.8495,
+      "eval_samples_per_second": 1424.374,
+      "eval_steps_per_second": 4.451,
+      "step": 18380
+    },
+    {
+      "epoch": 7.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1981,
+      "step": 18400
+    },
+    {
+      "epoch": 7.67,
+      "eval_accuracy": 0.9344166666666667,
+      "eval_loss": 0.1934199035167694,
+      "eval_runtime": 16.679,
+      "eval_samples_per_second": 1438.939,
+      "eval_steps_per_second": 4.497,
+      "step": 18400
+    },
+    {
+      "epoch": 7.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1237,
+      "step": 18420
+    },
+    {
+      "epoch": 7.67,
+      "eval_accuracy": 0.9329166666666666,
+      "eval_loss": 0.19676241278648376,
+      "eval_runtime": 16.3441,
+      "eval_samples_per_second": 1468.422,
+      "eval_steps_per_second": 4.589,
+      "step": 18420
+    },
+    {
+      "epoch": 7.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1649,
+      "step": 18440
+    },
+    {
+      "epoch": 7.68,
+      "eval_accuracy": 0.9333333333333333,
+      "eval_loss": 0.19533593952655792,
+      "eval_runtime": 16.5476,
+      "eval_samples_per_second": 1450.361,
+      "eval_steps_per_second": 4.532,
+      "step": 18440
+    },
+    {
+      "epoch": 7.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1832,
+      "step": 18460
+    },
+    {
+      "epoch": 7.69,
+      "eval_accuracy": 0.9344166666666667,
+      "eval_loss": 0.19351038336753845,
+      "eval_runtime": 16.5442,
+      "eval_samples_per_second": 1450.662,
+      "eval_steps_per_second": 4.533,
+      "step": 18460
+    },
+    {
+      "epoch": 7.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1337,
+      "step": 18480
+    },
+    {
+      "epoch": 7.7,
+      "eval_accuracy": 0.9326666666666666,
+      "eval_loss": 0.197190523147583,
+      "eval_runtime": 16.6092,
+      "eval_samples_per_second": 1444.979,
+      "eval_steps_per_second": 4.516,
+      "step": 18480
+    },
+    {
+      "epoch": 7.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1407,
+      "step": 18500
+    },
+    {
+      "epoch": 7.71,
+      "eval_accuracy": 0.9353333333333333,
+      "eval_loss": 0.19291214644908905,
+      "eval_runtime": 16.5801,
+      "eval_samples_per_second": 1447.52,
+      "eval_steps_per_second": 4.523,
+      "step": 18500
+    },
+    {
+      "epoch": 7.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1489,
+      "step": 18520
+    },
+    {
+      "epoch": 7.72,
+      "eval_accuracy": 0.9338333333333333,
+      "eval_loss": 0.19552947580814362,
+      "eval_runtime": 16.9798,
+      "eval_samples_per_second": 1413.444,
+      "eval_steps_per_second": 4.417,
+      "step": 18520
+    },
+    {
+      "epoch": 7.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1603,
+      "step": 18540
+    },
+    {
+      "epoch": 7.72,
+      "eval_accuracy": 0.9324583333333333,
+      "eval_loss": 0.19699038565158844,
+      "eval_runtime": 17.4729,
+      "eval_samples_per_second": 1373.558,
+      "eval_steps_per_second": 4.292,
+      "step": 18540
+    },
+    {
+      "epoch": 7.73,
+      "learning_rate": 1e-06,
+      "loss": 0.1468,
+      "step": 18560
+    },
+    {
+      "epoch": 7.73,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.19590060412883759,
+      "eval_runtime": 16.5983,
+      "eval_samples_per_second": 1445.927,
+      "eval_steps_per_second": 4.519,
+      "step": 18560
+    },
+    {
+      "epoch": 7.74,
+      "learning_rate": 1e-06,
+      "loss": 0.213,
+      "step": 18580
+    },
+    {
+      "epoch": 7.74,
+      "eval_accuracy": 0.934125,
+      "eval_loss": 0.1953480988740921,
+      "eval_runtime": 15.9179,
+      "eval_samples_per_second": 1507.733,
+      "eval_steps_per_second": 4.712,
+      "step": 18580
+    },
+    {
+      "epoch": 7.75,
+      "learning_rate": 1e-06,
+      "loss": 0.2005,
+      "step": 18600
+    },
+    {
+      "epoch": 7.75,
+      "eval_accuracy": 0.934875,
+      "eval_loss": 0.1933307647705078,
+      "eval_runtime": 16.5952,
+      "eval_samples_per_second": 1446.199,
+      "eval_steps_per_second": 4.519,
+      "step": 18600
+    },
+    {
+      "epoch": 7.76,
+      "learning_rate": 1e-06,
+      "loss": 0.1741,
+      "step": 18620
+    },
+    {
+      "epoch": 7.76,
+      "eval_accuracy": 0.9349583333333333,
+      "eval_loss": 0.19205695390701294,
+      "eval_runtime": 16.7425,
+      "eval_samples_per_second": 1433.48,
+      "eval_steps_per_second": 4.48,
+      "step": 18620
+    },
+    {
+      "epoch": 7.77,
+      "learning_rate": 1e-06,
+      "loss": 0.1618,
+      "step": 18640
+    },
+    {
+      "epoch": 7.77,
+      "eval_accuracy": 0.93375,
+      "eval_loss": 0.19280995428562164,
+      "eval_runtime": 16.6122,
+      "eval_samples_per_second": 1444.719,
+      "eval_steps_per_second": 4.515,
+      "step": 18640
+    },
+    {
+      "epoch": 7.78,
+      "learning_rate": 1e-06,
+      "loss": 0.2304,
+      "step": 18660
+    },
+    {
+      "epoch": 7.78,
+      "eval_accuracy": 0.93325,
+      "eval_loss": 0.1952376365661621,
+      "eval_runtime": 17.2502,
+      "eval_samples_per_second": 1391.287,
+      "eval_steps_per_second": 4.348,
+      "step": 18660
+    },
+    {
+      "epoch": 7.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1729,
+      "step": 18680
+    },
+    {
+      "epoch": 7.78,
+      "eval_accuracy": 0.933875,
+      "eval_loss": 0.1918487548828125,
+      "eval_runtime": 16.6725,
+      "eval_samples_per_second": 1439.495,
+      "eval_steps_per_second": 4.498,
+      "step": 18680
+    },
+    {
+      "epoch": 7.79,
+      "learning_rate": 1e-06,
+      "loss": 0.1632,
+      "step": 18700
+    },
+    {
+      "epoch": 7.79,
+      "eval_accuracy": 0.9339583333333333,
+      "eval_loss": 0.193745419383049,
+      "eval_runtime": 17.0655,
+      "eval_samples_per_second": 1406.35,
+      "eval_steps_per_second": 4.395,
+      "step": 18700
+    },
+    {
+      "epoch": 7.8,
+      "learning_rate": 1e-06,
+      "loss": 0.1381,
+      "step": 18720
+    },
+    {
+      "epoch": 7.8,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.19378027319908142,
+      "eval_runtime": 16.1642,
+      "eval_samples_per_second": 1484.763,
+      "eval_steps_per_second": 4.64,
+      "step": 18720
+    },
+    {
+      "epoch": 7.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1334,
+      "step": 18740
+    },
+    {
+      "epoch": 7.81,
+      "eval_accuracy": 0.9339166666666666,
+      "eval_loss": 0.19540290534496307,
+      "eval_runtime": 17.2682,
+      "eval_samples_per_second": 1389.836,
+      "eval_steps_per_second": 4.343,
+      "step": 18740
+    },
+    {
+      "epoch": 7.82,
+      "learning_rate": 1e-06,
+      "loss": 0.1706,
+      "step": 18760
+    },
+    {
+      "epoch": 7.82,
+      "eval_accuracy": 0.9347083333333334,
+      "eval_loss": 0.19170990586280823,
+      "eval_runtime": 17.5111,
+      "eval_samples_per_second": 1370.562,
+      "eval_steps_per_second": 4.283,
+      "step": 18760
+    },
+    {
+      "epoch": 7.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1774,
+      "step": 18780
+    },
+    {
+      "epoch": 7.83,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.19449305534362793,
+      "eval_runtime": 18.1073,
+      "eval_samples_per_second": 1325.429,
+      "eval_steps_per_second": 4.142,
+      "step": 18780
+    },
+    {
+      "epoch": 7.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1891,
+      "step": 18800
+    },
+    {
+      "epoch": 7.83,
+      "eval_accuracy": 0.935125,
+      "eval_loss": 0.19196221232414246,
+      "eval_runtime": 16.1253,
+      "eval_samples_per_second": 1488.341,
+      "eval_steps_per_second": 4.651,
+      "step": 18800
+    },
+    {
+      "epoch": 7.84,
+      "learning_rate": 1e-06,
+      "loss": 0.1949,
+      "step": 18820
+    },
+    {
+      "epoch": 7.84,
+      "eval_accuracy": 0.9347083333333334,
+      "eval_loss": 0.1915530413389206,
+      "eval_runtime": 16.928,
+      "eval_samples_per_second": 1417.773,
+      "eval_steps_per_second": 4.431,
+      "step": 18820
+    },
+    {
+      "epoch": 7.85,
+      "learning_rate": 1e-06,
+      "loss": 0.1511,
+      "step": 18840
+    },
+    {
+      "epoch": 7.85,
+      "eval_accuracy": 0.9354583333333333,
+      "eval_loss": 0.1908595710992813,
+      "eval_runtime": 16.4669,
+      "eval_samples_per_second": 1457.473,
+      "eval_steps_per_second": 4.555,
+      "step": 18840
+    },
+    {
+      "epoch": 7.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1501,
+      "step": 18860
+    },
+    {
+      "epoch": 7.86,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.19139742851257324,
+      "eval_runtime": 16.7079,
+      "eval_samples_per_second": 1436.446,
+      "eval_steps_per_second": 4.489,
+      "step": 18860
+    },
+    {
+      "epoch": 7.87,
+      "learning_rate": 1e-06,
+      "loss": 0.1367,
+      "step": 18880
+    },
+    {
+      "epoch": 7.87,
+      "eval_accuracy": 0.9359166666666666,
+      "eval_loss": 0.19081765413284302,
+      "eval_runtime": 16.8019,
+      "eval_samples_per_second": 1428.413,
+      "eval_steps_per_second": 4.464,
+      "step": 18880
+    },
+    {
+      "epoch": 7.88,
+      "learning_rate": 1e-06,
+      "loss": 0.179,
+      "step": 18900
+    },
+    {
+      "epoch": 7.88,
+      "eval_accuracy": 0.936,
+      "eval_loss": 0.1911996752023697,
+      "eval_runtime": 17.4687,
+      "eval_samples_per_second": 1373.888,
+      "eval_steps_per_second": 4.293,
+      "step": 18900
+    },
+    {
+      "epoch": 7.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1737,
+      "step": 18920
+    },
+    {
+      "epoch": 7.88,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.19142092764377594,
+      "eval_runtime": 17.9819,
+      "eval_samples_per_second": 1334.677,
+      "eval_steps_per_second": 4.171,
+      "step": 18920
+    },
+    {
+      "epoch": 7.89,
+      "learning_rate": 1e-06,
+      "loss": 0.1821,
+      "step": 18940
+    },
+    {
+      "epoch": 7.89,
+      "eval_accuracy": 0.9349166666666666,
+      "eval_loss": 0.19184531271457672,
+      "eval_runtime": 17.6224,
+      "eval_samples_per_second": 1361.903,
+      "eval_steps_per_second": 4.256,
+      "step": 18940
+    },
+    {
+      "epoch": 7.9,
+      "learning_rate": 1e-06,
+      "loss": 0.1611,
+      "step": 18960
+    },
+    {
+      "epoch": 7.9,
+      "eval_accuracy": 0.9360416666666667,
+      "eval_loss": 0.19008147716522217,
+      "eval_runtime": 16.936,
+      "eval_samples_per_second": 1417.103,
+      "eval_steps_per_second": 4.428,
+      "step": 18960
+    },
+    {
+      "epoch": 7.91,
+      "learning_rate": 1e-06,
+      "loss": 0.1389,
+      "step": 18980
+    },
+    {
+      "epoch": 7.91,
+      "eval_accuracy": 0.935875,
+      "eval_loss": 0.1900467723608017,
+      "eval_runtime": 17.5135,
+      "eval_samples_per_second": 1370.37,
+      "eval_steps_per_second": 4.282,
+      "step": 18980
+    },
+    {
+      "epoch": 7.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1751,
+      "step": 19000
+    },
+    {
+      "epoch": 7.92,
+      "eval_accuracy": 0.936875,
+      "eval_loss": 0.19042351841926575,
+      "eval_runtime": 17.8277,
+      "eval_samples_per_second": 1346.22,
+      "eval_steps_per_second": 4.207,
+      "step": 19000
+    },
+    {
+      "epoch": 7.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1955,
+      "step": 19020
+    },
+    {
+      "epoch": 7.92,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.19257663190364838,
+      "eval_runtime": 17.5409,
+      "eval_samples_per_second": 1368.229,
+      "eval_steps_per_second": 4.276,
+      "step": 19020
+    },
+    {
+      "epoch": 7.93,
+      "learning_rate": 1e-06,
+      "loss": 0.1762,
+      "step": 19040
+    },
+    {
+      "epoch": 7.93,
+      "eval_accuracy": 0.9362083333333333,
+      "eval_loss": 0.19056767225265503,
+      "eval_runtime": 17.5352,
+      "eval_samples_per_second": 1368.676,
+      "eval_steps_per_second": 4.277,
+      "step": 19040
+    },
+    {
+      "epoch": 7.94,
+      "learning_rate": 1e-06,
+      "loss": 0.1417,
+      "step": 19060
+    },
+    {
+      "epoch": 7.94,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.19115488231182098,
+      "eval_runtime": 17.4988,
+      "eval_samples_per_second": 1371.52,
+      "eval_steps_per_second": 4.286,
+      "step": 19060
+    },
+    {
+      "epoch": 7.95,
+      "learning_rate": 1e-06,
+      "loss": 0.1602,
+      "step": 19080
+    },
+    {
+      "epoch": 7.95,
+      "eval_accuracy": 0.9348333333333333,
+      "eval_loss": 0.19205690920352936,
+      "eval_runtime": 17.3557,
+      "eval_samples_per_second": 1382.829,
+      "eval_steps_per_second": 4.321,
+      "step": 19080
+    },
+    {
+      "epoch": 7.96,
+      "learning_rate": 1e-06,
+      "loss": 0.1355,
+      "step": 19100
+    },
+    {
+      "epoch": 7.96,
+      "eval_accuracy": 0.9320833333333334,
+      "eval_loss": 0.1974695324897766,
+      "eval_runtime": 17.4599,
+      "eval_samples_per_second": 1374.582,
+      "eval_steps_per_second": 4.296,
+      "step": 19100
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1488,
+      "step": 19120
+    },
+    {
+      "epoch": 7.97,
+      "eval_accuracy": 0.9339166666666666,
+      "eval_loss": 0.19483338296413422,
+      "eval_runtime": 17.4251,
+      "eval_samples_per_second": 1377.325,
+      "eval_steps_per_second": 4.304,
+      "step": 19120
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 1e-06,
+      "loss": 0.2128,
+      "step": 19140
+    },
+    {
+      "epoch": 7.97,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.19111751019954681,
+      "eval_runtime": 17.5678,
+      "eval_samples_per_second": 1366.137,
+      "eval_steps_per_second": 4.269,
+      "step": 19140
+    },
+    {
+      "epoch": 7.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1223,
+      "step": 19160
+    },
+    {
+      "epoch": 7.98,
+      "eval_accuracy": 0.9352083333333333,
+      "eval_loss": 0.19168196618556976,
+      "eval_runtime": 17.4393,
+      "eval_samples_per_second": 1376.199,
+      "eval_steps_per_second": 4.301,
+      "step": 19160
+    },
+    {
+      "epoch": 7.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1564,
+      "step": 19180
+    },
+    {
+      "epoch": 7.99,
+      "eval_accuracy": 0.934,
+      "eval_loss": 0.19453711807727814,
+      "eval_runtime": 16.8108,
+      "eval_samples_per_second": 1427.651,
+      "eval_steps_per_second": 4.461,
+      "step": 19180
+    },
+    {
+      "epoch": 8.0,
+      "learning_rate": 1e-06,
+      "loss": 0.1961,
+      "step": 19200
+    },
+    {
+      "epoch": 8.0,
+      "eval_accuracy": 0.934125,
+      "eval_loss": 0.19330745935440063,
+      "eval_runtime": 16.8924,
+      "eval_samples_per_second": 1420.757,
+      "eval_steps_per_second": 4.44,
+      "step": 19200
+    },
+    {
+      "epoch": 8.01,
+      "learning_rate": 1e-06,
+      "loss": 0.1791,
+      "step": 19220
+    },
+    {
+      "epoch": 8.01,
+      "eval_accuracy": 0.9350416666666667,
+      "eval_loss": 0.19044369459152222,
+      "eval_runtime": 16.3873,
+      "eval_samples_per_second": 1464.552,
+      "eval_steps_per_second": 4.577,
+      "step": 19220
+    },
+    {
+      "epoch": 8.02,
+      "learning_rate": 1e-06,
+      "loss": 0.1466,
+      "step": 19240
+    },
+    {
+      "epoch": 8.02,
+      "eval_accuracy": 0.935875,
+      "eval_loss": 0.19002403318881989,
+      "eval_runtime": 16.82,
+      "eval_samples_per_second": 1426.875,
+      "eval_steps_per_second": 4.459,
+      "step": 19240
+    },
+    {
+      "epoch": 8.03,
+      "learning_rate": 1e-06,
+      "loss": 0.135,
+      "step": 19260
+    },
+    {
+      "epoch": 8.03,
+      "eval_accuracy": 0.9357083333333334,
+      "eval_loss": 0.19046556949615479,
+      "eval_runtime": 16.49,
+      "eval_samples_per_second": 1455.424,
+      "eval_steps_per_second": 4.548,
+      "step": 19260
+    },
+    {
+      "epoch": 8.03,
+      "learning_rate": 1e-06,
+      "loss": 0.129,
+      "step": 19280
+    },
+    {
+      "epoch": 8.03,
+      "eval_accuracy": 0.935125,
+      "eval_loss": 0.1913367062807083,
+      "eval_runtime": 17.0223,
+      "eval_samples_per_second": 1409.919,
+      "eval_steps_per_second": 4.406,
+      "step": 19280
+    },
+    {
+      "epoch": 8.04,
+      "learning_rate": 1e-06,
+      "loss": 0.1778,
+      "step": 19300
+    },
+    {
+      "epoch": 8.04,
+      "eval_accuracy": 0.9329166666666666,
+      "eval_loss": 0.1969105303287506,
+      "eval_runtime": 16.1602,
+      "eval_samples_per_second": 1485.13,
+      "eval_steps_per_second": 4.641,
+      "step": 19300
+    },
+    {
+      "epoch": 8.05,
+      "learning_rate": 1e-06,
+      "loss": 0.1362,
+      "step": 19320
+    },
+    {
+      "epoch": 8.05,
+      "eval_accuracy": 0.9340416666666667,
+      "eval_loss": 0.19413326680660248,
+      "eval_runtime": 17.3298,
+      "eval_samples_per_second": 1384.896,
+      "eval_steps_per_second": 4.328,
+      "step": 19320
+    },
+    {
+      "epoch": 8.06,
+      "learning_rate": 1e-06,
+      "loss": 0.157,
+      "step": 19340
+    },
+    {
+      "epoch": 8.06,
+      "eval_accuracy": 0.93275,
+      "eval_loss": 0.19760599732398987,
+      "eval_runtime": 17.4331,
+      "eval_samples_per_second": 1376.69,
+      "eval_steps_per_second": 4.302,
+      "step": 19340
+    },
+    {
+      "epoch": 8.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1544,
+      "step": 19360
+    },
+    {
+      "epoch": 8.07,
+      "eval_accuracy": 0.935125,
+      "eval_loss": 0.1911655217409134,
+      "eval_runtime": 16.3075,
+      "eval_samples_per_second": 1471.715,
+      "eval_steps_per_second": 4.599,
+      "step": 19360
+    },
+    {
+      "epoch": 8.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1566,
+      "step": 19380
+    },
+    {
+      "epoch": 8.07,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.19158610701560974,
+      "eval_runtime": 16.834,
+      "eval_samples_per_second": 1425.685,
+      "eval_steps_per_second": 4.455,
+      "step": 19380
+    },
+    {
+      "epoch": 8.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1572,
+      "step": 19400
+    },
+    {
+      "epoch": 8.08,
+      "eval_accuracy": 0.93525,
+      "eval_loss": 0.19257643818855286,
+      "eval_runtime": 16.0408,
+      "eval_samples_per_second": 1496.18,
+      "eval_steps_per_second": 4.676,
+      "step": 19400
+    },
+    {
+      "epoch": 8.09,
+      "learning_rate": 1e-06,
+      "loss": 0.1077,
+      "step": 19420
+    },
+    {
+      "epoch": 8.09,
+      "eval_accuracy": 0.9349583333333333,
+      "eval_loss": 0.1925935298204422,
+      "eval_runtime": 17.6153,
+      "eval_samples_per_second": 1362.454,
+      "eval_steps_per_second": 4.258,
+      "step": 19420
+    },
+    {
+      "epoch": 8.1,
+      "learning_rate": 1e-06,
+      "loss": 0.1365,
+      "step": 19440
+    },
+    {
+      "epoch": 8.1,
+      "eval_accuracy": 0.9346666666666666,
+      "eval_loss": 0.19453419744968414,
+      "eval_runtime": 16.5289,
+      "eval_samples_per_second": 1451.999,
+      "eval_steps_per_second": 4.537,
+      "step": 19440
+    },
+    {
+      "epoch": 8.11,
+      "learning_rate": 1e-06,
+      "loss": 0.1526,
+      "step": 19460
+    },
+    {
+      "epoch": 8.11,
+      "eval_accuracy": 0.9355833333333333,
+      "eval_loss": 0.19276762008666992,
+      "eval_runtime": 16.3532,
+      "eval_samples_per_second": 1467.604,
+      "eval_steps_per_second": 4.586,
+      "step": 19460
+    },
+    {
+      "epoch": 8.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1794,
+      "step": 19480
+    },
+    {
+      "epoch": 8.12,
+      "eval_accuracy": 0.9357083333333334,
+      "eval_loss": 0.19133615493774414,
+      "eval_runtime": 16.7591,
+      "eval_samples_per_second": 1432.06,
+      "eval_steps_per_second": 4.475,
+      "step": 19480
+    },
+    {
+      "epoch": 8.12,
+      "learning_rate": 1e-06,
+      "loss": 0.127,
+      "step": 19500
+    },
+    {
+      "epoch": 8.12,
+      "eval_accuracy": 0.9350833333333334,
+      "eval_loss": 0.19240260124206543,
+      "eval_runtime": 17.3908,
+      "eval_samples_per_second": 1380.04,
+      "eval_steps_per_second": 4.313,
+      "step": 19500
+    },
+    {
+      "epoch": 8.13,
+      "learning_rate": 1e-06,
+      "loss": 0.1636,
+      "step": 19520
+    },
+    {
+      "epoch": 8.13,
+      "eval_accuracy": 0.9345,
+      "eval_loss": 0.19413629174232483,
+      "eval_runtime": 17.1984,
+      "eval_samples_per_second": 1395.482,
+      "eval_steps_per_second": 4.361,
+      "step": 19520
+    },
+    {
+      "epoch": 8.14,
+      "learning_rate": 1e-06,
+      "loss": 0.1826,
+      "step": 19540
+    },
+    {
+      "epoch": 8.14,
+      "eval_accuracy": 0.934625,
+      "eval_loss": 0.19350507855415344,
+      "eval_runtime": 16.3498,
+      "eval_samples_per_second": 1467.911,
+      "eval_steps_per_second": 4.587,
+      "step": 19540
+    },
+    {
+      "epoch": 8.15,
+      "learning_rate": 1e-06,
+      "loss": 0.1524,
+      "step": 19560
+    },
+    {
+      "epoch": 8.15,
+      "eval_accuracy": 0.933375,
+      "eval_loss": 0.19596153497695923,
+      "eval_runtime": 16.6776,
+      "eval_samples_per_second": 1439.057,
+      "eval_steps_per_second": 4.497,
+      "step": 19560
+    },
+    {
+      "epoch": 8.16,
+      "learning_rate": 1e-06,
+      "loss": 0.1208,
+      "step": 19580
+    },
+    {
+      "epoch": 8.16,
+      "eval_accuracy": 0.9338333333333333,
+      "eval_loss": 0.19459760189056396,
+      "eval_runtime": 16.3642,
+      "eval_samples_per_second": 1466.618,
+      "eval_steps_per_second": 4.583,
+      "step": 19580
+    },
+    {
+      "epoch": 8.17,
+      "learning_rate": 1e-06,
+      "loss": 0.1516,
+      "step": 19600
+    },
+    {
+      "epoch": 8.17,
+      "eval_accuracy": 0.9324583333333333,
+      "eval_loss": 0.19873768091201782,
+      "eval_runtime": 16.1804,
+      "eval_samples_per_second": 1483.275,
+      "eval_steps_per_second": 4.635,
+      "step": 19600
+    },
+    {
+      "epoch": 8.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1581,
+      "step": 19620
+    },
+    {
+      "epoch": 8.18,
+      "eval_accuracy": 0.935375,
+      "eval_loss": 0.19273337721824646,
+      "eval_runtime": 16.628,
+      "eval_samples_per_second": 1443.351,
+      "eval_steps_per_second": 4.51,
+      "step": 19620
+    },
+    {
+      "epoch": 8.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1537,
+      "step": 19640
+    },
+    {
+      "epoch": 8.18,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.1912853717803955,
+      "eval_runtime": 16.501,
+      "eval_samples_per_second": 1454.455,
+      "eval_steps_per_second": 4.545,
+      "step": 19640
+    },
+    {
+      "epoch": 8.19,
+      "learning_rate": 1e-06,
+      "loss": 0.1165,
+      "step": 19660
+    },
+    {
+      "epoch": 8.19,
+      "eval_accuracy": 0.9357083333333334,
+      "eval_loss": 0.19141745567321777,
+      "eval_runtime": 16.8771,
+      "eval_samples_per_second": 1422.043,
+      "eval_steps_per_second": 4.444,
+      "step": 19660
+    },
+    {
+      "epoch": 8.2,
+      "learning_rate": 1e-06,
+      "loss": 0.1374,
+      "step": 19680
+    },
+    {
+      "epoch": 8.2,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.1914907842874527,
+      "eval_runtime": 15.9508,
+      "eval_samples_per_second": 1504.626,
+      "eval_steps_per_second": 4.702,
+      "step": 19680
+    },
+    {
+      "epoch": 8.21,
+      "learning_rate": 1e-06,
+      "loss": 0.1806,
+      "step": 19700
+    },
+    {
+      "epoch": 8.21,
+      "eval_accuracy": 0.9340416666666667,
+      "eval_loss": 0.1957368105649948,
+      "eval_runtime": 17.0085,
+      "eval_samples_per_second": 1411.059,
+      "eval_steps_per_second": 4.41,
+      "step": 19700
+    },
+    {
+      "epoch": 8.22,
+      "learning_rate": 1e-06,
+      "loss": 0.2198,
+      "step": 19720
+    },
+    {
+      "epoch": 8.22,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.19590935111045837,
+      "eval_runtime": 17.4075,
+      "eval_samples_per_second": 1378.713,
+      "eval_steps_per_second": 4.308,
+      "step": 19720
+    },
+    {
+      "epoch": 8.22,
+      "learning_rate": 1e-06,
+      "loss": 0.177,
+      "step": 19740
+    },
+    {
+      "epoch": 8.22,
+      "eval_accuracy": 0.93525,
+      "eval_loss": 0.19214990735054016,
+      "eval_runtime": 17.3221,
+      "eval_samples_per_second": 1385.511,
+      "eval_steps_per_second": 4.33,
+      "step": 19740
+    },
+    {
+      "epoch": 8.23,
+      "learning_rate": 1e-06,
+      "loss": 0.1417,
+      "step": 19760
+    },
+    {
+      "epoch": 8.23,
+      "eval_accuracy": 0.9346666666666666,
+      "eval_loss": 0.1939082145690918,
+      "eval_runtime": 17.0832,
+      "eval_samples_per_second": 1404.892,
+      "eval_steps_per_second": 4.39,
+      "step": 19760
+    },
+    {
+      "epoch": 8.24,
+      "learning_rate": 1e-06,
+      "loss": 0.2049,
+      "step": 19780
+    },
+    {
+      "epoch": 8.24,
+      "eval_accuracy": 0.9353333333333333,
+      "eval_loss": 0.19216576218605042,
+      "eval_runtime": 16.5657,
+      "eval_samples_per_second": 1448.776,
+      "eval_steps_per_second": 4.527,
+      "step": 19780
+    },
+    {
+      "epoch": 8.25,
+      "learning_rate": 1e-06,
+      "loss": 0.1704,
+      "step": 19800
+    },
+    {
+      "epoch": 8.25,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.1919444501399994,
+      "eval_runtime": 17.0092,
+      "eval_samples_per_second": 1410.998,
+      "eval_steps_per_second": 4.409,
+      "step": 19800
+    },
+    {
+      "epoch": 8.26,
+      "learning_rate": 1e-06,
+      "loss": 0.1448,
+      "step": 19820
+    },
+    {
+      "epoch": 8.26,
+      "eval_accuracy": 0.9349166666666666,
+      "eval_loss": 0.19339194893836975,
+      "eval_runtime": 16.9285,
+      "eval_samples_per_second": 1417.726,
+      "eval_steps_per_second": 4.43,
+      "step": 19820
+    },
+    {
+      "epoch": 8.27,
+      "learning_rate": 1e-06,
+      "loss": 0.1578,
+      "step": 19840
+    },
+    {
+      "epoch": 8.27,
+      "eval_accuracy": 0.9350833333333334,
+      "eval_loss": 0.19213108718395233,
+      "eval_runtime": 16.9637,
+      "eval_samples_per_second": 1414.783,
+      "eval_steps_per_second": 4.421,
+      "step": 19840
+    },
+    {
+      "epoch": 8.28,
+      "learning_rate": 1e-06,
+      "loss": 0.2108,
+      "step": 19860
+    },
+    {
+      "epoch": 8.28,
+      "eval_accuracy": 0.9352916666666666,
+      "eval_loss": 0.1917588859796524,
+      "eval_runtime": 16.9832,
+      "eval_samples_per_second": 1413.159,
+      "eval_steps_per_second": 4.416,
+      "step": 19860
+    },
+    {
+      "epoch": 8.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1945,
+      "step": 19880
+    },
+    {
+      "epoch": 8.28,
+      "eval_accuracy": 0.9352916666666666,
+      "eval_loss": 0.1911618411540985,
+      "eval_runtime": 17.2548,
+      "eval_samples_per_second": 1390.917,
+      "eval_steps_per_second": 4.347,
+      "step": 19880
+    },
+    {
+      "epoch": 8.29,
+      "learning_rate": 1e-06,
+      "loss": 0.1808,
+      "step": 19900
+    },
+    {
+      "epoch": 8.29,
+      "eval_accuracy": 0.935125,
+      "eval_loss": 0.19140420854091644,
+      "eval_runtime": 16.7213,
+      "eval_samples_per_second": 1435.291,
+      "eval_steps_per_second": 4.485,
+      "step": 19900
+    },
+    {
+      "epoch": 8.3,
+      "learning_rate": 1e-06,
+      "loss": 0.153,
+      "step": 19920
+    },
+    {
+      "epoch": 8.3,
+      "eval_accuracy": 0.9332916666666666,
+      "eval_loss": 0.19377712905406952,
+      "eval_runtime": 17.2963,
+      "eval_samples_per_second": 1387.582,
+      "eval_steps_per_second": 4.336,
+      "step": 19920
+    },
+    {
+      "epoch": 8.31,
+      "learning_rate": 1e-06,
+      "loss": 0.1723,
+      "step": 19940
+    },
+    {
+      "epoch": 8.31,
+      "eval_accuracy": 0.9352083333333333,
+      "eval_loss": 0.19146102666854858,
+      "eval_runtime": 16.9128,
+      "eval_samples_per_second": 1419.047,
+      "eval_steps_per_second": 4.435,
+      "step": 19940
+    },
+    {
+      "epoch": 8.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1505,
+      "step": 19960
+    },
+    {
+      "epoch": 8.32,
+      "eval_accuracy": 0.9335,
+      "eval_loss": 0.1947011649608612,
+      "eval_runtime": 17.0812,
+      "eval_samples_per_second": 1405.052,
+      "eval_steps_per_second": 4.391,
+      "step": 19960
+    },
+    {
+      "epoch": 8.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1784,
+      "step": 19980
+    },
+    {
+      "epoch": 8.32,
+      "eval_accuracy": 0.935,
+      "eval_loss": 0.19037066400051117,
+      "eval_runtime": 16.5401,
+      "eval_samples_per_second": 1451.021,
+      "eval_steps_per_second": 4.534,
+      "step": 19980
+    },
+    {
+      "epoch": 8.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1611,
+      "step": 20000
+    },
+    {
+      "epoch": 8.33,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.1925501823425293,
+      "eval_runtime": 17.2658,
+      "eval_samples_per_second": 1390.034,
+      "eval_steps_per_second": 4.344,
+      "step": 20000
+    },
+    {
+      "epoch": 8.34,
+      "learning_rate": 1e-06,
+      "loss": 0.1505,
+      "step": 20020
+    },
+    {
+      "epoch": 8.34,
+      "eval_accuracy": 0.9349583333333333,
+      "eval_loss": 0.19143855571746826,
+      "eval_runtime": 16.5059,
+      "eval_samples_per_second": 1454.028,
+      "eval_steps_per_second": 4.544,
+      "step": 20020
+    },
+    {
+      "epoch": 8.35,
+      "learning_rate": 1e-06,
+      "loss": 0.148,
+      "step": 20040
+    },
+    {
+      "epoch": 8.35,
+      "eval_accuracy": 0.9352083333333333,
+      "eval_loss": 0.1917656511068344,
+      "eval_runtime": 16.2857,
+      "eval_samples_per_second": 1473.686,
+      "eval_steps_per_second": 4.605,
+      "step": 20040
+    },
+    {
+      "epoch": 8.36,
+      "learning_rate": 1e-06,
+      "loss": 0.1958,
+      "step": 20060
+    },
+    {
+      "epoch": 8.36,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.1908416450023651,
+      "eval_runtime": 16.8917,
+      "eval_samples_per_second": 1420.819,
+      "eval_steps_per_second": 4.44,
+      "step": 20060
+    },
+    {
+      "epoch": 8.37,
+      "learning_rate": 1e-06,
+      "loss": 0.1007,
+      "step": 20080
+    },
+    {
+      "epoch": 8.37,
+      "eval_accuracy": 0.93475,
+      "eval_loss": 0.19260205328464508,
+      "eval_runtime": 16.2197,
+      "eval_samples_per_second": 1479.681,
+      "eval_steps_per_second": 4.624,
+      "step": 20080
+    },
+    {
+      "epoch": 8.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1829,
+      "step": 20100
+    },
+    {
+      "epoch": 8.38,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.19410496950149536,
+      "eval_runtime": 16.7814,
+      "eval_samples_per_second": 1430.153,
+      "eval_steps_per_second": 4.469,
+      "step": 20100
+    },
+    {
+      "epoch": 8.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1413,
+      "step": 20120
+    },
+    {
+      "epoch": 8.38,
+      "eval_accuracy": 0.93475,
+      "eval_loss": 0.19318026304244995,
+      "eval_runtime": 16.4709,
+      "eval_samples_per_second": 1457.115,
+      "eval_steps_per_second": 4.553,
+      "step": 20120
+    },
+    {
+      "epoch": 8.39,
+      "learning_rate": 1e-06,
+      "loss": 0.1769,
+      "step": 20140
+    },
+    {
+      "epoch": 8.39,
+      "eval_accuracy": 0.9340833333333334,
+      "eval_loss": 0.19297830760478973,
+      "eval_runtime": 16.9359,
+      "eval_samples_per_second": 1417.108,
+      "eval_steps_per_second": 4.428,
+      "step": 20140
+    },
+    {
+      "epoch": 8.4,
+      "learning_rate": 1e-06,
+      "loss": 0.1843,
+      "step": 20160
+    },
+    {
+      "epoch": 8.4,
+      "eval_accuracy": 0.9353333333333333,
+      "eval_loss": 0.19084292650222778,
+      "eval_runtime": 17.2429,
+      "eval_samples_per_second": 1391.878,
+      "eval_steps_per_second": 4.35,
+      "step": 20160
+    },
+    {
+      "epoch": 8.41,
+      "learning_rate": 1e-06,
+      "loss": 0.2067,
+      "step": 20180
+    },
+    {
+      "epoch": 8.41,
+      "eval_accuracy": 0.9352916666666666,
+      "eval_loss": 0.19021664559841156,
+      "eval_runtime": 16.663,
+      "eval_samples_per_second": 1440.319,
+      "eval_steps_per_second": 4.501,
+      "step": 20180
+    },
+    {
+      "epoch": 8.42,
+      "learning_rate": 1e-06,
+      "loss": 0.1183,
+      "step": 20200
+    },
+    {
+      "epoch": 8.42,
+      "eval_accuracy": 0.9347083333333334,
+      "eval_loss": 0.19230937957763672,
+      "eval_runtime": 16.3196,
+      "eval_samples_per_second": 1470.622,
+      "eval_steps_per_second": 4.596,
+      "step": 20200
+    },
+    {
+      "epoch": 8.43,
+      "learning_rate": 1e-06,
+      "loss": 0.2065,
+      "step": 20220
+    },
+    {
+      "epoch": 8.43,
+      "eval_accuracy": 0.9350833333333334,
+      "eval_loss": 0.19128607213497162,
+      "eval_runtime": 16.6007,
+      "eval_samples_per_second": 1445.718,
+      "eval_steps_per_second": 4.518,
+      "step": 20220
+    },
+    {
+      "epoch": 8.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1466,
+      "step": 20240
+    },
+    {
+      "epoch": 8.43,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.19124051928520203,
+      "eval_runtime": 17.0198,
+      "eval_samples_per_second": 1410.12,
+      "eval_steps_per_second": 4.407,
+      "step": 20240
+    },
+    {
+      "epoch": 8.44,
+      "learning_rate": 1e-06,
+      "loss": 0.1557,
+      "step": 20260
+    },
+    {
+      "epoch": 8.44,
+      "eval_accuracy": 0.9355833333333333,
+      "eval_loss": 0.19068805873394012,
+      "eval_runtime": 16.4856,
+      "eval_samples_per_second": 1455.816,
+      "eval_steps_per_second": 4.549,
+      "step": 20260
+    },
+    {
+      "epoch": 8.45,
+      "learning_rate": 1e-06,
+      "loss": 0.1437,
+      "step": 20280
+    },
+    {
+      "epoch": 8.45,
+      "eval_accuracy": 0.934875,
+      "eval_loss": 0.1921066790819168,
+      "eval_runtime": 16.6454,
+      "eval_samples_per_second": 1441.842,
+      "eval_steps_per_second": 4.506,
+      "step": 20280
+    },
+    {
+      "epoch": 8.46,
+      "learning_rate": 1e-06,
+      "loss": 0.1567,
+      "step": 20300
+    },
+    {
+      "epoch": 8.46,
+      "eval_accuracy": 0.9345833333333333,
+      "eval_loss": 0.1937917321920395,
+      "eval_runtime": 16.4297,
+      "eval_samples_per_second": 1460.769,
+      "eval_steps_per_second": 4.565,
+      "step": 20300
+    },
+    {
+      "epoch": 8.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1363,
+      "step": 20320
+    },
+    {
+      "epoch": 8.47,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.19153279066085815,
+      "eval_runtime": 16.7967,
+      "eval_samples_per_second": 1428.851,
+      "eval_steps_per_second": 4.465,
+      "step": 20320
+    },
+    {
+      "epoch": 8.47,
+      "learning_rate": 1e-06,
+      "loss": 0.2041,
+      "step": 20340
+    },
+    {
+      "epoch": 8.47,
+      "eval_accuracy": 0.9349166666666666,
+      "eval_loss": 0.19226667284965515,
+      "eval_runtime": 16.8781,
+      "eval_samples_per_second": 1421.958,
+      "eval_steps_per_second": 4.444,
+      "step": 20340
+    },
+    {
+      "epoch": 8.48,
+      "learning_rate": 1e-06,
+      "loss": 0.1656,
+      "step": 20360
+    },
+    {
+      "epoch": 8.48,
+      "eval_accuracy": 0.933875,
+      "eval_loss": 0.19749757647514343,
+      "eval_runtime": 16.5317,
+      "eval_samples_per_second": 1451.756,
+      "eval_steps_per_second": 4.537,
+      "step": 20360
+    },
+    {
+      "epoch": 8.49,
+      "learning_rate": 1e-06,
+      "loss": 0.1774,
+      "step": 20380
+    },
+    {
+      "epoch": 8.49,
+      "eval_accuracy": 0.9335416666666667,
+      "eval_loss": 0.1959267109632492,
+      "eval_runtime": 16.9568,
+      "eval_samples_per_second": 1415.361,
+      "eval_steps_per_second": 4.423,
+      "step": 20380
+    },
+    {
+      "epoch": 8.5,
+      "learning_rate": 1e-06,
+      "loss": 0.1472,
+      "step": 20400
+    },
+    {
+      "epoch": 8.5,
+      "eval_accuracy": 0.9355833333333333,
+      "eval_loss": 0.19094142317771912,
+      "eval_runtime": 16.4125,
+      "eval_samples_per_second": 1462.3,
+      "eval_steps_per_second": 4.57,
+      "step": 20400
+    },
+    {
+      "epoch": 8.51,
+      "learning_rate": 1e-06,
+      "loss": 0.178,
+      "step": 20420
+    },
+    {
+      "epoch": 8.51,
+      "eval_accuracy": 0.9337083333333334,
+      "eval_loss": 0.19597363471984863,
+      "eval_runtime": 16.6336,
+      "eval_samples_per_second": 1442.864,
+      "eval_steps_per_second": 4.509,
+      "step": 20420
+    },
+    {
+      "epoch": 8.52,
+      "learning_rate": 1e-06,
+      "loss": 0.1561,
+      "step": 20440
+    },
+    {
+      "epoch": 8.52,
+      "eval_accuracy": 0.9354583333333333,
+      "eval_loss": 0.19258776307106018,
+      "eval_runtime": 17.0606,
+      "eval_samples_per_second": 1406.748,
+      "eval_steps_per_second": 4.396,
+      "step": 20440
+    },
+    {
+      "epoch": 8.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1726,
+      "step": 20460
+    },
+    {
+      "epoch": 8.53,
+      "eval_accuracy": 0.9355416666666667,
+      "eval_loss": 0.19182823598384857,
+      "eval_runtime": 17.3,
+      "eval_samples_per_second": 1387.283,
+      "eval_steps_per_second": 4.335,
+      "step": 20460
+    },
+    {
+      "epoch": 8.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1465,
+      "step": 20480
+    },
+    {
+      "epoch": 8.53,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.19172848761081696,
+      "eval_runtime": 16.287,
+      "eval_samples_per_second": 1473.569,
+      "eval_steps_per_second": 4.605,
+      "step": 20480
+    },
+    {
+      "epoch": 8.54,
+      "learning_rate": 1e-06,
+      "loss": 0.1504,
+      "step": 20500
+    },
+    {
+      "epoch": 8.54,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.19191785156726837,
+      "eval_runtime": 15.5845,
+      "eval_samples_per_second": 1539.99,
+      "eval_steps_per_second": 4.812,
+      "step": 20500
+    },
+    {
+      "epoch": 8.55,
+      "learning_rate": 1e-06,
+      "loss": 0.1593,
+      "step": 20520
+    },
+    {
+      "epoch": 8.55,
+      "eval_accuracy": 0.9358333333333333,
+      "eval_loss": 0.1908203661441803,
+      "eval_runtime": 16.4463,
+      "eval_samples_per_second": 1459.298,
+      "eval_steps_per_second": 4.56,
+      "step": 20520
+    },
+    {
+      "epoch": 8.56,
+      "learning_rate": 1e-06,
+      "loss": 0.1346,
+      "step": 20540
+    },
+    {
+      "epoch": 8.56,
+      "eval_accuracy": 0.9362916666666666,
+      "eval_loss": 0.19051861763000488,
+      "eval_runtime": 15.646,
+      "eval_samples_per_second": 1533.937,
+      "eval_steps_per_second": 4.794,
+      "step": 20540
+    },
+    {
+      "epoch": 8.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1509,
+      "step": 20560
+    },
+    {
+      "epoch": 8.57,
+      "eval_accuracy": 0.9364166666666667,
+      "eval_loss": 0.18970748782157898,
+      "eval_runtime": 15.4588,
+      "eval_samples_per_second": 1552.512,
+      "eval_steps_per_second": 4.852,
+      "step": 20560
+    },
+    {
+      "epoch": 8.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1419,
+      "step": 20580
+    },
+    {
+      "epoch": 8.57,
+      "eval_accuracy": 0.936375,
+      "eval_loss": 0.18967027962207794,
+      "eval_runtime": 15.484,
+      "eval_samples_per_second": 1549.985,
+      "eval_steps_per_second": 4.844,
+      "step": 20580
+    },
+    {
+      "epoch": 8.58,
+      "learning_rate": 1e-06,
+      "loss": 0.1477,
+      "step": 20600
+    },
+    {
+      "epoch": 8.58,
+      "eval_accuracy": 0.93525,
+      "eval_loss": 0.1920623481273651,
+      "eval_runtime": 15.7145,
+      "eval_samples_per_second": 1527.252,
+      "eval_steps_per_second": 4.773,
+      "step": 20600
+    },
+    {
+      "epoch": 8.59,
+      "learning_rate": 1e-06,
+      "loss": 0.1791,
+      "step": 20620
+    },
+    {
+      "epoch": 8.59,
+      "eval_accuracy": 0.9349583333333333,
+      "eval_loss": 0.19344042241573334,
+      "eval_runtime": 15.3045,
+      "eval_samples_per_second": 1568.164,
+      "eval_steps_per_second": 4.901,
+      "step": 20620
+    },
+    {
+      "epoch": 8.6,
+      "learning_rate": 1e-06,
+      "loss": 0.1848,
+      "step": 20640
+    },
+    {
+      "epoch": 8.6,
+      "eval_accuracy": 0.935,
+      "eval_loss": 0.1935572475194931,
+      "eval_runtime": 15.6596,
+      "eval_samples_per_second": 1532.606,
+      "eval_steps_per_second": 4.789,
+      "step": 20640
+    },
+    {
+      "epoch": 8.61,
+      "learning_rate": 1e-06,
+      "loss": 0.1561,
+      "step": 20660
+    },
+    {
+      "epoch": 8.61,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.19057251513004303,
+      "eval_runtime": 15.8036,
+      "eval_samples_per_second": 1518.644,
+      "eval_steps_per_second": 4.746,
+      "step": 20660
+    },
+    {
+      "epoch": 8.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1619,
+      "step": 20680
+    },
+    {
+      "epoch": 8.62,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.1917581558227539,
+      "eval_runtime": 15.4829,
+      "eval_samples_per_second": 1550.099,
+      "eval_steps_per_second": 4.844,
+      "step": 20680
+    },
+    {
+      "epoch": 8.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1778,
+      "step": 20700
+    },
+    {
+      "epoch": 8.62,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.18889638781547546,
+      "eval_runtime": 15.6817,
+      "eval_samples_per_second": 1530.446,
+      "eval_steps_per_second": 4.783,
+      "step": 20700
+    },
+    {
+      "epoch": 8.63,
+      "learning_rate": 1e-06,
+      "loss": 0.1892,
+      "step": 20720
+    },
+    {
+      "epoch": 8.63,
+      "eval_accuracy": 0.9360416666666667,
+      "eval_loss": 0.18932241201400757,
+      "eval_runtime": 15.6011,
+      "eval_samples_per_second": 1538.354,
+      "eval_steps_per_second": 4.807,
+      "step": 20720
+    },
+    {
+      "epoch": 8.64,
+      "learning_rate": 1e-06,
+      "loss": 0.1358,
+      "step": 20740
+    },
+    {
+      "epoch": 8.64,
+      "eval_accuracy": 0.9362083333333333,
+      "eval_loss": 0.18881624937057495,
+      "eval_runtime": 15.4873,
+      "eval_samples_per_second": 1549.656,
+      "eval_steps_per_second": 4.843,
+      "step": 20740
+    },
+    {
+      "epoch": 8.65,
+      "learning_rate": 1e-06,
+      "loss": 0.1466,
+      "step": 20760
+    },
+    {
+      "epoch": 8.65,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.19135870039463043,
+      "eval_runtime": 15.7754,
+      "eval_samples_per_second": 1521.355,
+      "eval_steps_per_second": 4.754,
+      "step": 20760
+    },
+    {
+      "epoch": 8.66,
+      "learning_rate": 1e-06,
+      "loss": 0.1536,
+      "step": 20780
+    },
+    {
+      "epoch": 8.66,
+      "eval_accuracy": 0.9358333333333333,
+      "eval_loss": 0.19162118434906006,
+      "eval_runtime": 15.6911,
+      "eval_samples_per_second": 1529.529,
+      "eval_steps_per_second": 4.78,
+      "step": 20780
+    },
+    {
+      "epoch": 8.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1417,
+      "step": 20800
+    },
+    {
+      "epoch": 8.67,
+      "eval_accuracy": 0.9351666666666667,
+      "eval_loss": 0.19261544942855835,
+      "eval_runtime": 15.318,
+      "eval_samples_per_second": 1566.781,
+      "eval_steps_per_second": 4.896,
+      "step": 20800
+    },
+    {
+      "epoch": 8.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1335,
+      "step": 20820
+    },
+    {
+      "epoch": 8.68,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.19034945964813232,
+      "eval_runtime": 15.7498,
+      "eval_samples_per_second": 1523.829,
+      "eval_steps_per_second": 4.762,
+      "step": 20820
+    },
+    {
+      "epoch": 8.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1698,
+      "step": 20840
+    },
+    {
+      "epoch": 8.68,
+      "eval_accuracy": 0.935,
+      "eval_loss": 0.19306336343288422,
+      "eval_runtime": 15.5793,
+      "eval_samples_per_second": 1540.501,
+      "eval_steps_per_second": 4.814,
+      "step": 20840
+    },
+    {
+      "epoch": 8.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1394,
+      "step": 20860
+    },
+    {
+      "epoch": 8.69,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.19079020619392395,
+      "eval_runtime": 15.4491,
+      "eval_samples_per_second": 1553.491,
+      "eval_steps_per_second": 4.855,
+      "step": 20860
+    },
+    {
+      "epoch": 8.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1467,
+      "step": 20880
+    },
+    {
+      "epoch": 8.7,
+      "eval_accuracy": 0.9353333333333333,
+      "eval_loss": 0.1910465806722641,
+      "eval_runtime": 15.5826,
+      "eval_samples_per_second": 1540.175,
+      "eval_steps_per_second": 4.813,
+      "step": 20880
+    },
+    {
+      "epoch": 8.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1307,
+      "step": 20900
+    },
+    {
+      "epoch": 8.71,
+      "eval_accuracy": 0.9344166666666667,
+      "eval_loss": 0.1931913048028946,
+      "eval_runtime": 15.7361,
+      "eval_samples_per_second": 1525.151,
+      "eval_steps_per_second": 4.766,
+      "step": 20900
+    },
+    {
+      "epoch": 8.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1566,
+      "step": 20920
+    },
+    {
+      "epoch": 8.72,
+      "eval_accuracy": 0.934375,
+      "eval_loss": 0.19242696464061737,
+      "eval_runtime": 16.0207,
+      "eval_samples_per_second": 1498.065,
+      "eval_steps_per_second": 4.681,
+      "step": 20920
+    },
+    {
+      "epoch": 8.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1916,
+      "step": 20940
+    },
+    {
+      "epoch": 8.72,
+      "eval_accuracy": 0.9332083333333333,
+      "eval_loss": 0.19626210629940033,
+      "eval_runtime": 15.5195,
+      "eval_samples_per_second": 1546.445,
+      "eval_steps_per_second": 4.833,
+      "step": 20940
+    },
+    {
+      "epoch": 8.73,
+      "learning_rate": 1e-06,
+      "loss": 0.1366,
+      "step": 20960
+    },
+    {
+      "epoch": 8.73,
+      "eval_accuracy": 0.934375,
+      "eval_loss": 0.1936860978603363,
+      "eval_runtime": 16.4371,
+      "eval_samples_per_second": 1460.113,
+      "eval_steps_per_second": 4.563,
+      "step": 20960
+    },
+    {
+      "epoch": 8.74,
+      "learning_rate": 1e-06,
+      "loss": 0.1532,
+      "step": 20980
+    },
+    {
+      "epoch": 8.74,
+      "eval_accuracy": 0.934,
+      "eval_loss": 0.19347819685935974,
+      "eval_runtime": 16.8336,
+      "eval_samples_per_second": 1425.717,
+      "eval_steps_per_second": 4.455,
+      "step": 20980
+    },
+    {
+      "epoch": 8.75,
+      "learning_rate": 1e-06,
+      "loss": 0.1524,
+      "step": 21000
+    },
+    {
+      "epoch": 8.75,
+      "eval_accuracy": 0.9351666666666667,
+      "eval_loss": 0.19147901237010956,
+      "eval_runtime": 17.5314,
+      "eval_samples_per_second": 1368.97,
+      "eval_steps_per_second": 4.278,
+      "step": 21000
+    },
+    {
+      "epoch": 8.76,
+      "learning_rate": 1e-06,
+      "loss": 0.1331,
+      "step": 21020
+    },
+    {
+      "epoch": 8.76,
+      "eval_accuracy": 0.9347083333333334,
+      "eval_loss": 0.19231154024600983,
+      "eval_runtime": 17.9038,
+      "eval_samples_per_second": 1340.496,
+      "eval_steps_per_second": 4.189,
+      "step": 21020
+    },
+    {
+      "epoch": 8.77,
+      "learning_rate": 1e-06,
+      "loss": 0.1776,
+      "step": 21040
+    },
+    {
+      "epoch": 8.77,
+      "eval_accuracy": 0.935125,
+      "eval_loss": 0.19299417734146118,
+      "eval_runtime": 17.3849,
+      "eval_samples_per_second": 1380.506,
+      "eval_steps_per_second": 4.314,
+      "step": 21040
+    },
+    {
+      "epoch": 8.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1772,
+      "step": 21060
+    },
+    {
+      "epoch": 8.78,
+      "eval_accuracy": 0.9354583333333333,
+      "eval_loss": 0.1907561719417572,
+      "eval_runtime": 17.0346,
+      "eval_samples_per_second": 1408.894,
+      "eval_steps_per_second": 4.403,
+      "step": 21060
+    },
+    {
+      "epoch": 8.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1501,
+      "step": 21080
+    },
+    {
+      "epoch": 8.78,
+      "eval_accuracy": 0.934125,
+      "eval_loss": 0.19375211000442505,
+      "eval_runtime": 17.0995,
+      "eval_samples_per_second": 1403.552,
+      "eval_steps_per_second": 4.386,
+      "step": 21080
+    },
+    {
+      "epoch": 8.79,
+      "learning_rate": 1e-06,
+      "loss": 0.1544,
+      "step": 21100
+    },
+    {
+      "epoch": 8.79,
+      "eval_accuracy": 0.936125,
+      "eval_loss": 0.18984746932983398,
+      "eval_runtime": 16.9332,
+      "eval_samples_per_second": 1417.338,
+      "eval_steps_per_second": 4.429,
+      "step": 21100
+    },
+    {
+      "epoch": 8.8,
+      "learning_rate": 1e-06,
+      "loss": 0.1852,
+      "step": 21120
+    },
+    {
+      "epoch": 8.8,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.18966814875602722,
+      "eval_runtime": 16.7173,
+      "eval_samples_per_second": 1435.64,
+      "eval_steps_per_second": 4.486,
+      "step": 21120
+    },
+    {
+      "epoch": 8.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1483,
+      "step": 21140
+    },
+    {
+      "epoch": 8.81,
+      "eval_accuracy": 0.9352083333333333,
+      "eval_loss": 0.19126974046230316,
+      "eval_runtime": 16.9887,
+      "eval_samples_per_second": 1412.7,
+      "eval_steps_per_second": 4.415,
+      "step": 21140
+    },
+    {
+      "epoch": 8.82,
+      "learning_rate": 1e-06,
+      "loss": 0.1434,
+      "step": 21160
+    },
+    {
+      "epoch": 8.82,
+      "eval_accuracy": 0.9355416666666667,
+      "eval_loss": 0.19145381450653076,
+      "eval_runtime": 17.2555,
+      "eval_samples_per_second": 1390.863,
+      "eval_steps_per_second": 4.346,
+      "step": 21160
+    },
+    {
+      "epoch": 8.82,
+      "learning_rate": 1e-06,
+      "loss": 0.2023,
+      "step": 21180
+    },
+    {
+      "epoch": 8.82,
+      "eval_accuracy": 0.9348333333333333,
+      "eval_loss": 0.19298754632472992,
+      "eval_runtime": 16.4729,
+      "eval_samples_per_second": 1456.937,
+      "eval_steps_per_second": 4.553,
+      "step": 21180
+    },
+    {
+      "epoch": 8.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1471,
+      "step": 21200
+    },
+    {
+      "epoch": 8.83,
+      "eval_accuracy": 0.933625,
+      "eval_loss": 0.19501623511314392,
+      "eval_runtime": 16.7339,
+      "eval_samples_per_second": 1434.212,
+      "eval_steps_per_second": 4.482,
+      "step": 21200
+    },
+    {
+      "epoch": 8.84,
+      "learning_rate": 1e-06,
+      "loss": 0.1606,
+      "step": 21220
+    },
+    {
+      "epoch": 8.84,
+      "eval_accuracy": 0.934375,
+      "eval_loss": 0.19614289700984955,
+      "eval_runtime": 16.3664,
+      "eval_samples_per_second": 1466.416,
+      "eval_steps_per_second": 4.583,
+      "step": 21220
+    },
+    {
+      "epoch": 8.85,
+      "learning_rate": 1e-06,
+      "loss": 0.1873,
+      "step": 21240
+    },
+    {
+      "epoch": 8.85,
+      "eval_accuracy": 0.9364166666666667,
+      "eval_loss": 0.19085697829723358,
+      "eval_runtime": 16.7737,
+      "eval_samples_per_second": 1430.812,
+      "eval_steps_per_second": 4.471,
+      "step": 21240
+    },
+    {
+      "epoch": 8.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1676,
+      "step": 21260
+    },
+    {
+      "epoch": 8.86,
+      "eval_accuracy": 0.936625,
+      "eval_loss": 0.18988537788391113,
+      "eval_runtime": 16.3918,
+      "eval_samples_per_second": 1464.148,
+      "eval_steps_per_second": 4.575,
+      "step": 21260
+    },
+    {
+      "epoch": 8.87,
+      "learning_rate": 1e-06,
+      "loss": 0.1421,
+      "step": 21280
+    },
+    {
+      "epoch": 8.87,
+      "eval_accuracy": 0.93625,
+      "eval_loss": 0.1904492825269699,
+      "eval_runtime": 16.5295,
+      "eval_samples_per_second": 1451.948,
+      "eval_steps_per_second": 4.537,
+      "step": 21280
+    },
+    {
+      "epoch": 8.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1475,
+      "step": 21300
+    },
+    {
+      "epoch": 8.88,
+      "eval_accuracy": 0.935,
+      "eval_loss": 0.1930071860551834,
+      "eval_runtime": 16.4595,
+      "eval_samples_per_second": 1458.128,
+      "eval_steps_per_second": 4.557,
+      "step": 21300
+    },
+    {
+      "epoch": 8.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1928,
+      "step": 21320
+    },
+    {
+      "epoch": 8.88,
+      "eval_accuracy": 0.934875,
+      "eval_loss": 0.19229546189308167,
+      "eval_runtime": 16.6828,
+      "eval_samples_per_second": 1438.611,
+      "eval_steps_per_second": 4.496,
+      "step": 21320
+    },
+    {
+      "epoch": 8.89,
+      "learning_rate": 1e-06,
+      "loss": 0.1633,
+      "step": 21340
+    },
+    {
+      "epoch": 8.89,
+      "eval_accuracy": 0.9347916666666667,
+      "eval_loss": 0.19193056225776672,
+      "eval_runtime": 16.5324,
+      "eval_samples_per_second": 1451.694,
+      "eval_steps_per_second": 4.537,
+      "step": 21340
+    },
+    {
+      "epoch": 8.9,
+      "learning_rate": 1e-06,
+      "loss": 0.1504,
+      "step": 21360
+    },
+    {
+      "epoch": 8.9,
+      "eval_accuracy": 0.935375,
+      "eval_loss": 0.1920657902956009,
+      "eval_runtime": 16.6038,
+      "eval_samples_per_second": 1445.452,
+      "eval_steps_per_second": 4.517,
+      "step": 21360
+    },
+    {
+      "epoch": 8.91,
+      "learning_rate": 1e-06,
+      "loss": 0.1489,
+      "step": 21380
+    },
+    {
+      "epoch": 8.91,
+      "eval_accuracy": 0.9335833333333333,
+      "eval_loss": 0.19495506584644318,
+      "eval_runtime": 17.9072,
+      "eval_samples_per_second": 1340.243,
+      "eval_steps_per_second": 4.188,
+      "step": 21380
+    },
+    {
+      "epoch": 8.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1646,
+      "step": 21400
+    },
+    {
+      "epoch": 8.92,
+      "eval_accuracy": 0.9352916666666666,
+      "eval_loss": 0.19053585827350616,
+      "eval_runtime": 17.2542,
+      "eval_samples_per_second": 1390.963,
+      "eval_steps_per_second": 4.347,
+      "step": 21400
+    },
+    {
+      "epoch": 8.93,
+      "learning_rate": 1e-06,
+      "loss": 0.161,
+      "step": 21420
+    },
+    {
+      "epoch": 8.93,
+      "eval_accuracy": 0.9343333333333333,
+      "eval_loss": 0.19274799525737762,
+      "eval_runtime": 16.8753,
+      "eval_samples_per_second": 1422.195,
+      "eval_steps_per_second": 4.444,
+      "step": 21420
+    },
+    {
+      "epoch": 8.93,
+      "learning_rate": 1e-06,
+      "loss": 0.1885,
+      "step": 21440
+    },
+    {
+      "epoch": 8.93,
+      "eval_accuracy": 0.9355833333333333,
+      "eval_loss": 0.18927617371082306,
+      "eval_runtime": 17.7571,
+      "eval_samples_per_second": 1351.572,
+      "eval_steps_per_second": 4.224,
+      "step": 21440
+    },
+    {
+      "epoch": 8.94,
+      "learning_rate": 1e-06,
+      "loss": 0.1789,
+      "step": 21460
+    },
+    {
+      "epoch": 8.94,
+      "eval_accuracy": 0.9364583333333333,
+      "eval_loss": 0.18856315314769745,
+      "eval_runtime": 16.8879,
+      "eval_samples_per_second": 1421.133,
+      "eval_steps_per_second": 4.441,
+      "step": 21460
+    },
+    {
+      "epoch": 8.95,
+      "learning_rate": 1e-06,
+      "loss": 0.1831,
+      "step": 21480
+    },
+    {
+      "epoch": 8.95,
+      "eval_accuracy": 0.9367083333333334,
+      "eval_loss": 0.18908809125423431,
+      "eval_runtime": 16.5935,
+      "eval_samples_per_second": 1446.346,
+      "eval_steps_per_second": 4.52,
+      "step": 21480
+    },
+    {
+      "epoch": 8.96,
+      "learning_rate": 1e-06,
+      "loss": 0.1553,
+      "step": 21500
+    },
+    {
+      "epoch": 8.96,
+      "eval_accuracy": 0.9363333333333334,
+      "eval_loss": 0.1887378990650177,
+      "eval_runtime": 17.029,
+      "eval_samples_per_second": 1409.364,
+      "eval_steps_per_second": 4.404,
+      "step": 21500
+    },
+    {
+      "epoch": 8.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1609,
+      "step": 21520
+    },
+    {
+      "epoch": 8.97,
+      "eval_accuracy": 0.936625,
+      "eval_loss": 0.1878993958234787,
+      "eval_runtime": 17.4532,
+      "eval_samples_per_second": 1375.109,
+      "eval_steps_per_second": 4.297,
+      "step": 21520
+    },
+    {
+      "epoch": 8.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1616,
+      "step": 21540
+    },
+    {
+      "epoch": 8.97,
+      "eval_accuracy": 0.93625,
+      "eval_loss": 0.18860691785812378,
+      "eval_runtime": 16.6603,
+      "eval_samples_per_second": 1440.553,
+      "eval_steps_per_second": 4.502,
+      "step": 21540
+    },
+    {
+      "epoch": 8.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1623,
+      "step": 21560
+    },
+    {
+      "epoch": 8.98,
+      "eval_accuracy": 0.934375,
+      "eval_loss": 0.19211074709892273,
+      "eval_runtime": 16.5367,
+      "eval_samples_per_second": 1451.315,
+      "eval_steps_per_second": 4.535,
+      "step": 21560
+    },
+    {
+      "epoch": 8.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1485,
+      "step": 21580
+    },
+    {
+      "epoch": 8.99,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.1881760060787201,
+      "eval_runtime": 16.5469,
+      "eval_samples_per_second": 1450.422,
+      "eval_steps_per_second": 4.533,
+      "step": 21580
+    },
+    {
+      "epoch": 9.0,
+      "learning_rate": 1e-06,
+      "loss": 0.1922,
+      "step": 21600
+    },
+    {
+      "epoch": 9.0,
+      "eval_accuracy": 0.9357916666666667,
+      "eval_loss": 0.18824529647827148,
+      "eval_runtime": 17.1765,
+      "eval_samples_per_second": 1397.261,
+      "eval_steps_per_second": 4.366,
+      "step": 21600
+    },
+    {
+      "epoch": 9.01,
+      "learning_rate": 1e-06,
+      "loss": 0.1314,
+      "step": 21620
+    },
+    {
+      "epoch": 9.01,
+      "eval_accuracy": 0.9365,
+      "eval_loss": 0.18808509409427643,
+      "eval_runtime": 16.3714,
+      "eval_samples_per_second": 1465.972,
+      "eval_steps_per_second": 4.581,
+      "step": 21620
+    },
+    {
+      "epoch": 9.02,
+      "learning_rate": 1e-06,
+      "loss": 0.1579,
+      "step": 21640
+    },
+    {
+      "epoch": 9.02,
+      "eval_accuracy": 0.935,
+      "eval_loss": 0.1919311285018921,
+      "eval_runtime": 16.3749,
+      "eval_samples_per_second": 1465.66,
+      "eval_steps_per_second": 4.58,
+      "step": 21640
+    },
+    {
+      "epoch": 9.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1449,
+      "step": 21660
+    },
+    {
+      "epoch": 9.03,
+      "eval_accuracy": 0.9358333333333333,
+      "eval_loss": 0.19011430442333221,
+      "eval_runtime": 16.428,
+      "eval_samples_per_second": 1460.921,
+      "eval_steps_per_second": 4.565,
+      "step": 21660
+    },
+    {
+      "epoch": 9.03,
+      "learning_rate": 1e-06,
+      "loss": 0.1394,
+      "step": 21680
+    },
+    {
+      "epoch": 9.03,
+      "eval_accuracy": 0.936875,
+      "eval_loss": 0.19003826379776,
+      "eval_runtime": 17.008,
+      "eval_samples_per_second": 1411.103,
+      "eval_steps_per_second": 4.41,
+      "step": 21680
+    },
+    {
+      "epoch": 9.04,
+      "learning_rate": 1e-06,
+      "loss": 0.1411,
+      "step": 21700
+    },
+    {
+      "epoch": 9.04,
+      "eval_accuracy": 0.9344583333333333,
+      "eval_loss": 0.19364295899868011,
+      "eval_runtime": 17.6973,
+      "eval_samples_per_second": 1356.136,
+      "eval_steps_per_second": 4.238,
+      "step": 21700
+    },
+    {
+      "epoch": 9.05,
+      "learning_rate": 1e-06,
+      "loss": 0.1622,
+      "step": 21720
+    },
+    {
+      "epoch": 9.05,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.190113365650177,
+      "eval_runtime": 17.0907,
+      "eval_samples_per_second": 1404.273,
+      "eval_steps_per_second": 4.388,
+      "step": 21720
+    },
+    {
+      "epoch": 9.06,
+      "learning_rate": 1e-06,
+      "loss": 0.1758,
+      "step": 21740
+    },
+    {
+      "epoch": 9.06,
+      "eval_accuracy": 0.9359166666666666,
+      "eval_loss": 0.19059090316295624,
+      "eval_runtime": 17.4852,
+      "eval_samples_per_second": 1372.587,
+      "eval_steps_per_second": 4.289,
+      "step": 21740
+    },
+    {
+      "epoch": 9.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1469,
+      "step": 21760
+    },
+    {
+      "epoch": 9.07,
+      "eval_accuracy": 0.9352916666666666,
+      "eval_loss": 0.19149552285671234,
+      "eval_runtime": 15.6633,
+      "eval_samples_per_second": 1532.24,
+      "eval_steps_per_second": 4.788,
+      "step": 21760
+    },
+    {
+      "epoch": 9.07,
+      "learning_rate": 1e-06,
+      "loss": 0.1543,
+      "step": 21780
+    },
+    {
+      "epoch": 9.07,
+      "eval_accuracy": 0.9346666666666666,
+      "eval_loss": 0.1922057569026947,
+      "eval_runtime": 15.6817,
+      "eval_samples_per_second": 1530.451,
+      "eval_steps_per_second": 4.783,
+      "step": 21780
+    },
+    {
+      "epoch": 9.08,
+      "learning_rate": 1e-06,
+      "loss": 0.1411,
+      "step": 21800
+    },
+    {
+      "epoch": 9.08,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.19251596927642822,
+      "eval_runtime": 15.8744,
+      "eval_samples_per_second": 1511.864,
+      "eval_steps_per_second": 4.725,
+      "step": 21800
+    },
+    {
+      "epoch": 9.09,
+      "learning_rate": 1e-06,
+      "loss": 0.1686,
+      "step": 21820
+    },
+    {
+      "epoch": 9.09,
+      "eval_accuracy": 0.937125,
+      "eval_loss": 0.1892337054014206,
+      "eval_runtime": 16.2971,
+      "eval_samples_per_second": 1472.653,
+      "eval_steps_per_second": 4.602,
+      "step": 21820
+    },
+    {
+      "epoch": 9.1,
+      "learning_rate": 1e-06,
+      "loss": 0.1332,
+      "step": 21840
+    },
+    {
+      "epoch": 9.1,
+      "eval_accuracy": 0.9366666666666666,
+      "eval_loss": 0.1890205591917038,
+      "eval_runtime": 15.4417,
+      "eval_samples_per_second": 1554.228,
+      "eval_steps_per_second": 4.857,
+      "step": 21840
+    },
+    {
+      "epoch": 9.11,
+      "learning_rate": 1e-06,
+      "loss": 0.1217,
+      "step": 21860
+    },
+    {
+      "epoch": 9.11,
+      "eval_accuracy": 0.9357083333333334,
+      "eval_loss": 0.1898655891418457,
+      "eval_runtime": 16.1103,
+      "eval_samples_per_second": 1489.727,
+      "eval_steps_per_second": 4.655,
+      "step": 21860
+    },
+    {
+      "epoch": 9.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1765,
+      "step": 21880
+    },
+    {
+      "epoch": 9.12,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.19171109795570374,
+      "eval_runtime": 16.5889,
+      "eval_samples_per_second": 1446.753,
+      "eval_steps_per_second": 4.521,
+      "step": 21880
+    },
+    {
+      "epoch": 9.12,
+      "learning_rate": 1e-06,
+      "loss": 0.1733,
+      "step": 21900
+    },
+    {
+      "epoch": 9.12,
+      "eval_accuracy": 0.9349583333333333,
+      "eval_loss": 0.19177711009979248,
+      "eval_runtime": 17.5638,
+      "eval_samples_per_second": 1366.449,
+      "eval_steps_per_second": 4.27,
+      "step": 21900
+    },
+    {
+      "epoch": 9.13,
+      "learning_rate": 1e-06,
+      "loss": 0.1424,
+      "step": 21920
+    },
+    {
+      "epoch": 9.13,
+      "eval_accuracy": 0.934625,
+      "eval_loss": 0.19338135421276093,
+      "eval_runtime": 17.301,
+      "eval_samples_per_second": 1387.204,
+      "eval_steps_per_second": 4.335,
+      "step": 21920
+    },
+    {
+      "epoch": 9.14,
+      "learning_rate": 1e-06,
+      "loss": 0.1389,
+      "step": 21940
+    },
+    {
+      "epoch": 9.14,
+      "eval_accuracy": 0.9360416666666667,
+      "eval_loss": 0.19054777920246124,
+      "eval_runtime": 16.9906,
+      "eval_samples_per_second": 1412.546,
+      "eval_steps_per_second": 4.414,
+      "step": 21940
+    },
+    {
+      "epoch": 9.15,
+      "learning_rate": 1e-06,
+      "loss": 0.1426,
+      "step": 21960
+    },
+    {
+      "epoch": 9.15,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.1908566951751709,
+      "eval_runtime": 17.0809,
+      "eval_samples_per_second": 1405.076,
+      "eval_steps_per_second": 4.391,
+      "step": 21960
+    },
+    {
+      "epoch": 9.16,
+      "learning_rate": 1e-06,
+      "loss": 0.1376,
+      "step": 21980
+    },
+    {
+      "epoch": 9.16,
+      "eval_accuracy": 0.936375,
+      "eval_loss": 0.18984192609786987,
+      "eval_runtime": 16.3175,
+      "eval_samples_per_second": 1470.81,
+      "eval_steps_per_second": 4.596,
+      "step": 21980
+    },
+    {
+      "epoch": 9.17,
+      "learning_rate": 1e-06,
+      "loss": 0.162,
+      "step": 22000
+    },
+    {
+      "epoch": 9.17,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.1910352259874344,
+      "eval_runtime": 16.1268,
+      "eval_samples_per_second": 1488.209,
+      "eval_steps_per_second": 4.651,
+      "step": 22000
+    },
+    {
+      "epoch": 9.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1555,
+      "step": 22020
+    },
+    {
+      "epoch": 9.18,
+      "eval_accuracy": 0.93475,
+      "eval_loss": 0.19323420524597168,
+      "eval_runtime": 16.5391,
+      "eval_samples_per_second": 1451.109,
+      "eval_steps_per_second": 4.535,
+      "step": 22020
+    },
+    {
+      "epoch": 9.18,
+      "learning_rate": 1e-06,
+      "loss": 0.1613,
+      "step": 22040
+    },
+    {
+      "epoch": 9.18,
+      "eval_accuracy": 0.9364583333333333,
+      "eval_loss": 0.1894889771938324,
+      "eval_runtime": 16.1517,
+      "eval_samples_per_second": 1485.912,
+      "eval_steps_per_second": 4.643,
+      "step": 22040
+    },
+    {
+      "epoch": 9.19,
+      "learning_rate": 1e-06,
+      "loss": 0.1407,
+      "step": 22060
+    },
+    {
+      "epoch": 9.19,
+      "eval_accuracy": 0.9354583333333333,
+      "eval_loss": 0.1917874664068222,
+      "eval_runtime": 17.701,
+      "eval_samples_per_second": 1355.857,
+      "eval_steps_per_second": 4.237,
+      "step": 22060
+    },
+    {
+      "epoch": 9.2,
+      "learning_rate": 1e-06,
+      "loss": 0.1709,
+      "step": 22080
+    },
+    {
+      "epoch": 9.2,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.19433771073818207,
+      "eval_runtime": 17.0414,
+      "eval_samples_per_second": 1408.333,
+      "eval_steps_per_second": 4.401,
+      "step": 22080
+    },
+    {
+      "epoch": 9.21,
+      "learning_rate": 1e-06,
+      "loss": 0.1435,
+      "step": 22100
+    },
+    {
+      "epoch": 9.21,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.19121021032333374,
+      "eval_runtime": 17.2755,
+      "eval_samples_per_second": 1389.249,
+      "eval_steps_per_second": 4.341,
+      "step": 22100
+    },
+    {
+      "epoch": 9.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1778,
+      "step": 22120
+    },
+    {
+      "epoch": 9.22,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.1902051866054535,
+      "eval_runtime": 16.2731,
+      "eval_samples_per_second": 1474.822,
+      "eval_steps_per_second": 4.609,
+      "step": 22120
+    },
+    {
+      "epoch": 9.22,
+      "learning_rate": 1e-06,
+      "loss": 0.1413,
+      "step": 22140
+    },
+    {
+      "epoch": 9.22,
+      "eval_accuracy": 0.9355416666666667,
+      "eval_loss": 0.1910925656557083,
+      "eval_runtime": 16.4198,
+      "eval_samples_per_second": 1461.653,
+      "eval_steps_per_second": 4.568,
+      "step": 22140
+    },
+    {
+      "epoch": 9.23,
+      "learning_rate": 1e-06,
+      "loss": 0.1772,
+      "step": 22160
+    },
+    {
+      "epoch": 9.23,
+      "eval_accuracy": 0.9365,
+      "eval_loss": 0.1890135258436203,
+      "eval_runtime": 16.5257,
+      "eval_samples_per_second": 1452.285,
+      "eval_steps_per_second": 4.538,
+      "step": 22160
+    },
+    {
+      "epoch": 9.24,
+      "learning_rate": 1e-06,
+      "loss": 0.1545,
+      "step": 22180
+    },
+    {
+      "epoch": 9.24,
+      "eval_accuracy": 0.9360416666666667,
+      "eval_loss": 0.19075444340705872,
+      "eval_runtime": 16.9041,
+      "eval_samples_per_second": 1419.773,
+      "eval_steps_per_second": 4.437,
+      "step": 22180
+    },
+    {
+      "epoch": 9.25,
+      "learning_rate": 1e-06,
+      "loss": 0.1434,
+      "step": 22200
+    },
+    {
+      "epoch": 9.25,
+      "eval_accuracy": 0.9331666666666667,
+      "eval_loss": 0.19980815052986145,
+      "eval_runtime": 16.2294,
+      "eval_samples_per_second": 1478.797,
+      "eval_steps_per_second": 4.621,
+      "step": 22200
+    },
+    {
+      "epoch": 9.26,
+      "learning_rate": 1e-06,
+      "loss": 0.1981,
+      "step": 22220
+    },
+    {
+      "epoch": 9.26,
+      "eval_accuracy": 0.9367916666666667,
+      "eval_loss": 0.1893180012702942,
+      "eval_runtime": 16.4467,
+      "eval_samples_per_second": 1459.263,
+      "eval_steps_per_second": 4.56,
+      "step": 22220
+    },
+    {
+      "epoch": 9.27,
+      "learning_rate": 1e-06,
+      "loss": 0.1585,
+      "step": 22240
+    },
+    {
+      "epoch": 9.27,
+      "eval_accuracy": 0.936,
+      "eval_loss": 0.18898563086986542,
+      "eval_runtime": 16.2988,
+      "eval_samples_per_second": 1472.504,
+      "eval_steps_per_second": 4.602,
+      "step": 22240
+    },
+    {
+      "epoch": 9.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1361,
+      "step": 22260
+    },
+    {
+      "epoch": 9.28,
+      "eval_accuracy": 0.9357916666666667,
+      "eval_loss": 0.1903260499238968,
+      "eval_runtime": 16.4778,
+      "eval_samples_per_second": 1456.508,
+      "eval_steps_per_second": 4.552,
+      "step": 22260
+    },
+    {
+      "epoch": 9.28,
+      "learning_rate": 1e-06,
+      "loss": 0.1666,
+      "step": 22280
+    },
+    {
+      "epoch": 9.28,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.19420108199119568,
+      "eval_runtime": 16.7685,
+      "eval_samples_per_second": 1431.255,
+      "eval_steps_per_second": 4.473,
+      "step": 22280
+    },
+    {
+      "epoch": 9.29,
+      "learning_rate": 1e-06,
+      "loss": 0.1855,
+      "step": 22300
+    },
+    {
+      "epoch": 9.29,
+      "eval_accuracy": 0.9355416666666667,
+      "eval_loss": 0.19144552946090698,
+      "eval_runtime": 16.1429,
+      "eval_samples_per_second": 1486.719,
+      "eval_steps_per_second": 4.646,
+      "step": 22300
+    },
+    {
+      "epoch": 9.3,
+      "learning_rate": 1e-06,
+      "loss": 0.126,
+      "step": 22320
+    },
+    {
+      "epoch": 9.3,
+      "eval_accuracy": 0.935875,
+      "eval_loss": 0.1904752403497696,
+      "eval_runtime": 16.2377,
+      "eval_samples_per_second": 1478.039,
+      "eval_steps_per_second": 4.619,
+      "step": 22320
+    },
+    {
+      "epoch": 9.31,
+      "learning_rate": 1e-06,
+      "loss": 0.1244,
+      "step": 22340
+    },
+    {
+      "epoch": 9.31,
+      "eval_accuracy": 0.9362916666666666,
+      "eval_loss": 0.19049489498138428,
+      "eval_runtime": 16.6183,
+      "eval_samples_per_second": 1444.194,
+      "eval_steps_per_second": 4.513,
+      "step": 22340
+    },
+    {
+      "epoch": 9.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1737,
+      "step": 22360
+    },
+    {
+      "epoch": 9.32,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.19159464538097382,
+      "eval_runtime": 16.5579,
+      "eval_samples_per_second": 1449.463,
+      "eval_steps_per_second": 4.53,
+      "step": 22360
+    },
+    {
+      "epoch": 9.32,
+      "learning_rate": 1e-06,
+      "loss": 0.1568,
+      "step": 22380
+    },
+    {
+      "epoch": 9.32,
+      "eval_accuracy": 0.9359166666666666,
+      "eval_loss": 0.19013747572898865,
+      "eval_runtime": 16.3587,
+      "eval_samples_per_second": 1467.107,
+      "eval_steps_per_second": 4.585,
+      "step": 22380
+    },
+    {
+      "epoch": 9.33,
+      "learning_rate": 1e-06,
+      "loss": 0.1501,
+      "step": 22400
+    },
+    {
+      "epoch": 9.33,
+      "eval_accuracy": 0.936625,
+      "eval_loss": 0.1898564100265503,
+      "eval_runtime": 16.4813,
+      "eval_samples_per_second": 1456.2,
+      "eval_steps_per_second": 4.551,
+      "step": 22400
+    },
+    {
+      "epoch": 9.34,
+      "learning_rate": 1e-06,
+      "loss": 0.133,
+      "step": 22420
+    },
+    {
+      "epoch": 9.34,
+      "eval_accuracy": 0.935875,
+      "eval_loss": 0.19129687547683716,
+      "eval_runtime": 17.9192,
+      "eval_samples_per_second": 1339.343,
+      "eval_steps_per_second": 4.185,
+      "step": 22420
+    },
+    {
+      "epoch": 9.35,
+      "learning_rate": 1e-06,
+      "loss": 0.1577,
+      "step": 22440
+    },
+    {
+      "epoch": 9.35,
+      "eval_accuracy": 0.9364583333333333,
+      "eval_loss": 0.19107209146022797,
+      "eval_runtime": 17.3207,
+      "eval_samples_per_second": 1385.625,
+      "eval_steps_per_second": 4.33,
+      "step": 22440
+    },
+    {
+      "epoch": 9.36,
+      "learning_rate": 1e-06,
+      "loss": 0.1463,
+      "step": 22460
+    },
+    {
+      "epoch": 9.36,
+      "eval_accuracy": 0.935875,
+      "eval_loss": 0.19182544946670532,
+      "eval_runtime": 17.9384,
+      "eval_samples_per_second": 1337.915,
+      "eval_steps_per_second": 4.181,
+      "step": 22460
+    },
+    {
+      "epoch": 9.37,
+      "learning_rate": 1e-06,
+      "loss": 0.1466,
+      "step": 22480
+    },
+    {
+      "epoch": 9.37,
+      "eval_accuracy": 0.93225,
+      "eval_loss": 0.20088885724544525,
+      "eval_runtime": 17.7162,
+      "eval_samples_per_second": 1354.693,
+      "eval_steps_per_second": 4.233,
+      "step": 22480
+    },
+    {
+      "epoch": 9.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1537,
+      "step": 22500
+    },
+    {
+      "epoch": 9.38,
+      "eval_accuracy": 0.9361666666666667,
+      "eval_loss": 0.19151511788368225,
+      "eval_runtime": 17.6,
+      "eval_samples_per_second": 1363.639,
+      "eval_steps_per_second": 4.261,
+      "step": 22500
+    },
+    {
+      "epoch": 9.38,
+      "learning_rate": 1e-06,
+      "loss": 0.1463,
+      "step": 22520
+    },
+    {
+      "epoch": 9.38,
+      "eval_accuracy": 0.933875,
+      "eval_loss": 0.1945749670267105,
+      "eval_runtime": 15.7657,
+      "eval_samples_per_second": 1522.29,
+      "eval_steps_per_second": 4.757,
+      "step": 22520
+    },
+    {
+      "epoch": 9.39,
+      "learning_rate": 1e-06,
+      "loss": 0.1763,
+      "step": 22540
+    },
+    {
+      "epoch": 9.39,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.19165822863578796,
+      "eval_runtime": 17.4199,
+      "eval_samples_per_second": 1377.737,
+      "eval_steps_per_second": 4.305,
+      "step": 22540
+    },
+    {
+      "epoch": 9.4,
+      "learning_rate": 1e-06,
+      "loss": 0.1375,
+      "step": 22560
+    },
+    {
+      "epoch": 9.4,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.19137227535247803,
+      "eval_runtime": 17.3107,
+      "eval_samples_per_second": 1386.428,
+      "eval_steps_per_second": 4.333,
+      "step": 22560
+    },
+    {
+      "epoch": 9.41,
+      "learning_rate": 1e-06,
+      "loss": 0.1616,
+      "step": 22580
+    },
+    {
+      "epoch": 9.41,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.19043239951133728,
+      "eval_runtime": 17.1156,
+      "eval_samples_per_second": 1402.228,
+      "eval_steps_per_second": 4.382,
+      "step": 22580
+    },
+    {
+      "epoch": 9.42,
+      "learning_rate": 1e-06,
+      "loss": 0.1809,
+      "step": 22600
+    },
+    {
+      "epoch": 9.42,
+      "eval_accuracy": 0.9346666666666666,
+      "eval_loss": 0.19357599318027496,
+      "eval_runtime": 17.8709,
+      "eval_samples_per_second": 1342.962,
+      "eval_steps_per_second": 4.197,
+      "step": 22600
+    },
+    {
+      "epoch": 9.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1436,
+      "step": 22620
+    },
+    {
+      "epoch": 9.43,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.18992185592651367,
+      "eval_runtime": 16.5938,
+      "eval_samples_per_second": 1446.323,
+      "eval_steps_per_second": 4.52,
+      "step": 22620
+    },
+    {
+      "epoch": 9.43,
+      "learning_rate": 1e-06,
+      "loss": 0.1831,
+      "step": 22640
+    },
+    {
+      "epoch": 9.43,
+      "eval_accuracy": 0.9361666666666667,
+      "eval_loss": 0.19047614932060242,
+      "eval_runtime": 17.0882,
+      "eval_samples_per_second": 1404.476,
+      "eval_steps_per_second": 4.389,
+      "step": 22640
+    },
+    {
+      "epoch": 9.44,
+      "learning_rate": 1e-06,
+      "loss": 0.135,
+      "step": 22660
+    },
+    {
+      "epoch": 9.44,
+      "eval_accuracy": 0.9363333333333334,
+      "eval_loss": 0.1894386261701584,
+      "eval_runtime": 17.1834,
+      "eval_samples_per_second": 1396.7,
+      "eval_steps_per_second": 4.365,
+      "step": 22660
+    },
+    {
+      "epoch": 9.45,
+      "learning_rate": 1e-06,
+      "loss": 0.198,
+      "step": 22680
+    },
+    {
+      "epoch": 9.45,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.18925711512565613,
+      "eval_runtime": 16.7997,
+      "eval_samples_per_second": 1428.594,
+      "eval_steps_per_second": 4.464,
+      "step": 22680
+    },
+    {
+      "epoch": 9.46,
+      "learning_rate": 1e-06,
+      "loss": 0.1595,
+      "step": 22700
+    },
+    {
+      "epoch": 9.46,
+      "eval_accuracy": 0.936,
+      "eval_loss": 0.19023475050926208,
+      "eval_runtime": 17.1467,
+      "eval_samples_per_second": 1399.688,
+      "eval_steps_per_second": 4.374,
+      "step": 22700
+    },
+    {
+      "epoch": 9.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1594,
+      "step": 22720
+    },
+    {
+      "epoch": 9.47,
+      "eval_accuracy": 0.9362083333333333,
+      "eval_loss": 0.18937428295612335,
+      "eval_runtime": 16.3859,
+      "eval_samples_per_second": 1464.678,
+      "eval_steps_per_second": 4.577,
+      "step": 22720
+    },
+    {
+      "epoch": 9.47,
+      "learning_rate": 1e-06,
+      "loss": 0.1741,
+      "step": 22740
+    },
+    {
+      "epoch": 9.47,
+      "eval_accuracy": 0.9365416666666667,
+      "eval_loss": 0.18718321621418,
+      "eval_runtime": 16.7434,
+      "eval_samples_per_second": 1433.4,
+      "eval_steps_per_second": 4.479,
+      "step": 22740
+    },
+    {
+      "epoch": 9.48,
+      "learning_rate": 1e-06,
+      "loss": 0.1526,
+      "step": 22760
+    },
+    {
+      "epoch": 9.48,
+      "eval_accuracy": 0.936125,
+      "eval_loss": 0.18912020325660706,
+      "eval_runtime": 16.0636,
+      "eval_samples_per_second": 1494.062,
+      "eval_steps_per_second": 4.669,
+      "step": 22760
+    },
+    {
+      "epoch": 9.49,
+      "learning_rate": 1e-06,
+      "loss": 0.1785,
+      "step": 22780
+    },
+    {
+      "epoch": 9.49,
+      "eval_accuracy": 0.9362083333333333,
+      "eval_loss": 0.1892082542181015,
+      "eval_runtime": 16.7318,
+      "eval_samples_per_second": 1434.395,
+      "eval_steps_per_second": 4.482,
+      "step": 22780
+    },
+    {
+      "epoch": 9.5,
+      "learning_rate": 1e-06,
+      "loss": 0.1487,
+      "step": 22800
+    },
+    {
+      "epoch": 9.5,
+      "eval_accuracy": 0.93375,
+      "eval_loss": 0.1960740089416504,
+      "eval_runtime": 16.2354,
+      "eval_samples_per_second": 1478.256,
+      "eval_steps_per_second": 4.62,
+      "step": 22800
+    },
+    {
+      "epoch": 9.51,
+      "learning_rate": 1e-06,
+      "loss": 0.151,
+      "step": 22820
+    },
+    {
+      "epoch": 9.51,
+      "eval_accuracy": 0.9355,
+      "eval_loss": 0.19206097722053528,
+      "eval_runtime": 16.4046,
+      "eval_samples_per_second": 1463.004,
+      "eval_steps_per_second": 4.572,
+      "step": 22820
+    },
+    {
+      "epoch": 9.52,
+      "learning_rate": 1e-06,
+      "loss": 0.1641,
+      "step": 22840
+    },
+    {
+      "epoch": 9.52,
+      "eval_accuracy": 0.9347083333333334,
+      "eval_loss": 0.1922009140253067,
+      "eval_runtime": 16.3419,
+      "eval_samples_per_second": 1468.617,
+      "eval_steps_per_second": 4.589,
+      "step": 22840
+    },
+    {
+      "epoch": 9.53,
+      "learning_rate": 1e-06,
+      "loss": 0.1403,
+      "step": 22860
+    },
+    {
+      "epoch": 9.53,
+      "eval_accuracy": 0.936625,
+      "eval_loss": 0.18889163434505463,
+      "eval_runtime": 16.6017,
+      "eval_samples_per_second": 1445.631,
+      "eval_steps_per_second": 4.518,
+      "step": 22860
+    },
+    {
+      "epoch": 9.53,
+      "learning_rate": 1e-06,
+      "loss": 0.182,
+      "step": 22880
+    },
+    {
+      "epoch": 9.53,
+      "eval_accuracy": 0.9352083333333333,
+      "eval_loss": 0.19116635620594025,
+      "eval_runtime": 16.607,
+      "eval_samples_per_second": 1445.171,
+      "eval_steps_per_second": 4.516,
+      "step": 22880
+    },
+    {
+      "epoch": 9.54,
+      "learning_rate": 1e-06,
+      "loss": 0.1655,
+      "step": 22900
+    },
+    {
+      "epoch": 9.54,
+      "eval_accuracy": 0.9345833333333333,
+      "eval_loss": 0.1919187605381012,
+      "eval_runtime": 16.4494,
+      "eval_samples_per_second": 1459.021,
+      "eval_steps_per_second": 4.559,
+      "step": 22900
+    },
+    {
+      "epoch": 9.55,
+      "learning_rate": 1e-06,
+      "loss": 0.1651,
+      "step": 22920
+    },
+    {
+      "epoch": 9.55,
+      "eval_accuracy": 0.9361666666666667,
+      "eval_loss": 0.1887436956167221,
+      "eval_runtime": 16.2982,
+      "eval_samples_per_second": 1472.556,
+      "eval_steps_per_second": 4.602,
+      "step": 22920
+    },
+    {
+      "epoch": 9.56,
+      "learning_rate": 1e-06,
+      "loss": 0.1493,
+      "step": 22940
+    },
+    {
+      "epoch": 9.56,
+      "eval_accuracy": 0.9370416666666667,
+      "eval_loss": 0.18585175275802612,
+      "eval_runtime": 16.7173,
+      "eval_samples_per_second": 1435.637,
+      "eval_steps_per_second": 4.486,
+      "step": 22940
+    },
+    {
+      "epoch": 9.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1563,
+      "step": 22960
+    },
+    {
+      "epoch": 9.57,
+      "eval_accuracy": 0.9357916666666667,
+      "eval_loss": 0.18925468623638153,
+      "eval_runtime": 17.5796,
+      "eval_samples_per_second": 1365.215,
+      "eval_steps_per_second": 4.266,
+      "step": 22960
+    },
+    {
+      "epoch": 9.57,
+      "learning_rate": 1e-06,
+      "loss": 0.1498,
+      "step": 22980
+    },
+    {
+      "epoch": 9.57,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.19032344222068787,
+      "eval_runtime": 15.9593,
+      "eval_samples_per_second": 1503.826,
+      "eval_steps_per_second": 4.699,
+      "step": 22980
+    },
+    {
+      "epoch": 9.58,
+      "learning_rate": 1e-06,
+      "loss": 0.131,
+      "step": 23000
+    },
+    {
+      "epoch": 9.58,
+      "eval_accuracy": 0.9360833333333334,
+      "eval_loss": 0.18864993751049042,
+      "eval_runtime": 15.7391,
+      "eval_samples_per_second": 1524.867,
+      "eval_steps_per_second": 4.765,
+      "step": 23000
+    },
+    {
+      "epoch": 9.59,
+      "learning_rate": 1e-06,
+      "loss": 0.1588,
+      "step": 23020
+    },
+    {
+      "epoch": 9.59,
+      "eval_accuracy": 0.9375416666666667,
+      "eval_loss": 0.18759743869304657,
+      "eval_runtime": 82.369,
+      "eval_samples_per_second": 291.372,
+      "eval_steps_per_second": 0.911,
+      "step": 23020
+    },
+    {
+      "epoch": 9.6,
+      "learning_rate": 1e-06,
+      "loss": 0.1555,
+      "step": 23040
+    },
+    {
+      "epoch": 9.6,
+      "eval_accuracy": 0.9354583333333333,
+      "eval_loss": 0.1895967274904251,
+      "eval_runtime": 16.3955,
+      "eval_samples_per_second": 1463.818,
+      "eval_steps_per_second": 4.574,
+      "step": 23040
+    },
+    {
+      "epoch": 9.61,
+      "learning_rate": 1e-06,
+      "loss": 0.1625,
+      "step": 23060
+    },
+    {
+      "epoch": 9.61,
+      "eval_accuracy": 0.9355416666666667,
+      "eval_loss": 0.1896596997976303,
+      "eval_runtime": 17.4896,
+      "eval_samples_per_second": 1372.245,
+      "eval_steps_per_second": 4.288,
+      "step": 23060
+    },
+    {
+      "epoch": 9.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1183,
+      "step": 23080
+    },
+    {
+      "epoch": 9.62,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.19340163469314575,
+      "eval_runtime": 17.2768,
+      "eval_samples_per_second": 1389.142,
+      "eval_steps_per_second": 4.341,
+      "step": 23080
+    },
+    {
+      "epoch": 9.62,
+      "learning_rate": 1e-06,
+      "loss": 0.1668,
+      "step": 23100
+    },
+    {
+      "epoch": 9.62,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.1944810450077057,
+      "eval_runtime": 17.3938,
+      "eval_samples_per_second": 1379.802,
+      "eval_steps_per_second": 4.312,
+      "step": 23100
+    },
+    {
+      "epoch": 9.63,
+      "learning_rate": 1e-06,
+      "loss": 0.1514,
+      "step": 23120
+    },
+    {
+      "epoch": 9.63,
+      "eval_accuracy": 0.9350416666666667,
+      "eval_loss": 0.19056230783462524,
+      "eval_runtime": 17.3545,
+      "eval_samples_per_second": 1382.923,
+      "eval_steps_per_second": 4.322,
+      "step": 23120
+    },
+    {
+      "epoch": 9.64,
+      "learning_rate": 1e-06,
+      "loss": 0.0979,
+      "step": 23140
+    },
+    {
+      "epoch": 9.64,
+      "eval_accuracy": 0.935375,
+      "eval_loss": 0.18961113691329956,
+      "eval_runtime": 17.9278,
+      "eval_samples_per_second": 1338.707,
+      "eval_steps_per_second": 4.183,
+      "step": 23140
+    },
+    {
+      "epoch": 9.65,
+      "learning_rate": 1e-06,
+      "loss": 0.1414,
+      "step": 23160
+    },
+    {
+      "epoch": 9.65,
+      "eval_accuracy": 0.93675,
+      "eval_loss": 0.18856357038021088,
+      "eval_runtime": 17.2651,
+      "eval_samples_per_second": 1390.087,
+      "eval_steps_per_second": 4.344,
+      "step": 23160
+    },
+    {
+      "epoch": 9.66,
+      "learning_rate": 1e-06,
+      "loss": 0.1359,
+      "step": 23180
+    },
+    {
+      "epoch": 9.66,
+      "eval_accuracy": 0.937625,
+      "eval_loss": 0.18765127658843994,
+      "eval_runtime": 17.5512,
+      "eval_samples_per_second": 1367.431,
+      "eval_steps_per_second": 4.273,
+      "step": 23180
+    },
+    {
+      "epoch": 9.67,
+      "learning_rate": 1e-06,
+      "loss": 0.1342,
+      "step": 23200
+    },
+    {
+      "epoch": 9.67,
+      "eval_accuracy": 0.9369166666666666,
+      "eval_loss": 0.1894044429063797,
+      "eval_runtime": 17.3459,
+      "eval_samples_per_second": 1383.612,
+      "eval_steps_per_second": 4.324,
+      "step": 23200
+    },
+    {
+      "epoch": 9.68,
+      "learning_rate": 1e-06,
+      "loss": 0.156,
+      "step": 23220
+    },
+    {
+      "epoch": 9.68,
+      "eval_accuracy": 0.9369166666666666,
+      "eval_loss": 0.18876871466636658,
+      "eval_runtime": 18.1165,
+      "eval_samples_per_second": 1324.757,
+      "eval_steps_per_second": 4.14,
+      "step": 23220
+    },
+    {
+      "epoch": 9.68,
+      "learning_rate": 1e-06,
+      "loss": 0.1527,
+      "step": 23240
+    },
+    {
+      "epoch": 9.68,
+      "eval_accuracy": 0.937375,
+      "eval_loss": 0.18748903274536133,
+      "eval_runtime": 17.4524,
+      "eval_samples_per_second": 1375.169,
+      "eval_steps_per_second": 4.297,
+      "step": 23240
+    },
+    {
+      "epoch": 9.69,
+      "learning_rate": 1e-06,
+      "loss": 0.1608,
+      "step": 23260
+    },
+    {
+      "epoch": 9.69,
+      "eval_accuracy": 0.9370416666666667,
+      "eval_loss": 0.18767432868480682,
+      "eval_runtime": 17.2496,
+      "eval_samples_per_second": 1391.339,
+      "eval_steps_per_second": 4.348,
+      "step": 23260
+    },
+    {
+      "epoch": 9.7,
+      "learning_rate": 1e-06,
+      "loss": 0.1576,
+      "step": 23280
+    },
+    {
+      "epoch": 9.7,
+      "eval_accuracy": 0.9365,
+      "eval_loss": 0.18888776004314423,
+      "eval_runtime": 17.5705,
+      "eval_samples_per_second": 1365.923,
+      "eval_steps_per_second": 4.269,
+      "step": 23280
+    },
+    {
+      "epoch": 9.71,
+      "learning_rate": 1e-06,
+      "loss": 0.1618,
+      "step": 23300
+    },
+    {
+      "epoch": 9.71,
+      "eval_accuracy": 0.9361666666666667,
+      "eval_loss": 0.19041982293128967,
+      "eval_runtime": 16.6527,
+      "eval_samples_per_second": 1441.209,
+      "eval_steps_per_second": 4.504,
+      "step": 23300
+    },
+    {
+      "epoch": 9.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1551,
+      "step": 23320
+    },
+    {
+      "epoch": 9.72,
+      "eval_accuracy": 0.9355833333333333,
+      "eval_loss": 0.19049932062625885,
+      "eval_runtime": 16.922,
+      "eval_samples_per_second": 1418.271,
+      "eval_steps_per_second": 4.432,
+      "step": 23320
+    },
+    {
+      "epoch": 9.72,
+      "learning_rate": 1e-06,
+      "loss": 0.1311,
+      "step": 23340
+    },
+    {
+      "epoch": 9.72,
+      "eval_accuracy": 0.9359166666666666,
+      "eval_loss": 0.18927621841430664,
+      "eval_runtime": 16.8033,
+      "eval_samples_per_second": 1428.291,
+      "eval_steps_per_second": 4.463,
+      "step": 23340
+    },
+    {
+      "epoch": 9.73,
+      "learning_rate": 1e-06,
+      "loss": 0.128,
+      "step": 23360
+    },
+    {
+      "epoch": 9.73,
+      "eval_accuracy": 0.93625,
+      "eval_loss": 0.1896050125360489,
+      "eval_runtime": 16.664,
+      "eval_samples_per_second": 1440.231,
+      "eval_steps_per_second": 4.501,
+      "step": 23360
+    },
+    {
+      "epoch": 9.74,
+      "learning_rate": 1e-06,
+      "loss": 0.1733,
+      "step": 23380
+    },
+    {
+      "epoch": 9.74,
+      "eval_accuracy": 0.934375,
+      "eval_loss": 0.19381864368915558,
+      "eval_runtime": 17.6268,
+      "eval_samples_per_second": 1361.562,
+      "eval_steps_per_second": 4.255,
+      "step": 23380
+    },
+    {
+      "epoch": 9.75,
+      "learning_rate": 1e-06,
+      "loss": 0.1546,
+      "step": 23400
+    },
+    {
+      "epoch": 9.75,
+      "eval_accuracy": 0.93575,
+      "eval_loss": 0.19130919873714447,
+      "eval_runtime": 17.4959,
+      "eval_samples_per_second": 1371.752,
+      "eval_steps_per_second": 4.287,
+      "step": 23400
+    },
+    {
+      "epoch": 9.76,
+      "learning_rate": 1e-06,
+      "loss": 0.2157,
+      "step": 23420
+    },
+    {
+      "epoch": 9.76,
+      "eval_accuracy": 0.9359166666666666,
+      "eval_loss": 0.18996796011924744,
+      "eval_runtime": 17.4322,
+      "eval_samples_per_second": 1376.762,
+      "eval_steps_per_second": 4.302,
+      "step": 23420
+    },
+    {
+      "epoch": 9.77,
+      "learning_rate": 1e-06,
+      "loss": 0.1345,
+      "step": 23440
+    },
+    {
+      "epoch": 9.77,
+      "eval_accuracy": 0.93425,
+      "eval_loss": 0.19452691078186035,
+      "eval_runtime": 17.4799,
+      "eval_samples_per_second": 1373.007,
+      "eval_steps_per_second": 4.291,
+      "step": 23440
+    },
+    {
+      "epoch": 9.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1598,
+      "step": 23460
+    },
+    {
+      "epoch": 9.78,
+      "eval_accuracy": 0.9352916666666666,
+      "eval_loss": 0.19014814496040344,
+      "eval_runtime": 16.5871,
+      "eval_samples_per_second": 1446.91,
+      "eval_steps_per_second": 4.522,
+      "step": 23460
+    },
+    {
+      "epoch": 9.78,
+      "learning_rate": 1e-06,
+      "loss": 0.1189,
+      "step": 23480
+    },
+    {
+      "epoch": 9.78,
+      "eval_accuracy": 0.936875,
+      "eval_loss": 0.1875523328781128,
+      "eval_runtime": 15.4296,
+      "eval_samples_per_second": 1555.449,
+      "eval_steps_per_second": 4.861,
+      "step": 23480
+    },
+    {
+      "epoch": 9.79,
+      "learning_rate": 1e-06,
+      "loss": 0.1406,
+      "step": 23500
+    },
+    {
+      "epoch": 9.79,
+      "eval_accuracy": 0.9364583333333333,
+      "eval_loss": 0.19084198772907257,
+      "eval_runtime": 15.6993,
+      "eval_samples_per_second": 1528.734,
+      "eval_steps_per_second": 4.777,
+      "step": 23500
+    },
+    {
+      "epoch": 9.8,
+      "learning_rate": 1e-06,
+      "loss": 0.1624,
+      "step": 23520
+    },
+    {
+      "epoch": 9.8,
+      "eval_accuracy": 0.9345,
+      "eval_loss": 0.19372619688510895,
+      "eval_runtime": 15.5172,
+      "eval_samples_per_second": 1546.671,
+      "eval_steps_per_second": 4.833,
+      "step": 23520
+    },
+    {
+      "epoch": 9.81,
+      "learning_rate": 1e-06,
+      "loss": 0.1422,
+      "step": 23540
+    },
+    {
+      "epoch": 9.81,
+      "eval_accuracy": 0.9336666666666666,
+      "eval_loss": 0.19716989994049072,
+      "eval_runtime": 15.6871,
+      "eval_samples_per_second": 1529.92,
+      "eval_steps_per_second": 4.781,
+      "step": 23540
+    },
+    {
+      "epoch": 9.82,
+      "learning_rate": 1e-06,
+      "loss": 0.149,
+      "step": 23560
+    },
+    {
+      "epoch": 9.82,
+      "eval_accuracy": 0.9368333333333333,
+      "eval_loss": 0.1898457407951355,
+      "eval_runtime": 15.5314,
+      "eval_samples_per_second": 1545.261,
+      "eval_steps_per_second": 4.829,
+      "step": 23560
+    },
+    {
+      "epoch": 9.82,
+      "learning_rate": 1e-06,
+      "loss": 0.1635,
+      "step": 23580
+    },
+    {
+      "epoch": 9.82,
+      "eval_accuracy": 0.935625,
+      "eval_loss": 0.19061589241027832,
+      "eval_runtime": 15.9229,
+      "eval_samples_per_second": 1507.261,
+      "eval_steps_per_second": 4.71,
+      "step": 23580
+    },
+    {
+      "epoch": 9.83,
+      "learning_rate": 1e-06,
+      "loss": 0.1648,
+      "step": 23600
+    },
+    {
+      "epoch": 9.83,
+      "eval_accuracy": 0.9361666666666667,
+      "eval_loss": 0.18975524604320526,
+      "eval_runtime": 15.3175,
+      "eval_samples_per_second": 1566.837,
+      "eval_steps_per_second": 4.896,
+      "step": 23600
+    },
+    {
+      "epoch": 9.84,
+      "learning_rate": 1e-06,
+      "loss": 0.1577,
+      "step": 23620
+    },
+    {
+      "epoch": 9.84,
+      "eval_accuracy": 0.9354166666666667,
+      "eval_loss": 0.1917409747838974,
+      "eval_runtime": 15.4166,
+      "eval_samples_per_second": 1556.767,
+      "eval_steps_per_second": 4.865,
+      "step": 23620
+    },
+    {
+      "epoch": 9.85,
+      "learning_rate": 1e-06,
+      "loss": 0.1604,
+      "step": 23640
+    },
+    {
+      "epoch": 9.85,
+      "eval_accuracy": 0.9365,
+      "eval_loss": 0.19033245742321014,
+      "eval_runtime": 15.7074,
+      "eval_samples_per_second": 1527.939,
+      "eval_steps_per_second": 4.775,
+      "step": 23640
+    },
+    {
+      "epoch": 9.86,
+      "learning_rate": 1e-06,
+      "loss": 0.1477,
+      "step": 23660
+    },
+    {
+      "epoch": 9.86,
+      "eval_accuracy": 0.9347916666666667,
+      "eval_loss": 0.19305865466594696,
+      "eval_runtime": 15.5818,
+      "eval_samples_per_second": 1540.26,
+      "eval_steps_per_second": 4.813,
+      "step": 23660
+    },
+    {
+      "epoch": 9.87,
+      "learning_rate": 1e-06,
+      "loss": 0.1374,
+      "step": 23680
+    },
+    {
+      "epoch": 9.87,
+      "eval_accuracy": 0.9369166666666666,
+      "eval_loss": 0.18935944139957428,
+      "eval_runtime": 15.6633,
+      "eval_samples_per_second": 1532.247,
+      "eval_steps_per_second": 4.788,
+      "step": 23680
+    },
+    {
+      "epoch": 9.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1524,
+      "step": 23700
+    },
+    {
+      "epoch": 9.88,
+      "eval_accuracy": 0.93625,
+      "eval_loss": 0.19033636152744293,
+      "eval_runtime": 15.5153,
+      "eval_samples_per_second": 1546.863,
+      "eval_steps_per_second": 4.834,
+      "step": 23700
+    },
+    {
+      "epoch": 9.88,
+      "learning_rate": 1e-06,
+      "loss": 0.1899,
+      "step": 23720
+    },
+    {
+      "epoch": 9.88,
+      "eval_accuracy": 0.937,
+      "eval_loss": 0.18940836191177368,
+      "eval_runtime": 15.9167,
+      "eval_samples_per_second": 1507.855,
+      "eval_steps_per_second": 4.712,
+      "step": 23720
+    },
+    {
+      "epoch": 9.89,
+      "learning_rate": 1e-06,
+      "loss": 0.1542,
+      "step": 23740
+    },
+    {
+      "epoch": 9.89,
+      "eval_accuracy": 0.9356666666666666,
+      "eval_loss": 0.1927276849746704,
+      "eval_runtime": 15.6406,
+      "eval_samples_per_second": 1534.465,
+      "eval_steps_per_second": 4.795,
+      "step": 23740
+    },
+    {
+      "epoch": 9.9,
+      "learning_rate": 1e-06,
+      "loss": 0.1709,
+      "step": 23760
+    },
+    {
+      "epoch": 9.9,
+      "eval_accuracy": 0.936,
+      "eval_loss": 0.19114099442958832,
+      "eval_runtime": 16.1487,
+      "eval_samples_per_second": 1486.188,
+      "eval_steps_per_second": 4.644,
+      "step": 23760
+    },
+    {
+      "epoch": 9.91,
+      "learning_rate": 1e-06,
+      "loss": 0.1555,
+      "step": 23780
+    },
+    {
+      "epoch": 9.91,
+      "eval_accuracy": 0.93725,
+      "eval_loss": 0.18706054985523224,
+      "eval_runtime": 15.697,
+      "eval_samples_per_second": 1528.954,
+      "eval_steps_per_second": 4.778,
+      "step": 23780
+    },
+    {
+      "epoch": 9.92,
+      "learning_rate": 1e-06,
+      "loss": 0.1723,
+      "step": 23800
+    },
+    {
+      "epoch": 9.92,
+      "eval_accuracy": 0.9372916666666666,
+      "eval_loss": 0.18747729063034058,
+      "eval_runtime": 15.4803,
+      "eval_samples_per_second": 1550.356,
+      "eval_steps_per_second": 4.845,
+      "step": 23800
+    },
+    {
+      "epoch": 9.93,
+      "learning_rate": 1e-06,
+      "loss": 0.1736,
+      "step": 23820
+    },
+    {
+      "epoch": 9.93,
+      "eval_accuracy": 0.935875,
+      "eval_loss": 0.19002044200897217,
+      "eval_runtime": 15.9218,
+      "eval_samples_per_second": 1507.368,
+      "eval_steps_per_second": 4.711,
+      "step": 23820
+    },
+    {
+      "epoch": 9.93,
+      "learning_rate": 1e-06,
+      "loss": 0.1354,
+      "step": 23840
+    },
+    {
+      "epoch": 9.93,
+      "eval_accuracy": 0.9373333333333334,
+      "eval_loss": 0.18683619797229767,
+      "eval_runtime": 15.4111,
+      "eval_samples_per_second": 1557.318,
+      "eval_steps_per_second": 4.867,
+      "step": 23840
+    },
+    {
+      "epoch": 9.94,
+      "learning_rate": 1e-06,
+      "loss": 0.1798,
+      "step": 23860
+    },
+    {
+      "epoch": 9.94,
+      "eval_accuracy": 0.937375,
+      "eval_loss": 0.1869436800479889,
+      "eval_runtime": 16.0901,
+      "eval_samples_per_second": 1491.604,
+      "eval_steps_per_second": 4.661,
+      "step": 23860
+    },
+    {
+      "epoch": 9.95,
+      "learning_rate": 1e-06,
+      "loss": 0.1258,
+      "step": 23880
+    },
+    {
+      "epoch": 9.95,
+      "eval_accuracy": 0.937,
+      "eval_loss": 0.18814752995967865,
+      "eval_runtime": 15.464,
+      "eval_samples_per_second": 1551.992,
+      "eval_steps_per_second": 4.85,
+      "step": 23880
+    },
+    {
+      "epoch": 9.96,
+      "learning_rate": 1e-06,
+      "loss": 0.1389,
+      "step": 23900
+    },
+    {
+      "epoch": 9.96,
+      "eval_accuracy": 0.9367083333333334,
+      "eval_loss": 0.18954576551914215,
+      "eval_runtime": 15.4402,
+      "eval_samples_per_second": 1554.384,
+      "eval_steps_per_second": 4.857,
+      "step": 23900
+    },
+    {
+      "epoch": 9.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1307,
+      "step": 23920
+    },
+    {
+      "epoch": 9.97,
+      "eval_accuracy": 0.93675,
+      "eval_loss": 0.189622163772583,
+      "eval_runtime": 15.4543,
+      "eval_samples_per_second": 1552.963,
+      "eval_steps_per_second": 4.853,
+      "step": 23920
+    },
+    {
+      "epoch": 9.97,
+      "learning_rate": 1e-06,
+      "loss": 0.1579,
+      "step": 23940
+    },
+    {
+      "epoch": 9.97,
+      "eval_accuracy": 0.937125,
+      "eval_loss": 0.1884654462337494,
+      "eval_runtime": 15.6522,
+      "eval_samples_per_second": 1533.332,
+      "eval_steps_per_second": 4.792,
+      "step": 23940
+    },
+    {
+      "epoch": 9.98,
+      "learning_rate": 1e-06,
+      "loss": 0.1667,
+      "step": 23960
+    },
+    {
+      "epoch": 9.98,
+      "eval_accuracy": 0.937,
+      "eval_loss": 0.18847951292991638,
+      "eval_runtime": 15.9741,
+      "eval_samples_per_second": 1502.433,
+      "eval_steps_per_second": 4.695,
+      "step": 23960
+    },
+    {
+      "epoch": 9.99,
+      "learning_rate": 1e-06,
+      "loss": 0.1407,
+      "step": 23980
+    },
+    {
+      "epoch": 9.99,
+      "eval_accuracy": 0.9372916666666666,
+      "eval_loss": 0.1892939805984497,
+      "eval_runtime": 15.4305,
+      "eval_samples_per_second": 1555.363,
+      "eval_steps_per_second": 4.861,
+      "step": 23980
+    },
+    {
+      "epoch": 10.0,
+      "learning_rate": 1e-06,
+      "loss": 0.1624,
+      "step": 24000
+    },
+    {
+      "epoch": 10.0,
+      "eval_accuracy": 0.9372083333333333,
+      "eval_loss": 0.18687449395656586,
+      "eval_runtime": 15.6942,
+      "eval_samples_per_second": 1529.227,
+      "eval_steps_per_second": 4.779,
+      "step": 24000
+    }
+  ],
+  "max_steps": 24000,
+  "num_train_epochs": 10,
+  "total_flos": 3.005791444826016e+16,
+  "trial_name": null,
+  "trial_params": null
+}