diff --git "a/trainer_state.json" "b/trainer_state.json"
--- "a/trainer_state.json"
+++ "b/trainer_state.json"
@@ -1,4744 +1,3379 @@
 {
-  "best_metric": 6.480716228485107,
-  "best_model_checkpoint": "/data1/attanasiog/babylm/roberta-tiny-10M/checkpoint-5900",
-  "epoch": 31.184852374839537,
-  "global_step": 6050,
+  "best_metric": 2.7371606826782227,
+  "best_model_checkpoint": "/data1/attanasiog/babylm/roberta-tiny-10M/checkpoint-4150",
+  "epoch": 89.57546563904945,
+  "global_step": 4300,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
   "log_history": [
     {
-      "epoch": 0.05,
-      "learning_rate": 4.000000000000001e-06,
-      "loss": 10.8666,
+      "epoch": 0.21,
+      "learning_rate": 8e-05,
+      "loss": 10.4287,
       "step": 10
     },
     {
-      "epoch": 0.1,
-      "learning_rate": 8.000000000000001e-06,
-      "loss": 10.8334,
+      "epoch": 0.41,
+      "learning_rate": 0.00016,
+      "loss": 9.0477,
       "step": 20
     },
     {
-      "epoch": 0.15,
-      "learning_rate": 1.2e-05,
-      "loss": 10.7554,
+      "epoch": 0.62,
+      "learning_rate": 0.00024,
+      "loss": 7.8228,
       "step": 30
     },
     {
-      "epoch": 0.21,
-      "learning_rate": 1.6000000000000003e-05,
-      "loss": 10.6494,
+      "epoch": 0.82,
+      "learning_rate": 0.00032,
+      "loss": 7.3343,
       "step": 40
     },
     {
-      "epoch": 0.26,
-      "learning_rate": 2e-05,
-      "loss": 10.5287,
+      "epoch": 1.04,
+      "learning_rate": 0.0004,
+      "loss": 7.8031,
       "step": 50
     },
     {
-      "epoch": 0.26,
-      "eval_accuracy": 0.048115526656560687,
-      "eval_loss": 10.44776725769043,
-      "eval_runtime": 168.8841,
-      "eval_samples_per_second": 142.435,
-      "eval_steps_per_second": 2.226,
+      "epoch": 1.04,
+      "eval_accuracy": 0.06061240850112075,
+      "eval_loss": 7.355990886688232,
+      "eval_runtime": 145.9523,
+      "eval_samples_per_second": 164.814,
+      "eval_steps_per_second": 5.152,
       "step": 50
     },
     {
-      "epoch": 0.31,
-      "learning_rate": 2e-05,
-      "loss": 10.4004,
+      "epoch": 1.25,
+      "learning_rate": 0.000399995625676045,
+      "loss": 7.2898,
       "step": 60
     },
     {
-      "epoch": 0.36,
-      "learning_rate": 2e-05,
-      "loss": 10.2909,
+      "epoch": 1.45,
+      "learning_rate": 0.0003999825028955268,
+      "loss": 7.1829,
       "step": 70
     },
     {
-      "epoch": 0.41,
-      "learning_rate": 2e-05,
-      "loss": 10.1876,
+      "epoch": 1.66,
+      "learning_rate": 0.0003999606322324786,
+      "loss": 7.0831,
       "step": 80
     },
     {
-      "epoch": 0.46,
-      "learning_rate": 2e-05,
-      "loss": 10.092,
+      "epoch": 1.86,
+      "learning_rate": 0.0003999300146435939,
+      "loss": 6.8807,
       "step": 90
     },
     {
-      "epoch": 0.51,
-      "learning_rate": 2e-05,
-      "loss": 10.0073,
+      "epoch": 2.08,
+      "learning_rate": 0.00039989065146818525,
+      "loss": 7.1948,
       "step": 100
     },
     {
-      "epoch": 0.51,
-      "eval_accuracy": 0.04883696521337363,
-      "eval_loss": 9.954952239990234,
-      "eval_runtime": 168.8432,
-      "eval_samples_per_second": 142.469,
-      "eval_steps_per_second": 2.227,
+      "epoch": 2.08,
+      "eval_accuracy": 0.11823707432860285,
+      "eval_loss": 6.737408638000488,
+      "eval_runtime": 145.8622,
+      "eval_samples_per_second": 164.916,
+      "eval_steps_per_second": 5.156,
       "step": 100
     },
     {
-      "epoch": 0.56,
-      "learning_rate": 2e-05,
-      "loss": 9.9271,
+      "epoch": 2.29,
+      "learning_rate": 0.0003998425444281255,
+      "loss": 6.659,
       "step": 110
     },
     {
-      "epoch": 0.62,
-      "learning_rate": 2e-05,
-      "loss": 9.8418,
+      "epoch": 2.49,
+      "learning_rate": 0.00039978569562777234,
+      "loss": 6.5924,
       "step": 120
     },
     {
-      "epoch": 0.67,
-      "learning_rate": 2e-05,
-      "loss": 9.7731,
+      "epoch": 2.7,
+      "learning_rate": 0.0003997201075538765,
+      "loss": 6.5237,
       "step": 130
     },
     {
-      "epoch": 0.72,
-      "learning_rate": 2e-05,
-      "loss": 9.6976,
+      "epoch": 2.9,
+      "learning_rate": 0.0003996457830754729,
+      "loss": 6.4927,
       "step": 140
     },
     {
-      "epoch": 0.77,
-      "learning_rate": 2e-05,
-      "loss": 9.6268,
+      "epoch": 3.12,
+      "learning_rate": 0.00039956272544375493,
+      "loss": 6.8927,
       "step": 150
     },
     {
-      "epoch": 0.77,
-      "eval_accuracy": 0.04877340177584069,
-      "eval_loss": 9.586450576782227,
-      "eval_runtime": 168.9917,
-      "eval_samples_per_second": 142.344,
-      "eval_steps_per_second": 2.225,
+      "epoch": 3.12,
+      "eval_accuracy": 0.1414597356195163,
+      "eval_loss": 6.502169609069824,
+      "eval_runtime": 145.9635,
+      "eval_samples_per_second": 164.801,
+      "eval_steps_per_second": 5.152,
       "step": 150
     },
     {
-      "epoch": 0.82,
-      "learning_rate": 2e-05,
-      "loss": 9.5519,
+      "epoch": 3.33,
+      "learning_rate": 0.00039947093829193245,
+      "loss": 6.4155,
       "step": 160
     },
     {
-      "epoch": 0.87,
-      "learning_rate": 2e-05,
-      "loss": 9.4822,
+      "epoch": 3.53,
+      "learning_rate": 0.00039937042563507283,
+      "loss": 6.4041,
       "step": 170
     },
     {
-      "epoch": 0.92,
-      "learning_rate": 2e-05,
-      "loss": 9.4239,
+      "epoch": 3.74,
+      "learning_rate": 0.00039926119186992537,
+      "loss": 6.3875,
       "step": 180
     },
     {
-      "epoch": 0.98,
-      "learning_rate": 2e-05,
-      "loss": 9.3531,
+      "epoch": 3.95,
+      "learning_rate": 0.0003991432417747288,
+      "loss": 6.3543,
       "step": 190
     },
     {
-      "epoch": 1.03,
-      "learning_rate": 2e-05,
-      "loss": 9.9837,
+      "epoch": 4.16,
+      "learning_rate": 0.0003990165805090023,
+      "loss": 6.7339,
       "step": 200
     },
     {
-      "epoch": 1.03,
-      "eval_accuracy": 0.04709288795319612,
-      "eval_loss": 9.250182151794434,
-      "eval_runtime": 169.1628,
-      "eval_samples_per_second": 142.2,
-      "eval_steps_per_second": 2.223,
+      "epoch": 4.16,
+      "eval_accuracy": 0.1482938589304516,
+      "eval_loss": 6.400519847869873,
+      "eval_runtime": 145.8639,
+      "eval_samples_per_second": 164.914,
+      "eval_steps_per_second": 5.155,
       "step": 200
     },
     {
-      "epoch": 1.08,
-      "learning_rate": 2e-05,
-      "loss": 9.2265,
+      "epoch": 4.37,
+      "learning_rate": 0.00039888121361332003,
+      "loss": 6.3085,
       "step": 210
     },
     {
-      "epoch": 1.13,
-      "learning_rate": 2e-05,
-      "loss": 9.1555,
+      "epoch": 4.58,
+      "learning_rate": 0.0003987371470090686,
+      "loss": 6.3213,
       "step": 220
     },
     {
-      "epoch": 1.18,
-      "learning_rate": 2e-05,
-      "loss": 9.0872,
+      "epoch": 4.78,
+      "learning_rate": 0.00039858438699818784,
+      "loss": 6.2931,
       "step": 230
     },
     {
-      "epoch": 1.24,
-      "learning_rate": 2e-05,
-      "loss": 9.0276,
+      "epoch": 4.99,
+      "learning_rate": 0.0003984229402628956,
+      "loss": 6.2716,
       "step": 240
     },
     {
-      "epoch": 1.29,
-      "learning_rate": 2e-05,
-      "loss": 8.9701,
+      "epoch": 5.21,
+      "learning_rate": 0.00039825281386539503,
+      "loss": 6.6609,
       "step": 250
     },
     {
-      "epoch": 1.29,
-      "eval_accuracy": 0.04664432564609462,
-      "eval_loss": 8.936966896057129,
-      "eval_runtime": 168.9806,
-      "eval_samples_per_second": 142.354,
-      "eval_steps_per_second": 2.225,
+      "epoch": 5.21,
+      "eval_accuracy": 0.1509599365008845,
+      "eval_loss": 6.3535308837890625,
+      "eval_runtime": 145.9161,
+      "eval_samples_per_second": 164.855,
+      "eval_steps_per_second": 5.154,
       "step": 250
     },
     {
-      "epoch": 1.34,
-      "learning_rate": 2e-05,
-      "loss": 8.913,
+      "epoch": 5.41,
+      "learning_rate": 0.000398074015247566,
+      "loss": 6.2501,
       "step": 260
     },
     {
-      "epoch": 1.39,
-      "learning_rate": 2e-05,
-      "loss": 8.8506,
+      "epoch": 5.62,
+      "learning_rate": 0.0003978865522306392,
+      "loss": 6.2436,
       "step": 270
     },
     {
-      "epoch": 1.44,
-      "learning_rate": 2e-05,
-      "loss": 8.7763,
+      "epoch": 5.82,
+      "learning_rate": 0.0003976904330148543,
+      "loss": 6.2418,
       "step": 280
     },
     {
-      "epoch": 1.49,
-      "learning_rate": 2e-05,
-      "loss": 8.7262,
+      "epoch": 6.04,
+      "learning_rate": 0.00039748566617910113,
+      "loss": 6.6426,
       "step": 290
     },
     {
-      "epoch": 1.54,
-      "learning_rate": 2e-05,
-      "loss": 8.6689,
+      "epoch": 6.25,
+      "learning_rate": 0.0003972722606805445,
+      "loss": 6.1972,
       "step": 300
     },
     {
-      "epoch": 1.54,
-      "eval_accuracy": 0.04727021760311789,
-      "eval_loss": 8.644742965698242,
-      "eval_runtime": 168.948,
-      "eval_samples_per_second": 142.381,
-      "eval_steps_per_second": 2.226,
+      "epoch": 6.25,
+      "eval_accuracy": 0.15188271193711186,
+      "eval_loss": 6.332435607910156,
+      "eval_runtime": 146.0198,
+      "eval_samples_per_second": 164.738,
+      "eval_steps_per_second": 5.15,
       "step": 300
     },
     {
-      "epoch": 1.6,
-      "learning_rate": 2e-05,
-      "loss": 8.6119,
+      "epoch": 6.45,
+      "learning_rate": 0.00039705022585423216,
+      "loss": 6.2183,
       "step": 310
     },
     {
-      "epoch": 1.65,
-      "learning_rate": 2e-05,
-      "loss": 8.5511,
+      "epoch": 6.66,
+      "learning_rate": 0.0003968195714126868,
+      "loss": 6.1899,
       "step": 320
     },
     {
-      "epoch": 1.7,
-      "learning_rate": 2e-05,
-      "loss": 8.503,
+      "epoch": 6.86,
+      "learning_rate": 0.00039658030744548075,
+      "loss": 6.192,
       "step": 330
     },
     {
-      "epoch": 1.75,
-      "learning_rate": 2e-05,
-      "loss": 8.4466,
+      "epoch": 7.08,
+      "learning_rate": 0.0003963324444187952,
+      "loss": 6.5971,
       "step": 340
     },
     {
-      "epoch": 1.8,
-      "learning_rate": 2e-05,
-      "loss": 8.3893,
+      "epoch": 7.29,
+      "learning_rate": 0.0003960759931749619,
+      "loss": 6.1685,
       "step": 350
     },
     {
-      "epoch": 1.8,
-      "eval_accuracy": 0.04733514138538159,
-      "eval_loss": 8.379368782043457,
-      "eval_runtime": 169.22,
-      "eval_samples_per_second": 142.152,
-      "eval_steps_per_second": 2.222,
+      "epoch": 7.29,
+      "eval_accuracy": 0.15276707185574287,
+      "eval_loss": 6.302943706512451,
+      "eval_runtime": 145.9601,
+      "eval_samples_per_second": 164.805,
+      "eval_steps_per_second": 5.152,
       "step": 350
     },
     {
-      "epoch": 1.85,
-      "learning_rate": 2e-05,
-      "loss": 8.3375,
+      "epoch": 7.49,
+      "learning_rate": 0.00039581096493198893,
+      "loss": 6.1653,
       "step": 360
     },
     {
-      "epoch": 1.9,
-      "learning_rate": 2e-05,
-      "loss": 8.2824,
+      "epoch": 7.7,
+      "learning_rate": 0.0003955373712830703,
+      "loss": 6.1623,
       "step": 370
     },
     {
-      "epoch": 1.96,
-      "learning_rate": 2e-05,
-      "loss": 8.2317,
+      "epoch": 7.9,
+      "learning_rate": 0.00039525522419607854,
+      "loss": 6.1604,
       "step": 380
     },
     {
-      "epoch": 2.01,
-      "learning_rate": 2e-05,
-      "loss": 8.8178,
+      "epoch": 8.12,
+      "learning_rate": 0.0003949645360130412,
+      "loss": 6.5496,
       "step": 390
     },
     {
-      "epoch": 2.06,
-      "learning_rate": 2e-05,
-      "loss": 8.1697,
+      "epoch": 8.33,
+      "learning_rate": 0.0003946653194496012,
+      "loss": 6.1302,
       "step": 400
     },
     {
-      "epoch": 2.06,
-      "eval_accuracy": 0.050607646781225514,
-      "eval_loss": 8.134186744689941,
-      "eval_runtime": 169.087,
-      "eval_samples_per_second": 142.264,
-      "eval_steps_per_second": 2.224,
+      "epoch": 8.33,
+      "eval_accuracy": 0.152128546451089,
+      "eval_loss": 6.2827558517456055,
+      "eval_runtime": 145.9935,
+      "eval_samples_per_second": 164.768,
+      "eval_steps_per_second": 5.151,
       "step": 400
     },
     {
-      "epoch": 2.11,
-      "learning_rate": 2e-05,
-      "loss": 8.0758,
+      "epoch": 8.53,
+      "learning_rate": 0.00039435758759446025,
+      "loss": 6.1368,
       "step": 410
     },
     {
-      "epoch": 2.16,
-      "learning_rate": 2e-05,
-      "loss": 8.048,
+      "epoch": 8.74,
+      "learning_rate": 0.00039404135390880664,
+      "loss": 6.1171,
       "step": 420
     },
     {
-      "epoch": 2.22,
-      "learning_rate": 2e-05,
-      "loss": 7.9864,
+      "epoch": 8.95,
+      "learning_rate": 0.0003937166322257262,
+      "loss": 6.1463,
       "step": 430
     },
     {
-      "epoch": 2.27,
-      "learning_rate": 2e-05,
-      "loss": 7.9734,
+      "epoch": 9.16,
+      "learning_rate": 0.00039338343674959745,
+      "loss": 6.537,
       "step": 440
     },
     {
-      "epoch": 2.32,
-      "learning_rate": 2e-05,
-      "loss": 7.926,
+      "epoch": 9.37,
+      "learning_rate": 0.00039304178205546976,
+      "loss": 6.093,
       "step": 450
     },
     {
-      "epoch": 2.32,
-      "eval_accuracy": 0.06168861888884143,
-      "eval_loss": 7.922093868255615,
-      "eval_runtime": 168.9407,
-      "eval_samples_per_second": 142.387,
-      "eval_steps_per_second": 2.226,
+      "epoch": 9.37,
+      "eval_accuracy": 0.15364162638834264,
+      "eval_loss": 6.256844520568848,
+      "eval_runtime": 146.039,
+      "eval_samples_per_second": 164.716,
+      "eval_steps_per_second": 5.149,
       "step": 450
     },
     {
-      "epoch": 2.37,
-      "learning_rate": 2e-05,
-      "loss": 7.8712,
+      "epoch": 9.58,
+      "learning_rate": 0.00039269168308842634,
+      "loss": 6.0973,
       "step": 460
     },
     {
-      "epoch": 2.42,
-      "learning_rate": 2e-05,
-      "loss": 7.8416,
+      "epoch": 9.78,
+      "learning_rate": 0.00039233315516293006,
+      "loss": 6.1012,
       "step": 470
     },
     {
-      "epoch": 2.47,
-      "learning_rate": 2e-05,
-      "loss": 7.8118,
+      "epoch": 9.99,
+      "learning_rate": 0.00039196621396215403,
+      "loss": 6.0809,
       "step": 480
     },
     {
-      "epoch": 2.52,
-      "learning_rate": 2e-05,
-      "loss": 7.7706,
+      "epoch": 10.21,
+      "learning_rate": 0.000391590875537295,
+      "loss": 6.4765,
       "step": 490
     },
     {
-      "epoch": 2.58,
-      "learning_rate": 2e-05,
-      "loss": 7.7329,
+      "epoch": 10.41,
+      "learning_rate": 0.00039120715630687155,
+      "loss": 6.0543,
       "step": 500
     },
     {
-      "epoch": 2.58,
-      "eval_accuracy": 0.06273108274751771,
-      "eval_loss": 7.739767074584961,
-      "eval_runtime": 169.0756,
-      "eval_samples_per_second": 142.274,
-      "eval_steps_per_second": 2.224,
+      "epoch": 10.41,
+      "eval_accuracy": 0.15444620739515735,
+      "eval_loss": 6.24298620223999,
+      "eval_runtime": 145.9243,
+      "eval_samples_per_second": 164.846,
+      "eval_steps_per_second": 5.153,
       "step": 500
     },
     {
-      "epoch": 2.63,
-      "learning_rate": 2e-05,
-      "loss": 7.7048,
+      "epoch": 10.62,
+      "learning_rate": 0.000390815073056006,
+      "loss": 6.0777,
       "step": 510
     },
     {
-      "epoch": 2.68,
-      "learning_rate": 2e-05,
-      "loss": 7.6478,
+      "epoch": 10.82,
+      "learning_rate": 0.00039041464293568983,
+      "loss": 6.0697,
       "step": 520
     },
     {
-      "epoch": 2.73,
-      "learning_rate": 2e-05,
-      "loss": 7.6647,
+      "epoch": 11.04,
+      "learning_rate": 0.00039000588346203374,
+      "loss": 6.4636,
       "step": 530
     },
     {
-      "epoch": 2.78,
-      "learning_rate": 2e-05,
-      "loss": 7.6017,
+      "epoch": 11.25,
+      "learning_rate": 0.0003895888125155014,
+      "loss": 6.0487,
       "step": 540
     },
     {
-      "epoch": 2.83,
-      "learning_rate": 2e-05,
-      "loss": 7.582,
+      "epoch": 11.45,
+      "learning_rate": 0.00038916344834012695,
+      "loss": 6.0479,
       "step": 550
     },
     {
-      "epoch": 2.83,
-      "eval_accuracy": 0.06907213036340595,
-      "eval_loss": 7.5844244956970215,
-      "eval_runtime": 169.0372,
-      "eval_samples_per_second": 142.306,
-      "eval_steps_per_second": 2.224,
+      "epoch": 11.45,
+      "eval_accuracy": 0.1541217862327054,
+      "eval_loss": 6.234572887420654,
+      "eval_runtime": 145.8799,
+      "eval_samples_per_second": 164.896,
+      "eval_steps_per_second": 5.155,
       "step": 550
     },
     {
-      "epoch": 2.88,
-      "learning_rate": 2e-05,
-      "loss": 7.5512,
+      "epoch": 11.66,
+      "learning_rate": 0.00038872980954271757,
+      "loss": 6.0617,
       "step": 560
     },
     {
-      "epoch": 2.93,
-      "learning_rate": 2e-05,
-      "loss": 7.505,
+      "epoch": 11.86,
+      "learning_rate": 0.00038828791509203895,
+      "loss": 6.0441,
       "step": 570
     },
     {
-      "epoch": 2.99,
-      "learning_rate": 2e-05,
-      "loss": 7.523,
+      "epoch": 12.08,
+      "learning_rate": 0.00038783778431798597,
+      "loss": 6.4461,
       "step": 580
     },
     {
-      "epoch": 3.04,
-      "learning_rate": 2e-05,
-      "loss": 8.0083,
+      "epoch": 12.29,
+      "learning_rate": 0.0003873794369107369,
+      "loss": 6.0258,
       "step": 590
     },
     {
-      "epoch": 3.09,
-      "learning_rate": 2e-05,
-      "loss": 7.4419,
+      "epoch": 12.49,
+      "learning_rate": 0.0003869128929198922,
+      "loss": 6.0372,
       "step": 600
     },
     {
-      "epoch": 3.09,
-      "eval_accuracy": 0.07293786059545417,
-      "eval_loss": 7.46204137802124,
-      "eval_runtime": 169.0193,
-      "eval_samples_per_second": 142.321,
-      "eval_steps_per_second": 2.225,
+      "epoch": 12.49,
+      "eval_accuracy": 0.1545538581772011,
+      "eval_loss": 6.223215103149414,
+      "eval_runtime": 145.9665,
+      "eval_samples_per_second": 164.798,
+      "eval_steps_per_second": 5.152,
       "step": 600
     },
     {
-      "epoch": 3.14,
-      "learning_rate": 2e-05,
-      "loss": 7.3961,
+      "epoch": 12.7,
+      "learning_rate": 0.0003864381727535973,
+      "loss": 6.0353,
       "step": 610
     },
     {
-      "epoch": 3.2,
-      "learning_rate": 2e-05,
-      "loss": 7.3728,
+      "epoch": 12.9,
+      "learning_rate": 0.00038595529717765027,
+      "loss": 6.041,
       "step": 620
     },
     {
-      "epoch": 3.25,
-      "learning_rate": 2e-05,
-      "loss": 7.3709,
+      "epoch": 13.12,
+      "learning_rate": 0.0003854642873145931,
+      "loss": 6.4207,
       "step": 630
     },
     {
-      "epoch": 3.3,
-      "learning_rate": 2e-05,
-      "loss": 7.3295,
+      "epoch": 13.33,
+      "learning_rate": 0.00038496516464278776,
+      "loss": 6.006,
       "step": 640
     },
     {
-      "epoch": 3.35,
-      "learning_rate": 2e-05,
-      "loss": 7.3658,
+      "epoch": 13.53,
+      "learning_rate": 0.00038445795099547697,
+      "loss": 6.0127,
       "step": 650
     },
     {
-      "epoch": 3.35,
-      "eval_accuracy": 0.07808497071718955,
-      "eval_loss": 7.373513698577881,
-      "eval_runtime": 169.1773,
-      "eval_samples_per_second": 142.188,
-      "eval_steps_per_second": 2.223,
+      "epoch": 13.53,
+      "eval_accuracy": 0.15411265298876436,
+      "eval_loss": 6.213912010192871,
+      "eval_runtime": 145.9328,
+      "eval_samples_per_second": 164.836,
+      "eval_steps_per_second": 5.153,
       "step": 650
     },
     {
-      "epoch": 3.4,
-      "learning_rate": 2e-05,
-      "loss": 7.342,
+      "epoch": 13.74,
+      "learning_rate": 0.0003839426685598287,
+      "loss": 6.0006,
       "step": 660
     },
     {
-      "epoch": 3.45,
-      "learning_rate": 2e-05,
-      "loss": 7.3369,
+      "epoch": 13.95,
+      "learning_rate": 0.000383419339875966,
+      "loss": 6.0152,
       "step": 670
     },
     {
-      "epoch": 3.5,
-      "learning_rate": 2e-05,
-      "loss": 7.292,
+      "epoch": 14.16,
+      "learning_rate": 0.00038288798783598087,
+      "loss": 6.3908,
       "step": 680
     },
     {
-      "epoch": 3.55,
-      "learning_rate": 2e-05,
-      "loss": 7.2974,
+      "epoch": 14.37,
+      "learning_rate": 0.0003823486356829329,
+      "loss": 5.9744,
       "step": 690
     },
     {
-      "epoch": 3.61,
-      "learning_rate": 2e-05,
-      "loss": 7.2857,
+      "epoch": 14.58,
+      "learning_rate": 0.0003818013070098325,
+      "loss": 5.968,
       "step": 700
     },
     {
-      "epoch": 3.61,
-      "eval_accuracy": 0.08012986753100795,
-      "eval_loss": 7.304928302764893,
-      "eval_runtime": 169.0533,
-      "eval_samples_per_second": 142.292,
-      "eval_steps_per_second": 2.224,
+      "epoch": 14.58,
+      "eval_accuracy": 0.15472111663693397,
+      "eval_loss": 6.20527458190918,
+      "eval_runtime": 145.9446,
+      "eval_samples_per_second": 164.823,
+      "eval_steps_per_second": 5.153,
       "step": 700
     },
     {
-      "epoch": 3.66,
-      "learning_rate": 2e-05,
-      "loss": 7.2677,
+      "epoch": 14.78,
+      "learning_rate": 0.0003812460257586089,
+      "loss": 5.9813,
       "step": 710
     },
     {
-      "epoch": 3.71,
-      "learning_rate": 2e-05,
-      "loss": 7.2592,
+      "epoch": 14.99,
+      "learning_rate": 0.000380682816219063,
+      "loss": 6.0108,
       "step": 720
     },
     {
-      "epoch": 3.76,
-      "learning_rate": 2e-05,
-      "loss": 7.2564,
+      "epoch": 15.21,
+      "learning_rate": 0.00038011170302780446,
+      "loss": 6.3495,
       "step": 730
     },
     {
-      "epoch": 3.81,
-      "learning_rate": 2e-05,
-      "loss": 7.2111,
+      "epoch": 15.41,
+      "learning_rate": 0.00037953271116717444,
+      "loss": 5.9708,
       "step": 740
     },
     {
-      "epoch": 3.86,
-      "learning_rate": 2e-05,
-      "loss": 7.224,
+      "epoch": 15.62,
+      "learning_rate": 0.0003789458659641527,
+      "loss": 5.9635,
       "step": 750
     },
     {
-      "epoch": 3.86,
-      "eval_accuracy": 0.08311436007920397,
-      "eval_loss": 7.2554144859313965,
-      "eval_runtime": 169.0481,
-      "eval_samples_per_second": 142.297,
-      "eval_steps_per_second": 2.224,
+      "epoch": 15.62,
+      "eval_accuracy": 0.15486276242328167,
+      "eval_loss": 6.199557781219482,
+      "eval_runtime": 145.9791,
+      "eval_samples_per_second": 164.784,
+      "eval_steps_per_second": 5.151,
       "step": 750
     },
     {
-      "epoch": 3.91,
-      "learning_rate": 2e-05,
-      "loss": 7.2082,
+      "epoch": 15.82,
+      "learning_rate": 0.0003783511930892495,
+      "loss": 5.9756,
       "step": 760
     },
     {
-      "epoch": 3.97,
-      "learning_rate": 2e-05,
-      "loss": 7.2165,
+      "epoch": 16.04,
+      "learning_rate": 0.00037774871855538275,
+      "loss": 6.3631,
       "step": 770
     },
     {
-      "epoch": 4.02,
-      "learning_rate": 2e-05,
-      "loss": 7.7143,
+      "epoch": 16.25,
+      "learning_rate": 0.00037713846871674045,
+      "loss": 5.9497,
       "step": 780
     },
     {
-      "epoch": 4.07,
-      "learning_rate": 2e-05,
-      "loss": 7.1994,
+      "epoch": 16.45,
+      "learning_rate": 0.0003765204702676274,
+      "loss": 5.9433,
       "step": 790
     },
     {
-      "epoch": 4.12,
-      "learning_rate": 2e-05,
-      "loss": 7.1851,
+      "epoch": 16.66,
+      "learning_rate": 0.0003758947502412978,
+      "loss": 5.9479,
       "step": 800
     },
     {
-      "epoch": 4.12,
-      "eval_accuracy": 0.0853165114983084,
-      "eval_loss": 7.208171844482422,
-      "eval_runtime": 169.0677,
-      "eval_samples_per_second": 142.28,
-      "eval_steps_per_second": 2.224,
+      "epoch": 16.66,
+      "eval_accuracy": 0.15478355696794033,
+      "eval_loss": 6.195274353027344,
+      "eval_runtime": 145.939,
+      "eval_samples_per_second": 164.829,
+      "eval_steps_per_second": 5.153,
       "step": 800
     },
     {
-      "epoch": 4.17,
-      "learning_rate": 2e-05,
-      "loss": 7.1016,
+      "epoch": 16.86,
+      "learning_rate": 0.0003752613360087727,
+      "loss": 5.9614,
       "step": 810
     },
     {
-      "epoch": 4.23,
-      "learning_rate": 2e-05,
-      "loss": 7.1461,
+      "epoch": 17.08,
+      "learning_rate": 0.00037462025527764265,
+      "loss": 6.326,
       "step": 820
     },
     {
-      "epoch": 4.28,
-      "learning_rate": 2e-05,
-      "loss": 7.1433,
+      "epoch": 17.29,
+      "learning_rate": 0.00037397153609085553,
+      "loss": 5.9293,
       "step": 830
     },
     {
-      "epoch": 4.33,
-      "learning_rate": 2e-05,
-      "loss": 7.1318,
+      "epoch": 17.49,
+      "learning_rate": 0.0003733152068254901,
+      "loss": 5.9305,
       "step": 840
     },
     {
-      "epoch": 4.38,
-      "learning_rate": 2e-05,
-      "loss": 7.1327,
+      "epoch": 17.7,
+      "learning_rate": 0.00037265129619151483,
+      "loss": 5.9371,
       "step": 850
     },
     {
-      "epoch": 4.38,
-      "eval_accuracy": 0.08777750406730751,
-      "eval_loss": 7.16784143447876,
-      "eval_runtime": 169.0166,
-      "eval_samples_per_second": 142.323,
-      "eval_steps_per_second": 2.225,
+      "epoch": 17.7,
+      "eval_accuracy": 0.15451778319531595,
+      "eval_loss": 6.1887054443359375,
+      "eval_runtime": 145.8431,
+      "eval_samples_per_second": 164.938,
+      "eval_steps_per_second": 5.156,
       "step": 850
     },
     {
-      "epoch": 4.43,
-      "learning_rate": 2e-05,
-      "loss": 7.1428,
+      "epoch": 17.9,
+      "learning_rate": 0.00037197983323053143,
+      "loss": 5.9348,
       "step": 860
     },
     {
-      "epoch": 4.48,
-      "learning_rate": 2e-05,
-      "loss": 7.1264,
+      "epoch": 18.12,
+      "learning_rate": 0.00037130084731450515,
+      "loss": 6.2994,
       "step": 870
     },
     {
-      "epoch": 4.53,
-      "learning_rate": 2e-05,
-      "loss": 7.1309,
+      "epoch": 18.33,
+      "learning_rate": 0.0003706143681444795,
+      "loss": 5.8969,
       "step": 880
     },
     {
-      "epoch": 4.59,
-      "learning_rate": 2e-05,
-      "loss": 7.1299,
+      "epoch": 18.53,
+      "learning_rate": 0.0003699204257492774,
+      "loss": 5.9219,
       "step": 890
     },
     {
-      "epoch": 4.64,
-      "learning_rate": 2e-05,
-      "loss": 7.0947,
+      "epoch": 18.74,
+      "learning_rate": 0.0003692190504841871,
+      "loss": 5.9046,
       "step": 900
     },
     {
-      "epoch": 4.64,
-      "eval_accuracy": 0.09090520258816484,
-      "eval_loss": 7.132561683654785,
-      "eval_runtime": 168.9513,
-      "eval_samples_per_second": 142.378,
-      "eval_steps_per_second": 2.225,
+      "epoch": 18.74,
+      "eval_accuracy": 0.1545486653674884,
+      "eval_loss": 6.161332130432129,
+      "eval_runtime": 145.9406,
+      "eval_samples_per_second": 164.827,
+      "eval_steps_per_second": 5.153,
       "step": 900
     },
     {
-      "epoch": 4.69,
-      "learning_rate": 2e-05,
-      "loss": 7.0953,
+      "epoch": 18.95,
+      "learning_rate": 0.00036851027302963493,
+      "loss": 5.9011,
       "step": 910
     },
     {
-      "epoch": 4.74,
-      "learning_rate": 2e-05,
-      "loss": 7.1143,
+      "epoch": 19.16,
+      "learning_rate": 0.00036779412438984294,
+      "loss": 6.2593,
       "step": 920
     },
     {
-      "epoch": 4.79,
-      "learning_rate": 2e-05,
-      "loss": 7.0972,
+      "epoch": 19.37,
+      "learning_rate": 0.0003670706358914725,
+      "loss": 5.8755,
       "step": 930
     },
     {
-      "epoch": 4.84,
-      "learning_rate": 2e-05,
-      "loss": 7.109,
+      "epoch": 19.58,
+      "learning_rate": 0.0003663398391822543,
+      "loss": 5.8396,
       "step": 940
     },
     {
-      "epoch": 4.89,
-      "learning_rate": 2e-05,
-      "loss": 7.0761,
+      "epoch": 19.78,
+      "learning_rate": 0.00036560176622960403,
+      "loss": 5.8368,
       "step": 950
     },
     {
-      "epoch": 4.89,
-      "eval_accuracy": 0.09188660007613132,
-      "eval_loss": 7.10692834854126,
-      "eval_runtime": 168.925,
-      "eval_samples_per_second": 142.4,
-      "eval_steps_per_second": 2.226,
+      "epoch": 19.78,
+      "eval_accuracy": 0.15570189218059025,
+      "eval_loss": 6.095159530639648,
+      "eval_runtime": 145.7599,
+      "eval_samples_per_second": 165.032,
+      "eval_steps_per_second": 5.159,
       "step": 950
     },
     {
-      "epoch": 4.94,
-      "learning_rate": 2e-05,
-      "loss": 7.0359,
+      "epoch": 19.99,
+      "learning_rate": 0.00036485644931922353,
+      "loss": 5.8184,
       "step": 960
     },
     {
-      "epoch": 5.0,
-      "learning_rate": 2e-05,
-      "loss": 7.0479,
+      "epoch": 20.21,
+      "learning_rate": 0.0003641039210536889,
+      "loss": 6.1866,
       "step": 970
     },
     {
-      "epoch": 5.05,
-      "learning_rate": 2e-05,
-      "loss": 7.5649,
+      "epoch": 20.41,
+      "learning_rate": 0.0003633442143510245,
+      "loss": 5.7848,
       "step": 980
     },
     {
-      "epoch": 5.1,
-      "learning_rate": 2e-05,
-      "loss": 7.0153,
+      "epoch": 20.62,
+      "learning_rate": 0.00036257736244326246,
+      "loss": 5.7807,
       "step": 990
     },
     {
-      "epoch": 5.15,
-      "learning_rate": 2e-05,
-      "loss": 7.0551,
+      "epoch": 20.82,
+      "learning_rate": 0.0003618033988749895,
+      "loss": 5.7914,
       "step": 1000
     },
     {
-      "epoch": 5.15,
-      "eval_accuracy": 0.0943313810914526,
-      "eval_loss": 7.080649375915527,
-      "eval_runtime": 169.1004,
-      "eval_samples_per_second": 142.253,
-      "eval_steps_per_second": 2.224,
+      "epoch": 20.82,
+      "eval_accuracy": 0.15694020859066315,
+      "eval_loss": 6.032991409301758,
+      "eval_runtime": 145.9881,
+      "eval_samples_per_second": 164.774,
+      "eval_steps_per_second": 5.151,
       "step": 1000
     },
     {
-      "epoch": 5.21,
-      "learning_rate": 2e-05,
-      "loss": 7.0119,
+      "epoch": 21.04,
+      "learning_rate": 0.0003610223575018795,
+      "loss": 6.1552,
       "step": 1010
     },
     {
-      "epoch": 5.26,
-      "learning_rate": 2e-05,
-      "loss": 7.0506,
+      "epoch": 21.25,
+      "learning_rate": 0.00036023427248921215,
+      "loss": 5.7428,
       "step": 1020
     },
     {
-      "epoch": 5.31,
-      "learning_rate": 2e-05,
-      "loss": 7.0266,
+      "epoch": 21.45,
+      "learning_rate": 0.0003594391783103792,
+      "loss": 5.7276,
       "step": 1030
     },
     {
-      "epoch": 5.36,
-      "learning_rate": 2e-05,
-      "loss": 7.0121,
+      "epoch": 21.66,
+      "learning_rate": 0.00035863710974537563,
+      "loss": 5.7289,
       "step": 1040
     },
     {
-      "epoch": 5.41,
-      "learning_rate": 2e-05,
-      "loss": 7.0389,
+      "epoch": 21.86,
+      "learning_rate": 0.00035782810187927875,
+      "loss": 5.7026,
       "step": 1050
     },
     {
-      "epoch": 5.41,
-      "eval_accuracy": 0.09524817664480986,
-      "eval_loss": 7.058758735656738,
-      "eval_runtime": 169.0231,
-      "eval_samples_per_second": 142.318,
-      "eval_steps_per_second": 2.225,
+      "epoch": 21.86,
+      "eval_accuracy": 0.16123595961673237,
+      "eval_loss": 5.942953109741211,
+      "eval_runtime": 145.9911,
+      "eval_samples_per_second": 164.77,
+      "eval_steps_per_second": 5.151,
       "step": 1050
     },
     {
-      "epoch": 5.46,
-      "learning_rate": 2e-05,
-      "loss": 7.002,
+      "epoch": 22.08,
+      "learning_rate": 0.0003570121901007136,
+      "loss": 6.0423,
       "step": 1060
     },
     {
-      "epoch": 5.51,
-      "learning_rate": 2e-05,
-      "loss": 7.0223,
+      "epoch": 22.29,
+      "learning_rate": 0.0003561894101003044,
+      "loss": 5.6495,
       "step": 1070
     },
     {
-      "epoch": 5.56,
-      "learning_rate": 2e-05,
-      "loss": 6.9822,
+      "epoch": 22.49,
+      "learning_rate": 0.00035535979786911396,
+      "loss": 5.6223,
       "step": 1080
     },
     {
-      "epoch": 5.62,
-      "learning_rate": 2e-05,
-      "loss": 6.9993,
+      "epoch": 22.7,
+      "learning_rate": 0.00035452338969706876,
+      "loss": 5.5675,
       "step": 1090
     },
     {
-      "epoch": 5.67,
-      "learning_rate": 2e-05,
-      "loss": 7.0226,
+      "epoch": 22.9,
+      "learning_rate": 0.00035368022217137184,
+      "loss": 5.491,
       "step": 1100
     },
     {
-      "epoch": 5.67,
-      "eval_accuracy": 0.09638965813145792,
-      "eval_loss": 7.037946701049805,
-      "eval_runtime": 169.0695,
-      "eval_samples_per_second": 142.279,
-      "eval_steps_per_second": 2.224,
+      "epoch": 22.9,
+      "eval_accuracy": 0.19736824293775215,
+      "eval_loss": 5.609994888305664,
+      "eval_runtime": 146.0961,
+      "eval_samples_per_second": 164.652,
+      "eval_steps_per_second": 5.147,
       "step": 1100
     },
     {
-      "epoch": 5.72,
-      "learning_rate": 2e-05,
-      "loss": 7.0049,
+      "epoch": 23.12,
+      "learning_rate": 0.00035283033217490227,
+      "loss": 5.6961,
       "step": 1110
     },
     {
-      "epoch": 5.77,
-      "learning_rate": 2e-05,
-      "loss": 6.9835,
+      "epoch": 23.33,
+      "learning_rate": 0.00035197375688460176,
+      "loss": 5.239,
       "step": 1120
     },
     {
-      "epoch": 5.82,
-      "learning_rate": 2e-05,
-      "loss": 6.9924,
+      "epoch": 23.53,
+      "learning_rate": 0.0003511105337698484,
+      "loss": 5.1252,
       "step": 1130
     },
     {
-      "epoch": 5.87,
-      "learning_rate": 2e-05,
-      "loss": 6.9933,
+      "epoch": 23.74,
+      "learning_rate": 0.0003502407005908177,
+      "loss": 5.0182,
       "step": 1140
     },
     {
-      "epoch": 5.92,
-      "learning_rate": 2e-05,
-      "loss": 6.9992,
+      "epoch": 23.95,
+      "learning_rate": 0.0003493642953968308,
+      "loss": 4.9289,
       "step": 1150
     },
     {
-      "epoch": 5.92,
-      "eval_accuracy": 0.09752125105444163,
-      "eval_loss": 7.01423454284668,
-      "eval_runtime": 169.1114,
-      "eval_samples_per_second": 142.244,
-      "eval_steps_per_second": 2.223,
+      "epoch": 23.95,
+      "eval_accuracy": 0.27019214299497635,
+      "eval_loss": 4.960735321044922,
+      "eval_runtime": 146.1516,
+      "eval_samples_per_second": 164.589,
+      "eval_steps_per_second": 5.145,
       "step": 1150
     },
     {
-      "epoch": 5.98,
-      "learning_rate": 2e-05,
-      "loss": 6.9667,
+      "epoch": 24.16,
+      "learning_rate": 0.00034848135652469,
+      "loss": 5.1346,
       "step": 1160
     },
     {
-      "epoch": 6.03,
-      "learning_rate": 2e-05,
-      "loss": 7.4753,
+      "epoch": 24.37,
+      "learning_rate": 0.00034759192259700196,
+      "loss": 4.7377,
       "step": 1170
     },
     {
-      "epoch": 6.08,
-      "learning_rate": 2e-05,
-      "loss": 6.9355,
+      "epoch": 24.58,
+      "learning_rate": 0.000346696032520488,
+      "loss": 4.6538,
       "step": 1180
     },
     {
-      "epoch": 6.13,
-      "learning_rate": 2e-05,
-      "loss": 6.9903,
+      "epoch": 24.78,
+      "learning_rate": 0.00034579372548428235,
+      "loss": 4.608,
       "step": 1190
     },
     {
-      "epoch": 6.18,
-      "learning_rate": 2e-05,
-      "loss": 6.9382,
+      "epoch": 24.99,
+      "learning_rate": 0.00034488504095821784,
+      "loss": 4.5214,
       "step": 1200
     },
     {
-      "epoch": 6.18,
-      "eval_accuracy": 0.09855111513654274,
-      "eval_loss": 6.997906684875488,
-      "eval_runtime": 169.2157,
-      "eval_samples_per_second": 142.156,
-      "eval_steps_per_second": 2.222,
+      "epoch": 24.99,
+      "eval_accuracy": 0.3050591252908655,
+      "eval_loss": 4.579548358917236,
+      "eval_runtime": 146.015,
+      "eval_samples_per_second": 164.743,
+      "eval_steps_per_second": 5.15,
       "step": 1200
     },
     {
-      "epoch": 6.24,
-      "learning_rate": 2e-05,
-      "loss": 6.9745,
+      "epoch": 25.21,
+      "learning_rate": 0.0003439700186910993,
+      "loss": 4.7508,
       "step": 1210
     },
     {
-      "epoch": 6.29,
-      "learning_rate": 2e-05,
-      "loss": 6.9784,
+      "epoch": 25.41,
+      "learning_rate": 0.00034304869870896513,
+      "loss": 4.4132,
       "step": 1220
     },
     {
-      "epoch": 6.34,
-      "learning_rate": 2e-05,
-      "loss": 6.9698,
+      "epoch": 25.62,
+      "learning_rate": 0.00034212112131333587,
+      "loss": 4.3489,
       "step": 1230
     },
     {
-      "epoch": 6.39,
-      "learning_rate": 2e-05,
-      "loss": 6.9203,
+      "epoch": 25.82,
+      "learning_rate": 0.0003411873270794518,
+      "loss": 4.3454,
       "step": 1240
     },
     {
-      "epoch": 6.44,
-      "learning_rate": 2e-05,
-      "loss": 6.956,
+      "epoch": 26.04,
+      "learning_rate": 0.00034024735685449773,
+      "loss": 4.5663,
       "step": 1250
     },
     {
-      "epoch": 6.44,
-      "eval_accuracy": 0.09872446648175658,
-      "eval_loss": 6.982813358306885,
-      "eval_runtime": 168.9414,
-      "eval_samples_per_second": 142.387,
-      "eval_steps_per_second": 2.226,
+      "epoch": 26.04,
+      "eval_accuracy": 0.32645309469898054,
+      "eval_loss": 4.345365047454834,
+      "eval_runtime": 146.0915,
+      "eval_samples_per_second": 164.657,
+      "eval_steps_per_second": 5.147,
       "step": 1250
     },
     {
-      "epoch": 6.49,
-      "learning_rate": 2e-05,
-      "loss": 6.9476,
+      "epoch": 26.25,
+      "learning_rate": 0.00033930125175581647,
+      "loss": 4.2188,
       "step": 1260
     },
     {
-      "epoch": 6.54,
-      "learning_rate": 2e-05,
-      "loss": 6.9329,
+      "epoch": 26.45,
+      "learning_rate": 0.0003383490531691099,
+      "loss": 4.1928,
       "step": 1270
     },
     {
-      "epoch": 6.6,
-      "learning_rate": 2e-05,
-      "loss": 6.9346,
+      "epoch": 26.66,
+      "learning_rate": 0.0003373908027466289,
+      "loss": 4.1575,
       "step": 1280
     },
     {
-      "epoch": 6.65,
-      "learning_rate": 2e-05,
-      "loss": 6.9047,
+      "epoch": 26.86,
+      "learning_rate": 0.00033642654240535134,
+      "loss": 4.1106,
       "step": 1290
     },
     {
-      "epoch": 6.7,
-      "learning_rate": 2e-05,
-      "loss": 6.9425,
+      "epoch": 27.08,
+      "learning_rate": 0.00033545631432514825,
+      "loss": 4.3717,
       "step": 1300
     },
     {
-      "epoch": 6.7,
-      "eval_accuracy": 0.1007779473820151,
-      "eval_loss": 6.961859703063965,
-      "eval_runtime": 169.104,
-      "eval_samples_per_second": 142.25,
-      "eval_steps_per_second": 2.223,
+      "epoch": 27.08,
+      "eval_accuracy": 0.3412254938630985,
+      "eval_loss": 4.1738104820251465,
+      "eval_runtime": 145.9707,
+      "eval_samples_per_second": 164.793,
+      "eval_steps_per_second": 5.152,
       "step": 1300
     },
     {
-      "epoch": 6.75,
-      "learning_rate": 2e-05,
-      "loss": 6.9086,
+      "epoch": 27.29,
+      "learning_rate": 0.00033448016094693895,
+      "loss": 4.007,
       "step": 1310
     },
     {
-      "epoch": 6.8,
-      "learning_rate": 2e-05,
-      "loss": 6.9056,
+      "epoch": 27.49,
+      "learning_rate": 0.0003334981249708345,
+      "loss": 4.003,
       "step": 1320
     },
     {
-      "epoch": 6.85,
-      "learning_rate": 2e-05,
-      "loss": 6.902,
+      "epoch": 27.7,
+      "learning_rate": 0.00033251024935427,
+      "loss": 3.9491,
       "step": 1330
     },
     {
-      "epoch": 6.9,
-      "learning_rate": 2e-05,
-      "loss": 6.9013,
+      "epoch": 27.9,
+      "learning_rate": 0.0003315165773101249,
+      "loss": 3.9411,
       "step": 1340
     },
     {
-      "epoch": 6.96,
-      "learning_rate": 2e-05,
-      "loss": 6.8872,
+      "epoch": 28.12,
+      "learning_rate": 0.00033051715230483374,
+      "loss": 4.1483,
       "step": 1350
     },
     {
-      "epoch": 6.96,
-      "eval_accuracy": 0.10138427920145159,
-      "eval_loss": 6.946805477142334,
-      "eval_runtime": 169.0,
-      "eval_samples_per_second": 142.337,
-      "eval_steps_per_second": 2.225,
+      "epoch": 28.12,
+      "eval_accuracy": 0.35552299245507185,
+      "eval_loss": 4.033575534820557,
+      "eval_runtime": 145.9738,
+      "eval_samples_per_second": 164.79,
+      "eval_steps_per_second": 5.152,
       "step": 1350
     },
     {
-      "epoch": 7.01,
-      "learning_rate": 2e-05,
-      "loss": 7.4327,
+      "epoch": 28.33,
+      "learning_rate": 0.0003295120180564838,
+      "loss": 3.8395,
       "step": 1360
     },
     {
-      "epoch": 7.06,
-      "learning_rate": 2e-05,
-      "loss": 6.8642,
+      "epoch": 28.53,
+      "learning_rate": 0.00032850121853290334,
+      "loss": 3.8271,
       "step": 1370
     },
     {
-      "epoch": 7.11,
-      "learning_rate": 2e-05,
-      "loss": 6.9062,
+      "epoch": 28.74,
+      "learning_rate": 0.000327484797949738,
+      "loss": 3.8272,
       "step": 1380
     },
     {
-      "epoch": 7.16,
-      "learning_rate": 2e-05,
-      "loss": 6.9209,
+      "epoch": 28.95,
+      "learning_rate": 0.00032646280076851684,
+      "loss": 3.7855,
       "step": 1390
     },
     {
-      "epoch": 7.22,
-      "learning_rate": 2e-05,
-      "loss": 6.8848,
+      "epoch": 29.16,
+      "learning_rate": 0.0003254352716947074,
+      "loss": 3.9988,
       "step": 1400
     },
     {
-      "epoch": 7.22,
-      "eval_accuracy": 0.1024037766438749,
-      "eval_loss": 6.931956768035889,
-      "eval_runtime": 169.0148,
-      "eval_samples_per_second": 142.325,
-      "eval_steps_per_second": 2.225,
+      "epoch": 29.16,
+      "eval_accuracy": 0.3677331361148426,
+      "eval_loss": 3.91800594329834,
+      "eval_runtime": 146.056,
+      "eval_samples_per_second": 164.697,
+      "eval_steps_per_second": 5.149,
       "step": 1400
     },
     {
-      "epoch": 7.27,
-      "learning_rate": 2e-05,
-      "loss": 6.9138,
+      "epoch": 29.37,
+      "learning_rate": 0.0003244022556757602,
+      "loss": 3.7379,
       "step": 1410
     },
     {
-      "epoch": 7.32,
-      "learning_rate": 2e-05,
-      "loss": 6.8736,
+      "epoch": 29.58,
+      "learning_rate": 0.0003233637978991422,
+      "loss": 3.6974,
       "step": 1420
     },
     {
-      "epoch": 7.37,
-      "learning_rate": 2e-05,
-      "loss": 6.9051,
+      "epoch": 29.78,
+      "learning_rate": 0.00032231994379036086,
+      "loss": 3.6966,
       "step": 1430
     },
     {
-      "epoch": 7.42,
-      "learning_rate": 2e-05,
-      "loss": 6.8833,
+      "epoch": 29.99,
+      "learning_rate": 0.0003212707390109765,
+      "loss": 3.6594,
       "step": 1440
     },
     {
-      "epoch": 7.47,
-      "learning_rate": 2e-05,
-      "loss": 6.8578,
+      "epoch": 30.21,
+      "learning_rate": 0.00032021622945660504,
+      "loss": 3.8695,
       "step": 1450
     },
     {
-      "epoch": 7.47,
-      "eval_accuracy": 0.10391318947463475,
-      "eval_loss": 6.9190216064453125,
-      "eval_runtime": 168.9483,
-      "eval_samples_per_second": 142.381,
-      "eval_steps_per_second": 2.226,
+      "epoch": 30.21,
+      "eval_accuracy": 0.37818666192863265,
+      "eval_loss": 3.81080961227417,
+      "eval_runtime": 146.0723,
+      "eval_samples_per_second": 164.679,
+      "eval_steps_per_second": 5.148,
       "step": 1450
     },
     {
-      "epoch": 7.52,
-      "learning_rate": 2e-05,
-      "loss": 6.8757,
+      "epoch": 30.41,
+      "learning_rate": 0.0003191564612549106,
+      "loss": 3.598,
       "step": 1460
     },
     {
-      "epoch": 7.58,
-      "learning_rate": 2e-05,
-      "loss": 6.8638,
+      "epoch": 30.62,
+      "learning_rate": 0.0003180914807635874,
+      "loss": 3.5942,
       "step": 1470
     },
     {
-      "epoch": 7.63,
-      "learning_rate": 2e-05,
-      "loss": 6.8717,
+      "epoch": 30.82,
+      "learning_rate": 0.00031702133456833236,
+      "loss": 3.585,
       "step": 1480
     },
     {
-      "epoch": 7.68,
-      "learning_rate": 2e-05,
-      "loss": 6.8967,
+      "epoch": 31.04,
+      "learning_rate": 0.00031594606948080663,
+      "loss": 3.7908,
       "step": 1490
     },
     {
-      "epoch": 7.73,
-      "learning_rate": 2e-05,
-      "loss": 6.8699,
+      "epoch": 31.25,
+      "learning_rate": 0.00031486573253658874,
+      "loss": 3.5017,
       "step": 1500
     },
     {
-      "epoch": 7.73,
-      "eval_accuracy": 0.10498492735216337,
-      "eval_loss": 6.902231216430664,
-      "eval_runtime": 168.9322,
-      "eval_samples_per_second": 142.394,
-      "eval_steps_per_second": 2.226,
+      "epoch": 31.25,
+      "eval_accuracy": 0.3878577124364749,
+      "eval_loss": 3.7240185737609863,
+      "eval_runtime": 145.8744,
+      "eval_samples_per_second": 164.902,
+      "eval_steps_per_second": 5.155,
       "step": 1500
     },
     {
-      "epoch": 7.78,
-      "learning_rate": 2e-05,
-      "loss": 6.867,
+      "epoch": 31.45,
+      "learning_rate": 0.00031378037099311627,
+      "loss": 3.5206,
       "step": 1510
     },
     {
-      "epoch": 7.83,
-      "learning_rate": 2e-05,
-      "loss": 6.8533,
+      "epoch": 31.66,
+      "learning_rate": 0.00031269003232761933,
+      "loss": 3.5049,
       "step": 1520
     },
     {
-      "epoch": 7.88,
-      "learning_rate": 2e-05,
-      "loss": 6.8669,
+      "epoch": 31.86,
+      "learning_rate": 0.0003115947642350433,
+      "loss": 3.4852,
       "step": 1530
     },
     {
-      "epoch": 7.93,
-      "learning_rate": 2e-05,
-      "loss": 6.862,
+      "epoch": 32.08,
+      "learning_rate": 0.00031049461462596267,
+      "loss": 3.6894,
       "step": 1540
     },
     {
-      "epoch": 7.99,
-      "learning_rate": 2e-05,
-      "loss": 6.8402,
+      "epoch": 32.29,
+      "learning_rate": 0.00030938963162448544,
+      "loss": 3.4311,
       "step": 1550
     },
     {
-      "epoch": 7.99,
-      "eval_accuracy": 0.10569349761970831,
-      "eval_loss": 6.8910064697265625,
-      "eval_runtime": 168.8729,
-      "eval_samples_per_second": 142.444,
-      "eval_steps_per_second": 2.227,
+      "epoch": 32.29,
+      "eval_accuracy": 0.3973612765821424,
+      "eval_loss": 3.6425790786743164,
+      "eval_runtime": 146.1194,
+      "eval_samples_per_second": 164.626,
+      "eval_steps_per_second": 5.146,
       "step": 1550
     },
     {
-      "epoch": 8.04,
-      "learning_rate": 2e-05,
-      "loss": 7.3336,
+      "epoch": 32.49,
+      "learning_rate": 0.0003082798635661476,
+      "loss": 3.4258,
       "step": 1560
     },
     {
-      "epoch": 8.09,
-      "learning_rate": 2e-05,
-      "loss": 6.8451,
+      "epoch": 32.7,
+      "learning_rate": 0.0003071653589957993,
+      "loss": 3.4076,
       "step": 1570
     },
     {
-      "epoch": 8.14,
-      "learning_rate": 2e-05,
-      "loss": 6.8418,
+      "epoch": 32.9,
+      "learning_rate": 0.000306046166665481,
+      "loss": 3.4117,
       "step": 1580
     },
     {
-      "epoch": 8.2,
-      "learning_rate": 2e-05,
-      "loss": 6.8648,
+      "epoch": 33.12,
+      "learning_rate": 0.00030492233553229076,
+      "loss": 3.5985,
       "step": 1590
     },
     {
-      "epoch": 8.25,
-      "learning_rate": 2e-05,
-      "loss": 6.8172,
+      "epoch": 33.33,
+      "learning_rate": 0.00030379391475624304,
+      "loss": 3.3517,
       "step": 1600
     },
     {
-      "epoch": 8.25,
-      "eval_accuracy": 0.1068832420259414,
-      "eval_loss": 6.8729939460754395,
-      "eval_runtime": 168.964,
-      "eval_samples_per_second": 142.368,
-      "eval_steps_per_second": 2.225,
+      "epoch": 33.33,
+      "eval_accuracy": 0.40682330082568596,
+      "eval_loss": 3.5615479946136475,
+      "eval_runtime": 146.0666,
+      "eval_samples_per_second": 164.685,
+      "eval_steps_per_second": 5.148,
       "step": 1600
     },
     {
-      "epoch": 8.3,
-      "learning_rate": 2e-05,
-      "loss": 6.832,
+      "epoch": 33.53,
+      "learning_rate": 0.0003026609536981183,
+      "loss": 3.3431,
       "step": 1610
     },
     {
-      "epoch": 8.35,
-      "learning_rate": 2e-05,
-      "loss": 6.8032,
+      "epoch": 33.74,
+      "learning_rate": 0.0003015235019173034,
+      "loss": 3.3546,
       "step": 1620
     },
     {
-      "epoch": 8.4,
-      "learning_rate": 2e-05,
-      "loss": 6.8527,
+      "epoch": 33.95,
+      "learning_rate": 0.00030038160916962404,
+      "loss": 3.3378,
       "step": 1630
     },
     {
-      "epoch": 8.45,
-      "learning_rate": 2e-05,
-      "loss": 6.8175,
+      "epoch": 34.16,
+      "learning_rate": 0.00029923532540516843,
+      "loss": 3.5305,
       "step": 1640
     },
     {
-      "epoch": 8.5,
-      "learning_rate": 2e-05,
-      "loss": 6.823,
+      "epoch": 34.37,
+      "learning_rate": 0.00029808470076610167,
+      "loss": 3.2856,
       "step": 1650
     },
     {
-      "epoch": 8.5,
-      "eval_accuracy": 0.10725955083624485,
-      "eval_loss": 6.86623477935791,
-      "eval_runtime": 168.9967,
-      "eval_samples_per_second": 142.34,
-      "eval_steps_per_second": 2.225,
+      "epoch": 34.37,
+      "eval_accuracy": 0.41555171151451314,
+      "eval_loss": 3.4914703369140625,
+      "eval_runtime": 146.1721,
+      "eval_samples_per_second": 164.566,
+      "eval_steps_per_second": 5.145,
       "step": 1650
     },
     {
-      "epoch": 8.55,
-      "learning_rate": 2e-05,
-      "loss": 6.8428,
+      "epoch": 34.58,
+      "learning_rate": 0.00029692978558447305,
+      "loss": 3.273,
       "step": 1660
     },
     {
-      "epoch": 8.61,
-      "learning_rate": 2e-05,
-      "loss": 6.8136,
+      "epoch": 34.78,
+      "learning_rate": 0.0002957706303800139,
+      "loss": 3.278,
       "step": 1670
     },
     {
-      "epoch": 8.66,
-      "learning_rate": 2e-05,
-      "loss": 6.828,
+      "epoch": 34.99,
+      "learning_rate": 0.0002946072858579282,
+      "loss": 3.2614,
       "step": 1680
     },
     {
-      "epoch": 8.71,
-      "learning_rate": 2e-05,
-      "loss": 6.8215,
+      "epoch": 35.21,
+      "learning_rate": 0.0002934398029066739,
+      "loss": 3.4456,
       "step": 1690
     },
     {
-      "epoch": 8.76,
-      "learning_rate": 2e-05,
-      "loss": 6.8028,
+      "epoch": 35.41,
+      "learning_rate": 0.0002922682325957376,
+      "loss": 3.227,
       "step": 1700
     },
     {
-      "epoch": 8.76,
-      "eval_accuracy": 0.1081789668246189,
-      "eval_loss": 6.848670482635498,
-      "eval_runtime": 168.985,
-      "eval_samples_per_second": 142.35,
-      "eval_steps_per_second": 2.225,
+      "epoch": 35.41,
+      "eval_accuracy": 0.4255488250901363,
+      "eval_loss": 3.41792893409729,
+      "eval_runtime": 146.0068,
+      "eval_samples_per_second": 164.753,
+      "eval_steps_per_second": 5.15,
       "step": 1700
     },
     {
-      "epoch": 8.81,
-      "learning_rate": 2e-05,
-      "loss": 6.8065,
+      "epoch": 35.62,
+      "learning_rate": 0.00029109262617339987,
+      "loss": 3.1995,
       "step": 1710
     },
     {
-      "epoch": 8.86,
-      "learning_rate": 2e-05,
-      "loss": 6.8185,
+      "epoch": 35.82,
+      "learning_rate": 0.0002899130350644941,
+      "loss": 3.2058,
       "step": 1720
     },
     {
-      "epoch": 8.91,
-      "learning_rate": 2e-05,
-      "loss": 6.8206,
+      "epoch": 36.04,
+      "learning_rate": 0.00028872951086815685,
+      "loss": 3.4183,
       "step": 1730
     },
     {
-      "epoch": 8.97,
-      "learning_rate": 2e-05,
-      "loss": 6.8051,
+      "epoch": 36.25,
+      "learning_rate": 0.00028754210535557036,
+      "loss": 3.1514,
       "step": 1740
     },
     {
-      "epoch": 9.02,
-      "learning_rate": 2e-05,
-      "loss": 7.3146,
+      "epoch": 36.45,
+      "learning_rate": 0.00028635087046769857,
+      "loss": 3.1675,
       "step": 1750
     },
     {
-      "epoch": 9.02,
-      "eval_accuracy": 0.10825639090130809,
-      "eval_loss": 6.8400139808654785,
-      "eval_runtime": 168.9057,
-      "eval_samples_per_second": 142.417,
-      "eval_steps_per_second": 2.226,
+      "epoch": 36.45,
+      "eval_accuracy": 0.43245804160401624,
+      "eval_loss": 3.3635590076446533,
+      "eval_runtime": 146.1639,
+      "eval_samples_per_second": 164.575,
+      "eval_steps_per_second": 5.145,
       "step": 1750
     },
     {
-      "epoch": 9.07,
-      "learning_rate": 2e-05,
-      "loss": 6.8451,
+      "epoch": 36.66,
+      "learning_rate": 0.00028515585831301456,
+      "loss": 3.1645,
       "step": 1760
     },
     {
-      "epoch": 9.12,
-      "learning_rate": 2e-05,
-      "loss": 6.8243,
+      "epoch": 36.86,
+      "learning_rate": 0.0002839571211652212,
+      "loss": 3.1617,
       "step": 1770
     },
     {
-      "epoch": 9.17,
-      "learning_rate": 2e-05,
-      "loss": 6.8415,
+      "epoch": 37.08,
+      "learning_rate": 0.00028275471146096466,
+      "loss": 3.3333,
       "step": 1780
     },
     {
-      "epoch": 9.23,
-      "learning_rate": 2e-05,
-      "loss": 6.7823,
+      "epoch": 37.29,
+      "learning_rate": 0.00028154868179754074,
+      "loss": 3.1167,
       "step": 1790
     },
     {
-      "epoch": 9.28,
-      "learning_rate": 2e-05,
-      "loss": 6.8014,
+      "epoch": 37.49,
+      "learning_rate": 0.0002803390849305939,
+      "loss": 3.0908,
       "step": 1800
     },
     {
-      "epoch": 9.28,
-      "eval_accuracy": 0.10921870585675643,
-      "eval_loss": 6.8303141593933105,
-      "eval_runtime": 169.0668,
-      "eval_samples_per_second": 142.281,
-      "eval_steps_per_second": 2.224,
+      "epoch": 37.49,
+      "eval_accuracy": 0.43940471782078516,
+      "eval_loss": 3.30828595161438,
+      "eval_runtime": 146.1043,
+      "eval_samples_per_second": 164.643,
+      "eval_steps_per_second": 5.147,
       "step": 1800
     },
     {
-      "epoch": 9.33,
-      "learning_rate": 2e-05,
-      "loss": 6.7885,
+      "epoch": 37.7,
+      "learning_rate": 0.0002791259737718097,
+      "loss": 3.1214,
       "step": 1810
     },
     {
-      "epoch": 9.38,
-      "learning_rate": 2e-05,
-      "loss": 6.7663,
+      "epoch": 37.9,
+      "learning_rate": 0.0002779094013866001,
+      "loss": 3.0987,
       "step": 1820
     },
     {
-      "epoch": 9.43,
-      "learning_rate": 2e-05,
-      "loss": 6.7476,
+      "epoch": 38.12,
+      "learning_rate": 0.00027668942099178234,
+      "loss": 3.2767,
       "step": 1830
     },
     {
-      "epoch": 9.48,
-      "learning_rate": 2e-05,
-      "loss": 6.8016,
+      "epoch": 38.33,
+      "learning_rate": 0.00027546608595325117,
+      "loss": 3.0716,
       "step": 1840
     },
     {
-      "epoch": 9.53,
-      "learning_rate": 2e-05,
-      "loss": 6.8028,
+      "epoch": 38.53,
+      "learning_rate": 0.00027423944978364416,
+      "loss": 3.0561,
       "step": 1850
     },
     {
-      "epoch": 9.53,
-      "eval_accuracy": 0.108833053639469,
-      "eval_loss": 6.822628021240234,
-      "eval_runtime": 169.002,
-      "eval_samples_per_second": 142.336,
-      "eval_steps_per_second": 2.225,
+      "epoch": 38.53,
+      "eval_accuracy": 0.44727625227121054,
+      "eval_loss": 3.25723934173584,
+      "eval_runtime": 145.9616,
+      "eval_samples_per_second": 164.804,
+      "eval_steps_per_second": 5.152,
       "step": 1850
     },
     {
-      "epoch": 9.59,
-      "learning_rate": 2e-05,
-      "loss": 6.7865,
+      "epoch": 38.74,
+      "learning_rate": 0.00027300956614000115,
+      "loss": 3.0564,
       "step": 1860
     },
     {
-      "epoch": 9.64,
-      "learning_rate": 2e-05,
-      "loss": 6.7951,
+      "epoch": 38.95,
+      "learning_rate": 0.00027177648882141704,
+      "loss": 3.0583,
       "step": 1870
     },
     {
-      "epoch": 9.69,
-      "learning_rate": 2e-05,
-      "loss": 6.7739,
+      "epoch": 39.16,
+      "learning_rate": 0.0002705402717666883,
+      "loss": 3.2319,
       "step": 1880
     },
     {
-      "epoch": 9.74,
-      "learning_rate": 2e-05,
-      "loss": 6.7718,
+      "epoch": 39.37,
+      "learning_rate": 0.00026930096905195363,
+      "loss": 3.0204,
       "step": 1890
     },
     {
-      "epoch": 9.79,
-      "learning_rate": 2e-05,
-      "loss": 6.7817,
+      "epoch": 39.58,
+      "learning_rate": 0.00026805863488832865,
+      "loss": 3.0139,
       "step": 1900
     },
     {
-      "epoch": 9.79,
-      "eval_accuracy": 0.11072009541672845,
-      "eval_loss": 6.807923316955566,
-      "eval_runtime": 168.9193,
-      "eval_samples_per_second": 142.405,
-      "eval_steps_per_second": 2.226,
+      "epoch": 39.58,
+      "eval_accuracy": 0.4525324485267982,
+      "eval_loss": 3.215851306915283,
+      "eval_runtime": 146.1327,
+      "eval_samples_per_second": 164.611,
+      "eval_steps_per_second": 5.146,
       "step": 1900
     },
     {
-      "epoch": 9.84,
-      "learning_rate": 2e-05,
-      "loss": 6.7725,
+      "epoch": 39.78,
+      "learning_rate": 0.00026681332361953424,
+      "loss": 3.0053,
       "step": 1910
     },
     {
-      "epoch": 9.89,
-      "learning_rate": 2e-05,
-      "loss": 6.7347,
+      "epoch": 39.99,
+      "learning_rate": 0.0002655650897195195,
+      "loss": 3.0171,
       "step": 1920
     },
     {
-      "epoch": 9.94,
-      "learning_rate": 2e-05,
-      "loss": 6.7559,
+      "epoch": 40.21,
+      "learning_rate": 0.0002643139877900791,
+      "loss": 3.1749,
       "step": 1930
     },
     {
-      "epoch": 10.0,
-      "learning_rate": 2e-05,
-      "loss": 6.7409,
+      "epoch": 40.41,
+      "learning_rate": 0.00026306007255846436,
+      "loss": 2.9764,
       "step": 1940
     },
     {
-      "epoch": 10.05,
-      "learning_rate": 2e-05,
-      "loss": 7.28,
+      "epoch": 40.62,
+      "learning_rate": 0.00026180339887498953,
+      "loss": 2.9837,
       "step": 1950
     },
     {
-      "epoch": 10.05,
-      "eval_accuracy": 0.11147970049664972,
-      "eval_loss": 6.802090167999268,
-      "eval_runtime": 169.0764,
-      "eval_samples_per_second": 142.273,
-      "eval_steps_per_second": 2.224,
+      "epoch": 40.62,
+      "eval_accuracy": 0.45754832554207525,
+      "eval_loss": 3.1789309978485107,
+      "eval_runtime": 146.1778,
+      "eval_samples_per_second": 164.56,
+      "eval_steps_per_second": 5.144,
       "step": 1950
     },
     {
-      "epoch": 10.1,
-      "learning_rate": 2e-05,
-      "loss": 6.7606,
+      "epoch": 40.82,
+      "learning_rate": 0.00026054402171063267,
+      "loss": 2.9752,
       "step": 1960
     },
     {
-      "epoch": 10.15,
-      "learning_rate": 2e-05,
-      "loss": 6.7658,
+      "epoch": 41.04,
+      "learning_rate": 0.0002592819961546308,
+      "loss": 3.1648,
       "step": 1970
     },
     {
-      "epoch": 10.21,
-      "learning_rate": 2e-05,
-      "loss": 6.7526,
+      "epoch": 41.25,
+      "learning_rate": 0.00025801737741207005,
+      "loss": 2.9438,
       "step": 1980
     },
     {
-      "epoch": 10.26,
-      "learning_rate": 2e-05,
-      "loss": 6.7531,
+      "epoch": 41.45,
+      "learning_rate": 0.000256750220801471,
+      "loss": 2.941,
       "step": 1990
     },
     {
-      "epoch": 10.31,
-      "learning_rate": 2e-05,
-      "loss": 6.7624,
+      "epoch": 41.66,
+      "learning_rate": 0.0002554805817523689,
+      "loss": 2.9387,
       "step": 2000
     },
     {
-      "epoch": 10.31,
-      "eval_accuracy": 0.11175787497505224,
-      "eval_loss": 6.793049335479736,
-      "eval_runtime": 169.0207,
-      "eval_samples_per_second": 142.32,
-      "eval_steps_per_second": 2.225,
+      "epoch": 41.66,
+      "eval_accuracy": 0.46179467604077673,
+      "eval_loss": 3.1430864334106445,
+      "eval_runtime": 146.0529,
+      "eval_samples_per_second": 164.701,
+      "eval_steps_per_second": 5.149,
       "step": 2000
     },
     {
-      "epoch": 10.36,
-      "learning_rate": 2e-05,
-      "loss": 6.7658,
+      "epoch": 41.86,
+      "learning_rate": 0.0002542085158028889,
+      "loss": 2.9371,
       "step": 2010
     },
     {
-      "epoch": 10.41,
-      "learning_rate": 2e-05,
-      "loss": 6.7513,
+      "epoch": 42.08,
+      "learning_rate": 0.00025293407859731633,
+      "loss": 3.1085,
       "step": 2020
     },
     {
-      "epoch": 10.46,
-      "learning_rate": 2e-05,
-      "loss": 6.7618,
+      "epoch": 42.29,
+      "learning_rate": 0.00025165732588366334,
+      "loss": 2.8999,
       "step": 2030
     },
     {
-      "epoch": 10.51,
-      "learning_rate": 2e-05,
-      "loss": 6.7146,
+      "epoch": 42.49,
+      "learning_rate": 0.00025037831351122967,
+      "loss": 2.9159,
       "step": 2040
     },
     {
-      "epoch": 10.56,
-      "learning_rate": 2e-05,
-      "loss": 6.7416,
+      "epoch": 42.7,
+      "learning_rate": 0.0002490970974281599,
+      "loss": 2.9034,
       "step": 2050
     },
     {
-      "epoch": 10.56,
-      "eval_accuracy": 0.11243676542273737,
-      "eval_loss": 6.786773681640625,
-      "eval_runtime": 168.9144,
-      "eval_samples_per_second": 142.409,
-      "eval_steps_per_second": 2.226,
+      "epoch": 42.7,
+      "eval_accuracy": 0.46535935335872575,
+      "eval_loss": 3.116283655166626,
+      "eval_runtime": 146.1195,
+      "eval_samples_per_second": 164.626,
+      "eval_steps_per_second": 5.146,
       "step": 2050
     },
     {
-      "epoch": 10.62,
-      "learning_rate": 2e-05,
-      "loss": 6.7629,
+      "epoch": 42.9,
+      "learning_rate": 0.00024781373367899597,
+      "loss": 2.8936,
       "step": 2060
     },
     {
-      "epoch": 10.67,
-      "learning_rate": 2e-05,
-      "loss": 6.7654,
+      "epoch": 43.12,
+      "learning_rate": 0.00024652827840222606,
+      "loss": 3.0697,
       "step": 2070
     },
     {
-      "epoch": 10.72,
-      "learning_rate": 2e-05,
-      "loss": 6.7179,
+      "epoch": 43.33,
+      "learning_rate": 0.00024524078782782807,
+      "loss": 2.8913,
       "step": 2080
     },
     {
-      "epoch": 10.77,
-      "learning_rate": 2e-05,
-      "loss": 6.7545,
+      "epoch": 43.53,
+      "learning_rate": 0.00024395131827481062,
+      "loss": 2.8624,
       "step": 2090
     },
     {
-      "epoch": 10.82,
-      "learning_rate": 2e-05,
-      "loss": 6.7288,
+      "epoch": 43.74,
+      "learning_rate": 0.0002426599261487494,
+      "loss": 2.8822,
       "step": 2100
     },
     {
-      "epoch": 10.82,
-      "eval_accuracy": 0.11334799297462662,
-      "eval_loss": 6.780516147613525,
-      "eval_runtime": 168.9613,
-      "eval_samples_per_second": 142.37,
-      "eval_steps_per_second": 2.225,
+      "epoch": 43.74,
+      "eval_accuracy": 0.46941429535485324,
+      "eval_loss": 3.0841524600982666,
+      "eval_runtime": 146.1268,
+      "eval_samples_per_second": 164.617,
+      "eval_steps_per_second": 5.146,
       "step": 2100
     },
     {
-      "epoch": 10.87,
-      "learning_rate": 2e-05,
-      "loss": 6.751,
+      "epoch": 43.95,
+      "learning_rate": 0.00024136666793931935,
+      "loss": 2.8655,
       "step": 2110
     },
     {
-      "epoch": 10.92,
-      "learning_rate": 2e-05,
-      "loss": 6.7746,
+      "epoch": 44.16,
+      "learning_rate": 0.00024007160021782427,
+      "loss": 3.0323,
       "step": 2120
     },
     {
-      "epoch": 10.98,
-      "learning_rate": 2e-05,
-      "loss": 6.7309,
+      "epoch": 44.37,
+      "learning_rate": 0.0002387747796347217,
+      "loss": 2.8446,
       "step": 2130
     },
     {
-      "epoch": 11.03,
-      "learning_rate": 2e-05,
-      "loss": 7.2015,
+      "epoch": 44.58,
+      "learning_rate": 0.00023747626291714498,
+      "loss": 2.8433,
       "step": 2140
     },
     {
-      "epoch": 11.08,
-      "learning_rate": 2e-05,
-      "loss": 6.7468,
+      "epoch": 44.78,
+      "learning_rate": 0.000236176106866422,
+      "loss": 2.836,
       "step": 2150
     },
     {
-      "epoch": 11.08,
-      "eval_accuracy": 0.11234059549981652,
-      "eval_loss": 6.772030353546143,
-      "eval_runtime": 169.0603,
-      "eval_samples_per_second": 142.287,
-      "eval_steps_per_second": 2.224,
+      "epoch": 44.78,
+      "eval_accuracy": 0.47268071006532664,
+      "eval_loss": 3.0583226680755615,
+      "eval_runtime": 145.9982,
+      "eval_samples_per_second": 164.762,
+      "eval_steps_per_second": 5.151,
       "step": 2150
     },
     {
-      "epoch": 11.13,
-      "learning_rate": 2e-05,
-      "loss": 6.6947,
+      "epoch": 44.99,
+      "learning_rate": 0.00023487436835559035,
+      "loss": 2.8457,
       "step": 2160
     },
     {
-      "epoch": 11.18,
-      "learning_rate": 2e-05,
-      "loss": 6.7154,
+      "epoch": 45.21,
+      "learning_rate": 0.00023357110432690954,
+      "loss": 2.9941,
       "step": 2170
     },
     {
-      "epoch": 11.24,
-      "learning_rate": 2e-05,
-      "loss": 6.6892,
+      "epoch": 45.41,
+      "learning_rate": 0.00023226637178937022,
+      "loss": 2.8208,
       "step": 2180
     },
     {
-      "epoch": 11.29,
-      "learning_rate": 2e-05,
-      "loss": 6.7256,
+      "epoch": 45.62,
+      "learning_rate": 0.00023096022781620034,
+      "loss": 2.8154,
       "step": 2190
     },
     {
-      "epoch": 11.34,
-      "learning_rate": 2e-05,
-      "loss": 6.7387,
+      "epoch": 45.82,
+      "learning_rate": 0.0002296527295423684,
+      "loss": 2.8129,
       "step": 2200
     },
     {
-      "epoch": 11.34,
-      "eval_accuracy": 0.11353738741762727,
-      "eval_loss": 6.7636189460754395,
-      "eval_runtime": 168.99,
-      "eval_samples_per_second": 142.346,
-      "eval_steps_per_second": 2.225,
+      "epoch": 45.82,
+      "eval_accuracy": 0.47600857452342976,
+      "eval_loss": 3.035902738571167,
+      "eval_runtime": 145.9849,
+      "eval_samples_per_second": 164.777,
+      "eval_steps_per_second": 5.151,
       "step": 2200
     },
     {
-      "epoch": 11.39,
-      "learning_rate": 2e-05,
-      "loss": 6.6785,
+      "epoch": 46.04,
+      "learning_rate": 0.00022834393416208486,
+      "loss": 2.9871,
       "step": 2210
     },
     {
-      "epoch": 11.44,
-      "learning_rate": 2e-05,
-      "loss": 6.6991,
+      "epoch": 46.25,
+      "learning_rate": 0.0002270338989262994,
+      "loss": 2.7892,
       "step": 2220
     },
     {
-      "epoch": 11.49,
-      "learning_rate": 2e-05,
-      "loss": 6.7317,
+      "epoch": 46.45,
+      "learning_rate": 0.00022572268114019726,
+      "loss": 2.7843,
       "step": 2230
     },
     {
-      "epoch": 11.54,
-      "learning_rate": 2e-05,
-      "loss": 6.7243,
+      "epoch": 46.66,
+      "learning_rate": 0.00022441033816069202,
+      "loss": 2.7867,
       "step": 2240
     },
     {
-      "epoch": 11.6,
-      "learning_rate": 2e-05,
-      "loss": 6.7242,
+      "epoch": 46.86,
+      "learning_rate": 0.00022309692739391727,
+      "loss": 2.7733,
       "step": 2250
     },
     {
-      "epoch": 11.6,
-      "eval_accuracy": 0.11342761785068739,
-      "eval_loss": 6.755679607391357,
-      "eval_runtime": 169.0043,
-      "eval_samples_per_second": 142.334,
-      "eval_steps_per_second": 2.225,
+      "epoch": 46.86,
+      "eval_accuracy": 0.47764141406488453,
+      "eval_loss": 3.017348051071167,
+      "eval_runtime": 146.0338,
+      "eval_samples_per_second": 164.722,
+      "eval_steps_per_second": 5.149,
       "step": 2250
     },
     {
-      "epoch": 11.65,
-      "learning_rate": 2e-05,
-      "loss": 6.7085,
+      "epoch": 47.08,
+      "learning_rate": 0.00022178250629271452,
+      "loss": 2.981,
       "step": 2260
     },
     {
-      "epoch": 11.7,
-      "learning_rate": 2e-05,
-      "loss": 6.7231,
+      "epoch": 47.29,
+      "learning_rate": 0.00022046713235412103,
+      "loss": 2.7598,
       "step": 2270
     },
     {
-      "epoch": 11.75,
-      "learning_rate": 2e-05,
-      "loss": 6.6973,
+      "epoch": 47.49,
+      "learning_rate": 0.00021915086311685404,
+      "loss": 2.7769,
       "step": 2280
     },
     {
-      "epoch": 11.8,
-      "learning_rate": 2e-05,
-      "loss": 6.6921,
+      "epoch": 47.7,
+      "learning_rate": 0.00021783375615879415,
+      "loss": 2.7753,
       "step": 2290
     },
     {
-      "epoch": 11.85,
-      "learning_rate": 2e-05,
-      "loss": 6.702,
+      "epoch": 47.9,
+      "learning_rate": 0.0002165158690944665,
+      "loss": 2.7589,
       "step": 2300
     },
     {
-      "epoch": 11.85,
-      "eval_accuracy": 0.11411896604340402,
-      "eval_loss": 6.749605178833008,
-      "eval_runtime": 169.2579,
-      "eval_samples_per_second": 142.12,
-      "eval_steps_per_second": 2.221,
+      "epoch": 47.9,
+      "eval_accuracy": 0.4811929413931917,
+      "eval_loss": 2.9977798461914062,
+      "eval_runtime": 146.0602,
+      "eval_samples_per_second": 164.692,
+      "eval_steps_per_second": 5.149,
       "step": 2300
     },
     {
-      "epoch": 11.9,
-      "learning_rate": 2e-05,
-      "loss": 6.7191,
+      "epoch": 48.12,
+      "learning_rate": 0.00021519725957252063,
+      "loss": 2.9409,
       "step": 2310
     },
     {
-      "epoch": 11.96,
-      "learning_rate": 2e-05,
-      "loss": 6.6818,
+      "epoch": 48.33,
+      "learning_rate": 0.00021387798527320882,
+      "loss": 2.7465,
       "step": 2320
     },
     {
-      "epoch": 12.01,
-      "learning_rate": 2e-05,
-      "loss": 7.236,
+      "epoch": 48.53,
+      "learning_rate": 0.0002125581039058627,
+      "loss": 2.7403,
       "step": 2330
     },
     {
-      "epoch": 12.06,
-      "learning_rate": 2e-05,
-      "loss": 6.6929,
+      "epoch": 48.74,
+      "learning_rate": 0.0002112376732063691,
+      "loss": 2.7284,
       "step": 2340
     },
     {
-      "epoch": 12.11,
-      "learning_rate": 2e-05,
-      "loss": 6.6662,
+      "epoch": 48.95,
+      "learning_rate": 0.00020991675093464448,
+      "loss": 2.7378,
       "step": 2350
     },
     {
-      "epoch": 12.11,
-      "eval_accuracy": 0.11496613116598788,
-      "eval_loss": 6.743268013000488,
-      "eval_runtime": 169.0391,
-      "eval_samples_per_second": 142.304,
-      "eval_steps_per_second": 2.224,
+      "epoch": 48.95,
+      "eval_accuracy": 0.4831324380858166,
+      "eval_loss": 2.9787769317626953,
+      "eval_runtime": 146.0148,
+      "eval_samples_per_second": 164.744,
+      "eval_steps_per_second": 5.15,
       "step": 2350
     },
     {
-      "epoch": 12.16,
-      "learning_rate": 2e-05,
-      "loss": 6.7133,
+      "epoch": 49.16,
+      "learning_rate": 0.00020859539487210813,
+      "loss": 2.9167,
       "step": 2360
     },
     {
-      "epoch": 12.22,
-      "learning_rate": 2e-05,
-      "loss": 6.7061,
+      "epoch": 49.37,
+      "learning_rate": 0.0002072736628191549,
+      "loss": 2.7203,
       "step": 2370
     },
     {
-      "epoch": 12.27,
-      "learning_rate": 2e-05,
-      "loss": 6.6505,
+      "epoch": 49.58,
+      "learning_rate": 0.0002059516125926265,
+      "loss": 2.7276,
       "step": 2380
     },
     {
-      "epoch": 12.32,
-      "learning_rate": 2e-05,
-      "loss": 6.7027,
+      "epoch": 49.78,
+      "learning_rate": 0.00020462930202328278,
+      "loss": 2.7001,
       "step": 2390
     },
     {
-      "epoch": 12.37,
-      "learning_rate": 2e-05,
-      "loss": 6.6781,
+      "epoch": 49.99,
+      "learning_rate": 0.00020330678895327174,
+      "loss": 2.7138,
       "step": 2400
     },
     {
-      "epoch": 12.37,
-      "eval_accuracy": 0.11483893825636936,
-      "eval_loss": 6.736194610595703,
-      "eval_runtime": 168.9398,
-      "eval_samples_per_second": 142.388,
-      "eval_steps_per_second": 2.226,
+      "epoch": 49.99,
+      "eval_accuracy": 0.4843915093446441,
+      "eval_loss": 2.967425584793091,
+      "eval_runtime": 146.0929,
+      "eval_samples_per_second": 164.655,
+      "eval_steps_per_second": 5.147,
       "step": 2400
     },
     {
-      "epoch": 12.42,
-      "learning_rate": 2e-05,
-      "loss": 6.7312,
+      "epoch": 50.21,
+      "learning_rate": 0.00020198413123359926,
+      "loss": 2.8865,
       "step": 2410
     },
     {
-      "epoch": 12.47,
-      "learning_rate": 2e-05,
-      "loss": 6.6969,
+      "epoch": 50.41,
+      "learning_rate": 0.00020066138672159903,
+      "loss": 2.698,
       "step": 2420
     },
     {
-      "epoch": 12.52,
-      "learning_rate": 2e-05,
-      "loss": 6.6962,
+      "epoch": 50.62,
+      "learning_rate": 0.00019933861327840098,
+      "loss": 2.6978,
       "step": 2430
     },
     {
-      "epoch": 12.58,
-      "learning_rate": 2e-05,
-      "loss": 6.729,
+      "epoch": 50.82,
+      "learning_rate": 0.00019801586876640073,
+      "loss": 2.704,
       "step": 2440
     },
     {
-      "epoch": 12.63,
-      "learning_rate": 2e-05,
-      "loss": 6.6743,
+      "epoch": 51.04,
+      "learning_rate": 0.0001966932110467283,
+      "loss": 2.8692,
       "step": 2450
     },
     {
-      "epoch": 12.63,
-      "eval_accuracy": 0.11607017718543403,
-      "eval_loss": 6.727517127990723,
-      "eval_runtime": 169.0176,
-      "eval_samples_per_second": 142.322,
-      "eval_steps_per_second": 2.225,
+      "epoch": 51.04,
+      "eval_accuracy": 0.4874163939573572,
+      "eval_loss": 2.9475862979888916,
+      "eval_runtime": 145.9737,
+      "eval_samples_per_second": 164.79,
+      "eval_steps_per_second": 5.152,
       "step": 2450
     },
     {
-      "epoch": 12.68,
-      "learning_rate": 2e-05,
-      "loss": 6.6952,
+      "epoch": 51.25,
+      "learning_rate": 0.00019537069797671724,
+      "loss": 2.6734,
       "step": 2460
     },
     {
-      "epoch": 12.73,
-      "learning_rate": 2e-05,
-      "loss": 6.6939,
+      "epoch": 51.45,
+      "learning_rate": 0.0001940483874073735,
+      "loss": 2.6636,
       "step": 2470
     },
     {
-      "epoch": 12.78,
-      "learning_rate": 2e-05,
-      "loss": 6.6848,
+      "epoch": 51.66,
+      "learning_rate": 0.00019272633718084517,
+      "loss": 2.6756,
       "step": 2480
     },
     {
-      "epoch": 12.83,
-      "learning_rate": 2e-05,
-      "loss": 6.6694,
+      "epoch": 51.86,
+      "learning_rate": 0.0001914046051278919,
+      "loss": 2.6808,
       "step": 2490
     },
     {
-      "epoch": 12.88,
-      "learning_rate": 2e-05,
-      "loss": 6.6843,
+      "epoch": 52.08,
+      "learning_rate": 0.00019008324906535554,
+      "loss": 2.8462,
       "step": 2500
     },
     {
-      "epoch": 12.88,
-      "eval_accuracy": 0.11647291428955145,
-      "eval_loss": 6.724733829498291,
-      "eval_runtime": 169.033,
-      "eval_samples_per_second": 142.31,
-      "eval_steps_per_second": 2.224,
+      "epoch": 52.08,
+      "eval_accuracy": 0.48931343115405407,
+      "eval_loss": 2.934227466583252,
+      "eval_runtime": 145.9977,
+      "eval_samples_per_second": 164.763,
+      "eval_steps_per_second": 5.151,
       "step": 2500
     },
     {
-      "epoch": 12.93,
-      "learning_rate": 2e-05,
-      "loss": 6.6913,
+      "epoch": 52.29,
+      "learning_rate": 0.0001887623267936309,
+      "loss": 2.6553,
       "step": 2510
     },
     {
-      "epoch": 12.99,
-      "learning_rate": 2e-05,
-      "loss": 6.6832,
+      "epoch": 52.49,
+      "learning_rate": 0.00018744189609413734,
+      "loss": 2.6559,
       "step": 2520
     },
     {
-      "epoch": 13.04,
-      "learning_rate": 2e-05,
-      "loss": 7.1899,
+      "epoch": 52.7,
+      "learning_rate": 0.0001861220147267912,
+      "loss": 2.6536,
       "step": 2530
     },
     {
-      "epoch": 13.09,
-      "learning_rate": 2e-05,
-      "loss": 6.683,
+      "epoch": 52.9,
+      "learning_rate": 0.0001848027404274794,
+      "loss": 2.6524,
       "step": 2540
     },
     {
-      "epoch": 13.14,
-      "learning_rate": 2e-05,
-      "loss": 6.6726,
+      "epoch": 53.12,
+      "learning_rate": 0.00018348413090553354,
+      "loss": 2.8312,
       "step": 2550
     },
     {
-      "epoch": 13.14,
-      "eval_accuracy": 0.11728262763830573,
-      "eval_loss": 6.712704658508301,
-      "eval_runtime": 169.1283,
-      "eval_samples_per_second": 142.229,
-      "eval_steps_per_second": 2.223,
+      "epoch": 53.12,
+      "eval_accuracy": 0.4900369570164547,
+      "eval_loss": 2.9268674850463867,
+      "eval_runtime": 146.0027,
+      "eval_samples_per_second": 164.757,
+      "eval_steps_per_second": 5.151,
       "step": 2550
     },
     {
-      "epoch": 13.2,
-      "learning_rate": 2e-05,
-      "loss": 6.6882,
+      "epoch": 53.33,
+      "learning_rate": 0.00018216624384120595,
+      "loss": 2.6306,
       "step": 2560
     },
     {
-      "epoch": 13.25,
-      "learning_rate": 2e-05,
-      "loss": 6.6761,
+      "epoch": 53.53,
+      "learning_rate": 0.00018084913688314597,
+      "loss": 2.6398,
       "step": 2570
     },
     {
-      "epoch": 13.3,
-      "learning_rate": 2e-05,
-      "loss": 6.6757,
+      "epoch": 53.74,
+      "learning_rate": 0.000179532867645879,
+      "loss": 2.6318,
       "step": 2580
     },
     {
-      "epoch": 13.35,
-      "learning_rate": 2e-05,
-      "loss": 6.6374,
+      "epoch": 53.95,
+      "learning_rate": 0.0001782174937072855,
+      "loss": 2.6358,
       "step": 2590
     },
     {
-      "epoch": 13.4,
-      "learning_rate": 2e-05,
-      "loss": 6.6656,
+      "epoch": 54.16,
+      "learning_rate": 0.00017690307260608278,
+      "loss": 2.7834,
       "step": 2600
     },
     {
-      "epoch": 13.4,
-      "eval_accuracy": 0.11703056957986009,
-      "eval_loss": 6.709805011749268,
-      "eval_runtime": 169.0648,
-      "eval_samples_per_second": 142.283,
-      "eval_steps_per_second": 2.224,
+      "epoch": 54.16,
+      "eval_accuracy": 0.4917280711401593,
+      "eval_loss": 2.911123037338257,
+      "eval_runtime": 146.0206,
+      "eval_samples_per_second": 164.737,
+      "eval_steps_per_second": 5.15,
       "step": 2600
     },
     {
-      "epoch": 13.45,
-      "learning_rate": 2e-05,
-      "loss": 6.6369,
+      "epoch": 54.37,
+      "learning_rate": 0.000175589661839308,
+      "loss": 2.6226,
       "step": 2610
     },
     {
-      "epoch": 13.5,
-      "learning_rate": 2e-05,
-      "loss": 6.6619,
+      "epoch": 54.58,
+      "learning_rate": 0.00017427731885980282,
+      "loss": 2.6183,
       "step": 2620
     },
     {
-      "epoch": 13.55,
-      "learning_rate": 2e-05,
-      "loss": 6.6576,
+      "epoch": 54.78,
+      "learning_rate": 0.0001729661010737007,
+      "loss": 2.6313,
       "step": 2630
     },
     {
-      "epoch": 13.61,
-      "learning_rate": 2e-05,
-      "loss": 6.6737,
+      "epoch": 54.99,
+      "learning_rate": 0.00017165606583791515,
+      "loss": 2.6366,
       "step": 2640
     },
     {
-      "epoch": 13.66,
-      "learning_rate": 2e-05,
-      "loss": 6.6428,
+      "epoch": 55.21,
+      "learning_rate": 0.00017034727045763158,
+      "loss": 2.7822,
       "step": 2650
     },
     {
-      "epoch": 13.66,
-      "eval_accuracy": 0.11852539903484209,
-      "eval_loss": 6.701896667480469,
-      "eval_runtime": 169.0251,
-      "eval_samples_per_second": 142.316,
-      "eval_steps_per_second": 2.225,
+      "epoch": 55.21,
+      "eval_accuracy": 0.4934482911572486,
+      "eval_loss": 2.8986542224884033,
+      "eval_runtime": 146.1152,
+      "eval_samples_per_second": 164.63,
+      "eval_steps_per_second": 5.147,
       "step": 2650
     },
     {
-      "epoch": 13.71,
-      "learning_rate": 2e-05,
-      "loss": 6.6634,
+      "epoch": 55.41,
+      "learning_rate": 0.00016903977218379974,
+      "loss": 2.5985,
       "step": 2660
     },
     {
-      "epoch": 13.76,
-      "learning_rate": 2e-05,
-      "loss": 6.6575,
+      "epoch": 55.62,
+      "learning_rate": 0.00016773362821062983,
+      "loss": 2.6059,
       "step": 2670
     },
     {
-      "epoch": 13.81,
-      "learning_rate": 2e-05,
-      "loss": 6.6788,
+      "epoch": 55.82,
+      "learning_rate": 0.00016642889567309048,
+      "loss": 2.6083,
       "step": 2680
     },
     {
-      "epoch": 13.86,
-      "learning_rate": 2e-05,
-      "loss": 6.659,
+      "epoch": 56.04,
+      "learning_rate": 0.0001651256316444097,
+      "loss": 2.7793,
       "step": 2690
     },
     {
-      "epoch": 13.91,
-      "learning_rate": 2e-05,
-      "loss": 6.6355,
+      "epoch": 56.25,
+      "learning_rate": 0.0001638238931335781,
+      "loss": 2.584,
       "step": 2700
     },
     {
-      "epoch": 13.91,
-      "eval_accuracy": 0.11754058144545079,
-      "eval_loss": 6.697854995727539,
-      "eval_runtime": 169.1175,
-      "eval_samples_per_second": 142.238,
-      "eval_steps_per_second": 2.223,
+      "epoch": 56.25,
+      "eval_accuracy": 0.49487679829418024,
+      "eval_loss": 2.8844311237335205,
+      "eval_runtime": 145.9294,
+      "eval_samples_per_second": 164.84,
+      "eval_steps_per_second": 5.153,
       "step": 2700
     },
     {
-      "epoch": 13.97,
-      "learning_rate": 2e-05,
-      "loss": 6.6451,
+      "epoch": 56.45,
+      "learning_rate": 0.00016252373708285504,
+      "loss": 2.5884,
       "step": 2710
     },
     {
-      "epoch": 14.02,
-      "learning_rate": 2e-05,
-      "loss": 7.1351,
+      "epoch": 56.66,
+      "learning_rate": 0.00016122522036527838,
+      "loss": 2.5881,
       "step": 2720
     },
     {
-      "epoch": 14.07,
-      "learning_rate": 2e-05,
-      "loss": 6.6533,
+      "epoch": 56.86,
+      "learning_rate": 0.00015992839978217578,
+      "loss": 2.5866,
       "step": 2730
     },
     {
-      "epoch": 14.12,
-      "learning_rate": 2e-05,
-      "loss": 6.6421,
+      "epoch": 57.08,
+      "learning_rate": 0.00015863333206068067,
+      "loss": 2.7644,
       "step": 2740
     },
     {
-      "epoch": 14.17,
-      "learning_rate": 2e-05,
-      "loss": 6.6521,
+      "epoch": 57.29,
+      "learning_rate": 0.00015734007385125067,
+      "loss": 2.5668,
       "step": 2750
     },
     {
-      "epoch": 14.17,
-      "eval_accuracy": 0.11875193333953445,
-      "eval_loss": 6.692318439483643,
-      "eval_runtime": 169.1423,
-      "eval_samples_per_second": 142.218,
-      "eval_steps_per_second": 2.223,
+      "epoch": 57.29,
+      "eval_accuracy": 0.49651714759851406,
+      "eval_loss": 2.880821704864502,
+      "eval_runtime": 146.1597,
+      "eval_samples_per_second": 164.58,
+      "eval_steps_per_second": 5.145,
       "step": 2750
     },
     {
-      "epoch": 14.23,
-      "learning_rate": 2e-05,
-      "loss": 6.6099,
+      "epoch": 57.49,
+      "learning_rate": 0.0001560486817251894,
+      "loss": 2.5728,
       "step": 2760
     },
     {
-      "epoch": 14.28,
-      "learning_rate": 2e-05,
-      "loss": 6.6254,
+      "epoch": 57.7,
+      "learning_rate": 0.000154759212172172,
+      "loss": 2.5765,
       "step": 2770
     },
     {
-      "epoch": 14.33,
-      "learning_rate": 2e-05,
-      "loss": 6.6545,
+      "epoch": 57.9,
+      "learning_rate": 0.00015347172159777396,
+      "loss": 2.5794,
       "step": 2780
     },
     {
-      "epoch": 14.38,
-      "learning_rate": 2e-05,
-      "loss": 6.6443,
+      "epoch": 58.12,
+      "learning_rate": 0.000152186266321004,
+      "loss": 2.7342,
       "step": 2790
     },
     {
-      "epoch": 14.43,
-      "learning_rate": 2e-05,
-      "loss": 6.6735,
+      "epoch": 58.33,
+      "learning_rate": 0.0001509029025718402,
+      "loss": 2.5536,
       "step": 2800
     },
     {
-      "epoch": 14.43,
-      "eval_accuracy": 0.1185968174100908,
-      "eval_loss": 6.684227466583252,
-      "eval_runtime": 169.1201,
-      "eval_samples_per_second": 142.236,
-      "eval_steps_per_second": 2.223,
+      "epoch": 58.33,
+      "eval_accuracy": 0.4981620698137741,
+      "eval_loss": 2.864001512527466,
+      "eval_runtime": 146.123,
+      "eval_samples_per_second": 164.622,
+      "eval_steps_per_second": 5.146,
       "step": 2800
     },
     {
-      "epoch": 14.48,
-      "learning_rate": 2e-05,
-      "loss": 6.6032,
+      "epoch": 58.53,
+      "learning_rate": 0.0001496216864887704,
+      "loss": 2.5466,
       "step": 2810
     },
     {
-      "epoch": 14.53,
-      "learning_rate": 2e-05,
-      "loss": 6.6217,
+      "epoch": 58.74,
+      "learning_rate": 0.00014834267411633674,
+      "loss": 2.553,
       "step": 2820
     },
     {
-      "epoch": 14.59,
-      "learning_rate": 2e-05,
-      "loss": 6.6212,
+      "epoch": 58.95,
+      "learning_rate": 0.0001470659214026837,
+      "loss": 2.5623,
       "step": 2830
     },
     {
-      "epoch": 14.64,
-      "learning_rate": 2e-05,
-      "loss": 6.636,
+      "epoch": 59.16,
+      "learning_rate": 0.00014579148419711119,
+      "loss": 2.727,
       "step": 2840
     },
     {
-      "epoch": 14.69,
-      "learning_rate": 2e-05,
-      "loss": 6.6151,
+      "epoch": 59.37,
+      "learning_rate": 0.00014451941824763113,
+      "loss": 2.5403,
       "step": 2850
     },
     {
-      "epoch": 14.69,
-      "eval_accuracy": 0.11951815218710449,
-      "eval_loss": 6.679075241088867,
-      "eval_runtime": 168.9877,
-      "eval_samples_per_second": 142.348,
-      "eval_steps_per_second": 2.225,
+      "epoch": 59.37,
+      "eval_accuracy": 0.49815218958648255,
+      "eval_loss": 2.860569953918457,
+      "eval_runtime": 146.132,
+      "eval_samples_per_second": 164.611,
+      "eval_steps_per_second": 5.146,
       "step": 2850
     },
     {
-      "epoch": 14.74,
-      "learning_rate": 2e-05,
-      "loss": 6.6269,
+      "epoch": 59.58,
+      "learning_rate": 0.000143249779198529,
+      "loss": 2.5441,
       "step": 2860
     },
     {
-      "epoch": 14.79,
-      "learning_rate": 2e-05,
-      "loss": 6.6356,
+      "epoch": 59.78,
+      "learning_rate": 0.00014198262258793002,
+      "loss": 2.5541,
       "step": 2870
     },
     {
-      "epoch": 14.84,
-      "learning_rate": 2e-05,
-      "loss": 6.6496,
+      "epoch": 59.99,
+      "learning_rate": 0.00014071800384536927,
+      "loss": 2.5482,
       "step": 2880
     },
     {
-      "epoch": 14.89,
-      "learning_rate": 2e-05,
-      "loss": 6.6702,
+      "epoch": 60.21,
+      "learning_rate": 0.00013945597828936737,
+      "loss": 2.6878,
       "step": 2890
     },
     {
-      "epoch": 14.94,
-      "learning_rate": 2e-05,
-      "loss": 6.6248,
+      "epoch": 60.41,
+      "learning_rate": 0.00013819660112501054,
+      "loss": 2.5294,
       "step": 2900
     },
     {
-      "epoch": 14.94,
-      "eval_accuracy": 0.11975263870094723,
-      "eval_loss": 6.675192832946777,
-      "eval_runtime": 169.15,
-      "eval_samples_per_second": 142.211,
-      "eval_steps_per_second": 2.223,
+      "epoch": 60.41,
+      "eval_accuracy": 0.5007705653773009,
+      "eval_loss": 2.8440916538238525,
+      "eval_runtime": 146.0675,
+      "eval_samples_per_second": 164.684,
+      "eval_steps_per_second": 5.148,
       "step": 2900
     },
     {
-      "epoch": 15.0,
-      "learning_rate": 2e-05,
-      "loss": 6.6325,
+      "epoch": 60.62,
+      "learning_rate": 0.00013693992744153572,
+      "loss": 2.5448,
       "step": 2910
     },
     {
-      "epoch": 15.05,
-      "learning_rate": 2e-05,
-      "loss": 7.124,
+      "epoch": 60.82,
+      "learning_rate": 0.00013568601220992097,
+      "loss": 2.5435,
       "step": 2920
     },
     {
-      "epoch": 15.1,
-      "learning_rate": 2e-05,
-      "loss": 6.6072,
+      "epoch": 61.04,
+      "learning_rate": 0.00013443491028048045,
+      "loss": 2.71,
       "step": 2930
     },
     {
-      "epoch": 15.15,
-      "learning_rate": 2e-05,
-      "loss": 6.6165,
+      "epoch": 61.25,
+      "learning_rate": 0.0001331866763804658,
+      "loss": 2.5199,
       "step": 2940
     },
     {
-      "epoch": 15.21,
-      "learning_rate": 2e-05,
-      "loss": 6.6427,
+      "epoch": 61.45,
+      "learning_rate": 0.0001319413651116714,
+      "loss": 2.513,
       "step": 2950
     },
     {
-      "epoch": 15.21,
-      "eval_accuracy": 0.12074793730646591,
-      "eval_loss": 6.666472434997559,
-      "eval_runtime": 169.0435,
-      "eval_samples_per_second": 142.301,
-      "eval_steps_per_second": 2.224,
+      "epoch": 61.45,
+      "eval_accuracy": 0.5013016714921779,
+      "eval_loss": 2.840217113494873,
+      "eval_runtime": 146.0072,
+      "eval_samples_per_second": 164.752,
+      "eval_steps_per_second": 5.15,
       "step": 2950
     },
     {
-      "epoch": 15.26,
-      "learning_rate": 2e-05,
-      "loss": 6.6282,
+      "epoch": 61.66,
+      "learning_rate": 0.00013069903094804644,
+      "loss": 2.5158,
       "step": 2960
     },
     {
-      "epoch": 15.31,
-      "learning_rate": 2e-05,
-      "loss": 6.6127,
+      "epoch": 61.86,
+      "learning_rate": 0.0001294597282333118,
+      "loss": 2.5292,
       "step": 2970
     },
     {
-      "epoch": 15.36,
-      "learning_rate": 2e-05,
-      "loss": 6.6067,
+      "epoch": 62.08,
+      "learning_rate": 0.00012822351117858303,
+      "loss": 2.6752,
       "step": 2980
     },
     {
-      "epoch": 15.41,
-      "learning_rate": 2e-05,
-      "loss": 6.6559,
+      "epoch": 62.29,
+      "learning_rate": 0.0001269904338599989,
+      "loss": 2.5094,
       "step": 2990
     },
     {
-      "epoch": 15.46,
-      "learning_rate": 2e-05,
-      "loss": 6.5947,
+      "epoch": 62.49,
+      "learning_rate": 0.0001257605502163558,
+      "loss": 2.5105,
       "step": 3000
     },
     {
-      "epoch": 15.46,
-      "eval_accuracy": 0.12065681925283153,
-      "eval_loss": 6.663944721221924,
-      "eval_runtime": 169.0216,
-      "eval_samples_per_second": 142.319,
-      "eval_steps_per_second": 2.225,
+      "epoch": 62.49,
+      "eval_accuracy": 0.5022339398713631,
+      "eval_loss": 2.8315513134002686,
+      "eval_runtime": 146.1095,
+      "eval_samples_per_second": 164.637,
+      "eval_steps_per_second": 5.147,
       "step": 3000
     },
     {
-      "epoch": 15.51,
-      "learning_rate": 2e-05,
-      "loss": 6.62,
+      "epoch": 62.7,
+      "learning_rate": 0.00012453391404674885,
+      "loss": 2.4981,
       "step": 3010
     },
     {
-      "epoch": 15.56,
-      "learning_rate": 2e-05,
-      "loss": 6.6387,
+      "epoch": 62.9,
+      "learning_rate": 0.00012331057900821768,
+      "loss": 2.5072,
       "step": 3020
     },
     {
-      "epoch": 15.62,
-      "learning_rate": 2e-05,
-      "loss": 6.6129,
+      "epoch": 63.12,
+      "learning_rate": 0.0001220905986134,
+      "loss": 2.6561,
       "step": 3030
     },
     {
-      "epoch": 15.67,
-      "learning_rate": 2e-05,
-      "loss": 6.6413,
+      "epoch": 63.33,
+      "learning_rate": 0.00012087402622819039,
+      "loss": 2.5062,
       "step": 3040
     },
     {
-      "epoch": 15.72,
-      "learning_rate": 2e-05,
-      "loss": 6.6199,
+      "epoch": 63.53,
+      "learning_rate": 0.00011966091506940616,
+      "loss": 2.4897,
       "step": 3050
     },
     {
-      "epoch": 15.72,
-      "eval_accuracy": 0.12169938085571898,
-      "eval_loss": 6.659843921661377,
-      "eval_runtime": 169.0987,
-      "eval_samples_per_second": 142.254,
-      "eval_steps_per_second": 2.224,
+      "epoch": 63.53,
+      "eval_accuracy": 0.502685487439774,
+      "eval_loss": 2.823685646057129,
+      "eval_runtime": 146.1084,
+      "eval_samples_per_second": 164.638,
+      "eval_steps_per_second": 5.147,
       "step": 3050
     },
     {
-      "epoch": 15.77,
-      "learning_rate": 2e-05,
-      "loss": 6.6067,
+      "epoch": 63.74,
+      "learning_rate": 0.00011845131820245934,
+      "loss": 2.4945,
       "step": 3060
     },
     {
-      "epoch": 15.82,
-      "learning_rate": 2e-05,
-      "loss": 6.6333,
+      "epoch": 63.95,
+      "learning_rate": 0.00011724528853903536,
+      "loss": 2.5023,
       "step": 3070
     },
     {
-      "epoch": 15.87,
-      "learning_rate": 2e-05,
-      "loss": 6.619,
+      "epoch": 64.16,
+      "learning_rate": 0.00011604287883477889,
+      "loss": 2.637,
       "step": 3080
     },
     {
-      "epoch": 15.92,
-      "learning_rate": 2e-05,
-      "loss": 6.6274,
+      "epoch": 64.37,
+      "learning_rate": 0.00011484414168698547,
+      "loss": 2.4841,
       "step": 3090
     },
     {
-      "epoch": 15.98,
-      "learning_rate": 2e-05,
-      "loss": 6.6127,
+      "epoch": 64.58,
+      "learning_rate": 0.00011364912953230145,
+      "loss": 2.4974,
       "step": 3100
     },
     {
-      "epoch": 15.98,
-      "eval_accuracy": 0.12192011294272531,
-      "eval_loss": 6.659284591674805,
-      "eval_runtime": 169.0589,
-      "eval_samples_per_second": 142.288,
-      "eval_steps_per_second": 2.224,
+      "epoch": 64.58,
+      "eval_accuracy": 0.5039655187362361,
+      "eval_loss": 2.818704605102539,
+      "eval_runtime": 146.0534,
+      "eval_samples_per_second": 164.7,
+      "eval_steps_per_second": 5.149,
       "step": 3100
     },
     {
-      "epoch": 16.03,
-      "learning_rate": 2e-05,
-      "loss": 7.1321,
+      "epoch": 64.78,
+      "learning_rate": 0.00011245789464442964,
+      "loss": 2.496,
       "step": 3110
     },
     {
-      "epoch": 16.08,
-      "learning_rate": 2e-05,
-      "loss": 6.5976,
+      "epoch": 64.99,
+      "learning_rate": 0.00011127048913184326,
+      "loss": 2.4902,
       "step": 3120
     },
     {
-      "epoch": 16.13,
-      "learning_rate": 2e-05,
-      "loss": 6.6261,
+      "epoch": 65.21,
+      "learning_rate": 0.00011008696493550599,
+      "loss": 2.6366,
       "step": 3130
     },
     {
-      "epoch": 16.18,
-      "learning_rate": 2e-05,
-      "loss": 6.6105,
+      "epoch": 65.41,
+      "learning_rate": 0.00010890737382660015,
+      "loss": 2.4739,
       "step": 3140
     },
     {
-      "epoch": 16.24,
-      "learning_rate": 2e-05,
-      "loss": 6.6031,
+      "epoch": 65.62,
+      "learning_rate": 0.00010773176740426248,
+      "loss": 2.4799,
       "step": 3150
     },
     {
-      "epoch": 16.24,
-      "eval_accuracy": 0.12264471521223672,
-      "eval_loss": 6.651196002960205,
-      "eval_runtime": 169.0792,
-      "eval_samples_per_second": 142.271,
-      "eval_steps_per_second": 2.224,
+      "epoch": 65.62,
+      "eval_accuracy": 0.5044451239477096,
+      "eval_loss": 2.8128514289855957,
+      "eval_runtime": 146.0215,
+      "eval_samples_per_second": 164.736,
+      "eval_steps_per_second": 5.15,
       "step": 3150
     },
     {
-      "epoch": 16.29,
-      "learning_rate": 2e-05,
-      "loss": 6.5796,
+      "epoch": 65.82,
+      "learning_rate": 0.00010656019709332606,
+      "loss": 2.4707,
       "step": 3160
     },
     {
-      "epoch": 16.34,
-      "learning_rate": 2e-05,
-      "loss": 6.5983,
+      "epoch": 66.04,
+      "learning_rate": 0.00010539271414207186,
+      "loss": 2.6249,
       "step": 3170
     },
     {
-      "epoch": 16.39,
-      "learning_rate": 2e-05,
-      "loss": 6.5812,
+      "epoch": 66.25,
+      "learning_rate": 0.00010422936961998609,
+      "loss": 2.4617,
       "step": 3180
     },
     {
-      "epoch": 16.44,
-      "learning_rate": 2e-05,
-      "loss": 6.6298,
+      "epoch": 66.45,
+      "learning_rate": 0.00010307021441552707,
+      "loss": 2.4508,
       "step": 3190
     },
     {
-      "epoch": 16.49,
-      "learning_rate": 2e-05,
-      "loss": 6.5742,
+      "epoch": 66.66,
+      "learning_rate": 0.00010191529923389845,
+      "loss": 2.4741,
       "step": 3200
     },
     {
-      "epoch": 16.49,
-      "eval_accuracy": 0.12269288211432908,
-      "eval_loss": 6.64845609664917,
-      "eval_runtime": 169.2018,
-      "eval_samples_per_second": 142.168,
-      "eval_steps_per_second": 2.222,
+      "epoch": 66.66,
+      "eval_accuracy": 0.5057173793056381,
+      "eval_loss": 2.805563449859619,
+      "eval_runtime": 146.0069,
+      "eval_samples_per_second": 164.752,
+      "eval_steps_per_second": 5.15,
       "step": 3200
     },
     {
-      "epoch": 16.54,
-      "learning_rate": 2e-05,
-      "loss": 6.6001,
+      "epoch": 66.86,
+      "learning_rate": 0.00010076467459483155,
+      "loss": 2.4658,
       "step": 3210
     },
     {
-      "epoch": 16.6,
-      "learning_rate": 2e-05,
-      "loss": 6.6124,
+      "epoch": 67.08,
+      "learning_rate": 9.961839083037592e-05,
+      "loss": 2.6267,
       "step": 3220
     },
     {
-      "epoch": 16.65,
-      "learning_rate": 2e-05,
-      "loss": 6.6131,
+      "epoch": 67.29,
+      "learning_rate": 9.847649808269658e-05,
+      "loss": 2.4656,
       "step": 3230
     },
     {
-      "epoch": 16.7,
-      "learning_rate": 2e-05,
-      "loss": 6.6051,
+      "epoch": 67.49,
+      "learning_rate": 9.733904630188176e-05,
+      "loss": 2.4421,
       "step": 3240
     },
     {
-      "epoch": 16.75,
-      "learning_rate": 2e-05,
-      "loss": 6.621,
+      "epoch": 67.7,
+      "learning_rate": 9.620608524375703e-05,
+      "loss": 2.4582,
       "step": 3250
     },
     {
-      "epoch": 16.75,
-      "eval_accuracy": 0.12212792135914549,
-      "eval_loss": 6.647208213806152,
-      "eval_runtime": 169.0564,
-      "eval_samples_per_second": 142.29,
-      "eval_steps_per_second": 2.224,
+      "epoch": 67.7,
+      "eval_accuracy": 0.506052237287108,
+      "eval_loss": 2.80246639251709,
+      "eval_runtime": 145.9985,
+      "eval_samples_per_second": 164.762,
+      "eval_steps_per_second": 5.151,
       "step": 3250
     },
     {
-      "epoch": 16.8,
-      "learning_rate": 2e-05,
-      "loss": 6.597,
+      "epoch": 67.9,
+      "learning_rate": 9.507766446770934e-05,
+      "loss": 2.456,
       "step": 3260
     },
     {
-      "epoch": 16.85,
-      "learning_rate": 2e-05,
-      "loss": 6.6102,
+      "epoch": 68.12,
+      "learning_rate": 9.39538333345191e-05,
+      "loss": 2.6204,
       "step": 3270
     },
     {
-      "epoch": 16.9,
-      "learning_rate": 2e-05,
-      "loss": 6.5693,
+      "epoch": 68.33,
+      "learning_rate": 9.283464100420063e-05,
+      "loss": 2.4513,
       "step": 3280
     },
     {
-      "epoch": 16.96,
-      "learning_rate": 2e-05,
-      "loss": 6.5589,
+      "epoch": 68.53,
+      "learning_rate": 9.17201364338524e-05,
+      "loss": 2.4486,
       "step": 3290
     },
     {
-      "epoch": 17.01,
-      "learning_rate": 2e-05,
-      "loss": 7.0655,
+      "epoch": 68.74,
+      "learning_rate": 9.061036837551466e-05,
+      "loss": 2.4389,
       "step": 3300
     },
     {
-      "epoch": 17.01,
-      "eval_accuracy": 0.123154659717467,
-      "eval_loss": 6.636902332305908,
-      "eval_runtime": 169.0684,
-      "eval_samples_per_second": 142.28,
-      "eval_steps_per_second": 2.224,
+      "epoch": 68.74,
+      "eval_accuracy": 0.5075605292045352,
+      "eval_loss": 2.791304111480713,
+      "eval_runtime": 146.0353,
+      "eval_samples_per_second": 164.72,
+      "eval_steps_per_second": 5.149,
       "step": 3300
     },
     {
-      "epoch": 17.06,
-      "learning_rate": 2e-05,
-      "loss": 6.5935,
+      "epoch": 68.95,
+      "learning_rate": 8.950538537403736e-05,
+      "loss": 2.4384,
       "step": 3310
     },
     {
-      "epoch": 17.11,
-      "learning_rate": 2e-05,
-      "loss": 6.5824,
+      "epoch": 69.16,
+      "learning_rate": 8.840523576495681e-05,
+      "loss": 2.5977,
       "step": 3320
     },
     {
-      "epoch": 17.16,
-      "learning_rate": 2e-05,
-      "loss": 6.5584,
+      "epoch": 69.37,
+      "learning_rate": 8.730996767238072e-05,
+      "loss": 2.4459,
       "step": 3330
     },
     {
-      "epoch": 17.22,
-      "learning_rate": 2e-05,
-      "loss": 6.5853,
+      "epoch": 69.58,
+      "learning_rate": 8.621962900688378e-05,
+      "loss": 2.4281,
       "step": 3340
     },
     {
-      "epoch": 17.27,
-      "learning_rate": 2e-05,
-      "loss": 6.5866,
+      "epoch": 69.78,
+      "learning_rate": 8.513426746341128e-05,
+      "loss": 2.4539,
       "step": 3350
     },
     {
-      "epoch": 17.27,
-      "eval_accuracy": 0.12344173360979259,
-      "eval_loss": 6.637628555297852,
-      "eval_runtime": 169.0659,
-      "eval_samples_per_second": 142.282,
-      "eval_steps_per_second": 2.224,
+      "epoch": 69.78,
+      "eval_accuracy": 0.5071934293322717,
+      "eval_loss": 2.7881319522857666,
+      "eval_runtime": 145.9867,
+      "eval_samples_per_second": 164.775,
+      "eval_steps_per_second": 5.151,
       "step": 3350
     },
     {
-      "epoch": 17.32,
-      "learning_rate": 2e-05,
-      "loss": 6.5515,
+      "epoch": 69.99,
+      "learning_rate": 8.405393051919333e-05,
+      "loss": 2.4298,
       "step": 3360
     },
     {
-      "epoch": 17.37,
-      "learning_rate": 2e-05,
-      "loss": 6.623,
+      "epoch": 70.21,
+      "learning_rate": 8.29786654316677e-05,
+      "loss": 2.5885,
       "step": 3370
     },
     {
-      "epoch": 17.42,
-      "learning_rate": 2e-05,
-      "loss": 6.5592,
+      "epoch": 70.41,
+      "learning_rate": 8.190851923641259e-05,
+      "loss": 2.4073,
       "step": 3380
     },
     {
-      "epoch": 17.47,
-      "learning_rate": 2e-05,
-      "loss": 6.5367,
+      "epoch": 70.62,
+      "learning_rate": 8.084353874508947e-05,
+      "loss": 2.4379,
       "step": 3390
     },
     {
-      "epoch": 17.52,
-      "learning_rate": 2e-05,
-      "loss": 6.6098,
+      "epoch": 70.82,
+      "learning_rate": 7.978377054339499e-05,
+      "loss": 2.4252,
       "step": 3400
     },
     {
-      "epoch": 17.52,
-      "eval_accuracy": 0.12519736165745915,
-      "eval_loss": 6.631270885467529,
-      "eval_runtime": 169.1186,
-      "eval_samples_per_second": 142.237,
-      "eval_steps_per_second": 2.223,
+      "epoch": 70.82,
+      "eval_accuracy": 0.5081794918909719,
+      "eval_loss": 2.7884321212768555,
+      "eval_runtime": 146.1195,
+      "eval_samples_per_second": 164.626,
+      "eval_steps_per_second": 5.146,
       "step": 3400
     },
     {
-      "epoch": 17.58,
-      "learning_rate": 2e-05,
-      "loss": 6.5829,
+      "epoch": 71.04,
+      "learning_rate": 7.872926098902358e-05,
+      "loss": 2.5932,
       "step": 3410
     },
     {
-      "epoch": 17.63,
-      "learning_rate": 2e-05,
-      "loss": 6.564,
+      "epoch": 71.25,
+      "learning_rate": 7.768005620963916e-05,
+      "loss": 2.4153,
       "step": 3420
     },
     {
-      "epoch": 17.68,
-      "learning_rate": 2e-05,
-      "loss": 6.5818,
+      "epoch": 71.45,
+      "learning_rate": 7.663620210085781e-05,
+      "loss": 2.4195,
       "step": 3430
     },
     {
-      "epoch": 17.73,
-      "learning_rate": 2e-05,
-      "loss": 6.5715,
+      "epoch": 71.66,
+      "learning_rate": 7.55977443242399e-05,
+      "loss": 2.4231,
       "step": 3440
     },
     {
-      "epoch": 17.78,
-      "learning_rate": 2e-05,
-      "loss": 6.5676,
+      "epoch": 71.86,
+      "learning_rate": 7.456472830529259e-05,
+      "loss": 2.4287,
       "step": 3450
     },
     {
-      "epoch": 17.78,
-      "eval_accuracy": 0.124823313950876,
-      "eval_loss": 6.625356674194336,
-      "eval_runtime": 169.1641,
-      "eval_samples_per_second": 142.199,
-      "eval_steps_per_second": 2.223,
+      "epoch": 71.86,
+      "eval_accuracy": 0.5093288685486723,
+      "eval_loss": 2.778383493423462,
+      "eval_runtime": 145.9882,
+      "eval_samples_per_second": 164.774,
+      "eval_steps_per_second": 5.151,
       "step": 3450
     },
     {
-      "epoch": 17.83,
-      "learning_rate": 2e-05,
-      "loss": 6.5796,
+      "epoch": 72.08,
+      "learning_rate": 7.353719923148324e-05,
+      "loss": 2.5804,
       "step": 3460
     },
     {
-      "epoch": 17.88,
-      "learning_rate": 2e-05,
-      "loss": 6.595,
+      "epoch": 72.29,
+      "learning_rate": 7.251520205026205e-05,
+      "loss": 2.4048,
       "step": 3470
     },
     {
-      "epoch": 17.93,
-      "learning_rate": 2e-05,
-      "loss": 6.5946,
+      "epoch": 72.49,
+      "learning_rate": 7.149878146709676e-05,
+      "loss": 2.4008,
       "step": 3480
     },
     {
-      "epoch": 17.99,
-      "learning_rate": 2e-05,
-      "loss": 6.5585,
+      "epoch": 72.7,
+      "learning_rate": 7.048798194351625e-05,
+      "loss": 2.41,
       "step": 3490
     },
     {
-      "epoch": 18.04,
-      "learning_rate": 2e-05,
-      "loss": 7.0636,
+      "epoch": 72.9,
+      "learning_rate": 6.948284769516627e-05,
+      "loss": 2.4131,
       "step": 3500
     },
     {
-      "epoch": 18.04,
-      "eval_accuracy": 0.1255522859600729,
-      "eval_loss": 6.622625350952148,
-      "eval_runtime": 169.0712,
-      "eval_samples_per_second": 142.277,
-      "eval_steps_per_second": 2.224,
+      "epoch": 72.9,
+      "eval_accuracy": 0.5098879891877023,
+      "eval_loss": 2.7781522274017334,
+      "eval_runtime": 146.0156,
+      "eval_samples_per_second": 164.743,
+      "eval_steps_per_second": 5.15,
       "step": 3500
     },
     {
-      "epoch": 18.09,
-      "learning_rate": 2e-05,
-      "loss": 6.5962,
+      "epoch": 73.12,
+      "learning_rate": 6.848342268987511e-05,
+      "loss": 2.5661,
       "step": 3510
     },
     {
-      "epoch": 18.14,
-      "learning_rate": 2e-05,
-      "loss": 6.544,
+      "epoch": 73.33,
+      "learning_rate": 6.748975064573007e-05,
+      "loss": 2.3994,
       "step": 3520
     },
     {
-      "epoch": 18.2,
-      "learning_rate": 2e-05,
-      "loss": 6.5729,
+      "epoch": 73.53,
+      "learning_rate": 6.650187502916552e-05,
+      "loss": 2.4078,
       "step": 3530
     },
     {
-      "epoch": 18.25,
-      "learning_rate": 2e-05,
-      "loss": 6.5665,
+      "epoch": 73.74,
+      "learning_rate": 6.551983905306107e-05,
+      "loss": 2.4168,
       "step": 3540
     },
     {
-      "epoch": 18.3,
-      "learning_rate": 2e-05,
-      "loss": 6.5444,
+      "epoch": 73.95,
+      "learning_rate": 6.454368567485183e-05,
+      "loss": 2.4016,
       "step": 3550
     },
     {
-      "epoch": 18.3,
-      "eval_accuracy": 0.12534055374011083,
-      "eval_loss": 6.616397857666016,
-      "eval_runtime": 169.0563,
-      "eval_samples_per_second": 142.29,
-      "eval_steps_per_second": 2.224,
+      "epoch": 73.95,
+      "eval_accuracy": 0.5097699735946659,
+      "eval_loss": 2.772381544113159,
+      "eval_runtime": 146.0586,
+      "eval_samples_per_second": 164.694,
+      "eval_steps_per_second": 5.149,
       "step": 3550
     },
     {
-      "epoch": 18.35,
-      "learning_rate": 2e-05,
-      "loss": 6.6036,
+      "epoch": 74.16,
+      "learning_rate": 6.35734575946487e-05,
+      "loss": 2.5732,
       "step": 3560
     },
     {
-      "epoch": 18.4,
-      "learning_rate": 2e-05,
-      "loss": 6.5666,
+      "epoch": 74.37,
+      "learning_rate": 6.260919725337109e-05,
+      "loss": 2.3961,
       "step": 3570
     },
     {
-      "epoch": 18.45,
-      "learning_rate": 2e-05,
-      "loss": 6.5536,
+      "epoch": 74.58,
+      "learning_rate": 6.165094683089015e-05,
+      "loss": 2.4073,
       "step": 3580
     },
     {
-      "epoch": 18.5,
-      "learning_rate": 2e-05,
-      "loss": 6.5631,
+      "epoch": 74.78,
+      "learning_rate": 6.069874824418356e-05,
+      "loss": 2.3997,
       "step": 3590
     },
     {
-      "epoch": 18.55,
-      "learning_rate": 2e-05,
-      "loss": 6.561,
+      "epoch": 74.99,
+      "learning_rate": 5.975264314550229e-05,
+      "loss": 2.3998,
       "step": 3600
     },
     {
-      "epoch": 18.55,
-      "eval_accuracy": 0.12544435584844563,
-      "eval_loss": 6.615684986114502,
-      "eval_runtime": 169.0849,
-      "eval_samples_per_second": 142.266,
-      "eval_steps_per_second": 2.224,
+      "epoch": 74.99,
+      "eval_accuracy": 0.5110515365958426,
+      "eval_loss": 2.7658748626708984,
+      "eval_runtime": 146.0874,
+      "eval_samples_per_second": 164.662,
+      "eval_steps_per_second": 5.148,
       "step": 3600
     },
     {
-      "epoch": 18.61,
-      "learning_rate": 2e-05,
-      "loss": 6.5385,
+      "epoch": 75.21,
+      "learning_rate": 5.881267292054828e-05,
+      "loss": 2.5492,
       "step": 3610
     },
     {
-      "epoch": 18.66,
-      "learning_rate": 2e-05,
-      "loss": 6.5558,
+      "epoch": 75.41,
+      "learning_rate": 5.787887868666417e-05,
+      "loss": 2.3838,
       "step": 3620
     },
     {
-      "epoch": 18.71,
-      "learning_rate": 2e-05,
-      "loss": 6.5936,
+      "epoch": 75.62,
+      "learning_rate": 5.6951301291034945e-05,
+      "loss": 2.398,
       "step": 3630
     },
     {
-      "epoch": 18.76,
-      "learning_rate": 2e-05,
-      "loss": 6.5451,
+      "epoch": 75.82,
+      "learning_rate": 5.602998130890065e-05,
+      "loss": 2.4025,
       "step": 3640
     },
     {
-      "epoch": 18.81,
-      "learning_rate": 2e-05,
-      "loss": 6.5882,
+      "epoch": 76.04,
+      "learning_rate": 5.511495904178221e-05,
+      "loss": 2.5475,
       "step": 3650
     },
     {
-      "epoch": 18.81,
-      "eval_accuracy": 0.12569411736714234,
-      "eval_loss": 6.607241153717041,
-      "eval_runtime": 169.0962,
-      "eval_samples_per_second": 142.256,
-      "eval_steps_per_second": 2.224,
+      "epoch": 76.04,
+      "eval_accuracy": 0.510823536539714,
+      "eval_loss": 2.7650203704833984,
+      "eval_runtime": 146.0073,
+      "eval_samples_per_second": 164.752,
+      "eval_steps_per_second": 5.15,
       "step": 3650
     },
     {
-      "epoch": 18.86,
-      "learning_rate": 2e-05,
-      "loss": 6.5568,
+      "epoch": 76.25,
+      "learning_rate": 5.4206274515717736e-05,
+      "loss": 2.4011,
       "step": 3660
     },
     {
-      "epoch": 18.91,
-      "learning_rate": 2e-05,
-      "loss": 6.5808,
+      "epoch": 76.45,
+      "learning_rate": 5.330396747951205e-05,
+      "loss": 2.3818,
       "step": 3670
     },
     {
-      "epoch": 18.97,
-      "learning_rate": 2e-05,
-      "loss": 6.5542,
+      "epoch": 76.66,
+      "learning_rate": 5.240807740299811e-05,
+      "loss": 2.3911,
       "step": 3680
     },
     {
-      "epoch": 19.02,
-      "learning_rate": 2e-05,
-      "loss": 7.0483,
+      "epoch": 76.86,
+      "learning_rate": 5.1518643475310034e-05,
+      "loss": 2.389,
       "step": 3690
     },
     {
-      "epoch": 19.07,
-      "learning_rate": 2e-05,
-      "loss": 6.5518,
+      "epoch": 77.08,
+      "learning_rate": 5.0635704603169287e-05,
+      "loss": 2.5443,
       "step": 3700
     },
     {
-      "epoch": 19.07,
-      "eval_accuracy": 0.12669918361249846,
-      "eval_loss": 6.6064453125,
-      "eval_runtime": 169.0526,
-      "eval_samples_per_second": 142.293,
-      "eval_steps_per_second": 2.224,
+      "epoch": 77.08,
+      "eval_accuracy": 0.5117344133064243,
+      "eval_loss": 2.7620205879211426,
+      "eval_runtime": 146.0022,
+      "eval_samples_per_second": 164.758,
+      "eval_steps_per_second": 5.151,
       "step": 3700
     },
     {
-      "epoch": 19.12,
-      "learning_rate": 2e-05,
-      "loss": 6.5397,
+      "epoch": 77.29,
+      "learning_rate": 4.975929940918236e-05,
+      "loss": 2.38,
       "step": 3710
     },
     {
-      "epoch": 19.17,
-      "learning_rate": 2e-05,
-      "loss": 6.5834,
+      "epoch": 77.49,
+      "learning_rate": 4.8889466230151646e-05,
+      "loss": 2.3758,
       "step": 3720
     },
     {
-      "epoch": 19.23,
-      "learning_rate": 2e-05,
-      "loss": 6.5657,
+      "epoch": 77.7,
+      "learning_rate": 4.8026243115398314e-05,
+      "loss": 2.3744,
       "step": 3730
     },
     {
-      "epoch": 19.28,
-      "learning_rate": 2e-05,
-      "loss": 6.5657,
+      "epoch": 77.9,
+      "learning_rate": 4.7169667825097775e-05,
+      "loss": 2.3784,
       "step": 3740
     },
     {
-      "epoch": 19.33,
-      "learning_rate": 2e-05,
-      "loss": 6.5599,
+      "epoch": 78.12,
+      "learning_rate": 4.631977782862824e-05,
+      "loss": 2.5381,
       "step": 3750
     },
     {
-      "epoch": 19.33,
-      "eval_accuracy": 0.12710447874336364,
-      "eval_loss": 6.60552453994751,
-      "eval_runtime": 169.0746,
-      "eval_samples_per_second": 142.274,
-      "eval_steps_per_second": 2.224,
+      "epoch": 78.12,
+      "eval_accuracy": 0.5115312635222847,
+      "eval_loss": 2.76308274269104,
+      "eval_runtime": 146.1953,
+      "eval_samples_per_second": 164.54,
+      "eval_steps_per_second": 5.144,
       "step": 3750
     },
     {
-      "epoch": 19.38,
-      "learning_rate": 2e-05,
-      "loss": 6.5841,
+      "epoch": 78.33,
+      "learning_rate": 4.547661030293129e-05,
+      "loss": 2.3771,
       "step": 3760
     },
     {
-      "epoch": 19.43,
-      "learning_rate": 2e-05,
-      "loss": 6.5189,
+      "epoch": 78.53,
+      "learning_rate": 4.464020213088611e-05,
+      "loss": 2.3786,
       "step": 3770
     },
     {
-      "epoch": 19.48,
-      "learning_rate": 2e-05,
-      "loss": 6.551,
+      "epoch": 78.74,
+      "learning_rate": 4.381058989969564e-05,
+      "loss": 2.3688,
       "step": 3780
     },
     {
-      "epoch": 19.53,
-      "learning_rate": 2e-05,
-      "loss": 6.5466,
+      "epoch": 78.95,
+      "learning_rate": 4.298780989928646e-05,
+      "loss": 2.3792,
       "step": 3790
     },
     {
-      "epoch": 19.59,
-      "learning_rate": 2e-05,
-      "loss": 6.5407,
+      "epoch": 79.16,
+      "learning_rate": 4.217189812072131e-05,
+      "loss": 2.5269,
       "step": 3800
     },
     {
-      "epoch": 19.59,
-      "eval_accuracy": 0.12744788950670777,
-      "eval_loss": 6.598654270172119,
-      "eval_runtime": 169.0329,
-      "eval_samples_per_second": 142.31,
-      "eval_steps_per_second": 2.224,
+      "epoch": 79.16,
+      "eval_accuracy": 0.5122286175796967,
+      "eval_loss": 2.7577943801879883,
+      "eval_runtime": 146.122,
+      "eval_samples_per_second": 164.623,
+      "eval_steps_per_second": 5.146,
       "step": 3800
     },
     {
-      "epoch": 19.64,
-      "learning_rate": 2e-05,
-      "loss": 6.5842,
+      "epoch": 79.37,
+      "learning_rate": 4.136289025462443e-05,
+      "loss": 2.3679,
       "step": 3810
     },
     {
-      "epoch": 19.69,
-      "learning_rate": 2e-05,
-      "loss": 6.5077,
+      "epoch": 79.58,
+      "learning_rate": 4.0560821689620856e-05,
+      "loss": 2.3749,
       "step": 3820
     },
     {
-      "epoch": 19.74,
-      "learning_rate": 2e-05,
-      "loss": 6.5263,
+      "epoch": 79.78,
+      "learning_rate": 3.976572751078782e-05,
+      "loss": 2.3605,
       "step": 3830
     },
     {
-      "epoch": 19.79,
-      "learning_rate": 2e-05,
-      "loss": 6.5459,
+      "epoch": 79.99,
+      "learning_rate": 3.8977642498120594e-05,
+      "loss": 2.3747,
       "step": 3840
     },
     {
-      "epoch": 19.84,
-      "learning_rate": 2e-05,
-      "loss": 6.5373,
+      "epoch": 80.21,
+      "learning_rate": 3.819660112501053e-05,
+      "loss": 2.5288,
       "step": 3850
     },
     {
-      "epoch": 19.84,
-      "eval_accuracy": 0.12802956265143414,
-      "eval_loss": 6.595397472381592,
-      "eval_runtime": 169.0397,
-      "eval_samples_per_second": 142.304,
-      "eval_steps_per_second": 2.224,
+      "epoch": 80.21,
+      "eval_accuracy": 0.5124386898610601,
+      "eval_loss": 2.754046678543091,
+      "eval_runtime": 146.1606,
+      "eval_samples_per_second": 164.579,
+      "eval_steps_per_second": 5.145,
       "step": 3850
     },
     {
-      "epoch": 19.89,
-      "learning_rate": 2e-05,
-      "loss": 6.5126,
+      "epoch": 80.41,
+      "learning_rate": 3.742263755673758e-05,
+      "loss": 2.367,
       "step": 3860
     },
     {
-      "epoch": 19.94,
-      "learning_rate": 2e-05,
-      "loss": 6.5966,
+      "epoch": 80.62,
+      "learning_rate": 3.6655785648975585e-05,
+      "loss": 2.3667,
       "step": 3870
     },
     {
-      "epoch": 20.0,
-      "learning_rate": 2e-05,
-      "loss": 6.5305,
+      "epoch": 80.82,
+      "learning_rate": 3.589607894631111e-05,
+      "loss": 2.3717,
       "step": 3880
     },
     {
-      "epoch": 20.05,
-      "learning_rate": 2e-05,
-      "loss": 7.0605,
+      "epoch": 81.04,
+      "learning_rate": 3.514355068077655e-05,
+      "loss": 2.5195,
       "step": 3890
     },
     {
-      "epoch": 20.1,
-      "learning_rate": 2e-05,
-      "loss": 6.5381,
+      "epoch": 81.25,
+      "learning_rate": 3.439823377039599e-05,
+      "loss": 2.3669,
       "step": 3900
     },
     {
-      "epoch": 20.1,
-      "eval_accuracy": 0.1281530363855393,
-      "eval_loss": 6.589879989624023,
-      "eval_runtime": 169.1405,
-      "eval_samples_per_second": 142.219,
-      "eval_steps_per_second": 2.223,
+      "epoch": 81.25,
+      "eval_accuracy": 0.5124800918682825,
+      "eval_loss": 2.752890110015869,
+      "eval_runtime": 145.9521,
+      "eval_samples_per_second": 164.814,
+      "eval_steps_per_second": 5.152,
       "step": 3900
     },
     {
-      "epoch": 20.15,
-      "learning_rate": 2e-05,
-      "loss": 6.5167,
+      "epoch": 81.45,
+      "learning_rate": 3.36601608177457e-05,
+      "loss": 2.3595,
       "step": 3910
     },
     {
-      "epoch": 20.21,
-      "learning_rate": 2e-05,
-      "loss": 6.5289,
+      "epoch": 81.66,
+      "learning_rate": 3.292936410852754e-05,
+      "loss": 2.3727,
       "step": 3920
     },
     {
-      "epoch": 20.26,
-      "learning_rate": 2e-05,
-      "loss": 6.5296,
+      "epoch": 81.86,
+      "learning_rate": 3.220587561015709e-05,
+      "loss": 2.3707,
       "step": 3930
     },
     {
-      "epoch": 20.31,
-      "learning_rate": 2e-05,
-      "loss": 6.4745,
+      "epoch": 82.08,
+      "learning_rate": 3.148972697036507e-05,
+      "loss": 2.508,
       "step": 3940
     },
     {
-      "epoch": 20.36,
-      "learning_rate": 2e-05,
-      "loss": 6.5517,
+      "epoch": 82.29,
+      "learning_rate": 3.078094951581289e-05,
+      "loss": 2.3631,
       "step": 3950
     },
     {
-      "epoch": 20.36,
-      "eval_accuracy": 0.12833386885114662,
-      "eval_loss": 6.58883810043335,
-      "eval_runtime": 168.9823,
-      "eval_samples_per_second": 142.352,
-      "eval_steps_per_second": 2.225,
+      "epoch": 82.29,
+      "eval_accuracy": 0.5132219293707184,
+      "eval_loss": 2.749772071838379,
+      "eval_runtime": 146.0679,
+      "eval_samples_per_second": 164.684,
+      "eval_steps_per_second": 5.148,
       "step": 3950
     },
     {
-      "epoch": 20.41,
-      "learning_rate": 2e-05,
-      "loss": 6.5466,
+      "epoch": 82.49,
+      "learning_rate": 3.007957425072265e-05,
+      "loss": 2.3568,
       "step": 3960
     },
     {
-      "epoch": 20.46,
-      "learning_rate": 2e-05,
-      "loss": 6.5482,
+      "epoch": 82.7,
+      "learning_rate": 2.9385631855520546e-05,
+      "loss": 2.3679,
       "step": 3970
     },
     {
-      "epoch": 20.51,
-      "learning_rate": 2e-05,
-      "loss": 6.5047,
+      "epoch": 82.9,
+      "learning_rate": 2.8699152685494925e-05,
+      "loss": 2.3504,
       "step": 3980
     },
     {
-      "epoch": 20.56,
-      "learning_rate": 2e-05,
-      "loss": 6.499,
+      "epoch": 83.12,
+      "learning_rate": 2.8020166769468616e-05,
+      "loss": 2.5054,
       "step": 3990
     },
     {
-      "epoch": 20.62,
-      "learning_rate": 2e-05,
-      "loss": 6.5371,
+      "epoch": 83.33,
+      "learning_rate": 2.7348703808485223e-05,
+      "loss": 2.3499,
       "step": 4000
     },
     {
-      "epoch": 20.62,
-      "eval_accuracy": 0.1295166100847349,
-      "eval_loss": 6.585357189178467,
-      "eval_runtime": 169.1057,
-      "eval_samples_per_second": 142.248,
-      "eval_steps_per_second": 2.223,
+      "epoch": 83.33,
+      "eval_accuracy": 0.5135577468816207,
+      "eval_loss": 2.7453861236572266,
+      "eval_runtime": 146.0782,
+      "eval_samples_per_second": 164.672,
+      "eval_steps_per_second": 5.148,
       "step": 4000
     },
     {
-      "epoch": 20.67,
-      "learning_rate": 2e-05,
-      "loss": 6.5112,
+      "epoch": 83.53,
+      "learning_rate": 2.6684793174509915e-05,
+      "loss": 2.3478,
       "step": 4010
     },
     {
-      "epoch": 20.72,
-      "learning_rate": 2e-05,
-      "loss": 6.538,
+      "epoch": 83.74,
+      "learning_rate": 2.6028463909144574e-05,
+      "loss": 2.3686,
       "step": 4020
     },
     {
-      "epoch": 20.77,
-      "learning_rate": 2e-05,
-      "loss": 6.5547,
+      "epoch": 83.95,
+      "learning_rate": 2.5379744722357403e-05,
+      "loss": 2.3636,
       "step": 4030
     },
     {
-      "epoch": 20.82,
-      "learning_rate": 2e-05,
-      "loss": 6.5062,
+      "epoch": 84.16,
+      "learning_rate": 2.473866399122733e-05,
+      "loss": 2.5195,
       "step": 4040
     },
     {
-      "epoch": 20.87,
-      "learning_rate": 2e-05,
-      "loss": 6.5819,
+      "epoch": 84.37,
+      "learning_rate": 2.410524975870221e-05,
+      "loss": 2.3726,
       "step": 4050
     },
     {
-      "epoch": 20.87,
-      "eval_accuracy": 0.12822338785929896,
-      "eval_loss": 6.582521915435791,
-      "eval_runtime": 169.044,
-      "eval_samples_per_second": 142.3,
-      "eval_steps_per_second": 2.224,
+      "epoch": 84.37,
+      "eval_accuracy": 0.5140964497348997,
+      "eval_loss": 2.7446117401123047,
+      "eval_runtime": 146.09,
+      "eval_samples_per_second": 164.659,
+      "eval_steps_per_second": 5.148,
       "step": 4050
     },
     {
-      "epoch": 20.92,
-      "learning_rate": 2e-05,
-      "loss": 6.4947,
+      "epoch": 84.58,
+      "learning_rate": 2.347952973237262e-05,
+      "loss": 2.3504,
       "step": 4060
     },
     {
-      "epoch": 20.98,
-      "learning_rate": 2e-05,
-      "loss": 6.5279,
+      "epoch": 84.78,
+      "learning_rate": 2.286153128325954e-05,
+      "loss": 2.351,
       "step": 4070
     },
     {
-      "epoch": 21.03,
-      "learning_rate": 2e-05,
-      "loss": 7.0033,
+      "epoch": 84.99,
+      "learning_rate": 2.2251281444617257e-05,
+      "loss": 2.3506,
       "step": 4080
     },
     {
-      "epoch": 21.08,
-      "learning_rate": 2e-05,
-      "loss": 6.5198,
+      "epoch": 85.21,
+      "learning_rate": 2.1648806910750575e-05,
+      "loss": 2.5104,
       "step": 4090
     },
     {
-      "epoch": 21.13,
-      "learning_rate": 2e-05,
-      "loss": 6.5425,
+      "epoch": 85.41,
+      "learning_rate": 2.1054134035847307e-05,
+      "loss": 2.3411,
       "step": 4100
     },
     {
-      "epoch": 21.13,
-      "eval_accuracy": 0.12887030827275436,
-      "eval_loss": 6.579444408416748,
-      "eval_runtime": 169.204,
-      "eval_samples_per_second": 142.166,
-      "eval_steps_per_second": 2.222,
+      "epoch": 85.41,
+      "eval_accuracy": 0.5143741932133077,
+      "eval_loss": 2.740255355834961,
+      "eval_runtime": 146.0438,
+      "eval_samples_per_second": 164.711,
+      "eval_steps_per_second": 5.149,
       "step": 4100
     },
     {
-      "epoch": 21.18,
-      "learning_rate": 2e-05,
-      "loss": 6.4964,
+      "epoch": 85.62,
+      "learning_rate": 2.0467288832825583e-05,
+      "loss": 2.3666,
       "step": 4110
     },
     {
-      "epoch": 21.24,
-      "learning_rate": 2e-05,
-      "loss": 6.5506,
+      "epoch": 85.82,
+      "learning_rate": 1.9888296972195587e-05,
+      "loss": 2.3451,
       "step": 4120
     },
     {
-      "epoch": 21.29,
-      "learning_rate": 2e-05,
-      "loss": 6.5544,
+      "epoch": 86.04,
+      "learning_rate": 1.931718378093703e-05,
+      "loss": 2.5151,
       "step": 4130
     },
     {
-      "epoch": 21.34,
-      "learning_rate": 2e-05,
-      "loss": 6.5089,
+      "epoch": 86.25,
+      "learning_rate": 1.875397424139109e-05,
+      "loss": 2.3539,
       "step": 4140
     },
     {
-      "epoch": 21.39,
-      "learning_rate": 2e-05,
-      "loss": 6.5372,
+      "epoch": 86.45,
+      "learning_rate": 1.81986929901675e-05,
+      "loss": 2.3321,
       "step": 4150
     },
     {
-      "epoch": 21.39,
-      "eval_accuracy": 0.12997385168937536,
-      "eval_loss": 6.575997829437256,
-      "eval_runtime": 169.1292,
-      "eval_samples_per_second": 142.229,
-      "eval_steps_per_second": 2.223,
+      "epoch": 86.45,
+      "eval_accuracy": 0.5146461086764289,
+      "eval_loss": 2.7371606826782227,
+      "eval_runtime": 146.1164,
+      "eval_samples_per_second": 164.629,
+      "eval_steps_per_second": 5.147,
       "step": 4150
     },
     {
-      "epoch": 21.44,
-      "learning_rate": 2e-05,
-      "loss": 6.5305,
+      "epoch": 86.66,
+      "learning_rate": 1.765136431706711e-05,
+      "loss": 2.3573,
       "step": 4160
     },
     {
-      "epoch": 21.49,
-      "learning_rate": 2e-05,
-      "loss": 6.5443,
+      "epoch": 86.86,
+      "learning_rate": 1.711201216401912e-05,
+      "loss": 2.3422,
       "step": 4170
     },
     {
-      "epoch": 21.54,
-      "learning_rate": 2e-05,
-      "loss": 6.5214,
+      "epoch": 87.08,
+      "learning_rate": 1.6580660124034032e-05,
+      "loss": 2.5055,
       "step": 4180
     },
     {
-      "epoch": 21.6,
-      "learning_rate": 2e-05,
-      "loss": 6.4939,
+      "epoch": 87.29,
+      "learning_rate": 1.605733144017132e-05,
+      "loss": 2.3429,
       "step": 4190
     },
     {
-      "epoch": 21.65,
-      "learning_rate": 2e-05,
-      "loss": 6.544,
+      "epoch": 87.49,
+      "learning_rate": 1.5542049004523053e-05,
+      "loss": 2.3456,
       "step": 4200
     },
     {
-      "epoch": 21.65,
-      "eval_accuracy": 0.13029984560795255,
-      "eval_loss": 6.571804046630859,
-      "eval_runtime": 168.9481,
-      "eval_samples_per_second": 142.381,
-      "eval_steps_per_second": 2.226,
+      "epoch": 87.49,
+      "eval_accuracy": 0.5146212850149416,
+      "eval_loss": 2.7389299869537354,
+      "eval_runtime": 146.0012,
+      "eval_samples_per_second": 164.759,
+      "eval_steps_per_second": 5.151,
       "step": 4200
     },
     {
-      "epoch": 21.7,
-      "learning_rate": 2e-05,
-      "loss": 6.5006,
+      "epoch": 87.7,
+      "learning_rate": 1.503483535721224e-05,
+      "loss": 2.3608,
       "step": 4210
     },
     {
-      "epoch": 21.75,
-      "learning_rate": 2e-05,
-      "loss": 6.5257,
+      "epoch": 87.9,
+      "learning_rate": 1.4535712685406921e-05,
+      "loss": 2.3466,
       "step": 4220
     },
     {
-      "epoch": 21.8,
-      "learning_rate": 2e-05,
-      "loss": 6.5282,
+      "epoch": 88.12,
+      "learning_rate": 1.4044702822349731e-05,
+      "loss": 2.4892,
       "step": 4230
     },
     {
-      "epoch": 21.85,
-      "learning_rate": 2e-05,
-      "loss": 6.5319,
+      "epoch": 88.33,
+      "learning_rate": 1.3561827246402692e-05,
+      "loss": 2.3418,
       "step": 4240
     },
     {
-      "epoch": 21.9,
-      "learning_rate": 2e-05,
-      "loss": 6.5129,
+      "epoch": 88.53,
+      "learning_rate": 1.3087107080107853e-05,
+      "loss": 2.3372,
       "step": 4250
     },
     {
-      "epoch": 21.9,
-      "eval_accuracy": 0.13096158217776768,
-      "eval_loss": 6.565962314605713,
-      "eval_runtime": 168.9477,
-      "eval_samples_per_second": 142.381,
-      "eval_steps_per_second": 2.226,
+      "epoch": 88.53,
+      "eval_accuracy": 0.515111201963272,
+      "eval_loss": 2.7384002208709717,
+      "eval_runtime": 145.8226,
+      "eval_samples_per_second": 164.961,
+      "eval_steps_per_second": 5.157,
       "step": 4250
     },
     {
-      "epoch": 21.96,
-      "learning_rate": 2e-05,
-      "loss": 6.4916,
+      "epoch": 88.74,
+      "learning_rate": 1.2620563089263093e-05,
+      "loss": 2.3411,
       "step": 4260
     },
     {
-      "epoch": 22.01,
-      "learning_rate": 2e-05,
-      "loss": 6.9742,
+      "epoch": 88.95,
+      "learning_rate": 1.2162215682014012e-05,
+      "loss": 2.3637,
       "step": 4270
     },
     {
-      "epoch": 22.06,
-      "learning_rate": 2e-05,
-      "loss": 6.4991,
+      "epoch": 89.16,
+      "learning_rate": 1.1712084907961053e-05,
+      "loss": 2.4971,
       "step": 4280
     },
     {
-      "epoch": 22.11,
-      "learning_rate": 2e-05,
-      "loss": 6.516,
+      "epoch": 89.37,
+      "learning_rate": 1.127019045728246e-05,
+      "loss": 2.3476,
       "step": 4290
     },
     {
-      "epoch": 22.16,
-      "learning_rate": 2e-05,
-      "loss": 6.4798,
+      "epoch": 89.58,
+      "learning_rate": 1.0836551659873074e-05,
+      "loss": 2.343,
       "step": 4300
     },
     {
-      "epoch": 22.16,
-      "eval_accuracy": 0.1304970998721391,
-      "eval_loss": 6.56817102432251,
-      "eval_runtime": 168.9692,
-      "eval_samples_per_second": 142.363,
-      "eval_steps_per_second": 2.225,
+      "epoch": 89.58,
+      "eval_accuracy": 0.5144067649722459,
+      "eval_loss": 2.7397918701171875,
+      "eval_runtime": 146.0005,
+      "eval_samples_per_second": 164.76,
+      "eval_steps_per_second": 5.151,
       "step": 4300
     },
     {
-      "epoch": 22.22,
-      "learning_rate": 2e-05,
-      "loss": 6.5255,
-      "step": 4310
-    },
-    {
-      "epoch": 22.27,
-      "learning_rate": 2e-05,
-      "loss": 6.4953,
-      "step": 4320
-    },
-    {
-      "epoch": 22.32,
-      "learning_rate": 2e-05,
-      "loss": 6.5558,
-      "step": 4330
-    },
-    {
-      "epoch": 22.37,
-      "learning_rate": 2e-05,
-      "loss": 6.5029,
-      "step": 4340
-    },
-    {
-      "epoch": 22.42,
-      "learning_rate": 2e-05,
-      "loss": 6.5556,
-      "step": 4350
-    },
-    {
-      "epoch": 22.42,
-      "eval_accuracy": 0.13152063993455215,
-      "eval_loss": 6.561850547790527,
-      "eval_runtime": 168.9174,
-      "eval_samples_per_second": 142.407,
-      "eval_steps_per_second": 2.226,
-      "step": 4350
-    },
-    {
-      "epoch": 22.47,
-      "learning_rate": 2e-05,
-      "loss": 6.4888,
-      "step": 4360
-    },
-    {
-      "epoch": 22.52,
-      "learning_rate": 2e-05,
-      "loss": 6.4516,
-      "step": 4370
-    },
-    {
-      "epoch": 22.58,
-      "learning_rate": 2e-05,
-      "loss": 6.5203,
-      "step": 4380
-    },
-    {
-      "epoch": 22.63,
-      "learning_rate": 2e-05,
-      "loss": 6.502,
-      "step": 4390
-    },
-    {
-      "epoch": 22.68,
-      "learning_rate": 2e-05,
-      "loss": 6.4946,
-      "step": 4400
-    },
-    {
-      "epoch": 22.68,
-      "eval_accuracy": 0.13143446792084687,
-      "eval_loss": 6.558900833129883,
-      "eval_runtime": 168.9992,
-      "eval_samples_per_second": 142.338,
-      "eval_steps_per_second": 2.225,
-      "step": 4400
-    },
-    {
-      "epoch": 22.73,
-      "learning_rate": 2e-05,
-      "loss": 6.4919,
-      "step": 4410
-    },
-    {
-      "epoch": 22.78,
-      "learning_rate": 2e-05,
-      "loss": 6.482,
-      "step": 4420
-    },
-    {
-      "epoch": 22.83,
-      "learning_rate": 2e-05,
-      "loss": 6.5231,
-      "step": 4430
-    },
-    {
-      "epoch": 22.88,
-      "learning_rate": 2e-05,
-      "loss": 6.5293,
-      "step": 4440
-    },
-    {
-      "epoch": 22.93,
-      "learning_rate": 2e-05,
-      "loss": 6.5212,
-      "step": 4450
-    },
-    {
-      "epoch": 22.93,
-      "eval_accuracy": 0.13184165697753022,
-      "eval_loss": 6.559324264526367,
-      "eval_runtime": 168.9331,
-      "eval_samples_per_second": 142.394,
-      "eval_steps_per_second": 2.226,
-      "step": 4450
-    },
-    {
-      "epoch": 22.99,
-      "learning_rate": 2e-05,
-      "loss": 6.4944,
-      "step": 4460
-    },
-    {
-      "epoch": 23.04,
-      "learning_rate": 2e-05,
-      "loss": 6.9696,
-      "step": 4470
-    },
-    {
-      "epoch": 23.09,
-      "learning_rate": 2e-05,
-      "loss": 6.5051,
-      "step": 4480
-    },
-    {
-      "epoch": 23.14,
-      "learning_rate": 2e-05,
-      "loss": 6.4822,
-      "step": 4490
-    },
-    {
-      "epoch": 23.2,
-      "learning_rate": 2e-05,
-      "loss": 6.5055,
-      "step": 4500
-    },
-    {
-      "epoch": 23.2,
-      "eval_accuracy": 0.1310761376111694,
-      "eval_loss": 6.5552473068237305,
-      "eval_runtime": 169.0117,
-      "eval_samples_per_second": 142.327,
-      "eval_steps_per_second": 2.225,
-      "step": 4500
-    },
-    {
-      "epoch": 23.25,
-      "learning_rate": 2e-05,
-      "loss": 6.5152,
-      "step": 4510
-    },
-    {
-      "epoch": 23.3,
-      "learning_rate": 2e-05,
-      "loss": 6.504,
-      "step": 4520
-    },
-    {
-      "epoch": 23.35,
-      "learning_rate": 2e-05,
-      "loss": 6.5149,
-      "step": 4530
-    },
-    {
-      "epoch": 23.4,
-      "learning_rate": 2e-05,
-      "loss": 6.4913,
-      "step": 4540
-    },
-    {
-      "epoch": 23.45,
-      "learning_rate": 2e-05,
-      "loss": 6.4693,
-      "step": 4550
-    },
-    {
-      "epoch": 23.45,
-      "eval_accuracy": 0.13253956813479448,
-      "eval_loss": 6.548061847686768,
-      "eval_runtime": 169.0655,
-      "eval_samples_per_second": 142.282,
-      "eval_steps_per_second": 2.224,
-      "step": 4550
-    },
-    {
-      "epoch": 23.5,
-      "learning_rate": 2e-05,
-      "loss": 6.4741,
-      "step": 4560
-    },
-    {
-      "epoch": 23.55,
-      "learning_rate": 2e-05,
-      "loss": 6.5033,
-      "step": 4570
-    },
-    {
-      "epoch": 23.61,
-      "learning_rate": 2e-05,
-      "loss": 6.5313,
-      "step": 4580
-    },
-    {
-      "epoch": 23.66,
-      "learning_rate": 2e-05,
-      "loss": 6.4789,
-      "step": 4590
-    },
-    {
-      "epoch": 23.71,
-      "learning_rate": 2e-05,
-      "loss": 6.4706,
-      "step": 4600
-    },
-    {
-      "epoch": 23.71,
-      "eval_accuracy": 0.13172442724332517,
-      "eval_loss": 6.54693078994751,
-      "eval_runtime": 169.0562,
-      "eval_samples_per_second": 142.29,
-      "eval_steps_per_second": 2.224,
-      "step": 4600
-    },
-    {
-      "epoch": 23.76,
-      "learning_rate": 2e-05,
-      "loss": 6.5044,
-      "step": 4610
-    },
-    {
-      "epoch": 23.81,
-      "learning_rate": 2e-05,
-      "loss": 6.5038,
-      "step": 4620
-    },
-    {
-      "epoch": 23.86,
-      "learning_rate": 2e-05,
-      "loss": 6.5173,
-      "step": 4630
-    },
-    {
-      "epoch": 23.91,
-      "learning_rate": 2e-05,
-      "loss": 6.5048,
-      "step": 4640
-    },
-    {
-      "epoch": 23.97,
-      "learning_rate": 2e-05,
-      "loss": 6.495,
-      "step": 4650
-    },
-    {
-      "epoch": 23.97,
-      "eval_accuracy": 0.13237315726859447,
-      "eval_loss": 6.546177387237549,
-      "eval_runtime": 168.9825,
-      "eval_samples_per_second": 142.352,
-      "eval_steps_per_second": 2.225,
-      "step": 4650
-    },
-    {
-      "epoch": 24.02,
-      "learning_rate": 2e-05,
-      "loss": 6.9878,
-      "step": 4660
-    },
-    {
-      "epoch": 24.07,
-      "learning_rate": 2e-05,
-      "loss": 6.4801,
-      "step": 4670
-    },
-    {
-      "epoch": 24.12,
-      "learning_rate": 2e-05,
-      "loss": 6.5032,
-      "step": 4680
-    },
-    {
-      "epoch": 24.17,
-      "learning_rate": 2e-05,
-      "loss": 6.4802,
-      "step": 4690
-    },
-    {
-      "epoch": 24.23,
-      "learning_rate": 2e-05,
-      "loss": 6.4901,
-      "step": 4700
-    },
-    {
-      "epoch": 24.23,
-      "eval_accuracy": 0.13279457912397785,
-      "eval_loss": 6.541420936584473,
-      "eval_runtime": 168.8748,
-      "eval_samples_per_second": 142.443,
-      "eval_steps_per_second": 2.227,
-      "step": 4700
-    },
-    {
-      "epoch": 24.28,
-      "learning_rate": 2e-05,
-      "loss": 6.4895,
-      "step": 4710
-    },
-    {
-      "epoch": 24.33,
-      "learning_rate": 2e-05,
-      "loss": 6.5127,
-      "step": 4720
-    },
-    {
-      "epoch": 24.38,
-      "learning_rate": 2e-05,
-      "loss": 6.5062,
-      "step": 4730
-    },
-    {
-      "epoch": 24.43,
-      "learning_rate": 2e-05,
-      "loss": 6.4874,
-      "step": 4740
-    },
-    {
-      "epoch": 24.48,
-      "learning_rate": 2e-05,
-      "loss": 6.4936,
-      "step": 4750
-    },
-    {
-      "epoch": 24.48,
-      "eval_accuracy": 0.13342969669901336,
-      "eval_loss": 6.538527011871338,
-      "eval_runtime": 168.9618,
-      "eval_samples_per_second": 142.369,
-      "eval_steps_per_second": 2.225,
-      "step": 4750
-    },
-    {
-      "epoch": 24.53,
-      "learning_rate": 2e-05,
-      "loss": 6.4868,
-      "step": 4760
-    },
-    {
-      "epoch": 24.59,
-      "learning_rate": 2e-05,
-      "loss": 6.4657,
-      "step": 4770
-    },
-    {
-      "epoch": 24.64,
-      "learning_rate": 2e-05,
-      "loss": 6.4732,
-      "step": 4780
-    },
-    {
-      "epoch": 24.69,
-      "learning_rate": 2e-05,
-      "loss": 6.5089,
-      "step": 4790
-    },
-    {
-      "epoch": 24.74,
-      "learning_rate": 2e-05,
-      "loss": 6.481,
-      "step": 4800
-    },
-    {
-      "epoch": 24.74,
-      "eval_accuracy": 0.13314294552046066,
-      "eval_loss": 6.536244869232178,
-      "eval_runtime": 168.9165,
-      "eval_samples_per_second": 142.408,
-      "eval_steps_per_second": 2.226,
-      "step": 4800
-    },
-    {
-      "epoch": 24.79,
-      "learning_rate": 2e-05,
-      "loss": 6.4942,
-      "step": 4810
-    },
-    {
-      "epoch": 24.84,
-      "learning_rate": 2e-05,
-      "loss": 6.4739,
-      "step": 4820
-    },
-    {
-      "epoch": 24.89,
-      "learning_rate": 2e-05,
-      "loss": 6.4646,
-      "step": 4830
-    },
-    {
-      "epoch": 24.94,
-      "learning_rate": 2e-05,
-      "loss": 6.4478,
-      "step": 4840
-    },
-    {
-      "epoch": 25.0,
-      "learning_rate": 2e-05,
-      "loss": 6.5186,
-      "step": 4850
-    },
-    {
-      "epoch": 25.0,
-      "eval_accuracy": 0.13348454019073938,
-      "eval_loss": 6.535652160644531,
-      "eval_runtime": 168.9724,
-      "eval_samples_per_second": 142.361,
-      "eval_steps_per_second": 2.225,
-      "step": 4850
-    },
-    {
-      "epoch": 25.05,
-      "learning_rate": 2e-05,
-      "loss": 6.9644,
-      "step": 4860
-    },
-    {
-      "epoch": 25.1,
-      "learning_rate": 2e-05,
-      "loss": 6.4549,
-      "step": 4870
-    },
-    {
-      "epoch": 25.15,
-      "learning_rate": 2e-05,
-      "loss": 6.4725,
-      "step": 4880
-    },
-    {
-      "epoch": 25.21,
-      "learning_rate": 2e-05,
-      "loss": 6.467,
-      "step": 4890
-    },
-    {
-      "epoch": 25.26,
-      "learning_rate": 2e-05,
-      "loss": 6.4711,
-      "step": 4900
-    },
-    {
-      "epoch": 25.26,
-      "eval_accuracy": 0.1339031847051157,
-      "eval_loss": 6.53087854385376,
-      "eval_runtime": 168.9489,
-      "eval_samples_per_second": 142.38,
-      "eval_steps_per_second": 2.226,
-      "step": 4900
-    },
-    {
-      "epoch": 25.31,
-      "learning_rate": 2e-05,
-      "loss": 6.4475,
-      "step": 4910
-    },
-    {
-      "epoch": 25.36,
-      "learning_rate": 2e-05,
-      "loss": 6.4674,
-      "step": 4920
-    },
-    {
-      "epoch": 25.41,
-      "learning_rate": 2e-05,
-      "loss": 6.4737,
-      "step": 4930
-    },
-    {
-      "epoch": 25.46,
-      "learning_rate": 2e-05,
-      "loss": 6.5061,
-      "step": 4940
-    },
-    {
-      "epoch": 25.51,
-      "learning_rate": 2e-05,
-      "loss": 6.4513,
-      "step": 4950
-    },
-    {
-      "epoch": 25.51,
-      "eval_accuracy": 0.13372873418009035,
-      "eval_loss": 6.528397083282471,
-      "eval_runtime": 168.9948,
-      "eval_samples_per_second": 142.342,
-      "eval_steps_per_second": 2.225,
-      "step": 4950
-    },
-    {
-      "epoch": 25.56,
-      "learning_rate": 2e-05,
-      "loss": 6.4663,
-      "step": 4960
-    },
-    {
-      "epoch": 25.62,
-      "learning_rate": 2e-05,
-      "loss": 6.4488,
-      "step": 4970
-    },
-    {
-      "epoch": 25.67,
-      "learning_rate": 2e-05,
-      "loss": 6.4882,
-      "step": 4980
-    },
-    {
-      "epoch": 25.72,
-      "learning_rate": 2e-05,
-      "loss": 6.4644,
-      "step": 4990
-    },
-    {
-      "epoch": 25.77,
-      "learning_rate": 2e-05,
-      "loss": 6.4652,
-      "step": 5000
-    },
-    {
-      "epoch": 25.77,
-      "eval_accuracy": 0.1343316375042143,
-      "eval_loss": 6.524177074432373,
-      "eval_runtime": 168.9768,
-      "eval_samples_per_second": 142.357,
-      "eval_steps_per_second": 2.225,
-      "step": 5000
-    },
-    {
-      "epoch": 25.82,
-      "learning_rate": 2e-05,
-      "loss": 6.5203,
-      "step": 5010
-    },
-    {
-      "epoch": 25.87,
-      "learning_rate": 2e-05,
-      "loss": 6.4518,
-      "step": 5020
-    },
-    {
-      "epoch": 25.92,
-      "learning_rate": 2e-05,
-      "loss": 6.4791,
-      "step": 5030
-    },
-    {
-      "epoch": 25.98,
-      "learning_rate": 2e-05,
-      "loss": 6.4843,
-      "step": 5040
-    },
-    {
-      "epoch": 26.03,
-      "learning_rate": 2e-05,
-      "loss": 6.9335,
-      "step": 5050
-    },
-    {
-      "epoch": 26.03,
-      "eval_accuracy": 0.13449298626675552,
-      "eval_loss": 6.521671772003174,
-      "eval_runtime": 168.9858,
-      "eval_samples_per_second": 142.349,
-      "eval_steps_per_second": 2.225,
-      "step": 5050
-    },
-    {
-      "epoch": 26.08,
-      "learning_rate": 2e-05,
-      "loss": 6.4677,
-      "step": 5060
-    },
-    {
-      "epoch": 26.13,
-      "learning_rate": 2e-05,
-      "loss": 6.4347,
-      "step": 5070
-    },
-    {
-      "epoch": 26.18,
-      "learning_rate": 2e-05,
-      "loss": 6.446,
-      "step": 5080
-    },
-    {
-      "epoch": 26.24,
-      "learning_rate": 2e-05,
-      "loss": 6.4386,
-      "step": 5090
-    },
-    {
-      "epoch": 26.29,
-      "learning_rate": 2e-05,
-      "loss": 6.4747,
-      "step": 5100
-    },
-    {
-      "epoch": 26.29,
-      "eval_accuracy": 0.13449609380922367,
-      "eval_loss": 6.520633697509766,
-      "eval_runtime": 168.9534,
-      "eval_samples_per_second": 142.377,
-      "eval_steps_per_second": 2.225,
-      "step": 5100
-    },
-    {
-      "epoch": 26.34,
-      "learning_rate": 2e-05,
-      "loss": 6.4392,
-      "step": 5110
-    },
-    {
-      "epoch": 26.39,
-      "learning_rate": 2e-05,
-      "loss": 6.4715,
-      "step": 5120
-    },
-    {
-      "epoch": 26.44,
-      "learning_rate": 2e-05,
-      "loss": 6.4502,
-      "step": 5130
-    },
-    {
-      "epoch": 26.49,
-      "learning_rate": 2e-05,
-      "loss": 6.4566,
-      "step": 5140
-    },
-    {
-      "epoch": 26.54,
-      "learning_rate": 2e-05,
-      "loss": 6.4702,
-      "step": 5150
-    },
-    {
-      "epoch": 26.54,
-      "eval_accuracy": 0.13501667146291277,
-      "eval_loss": 6.52007532119751,
-      "eval_runtime": 168.9331,
-      "eval_samples_per_second": 142.394,
-      "eval_steps_per_second": 2.226,
-      "step": 5150
-    },
-    {
-      "epoch": 26.6,
-      "learning_rate": 2e-05,
-      "loss": 6.4778,
-      "step": 5160
-    },
-    {
-      "epoch": 26.65,
-      "learning_rate": 2e-05,
-      "loss": 6.4631,
-      "step": 5170
-    },
-    {
-      "epoch": 26.7,
-      "learning_rate": 2e-05,
-      "loss": 6.453,
-      "step": 5180
-    },
-    {
-      "epoch": 26.75,
-      "learning_rate": 2e-05,
-      "loss": 6.4477,
-      "step": 5190
-    },
-    {
-      "epoch": 26.8,
-      "learning_rate": 2e-05,
-      "loss": 6.4524,
-      "step": 5200
-    },
-    {
-      "epoch": 26.8,
-      "eval_accuracy": 0.13524723051481435,
-      "eval_loss": 6.515605926513672,
-      "eval_runtime": 168.9736,
-      "eval_samples_per_second": 142.359,
-      "eval_steps_per_second": 2.225,
-      "step": 5200
-    },
-    {
-      "epoch": 26.85,
-      "learning_rate": 2e-05,
-      "loss": 6.4479,
-      "step": 5210
-    },
-    {
-      "epoch": 26.9,
-      "learning_rate": 2e-05,
-      "loss": 6.4726,
-      "step": 5220
-    },
-    {
-      "epoch": 26.96,
-      "learning_rate": 2e-05,
-      "loss": 6.4736,
-      "step": 5230
-    },
-    {
-      "epoch": 27.01,
-      "learning_rate": 2e-05,
-      "loss": 6.9569,
-      "step": 5240
-    },
-    {
-      "epoch": 27.06,
-      "learning_rate": 2e-05,
-      "loss": 6.4225,
-      "step": 5250
-    },
-    {
-      "epoch": 27.06,
-      "eval_accuracy": 0.13492381162327038,
-      "eval_loss": 6.515013694763184,
-      "eval_runtime": 168.8677,
-      "eval_samples_per_second": 142.449,
-      "eval_steps_per_second": 2.227,
-      "step": 5250
-    },
-    {
-      "epoch": 27.11,
-      "learning_rate": 2e-05,
-      "loss": 6.4502,
-      "step": 5260
-    },
-    {
-      "epoch": 27.16,
-      "learning_rate": 2e-05,
-      "loss": 6.4587,
-      "step": 5270
-    },
-    {
-      "epoch": 27.22,
-      "learning_rate": 2e-05,
-      "loss": 6.4512,
-      "step": 5280
-    },
-    {
-      "epoch": 27.27,
-      "learning_rate": 2e-05,
-      "loss": 6.4495,
-      "step": 5290
-    },
-    {
-      "epoch": 27.32,
-      "learning_rate": 2e-05,
-      "loss": 6.4599,
-      "step": 5300
-    },
-    {
-      "epoch": 27.32,
-      "eval_accuracy": 0.13545740402985318,
-      "eval_loss": 6.511618137359619,
-      "eval_runtime": 168.8908,
-      "eval_samples_per_second": 142.429,
-      "eval_steps_per_second": 2.226,
-      "step": 5300
-    },
-    {
-      "epoch": 27.37,
-      "learning_rate": 2e-05,
-      "loss": 6.4499,
-      "step": 5310
-    },
-    {
-      "epoch": 27.42,
-      "learning_rate": 2e-05,
-      "loss": 6.4468,
-      "step": 5320
-    },
-    {
-      "epoch": 27.47,
-      "learning_rate": 2e-05,
-      "loss": 6.4448,
-      "step": 5330
-    },
-    {
-      "epoch": 27.52,
-      "learning_rate": 2e-05,
-      "loss": 6.4436,
-      "step": 5340
-    },
-    {
-      "epoch": 27.58,
-      "learning_rate": 2e-05,
-      "loss": 6.4591,
-      "step": 5350
-    },
-    {
-      "epoch": 27.58,
-      "eval_accuracy": 0.1358384989998707,
-      "eval_loss": 6.509791374206543,
-      "eval_runtime": 169.0437,
-      "eval_samples_per_second": 142.301,
-      "eval_steps_per_second": 2.224,
-      "step": 5350
-    },
-    {
-      "epoch": 27.63,
-      "learning_rate": 2e-05,
-      "loss": 6.4593,
-      "step": 5360
-    },
-    {
-      "epoch": 27.68,
-      "learning_rate": 2e-05,
-      "loss": 6.4223,
-      "step": 5370
-    },
-    {
-      "epoch": 27.73,
-      "learning_rate": 2e-05,
-      "loss": 6.4278,
-      "step": 5380
-    },
-    {
-      "epoch": 27.78,
-      "learning_rate": 2e-05,
-      "loss": 6.4607,
-      "step": 5390
-    },
-    {
-      "epoch": 27.83,
-      "learning_rate": 2e-05,
-      "loss": 6.4184,
-      "step": 5400
-    },
-    {
-      "epoch": 27.83,
-      "eval_accuracy": 0.1352895741192505,
-      "eval_loss": 6.509643077850342,
-      "eval_runtime": 169.086,
-      "eval_samples_per_second": 142.265,
-      "eval_steps_per_second": 2.224,
-      "step": 5400
-    },
-    {
-      "epoch": 27.88,
-      "learning_rate": 2e-05,
-      "loss": 6.455,
-      "step": 5410
-    },
-    {
-      "epoch": 27.93,
-      "learning_rate": 2e-05,
-      "loss": 6.4447,
-      "step": 5420
-    },
-    {
-      "epoch": 27.99,
-      "learning_rate": 2e-05,
-      "loss": 6.3932,
-      "step": 5430
-    },
-    {
-      "epoch": 28.04,
-      "learning_rate": 2e-05,
-      "loss": 6.953,
-      "step": 5440
-    },
-    {
-      "epoch": 28.09,
-      "learning_rate": 2e-05,
-      "loss": 6.43,
-      "step": 5450
-    },
-    {
-      "epoch": 28.09,
-      "eval_accuracy": 0.1360815394195743,
-      "eval_loss": 6.507402420043945,
-      "eval_runtime": 169.2594,
-      "eval_samples_per_second": 142.119,
-      "eval_steps_per_second": 2.221,
-      "step": 5450
-    },
-    {
-      "epoch": 28.14,
-      "learning_rate": 2e-05,
-      "loss": 6.4622,
-      "step": 5460
-    },
-    {
-      "epoch": 28.2,
-      "learning_rate": 2e-05,
-      "loss": 6.4898,
-      "step": 5470
-    },
-    {
-      "epoch": 28.25,
-      "learning_rate": 2e-05,
-      "loss": 6.4297,
-      "step": 5480
-    },
-    {
-      "epoch": 28.3,
-      "learning_rate": 2e-05,
-      "loss": 6.4151,
-      "step": 5490
-    },
-    {
-      "epoch": 28.35,
-      "learning_rate": 2e-05,
-      "loss": 6.4604,
-      "step": 5500
-    },
-    {
-      "epoch": 28.35,
-      "eval_accuracy": 0.136740573949904,
-      "eval_loss": 6.4999098777771,
-      "eval_runtime": 169.0085,
-      "eval_samples_per_second": 142.33,
-      "eval_steps_per_second": 2.225,
-      "step": 5500
-    },
-    {
-      "epoch": 28.4,
-      "learning_rate": 2e-05,
-      "loss": 6.4254,
-      "step": 5510
-    },
-    {
-      "epoch": 28.45,
-      "learning_rate": 2e-05,
-      "loss": 6.4188,
-      "step": 5520
-    },
-    {
-      "epoch": 28.5,
-      "learning_rate": 2e-05,
-      "loss": 6.4418,
-      "step": 5530
-    },
-    {
-      "epoch": 28.55,
-      "learning_rate": 2e-05,
-      "loss": 6.4201,
-      "step": 5540
-    },
-    {
-      "epoch": 28.61,
-      "learning_rate": 2e-05,
-      "loss": 6.4593,
-      "step": 5550
-    },
-    {
-      "epoch": 28.61,
-      "eval_accuracy": 0.13586421209634997,
-      "eval_loss": 6.499355792999268,
-      "eval_runtime": 169.0692,
-      "eval_samples_per_second": 142.279,
-      "eval_steps_per_second": 2.224,
-      "step": 5550
-    },
-    {
-      "epoch": 28.66,
-      "learning_rate": 2e-05,
-      "loss": 6.4249,
-      "step": 5560
-    },
-    {
-      "epoch": 28.71,
-      "learning_rate": 2e-05,
-      "loss": 6.4335,
-      "step": 5570
-    },
-    {
-      "epoch": 28.76,
-      "learning_rate": 2e-05,
-      "loss": 6.4411,
-      "step": 5580
-    },
-    {
-      "epoch": 28.81,
-      "learning_rate": 2e-05,
-      "loss": 6.4238,
-      "step": 5590
-    },
-    {
-      "epoch": 28.86,
-      "learning_rate": 2e-05,
-      "loss": 6.4648,
-      "step": 5600
-    },
-    {
-      "epoch": 28.86,
-      "eval_accuracy": 0.13563087896124043,
-      "eval_loss": 6.498055934906006,
-      "eval_runtime": 168.9957,
-      "eval_samples_per_second": 142.341,
-      "eval_steps_per_second": 2.225,
-      "step": 5600
-    },
-    {
-      "epoch": 28.91,
-      "learning_rate": 2e-05,
-      "loss": 6.4051,
-      "step": 5610
-    },
-    {
-      "epoch": 28.97,
-      "learning_rate": 2e-05,
-      "loss": 6.4433,
-      "step": 5620
-    },
-    {
-      "epoch": 29.02,
-      "learning_rate": 2e-05,
-      "loss": 6.9371,
-      "step": 5630
-    },
-    {
-      "epoch": 29.07,
-      "learning_rate": 2e-05,
-      "loss": 6.4164,
-      "step": 5640
-    },
-    {
-      "epoch": 29.12,
-      "learning_rate": 2e-05,
-      "loss": 6.4453,
-      "step": 5650
-    },
-    {
-      "epoch": 29.12,
-      "eval_accuracy": 0.1374106846803503,
-      "eval_loss": 6.494868755340576,
-      "eval_runtime": 169.1796,
-      "eval_samples_per_second": 142.186,
-      "eval_steps_per_second": 2.222,
-      "step": 5650
-    },
-    {
-      "epoch": 29.17,
-      "learning_rate": 2e-05,
-      "loss": 6.4349,
-      "step": 5660
-    },
-    {
-      "epoch": 29.23,
-      "learning_rate": 2e-05,
-      "loss": 6.468,
-      "step": 5670
-    },
-    {
-      "epoch": 29.28,
-      "learning_rate": 2e-05,
-      "loss": 6.4532,
-      "step": 5680
-    },
-    {
-      "epoch": 29.33,
-      "learning_rate": 2e-05,
-      "loss": 6.4349,
-      "step": 5690
-    },
-    {
-      "epoch": 29.38,
-      "learning_rate": 2e-05,
-      "loss": 6.4275,
-      "step": 5700
-    },
-    {
-      "epoch": 29.38,
-      "eval_accuracy": 0.13618291226339158,
-      "eval_loss": 6.49544095993042,
-      "eval_runtime": 169.054,
-      "eval_samples_per_second": 142.292,
-      "eval_steps_per_second": 2.224,
-      "step": 5700
-    },
-    {
-      "epoch": 29.43,
-      "learning_rate": 2e-05,
-      "loss": 6.4304,
-      "step": 5710
-    },
-    {
-      "epoch": 29.48,
-      "learning_rate": 2e-05,
-      "loss": 6.4465,
-      "step": 5720
-    },
-    {
-      "epoch": 29.53,
-      "learning_rate": 2e-05,
-      "loss": 6.4144,
-      "step": 5730
-    },
-    {
-      "epoch": 29.59,
-      "learning_rate": 2e-05,
-      "loss": 6.4441,
-      "step": 5740
-    },
-    {
-      "epoch": 29.64,
-      "learning_rate": 2e-05,
-      "loss": 6.4165,
-      "step": 5750
-    },
-    {
-      "epoch": 29.64,
-      "eval_accuracy": 0.1368880629366925,
-      "eval_loss": 6.493827819824219,
-      "eval_runtime": 169.0908,
-      "eval_samples_per_second": 142.261,
-      "eval_steps_per_second": 2.224,
-      "step": 5750
-    },
-    {
-      "epoch": 29.69,
-      "learning_rate": 2e-05,
-      "loss": 6.4493,
-      "step": 5760
-    },
-    {
-      "epoch": 29.74,
-      "learning_rate": 2e-05,
-      "loss": 6.4144,
-      "step": 5770
-    },
-    {
-      "epoch": 29.79,
-      "learning_rate": 2e-05,
-      "loss": 6.4323,
-      "step": 5780
-    },
-    {
-      "epoch": 29.84,
-      "learning_rate": 2e-05,
-      "loss": 6.4063,
-      "step": 5790
-    },
-    {
-      "epoch": 29.89,
-      "learning_rate": 2e-05,
-      "loss": 6.4211,
-      "step": 5800
-    },
-    {
-      "epoch": 29.89,
-      "eval_accuracy": 0.1375700871847227,
-      "eval_loss": 6.4911370277404785,
-      "eval_runtime": 169.0097,
-      "eval_samples_per_second": 142.329,
-      "eval_steps_per_second": 2.225,
-      "step": 5800
-    },
-    {
-      "epoch": 29.94,
-      "learning_rate": 2e-05,
-      "loss": 6.3907,
-      "step": 5810
-    },
-    {
-      "epoch": 30.0,
-      "learning_rate": 2e-05,
-      "loss": 6.4137,
-      "step": 5820
-    },
-    {
-      "epoch": 30.05,
-      "learning_rate": 2e-05,
-      "loss": 6.9177,
-      "step": 5830
-    },
-    {
-      "epoch": 30.1,
-      "learning_rate": 2e-05,
-      "loss": 6.4397,
-      "step": 5840
-    },
-    {
-      "epoch": 30.15,
-      "learning_rate": 2e-05,
-      "loss": 6.4188,
-      "step": 5850
-    },
-    {
-      "epoch": 30.15,
-      "eval_accuracy": 0.13739219384676168,
-      "eval_loss": 6.486028671264648,
-      "eval_runtime": 169.0403,
-      "eval_samples_per_second": 142.303,
-      "eval_steps_per_second": 2.224,
-      "step": 5850
-    },
-    {
-      "epoch": 30.21,
-      "learning_rate": 2e-05,
-      "loss": 6.395,
-      "step": 5860
-    },
-    {
-      "epoch": 30.26,
-      "learning_rate": 2e-05,
-      "loss": 6.4017,
-      "step": 5870
-    },
-    {
-      "epoch": 30.31,
-      "learning_rate": 2e-05,
-      "loss": 6.4429,
-      "step": 5880
-    },
-    {
-      "epoch": 30.36,
-      "learning_rate": 2e-05,
-      "loss": 6.3932,
-      "step": 5890
-    },
-    {
-      "epoch": 30.41,
-      "learning_rate": 2e-05,
-      "loss": 6.4337,
-      "step": 5900
-    },
-    {
-      "epoch": 30.41,
-      "eval_accuracy": 0.13796805623574424,
-      "eval_loss": 6.480716228485107,
-      "eval_runtime": 169.1216,
-      "eval_samples_per_second": 142.235,
-      "eval_steps_per_second": 2.223,
-      "step": 5900
-    },
-    {
-      "epoch": 30.46,
-      "learning_rate": 2e-05,
-      "loss": 6.4171,
-      "step": 5910
-    },
-    {
-      "epoch": 30.51,
-      "learning_rate": 2e-05,
-      "loss": 6.42,
-      "step": 5920
-    },
-    {
-      "epoch": 30.56,
-      "learning_rate": 2e-05,
-      "loss": 6.4512,
-      "step": 5930
-    },
-    {
-      "epoch": 30.62,
-      "learning_rate": 2e-05,
-      "loss": 6.4306,
-      "step": 5940
-    },
-    {
-      "epoch": 30.67,
-      "learning_rate": 2e-05,
-      "loss": 6.4228,
-      "step": 5950
-    },
-    {
-      "epoch": 30.67,
-      "eval_accuracy": 0.1374551952909284,
-      "eval_loss": 6.487619400024414,
-      "eval_runtime": 169.1032,
-      "eval_samples_per_second": 142.25,
-      "eval_steps_per_second": 2.223,
-      "step": 5950
-    },
-    {
-      "epoch": 30.72,
-      "learning_rate": 2e-05,
-      "loss": 6.4332,
-      "step": 5960
-    },
-    {
-      "epoch": 30.77,
-      "learning_rate": 2e-05,
-      "loss": 6.413,
-      "step": 5970
-    },
-    {
-      "epoch": 30.82,
-      "learning_rate": 2e-05,
-      "loss": 6.3904,
-      "step": 5980
-    },
-    {
-      "epoch": 30.87,
-      "learning_rate": 2e-05,
-      "loss": 6.4198,
-      "step": 5990
-    },
-    {
-      "epoch": 30.92,
-      "learning_rate": 2e-05,
-      "loss": 6.3841,
-      "step": 6000
-    },
-    {
-      "epoch": 30.92,
-      "eval_accuracy": 0.1375611450497736,
-      "eval_loss": 6.4811110496521,
-      "eval_runtime": 169.0873,
-      "eval_samples_per_second": 142.264,
-      "eval_steps_per_second": 2.224,
-      "step": 6000
-    },
-    {
-      "epoch": 30.98,
-      "learning_rate": 2e-05,
-      "loss": 6.4452,
-      "step": 6010
-    },
-    {
-      "epoch": 31.03,
-      "learning_rate": 2e-05,
-      "loss": 6.9049,
-      "step": 6020
-    },
-    {
-      "epoch": 31.08,
-      "learning_rate": 2e-05,
-      "loss": 6.4161,
-      "step": 6030
-    },
-    {
-      "epoch": 31.13,
-      "learning_rate": 2e-05,
-      "loss": 6.3858,
-      "step": 6040
-    },
-    {
-      "epoch": 31.18,
-      "learning_rate": 2e-05,
-      "loss": 6.4383,
-      "step": 6050
-    },
-    {
-      "epoch": 31.18,
-      "eval_accuracy": 0.13791837603204882,
-      "eval_loss": 6.483151912689209,
-      "eval_runtime": 169.0866,
-      "eval_samples_per_second": 142.264,
-      "eval_steps_per_second": 2.224,
-      "step": 6050
-    },
-    {
-      "epoch": 31.18,
-      "step": 6050,
-      "total_flos": 1.125873723276288e+16,
-      "train_loss": 6.89371246590102,
-      "train_runtime": 29548.4223,
-      "train_samples_per_second": 33.721,
-      "train_steps_per_second": 0.263
+      "epoch": 89.58,
+      "step": 4300,
+      "total_flos": 2.954083328682332e+17,
+      "train_loss": 3.7431876293448516,
+      "train_runtime": 42244.2763,
+      "train_samples_per_second": 58.967,
+      "train_steps_per_second": 0.114
     }
   ],
-  "max_steps": 7760,
-  "num_train_epochs": 40,
-  "total_flos": 1.125873723276288e+16,
+  "max_steps": 4800,
+  "num_train_epochs": 100,
+  "total_flos": 2.954083328682332e+17,
   "trial_name": null,
   "trial_params": null
 }