End of training

all_results.json

@@ -1,12 +1,12 @@
 {
     "epoch": 20.0,
-    "eval_accuracy": 0.9168873197999412,
-    "eval_loss": 0.2988849878311157,
-    "eval_runtime": 141.4021,
-    "eval_samples_per_second": 48.076,
-    "eval_steps_per_second": 0.757,
-    "train_loss": 0.8135456519344858,
-    "train_runtime": 27591.4491,
-    "train_samples_per_second": 37.036,
-    "train_steps_per_second": 1.158
+    "eval_accuracy": 0.9149749926448956,
+    "eval_loss": 0.30149412155151367,
+    "eval_runtime": 149.0236,
+    "eval_samples_per_second": 45.617,
+    "eval_steps_per_second": 0.718,
+    "train_loss": 0.8136612295581911,
+    "train_runtime": 28286.4741,
+    "train_samples_per_second": 36.126,
+    "train_steps_per_second": 1.129
 }

eval_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 20.0,
-    "eval_accuracy": 0.9168873197999412,
-    "eval_loss": 0.2988849878311157,
-    "eval_runtime": 141.4021,
-    "eval_samples_per_second": 48.076,
-    "eval_steps_per_second": 0.757
+    "eval_accuracy": 0.9149749926448956,
+    "eval_loss": 0.30149412155151367,
+    "eval_runtime": 149.0236,
+    "eval_samples_per_second": 45.617,
+    "eval_steps_per_second": 0.718
 }

train_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 20.0,
-    "train_loss": 0.8135456519344858,
-    "train_runtime": 27591.4491,
-    "train_samples_per_second": 37.036,
-    "train_steps_per_second": 1.158
+    "train_loss": 0.8136612295581911,
+    "train_runtime": 28286.4741,
+    "train_samples_per_second": 36.126,
+    "train_steps_per_second": 1.129
 }

trainer_state.json

@@ -1,6 +1,6 @@
 {
-  "best_metric": 0.9168873197999412,
-  "best_model_checkpoint": "trillsson3-ft-keyword-spotting-12/checkpoint-27149",
+  "best_metric": 0.9149749926448956,
+  "best_model_checkpoint": "trillsson3-ft-keyword-spotting-12/checkpoint-23955",
   "epoch": 20.0,
   "global_step": 31940,
   "is_hyper_param_search": false,
@@ -10,2105 +10,2105 @@
     {
       "epoch": 0.06,
       "learning_rate": 8.73512836568566e-06,
-      "loss": 8.0466,
+      "loss": 8.0793,
       "step": 100
     },
     {
       "epoch": 0.13,
       "learning_rate": 1.812773951158422e-05,
-      "loss": 7.1478,
+      "loss": 7.1754,
       "step": 200
     },
     {
       "epoch": 0.19,
       "learning_rate": 2.7520350657482777e-05,
-      "loss": 5.5962,
+      "loss": 5.6723,
       "step": 300
     },
     {
       "epoch": 0.25,
       "learning_rate": 3.691296180338134e-05,
-      "loss": 4.2177,
+      "loss": 4.1468,
       "step": 400
     },
     {
       "epoch": 0.31,
       "learning_rate": 4.6305572949279896e-05,
-      "loss": 3.3186,
+      "loss": 3.3576,
       "step": 500
     },
     {
       "epoch": 0.38,
       "learning_rate": 5.5698184095178454e-05,
-      "loss": 2.986,
+      "loss": 2.955,
       "step": 600
     },
     {
       "epoch": 0.44,
       "learning_rate": 6.509079524107701e-05,
-      "loss": 2.7112,
+      "loss": 2.7689,
       "step": 700
     },
     {
       "epoch": 0.5,
       "learning_rate": 7.448340638697557e-05,
-      "loss": 2.5722,
+      "loss": 2.5858,
       "step": 800
     },
     {
       "epoch": 0.56,
       "learning_rate": 8.387601753287413e-05,
-      "loss": 2.3806,
+      "loss": 2.3275,
       "step": 900
     },
     {
       "epoch": 0.63,
       "learning_rate": 9.326862867877268e-05,
-      "loss": 2.146,
+      "loss": 2.0771,
       "step": 1000
     },
     {
       "epoch": 0.69,
       "learning_rate": 0.00010266123982467126,
-      "loss": 1.8771,
+      "loss": 1.8958,
       "step": 1100
     },
     {
       "epoch": 0.75,
       "learning_rate": 0.0001120538509705698,
-      "loss": 1.6311,
+      "loss": 1.6778,
       "step": 1200
     },
     {
       "epoch": 0.81,
       "learning_rate": 0.00012144646211646837,
-      "loss": 1.5842,
+      "loss": 1.5346,
       "step": 1300
     },
     {
       "epoch": 0.88,
       "learning_rate": 0.00013083907326236693,
-      "loss": 1.4234,
+      "loss": 1.4238,
       "step": 1400
     },
     {
       "epoch": 0.94,
       "learning_rate": 0.0001402316844082655,
-      "loss": 1.2304,
+      "loss": 1.2824,
       "step": 1500
     },
     {
       "epoch": 1.0,
-      "eval_accuracy": 0.6844660194174758,
-      "eval_loss": 0.7905995845794678,
-      "eval_runtime": 146.1452,
-      "eval_samples_per_second": 46.515,
-      "eval_steps_per_second": 0.732,
+      "eval_accuracy": 0.6891732862606649,
+      "eval_loss": 0.7817752957344055,
+      "eval_runtime": 166.2325,
+      "eval_samples_per_second": 40.895,
+      "eval_steps_per_second": 0.644,
       "step": 1597
     },
     {
       "epoch": 1.0,
       "learning_rate": 0.00014962429555416404,
-      "loss": 1.2104,
+      "loss": 1.2121,
       "step": 1600
     },
     {
       "epoch": 1.06,
       "learning_rate": 0.0001590169067000626,
-      "loss": 1.1652,
+      "loss": 1.1391,
       "step": 1700
     },
     {
       "epoch": 1.13,
       "learning_rate": 0.00016840951784596116,
-      "loss": 1.1149,
+      "loss": 1.12,
       "step": 1800
     },
     {
       "epoch": 1.19,
       "learning_rate": 0.00017780212899185972,
-      "loss": 1.0575,
+      "loss": 1.053,
       "step": 1900
     },
     {
       "epoch": 1.25,
       "learning_rate": 0.0001871947401377583,
-      "loss": 1.0265,
+      "loss": 1.0245,
       "step": 2000
     },
     {
       "epoch": 1.31,
       "learning_rate": 0.00019658735128365683,
-      "loss": 0.9752,
+      "loss": 0.9887,
       "step": 2100
     },
     {
       "epoch": 1.38,
       "learning_rate": 0.0002059799624295554,
-      "loss": 0.9388,
+      "loss": 0.9671,
       "step": 2200
     },
     {
       "epoch": 1.44,
       "learning_rate": 0.00021537257357545395,
-      "loss": 0.9153,
+      "loss": 0.8912,
       "step": 2300
     },
     {
       "epoch": 1.5,
       "learning_rate": 0.00022476518472135253,
-      "loss": 0.9111,
+      "loss": 0.9015,
       "step": 2400
     },
     {
       "epoch": 1.57,
       "learning_rate": 0.0002341577958672511,
-      "loss": 0.9077,
+      "loss": 0.89,
       "step": 2500
     },
     {
       "epoch": 1.63,
       "learning_rate": 0.00024355040701314962,
-      "loss": 0.8233,
+      "loss": 0.8173,
       "step": 2600
     },
     {
       "epoch": 1.69,
       "learning_rate": 0.0002529430181590482,
-      "loss": 0.8334,
+      "loss": 0.8338,
       "step": 2700
     },
     {
       "epoch": 1.75,
       "learning_rate": 0.00026233562930494674,
-      "loss": 0.8003,
+      "loss": 0.8313,
       "step": 2800
     },
     {
       "epoch": 1.82,
       "learning_rate": 0.0002717282404508453,
-      "loss": 0.838,
+      "loss": 0.8395,
       "step": 2900
     },
     {
       "epoch": 1.88,
       "learning_rate": 0.00028112085159674385,
-      "loss": 0.7956,
+      "loss": 0.8036,
       "step": 3000
     },
     {
       "epoch": 1.94,
       "learning_rate": 0.00029051346274264244,
-      "loss": 0.8117,
+      "loss": 0.8003,
       "step": 3100
     },
     {
       "epoch": 2.0,
-      "eval_accuracy": 0.8673139158576052,
-      "eval_loss": 0.435994416475296,
-      "eval_runtime": 148.4175,
-      "eval_samples_per_second": 45.803,
-      "eval_steps_per_second": 0.721,
+      "eval_accuracy": 0.8734922035892909,
+      "eval_loss": 0.44425612688064575,
+      "eval_runtime": 158.3963,
+      "eval_samples_per_second": 42.918,
+      "eval_steps_per_second": 0.676,
       "step": 3194
     },
     {
       "epoch": 2.0,
       "learning_rate": 0.00029990607388854097,
-      "loss": 0.7923,
+      "loss": 0.7951,
       "step": 3200
     },
     {
       "epoch": 2.07,
       "learning_rate": 0.0002989668127739511,
-      "loss": 0.8091,
+      "loss": 0.8114,
       "step": 3300
     },
     {
       "epoch": 2.13,
       "learning_rate": 0.00029792318931329574,
-      "loss": 0.808,
+      "loss": 0.8343,
       "step": 3400
     },
     {
       "epoch": 2.19,
       "learning_rate": 0.00029687956585264035,
-      "loss": 0.768,
+      "loss": 0.7617,
       "step": 3500
     },
     {
       "epoch": 2.25,
       "learning_rate": 0.00029583594239198497,
-      "loss": 0.7865,
+      "loss": 0.7779,
       "step": 3600
     },
     {
       "epoch": 2.32,
       "learning_rate": 0.0002947923189313295,
-      "loss": 0.745,
+      "loss": 0.7512,
       "step": 3700
     },
     {
       "epoch": 2.38,
       "learning_rate": 0.00029374869547067414,
-      "loss": 0.7788,
+      "loss": 0.7962,
       "step": 3800
     },
     {
       "epoch": 2.44,
       "learning_rate": 0.00029270507201001875,
-      "loss": 0.73,
+      "loss": 0.7134,
       "step": 3900
     },
     {
       "epoch": 2.5,
       "learning_rate": 0.00029166144854936337,
-      "loss": 0.7761,
+      "loss": 0.786,
       "step": 4000
     },
     {
       "epoch": 2.57,
       "learning_rate": 0.000290617825088708,
-      "loss": 0.7531,
+      "loss": 0.7808,
       "step": 4100
     },
     {
       "epoch": 2.63,
       "learning_rate": 0.0002895742016280526,
-      "loss": 0.7449,
+      "loss": 0.7379,
       "step": 4200
     },
     {
       "epoch": 2.69,
       "learning_rate": 0.00028853057816739715,
-      "loss": 0.7532,
+      "loss": 0.7545,
       "step": 4300
     },
     {
       "epoch": 2.76,
       "learning_rate": 0.00028748695470674177,
-      "loss": 0.7488,
+      "loss": 0.7557,
       "step": 4400
     },
     {
       "epoch": 2.82,
       "learning_rate": 0.0002864433312460864,
-      "loss": 0.7766,
+      "loss": 0.7633,
       "step": 4500
     },
     {
       "epoch": 2.88,
       "learning_rate": 0.000285399707785431,
-      "loss": 0.7362,
+      "loss": 0.7414,
       "step": 4600
     },
     {
       "epoch": 2.94,
       "learning_rate": 0.0002843560843247756,
-      "loss": 0.7322,
+      "loss": 0.7232,
       "step": 4700
     },
     {
       "epoch": 3.0,
-      "eval_accuracy": 0.884377758164166,
-      "eval_loss": 0.3766401708126068,
-      "eval_runtime": 145.2147,
-      "eval_samples_per_second": 46.813,
-      "eval_steps_per_second": 0.737,
+      "eval_accuracy": 0.8833480435422183,
+      "eval_loss": 0.372787207365036,
+      "eval_runtime": 149.856,
+      "eval_samples_per_second": 45.364,
+      "eval_steps_per_second": 0.714,
       "step": 4791
     },
     {
       "epoch": 3.01,
       "learning_rate": 0.00028331246086412017,
-      "loss": 0.7419,
+      "loss": 0.7661,
       "step": 4800
     },
     {
       "epoch": 3.07,
       "learning_rate": 0.00028226883740346484,
-      "loss": 0.75,
+      "loss": 0.7224,
       "step": 4900
     },
     {
       "epoch": 3.13,
       "learning_rate": 0.0002812252139428094,
-      "loss": 0.7245,
+      "loss": 0.7186,
       "step": 5000
     },
     {
       "epoch": 3.19,
-      "learning_rate": 0.000280181590482154,
-      "loss": 0.7237,
+      "learning_rate": 0.0002801920267167606,
+      "loss": 0.7151,
       "step": 5100
     },
     {
       "epoch": 3.26,
-      "learning_rate": 0.0002791379670214986,
-      "loss": 0.6822,
+      "learning_rate": 0.00027914840325610516,
+      "loss": 0.7175,
       "step": 5200
     },
     {
       "epoch": 3.32,
-      "learning_rate": 0.00027809434356084324,
-      "loss": 0.6956,
+      "learning_rate": 0.0002781047797954498,
+      "loss": 0.6871,
       "step": 5300
     },
     {
       "epoch": 3.38,
-      "learning_rate": 0.0002770507201001878,
-      "loss": 0.7233,
+      "learning_rate": 0.0002770611563347944,
+      "loss": 0.7446,
       "step": 5400
     },
     {
       "epoch": 3.44,
-      "learning_rate": 0.00027600709663953247,
-      "loss": 0.6914,
+      "learning_rate": 0.000276017532874139,
+      "loss": 0.7128,
       "step": 5500
     },
     {
       "epoch": 3.51,
-      "learning_rate": 0.000274963473178877,
-      "loss": 0.6867,
+      "learning_rate": 0.0002749739094134836,
+      "loss": 0.6896,
       "step": 5600
     },
     {
       "epoch": 3.57,
-      "learning_rate": 0.00027391984971822164,
-      "loss": 0.7331,
+      "learning_rate": 0.0002739302859528282,
+      "loss": 0.714,
       "step": 5700
     },
     {
       "epoch": 3.63,
-      "learning_rate": 0.00027287622625756625,
-      "loss": 0.742,
+      "learning_rate": 0.0002728866624921728,
+      "loss": 0.7478,
       "step": 5800
     },
     {
       "epoch": 3.69,
-      "learning_rate": 0.00027183260279691087,
-      "loss": 0.7081,
+      "learning_rate": 0.0002718430390315174,
+      "loss": 0.6853,
       "step": 5900
     },
     {
       "epoch": 3.76,
-      "learning_rate": 0.0002707889793362554,
-      "loss": 0.7142,
+      "learning_rate": 0.000270799415570862,
+      "loss": 0.7165,
       "step": 6000
     },
     {
       "epoch": 3.82,
       "learning_rate": 0.0002697557921102066,
-      "loss": 0.7088,
+      "loss": 0.704,
       "step": 6100
     },
     {
       "epoch": 3.88,
       "learning_rate": 0.00026871216864955125,
-      "loss": 0.7029,
+      "loss": 0.6954,
       "step": 6200
     },
     {
       "epoch": 3.94,
       "learning_rate": 0.0002676685451888958,
-      "loss": 0.7458,
+      "loss": 0.73,
       "step": 6300
     },
     {
       "epoch": 4.0,
-      "eval_accuracy": 0.8986466607825832,
-      "eval_loss": 0.34493356943130493,
-      "eval_runtime": 145.403,
-      "eval_samples_per_second": 46.753,
-      "eval_steps_per_second": 0.736,
+      "eval_accuracy": 0.8973227419829362,
+      "eval_loss": 0.346542090177536,
+      "eval_runtime": 149.5904,
+      "eval_samples_per_second": 45.444,
+      "eval_steps_per_second": 0.715,
       "step": 6388
     },
     {
       "epoch": 4.01,
       "learning_rate": 0.0002666249217282404,
-      "loss": 0.7236,
+      "loss": 0.7149,
       "step": 6400
     },
     {
       "epoch": 4.07,
       "learning_rate": 0.00026558129826758503,
-      "loss": 0.7012,
+      "loss": 0.7107,
       "step": 6500
     },
     {
       "epoch": 4.13,
       "learning_rate": 0.00026453767480692965,
-      "loss": 0.7051,
+      "loss": 0.6744,
       "step": 6600
     },
     {
       "epoch": 4.2,
       "learning_rate": 0.0002634940513462742,
-      "loss": 0.7298,
+      "loss": 0.727,
       "step": 6700
     },
     {
       "epoch": 4.26,
       "learning_rate": 0.0002624504278856189,
-      "loss": 0.6991,
+      "loss": 0.7,
       "step": 6800
     },
     {
       "epoch": 4.32,
-      "learning_rate": 0.00026141724065957,
-      "loss": 0.7047,
+      "learning_rate": 0.00026140680442496343,
+      "loss": 0.7239,
       "step": 6900
     },
     {
       "epoch": 4.38,
-      "learning_rate": 0.0002603736171989146,
-      "loss": 0.6833,
+      "learning_rate": 0.00026036318096430805,
+      "loss": 0.6847,
       "step": 7000
     },
     {
       "epoch": 4.45,
-      "learning_rate": 0.0002593299937382592,
-      "loss": 0.7374,
+      "learning_rate": 0.00025931955750365266,
+      "loss": 0.7151,
       "step": 7100
     },
     {
       "epoch": 4.51,
-      "learning_rate": 0.0002582863702776038,
-      "loss": 0.7404,
+      "learning_rate": 0.0002582759340429973,
+      "loss": 0.7228,
       "step": 7200
     },
     {
       "epoch": 4.57,
       "learning_rate": 0.0002572427468169484,
-      "loss": 0.6973,
+      "loss": 0.6925,
       "step": 7300
     },
     {
       "epoch": 4.63,
       "learning_rate": 0.000256199123356293,
-      "loss": 0.7085,
+      "loss": 0.7064,
       "step": 7400
     },
     {
       "epoch": 4.7,
       "learning_rate": 0.00025515549989563765,
-      "loss": 0.7029,
+      "loss": 0.6861,
       "step": 7500
     },
     {
       "epoch": 4.76,
       "learning_rate": 0.0002541118764349822,
-      "loss": 0.7181,
+      "loss": 0.7057,
       "step": 7600
     },
     {
       "epoch": 4.82,
       "learning_rate": 0.0002530682529743269,
-      "loss": 0.7089,
+      "loss": 0.6811,
       "step": 7700
     },
     {
       "epoch": 4.88,
       "learning_rate": 0.00025202462951367144,
-      "loss": 0.693,
+      "loss": 0.6676,
       "step": 7800
     },
     {
       "epoch": 4.95,
       "learning_rate": 0.00025098100605301605,
-      "loss": 0.7312,
+      "loss": 0.7015,
       "step": 7900
     },
     {
       "epoch": 5.0,
-      "eval_accuracy": 0.913503971756399,
-      "eval_loss": 0.32345056533813477,
-      "eval_runtime": 144.554,
-      "eval_samples_per_second": 47.027,
-      "eval_steps_per_second": 0.74,
+      "eval_accuracy": 0.910856134157105,
+      "eval_loss": 0.3211327791213989,
+      "eval_runtime": 150.7528,
+      "eval_samples_per_second": 45.094,
+      "eval_steps_per_second": 0.71,
       "step": 7985
     },
     {
       "epoch": 5.01,
       "learning_rate": 0.00024993738259236067,
-      "loss": 0.7072,
+      "loss": 0.708,
       "step": 8000
     },
     {
       "epoch": 5.07,
       "learning_rate": 0.0002488937591317053,
-      "loss": 0.7333,
+      "loss": 0.7372,
       "step": 8100
     },
     {
       "epoch": 5.13,
       "learning_rate": 0.00024785013567104984,
-      "loss": 0.7247,
+      "loss": 0.7089,
       "step": 8200
     },
     {
       "epoch": 5.2,
       "learning_rate": 0.00024680651221039445,
-      "loss": 0.6973,
+      "loss": 0.6962,
       "step": 8300
     },
     {
       "epoch": 5.26,
       "learning_rate": 0.00024576288874973907,
-      "loss": 0.7303,
+      "loss": 0.7209,
       "step": 8400
     },
     {
       "epoch": 5.32,
       "learning_rate": 0.0002447192652890837,
-      "loss": 0.7084,
+      "loss": 0.7091,
       "step": 8500
     },
     {
       "epoch": 5.39,
       "learning_rate": 0.0002436756418284283,
-      "loss": 0.6678,
+      "loss": 0.6898,
       "step": 8600
     },
     {
       "epoch": 5.45,
       "learning_rate": 0.00024263201836777288,
-      "loss": 0.7386,
+      "loss": 0.726,
       "step": 8700
     },
     {
       "epoch": 5.51,
       "learning_rate": 0.00024158839490711747,
-      "loss": 0.665,
+      "loss": 0.682,
       "step": 8800
     },
     {
       "epoch": 5.57,
       "learning_rate": 0.0002405447714464621,
-      "loss": 0.7161,
+      "loss": 0.6855,
       "step": 8900
     },
     {
       "epoch": 5.64,
       "learning_rate": 0.0002395011479858067,
-      "loss": 0.6969,
+      "loss": 0.7108,
       "step": 9000
     },
     {
       "epoch": 5.7,
-      "learning_rate": 0.00023846796075975785,
-      "loss": 0.6958,
+      "learning_rate": 0.00023845752452515128,
+      "loss": 0.6945,
       "step": 9100
     },
     {
       "epoch": 5.76,
-      "learning_rate": 0.00023742433729910246,
-      "loss": 0.6686,
+      "learning_rate": 0.00023741390106449592,
+      "loss": 0.6621,
       "step": 9200
     },
     {
       "epoch": 5.82,
-      "learning_rate": 0.00023638071383844708,
-      "loss": 0.7263,
+      "learning_rate": 0.0002363702776038405,
+      "loss": 0.7176,
       "step": 9300
     },
     {
       "epoch": 5.89,
-      "learning_rate": 0.00023533709037779166,
-      "loss": 0.671,
+      "learning_rate": 0.00023532665414318512,
+      "loss": 0.6824,
       "step": 9400
     },
     {
       "epoch": 5.95,
-      "learning_rate": 0.00023429346691713628,
-      "loss": 0.7199,
+      "learning_rate": 0.00023428303068252974,
+      "loss": 0.6981,
       "step": 9500
     },
     {
       "epoch": 6.0,
-      "eval_accuracy": 0.9058546631362165,
-      "eval_loss": 0.3211215138435364,
-      "eval_runtime": 144.8271,
-      "eval_samples_per_second": 46.939,
-      "eval_steps_per_second": 0.739,
+      "eval_accuracy": 0.9080611944689615,
+      "eval_loss": 0.3200249671936035,
+      "eval_runtime": 151.3502,
+      "eval_samples_per_second": 44.916,
+      "eval_steps_per_second": 0.707,
       "step": 9582
     },
     {
       "epoch": 6.01,
-      "learning_rate": 0.0002332498434564809,
-      "loss": 0.697,
+      "learning_rate": 0.00023323940722187433,
+      "loss": 0.6934,
       "step": 9600
     },
     {
       "epoch": 6.07,
-      "learning_rate": 0.00023220621999582548,
-      "loss": 0.7058,
+      "learning_rate": 0.00023219578376121894,
+      "loss": 0.7101,
       "step": 9700
     },
     {
       "epoch": 6.14,
-      "learning_rate": 0.00023116259653517006,
-      "loss": 0.6958,
+      "learning_rate": 0.00023115216030056353,
+      "loss": 0.6922,
       "step": 9800
     },
     {
       "epoch": 6.2,
-      "learning_rate": 0.0002301189730745147,
-      "loss": 0.7128,
+      "learning_rate": 0.00023010853683990814,
+      "loss": 0.7027,
       "step": 9900
     },
     {
       "epoch": 6.26,
-      "learning_rate": 0.0002290753496138593,
-      "loss": 0.6806,
+      "learning_rate": 0.00022906491337925275,
+      "loss": 0.6931,
       "step": 10000
     },
     {
       "epoch": 6.32,
-      "learning_rate": 0.00022803172615320393,
-      "loss": 0.6512,
+      "learning_rate": 0.00022802128991859734,
+      "loss": 0.6782,
       "step": 10100
     },
     {
       "epoch": 6.39,
-      "learning_rate": 0.00022698810269254852,
-      "loss": 0.682,
+      "learning_rate": 0.00022697766645794195,
+      "loss": 0.6842,
       "step": 10200
     },
     {
       "epoch": 6.45,
-      "learning_rate": 0.0002259444792318931,
-      "loss": 0.739,
+      "learning_rate": 0.00022593404299728657,
+      "loss": 0.7295,
       "step": 10300
     },
     {
       "epoch": 6.51,
-      "learning_rate": 0.00022490085577123772,
-      "loss": 0.7002,
+      "learning_rate": 0.00022489041953663115,
+      "loss": 0.7145,
       "step": 10400
     },
     {
       "epoch": 6.57,
-      "learning_rate": 0.00022385723231058233,
-      "loss": 0.6605,
+      "learning_rate": 0.00022384679607597577,
+      "loss": 0.6637,
       "step": 10500
     },
     {
       "epoch": 6.64,
-      "learning_rate": 0.00022281360884992692,
-      "loss": 0.695,
+      "learning_rate": 0.00022280317261532038,
+      "loss": 0.6666,
       "step": 10600
     },
     {
       "epoch": 6.7,
-      "learning_rate": 0.00022176998538927153,
-      "loss": 0.7051,
+      "learning_rate": 0.00022175954915466497,
+      "loss": 0.7068,
       "step": 10700
     },
     {
       "epoch": 6.76,
-      "learning_rate": 0.00022072636192861615,
-      "loss": 0.6833,
+      "learning_rate": 0.00022071592569400955,
+      "loss": 0.6759,
       "step": 10800
     },
     {
       "epoch": 6.83,
-      "learning_rate": 0.00021968273846796073,
-      "loss": 0.6494,
+      "learning_rate": 0.0002196723022333542,
+      "loss": 0.6926,
       "step": 10900
     },
     {
       "epoch": 6.89,
-      "learning_rate": 0.00021863911500730535,
-      "loss": 0.6715,
+      "learning_rate": 0.00021862867877269878,
+      "loss": 0.6825,
       "step": 11000
     },
     {
       "epoch": 6.95,
-      "learning_rate": 0.00021759549154664993,
-      "loss": 0.72,
+      "learning_rate": 0.00021758505531204342,
+      "loss": 0.6807,
       "step": 11100
     },
     {
       "epoch": 7.0,
-      "eval_accuracy": 0.9020300088261254,
-      "eval_loss": 0.32632216811180115,
-      "eval_runtime": 144.3277,
-      "eval_samples_per_second": 47.101,
-      "eval_steps_per_second": 0.741,
+      "eval_accuracy": 0.9058546631362165,
+      "eval_loss": 0.3208906650543213,
+      "eval_runtime": 150.3488,
+      "eval_samples_per_second": 45.215,
+      "eval_steps_per_second": 0.712,
       "step": 11179
     },
     {
       "epoch": 7.01,
-      "learning_rate": 0.00021655186808599455,
-      "loss": 0.6672,
+      "learning_rate": 0.000216541431851388,
+      "loss": 0.677,
       "step": 11200
     },
     {
       "epoch": 7.08,
       "learning_rate": 0.00021550824462533916,
-      "loss": 0.7304,
+      "loss": 0.7379,
       "step": 11300
     },
     {
       "epoch": 7.14,
-      "learning_rate": 0.00021446462116468375,
-      "loss": 0.6994,
+      "learning_rate": 0.00021447505739929034,
+      "loss": 0.7156,
       "step": 11400
     },
     {
       "epoch": 7.2,
-      "learning_rate": 0.00021342099770402836,
-      "loss": 0.6855,
+      "learning_rate": 0.00021343143393863493,
+      "loss": 0.6406,
       "step": 11500
     },
     {
       "epoch": 7.26,
-      "learning_rate": 0.00021237737424337298,
-      "loss": 0.6865,
+      "learning_rate": 0.0002123878104779795,
+      "loss": 0.6888,
       "step": 11600
     },
     {
       "epoch": 7.33,
-      "learning_rate": 0.00021133375078271756,
-      "loss": 0.6766,
+      "learning_rate": 0.00021134418701732413,
+      "loss": 0.6968,
       "step": 11700
     },
     {
       "epoch": 7.39,
-      "learning_rate": 0.0002102901273220622,
-      "loss": 0.6909,
+      "learning_rate": 0.00021030056355666874,
+      "loss": 0.6989,
       "step": 11800
     },
     {
       "epoch": 7.45,
-      "learning_rate": 0.0002092465038614068,
-      "loss": 0.6816,
+      "learning_rate": 0.00020925694009601333,
+      "loss": 0.6751,
       "step": 11900
     },
     {
       "epoch": 7.51,
-      "learning_rate": 0.00020820288040075138,
-      "loss": 0.6818,
+      "learning_rate": 0.00020821331663535794,
+      "loss": 0.6879,
       "step": 12000
     },
     {
       "epoch": 7.58,
-      "learning_rate": 0.00020715925694009602,
-      "loss": 0.6503,
+      "learning_rate": 0.00020716969317470255,
+      "loss": 0.678,
       "step": 12100
     },
     {
       "epoch": 7.64,
-      "learning_rate": 0.0002061156334794406,
-      "loss": 0.6734,
+      "learning_rate": 0.00020612606971404714,
+      "loss": 0.6501,
       "step": 12200
     },
     {
       "epoch": 7.7,
-      "learning_rate": 0.0002050720100187852,
-      "loss": 0.6815,
+      "learning_rate": 0.00020508244625339176,
+      "loss": 0.6679,
       "step": 12300
     },
     {
       "epoch": 7.76,
-      "learning_rate": 0.00020402838655812983,
-      "loss": 0.719,
+      "learning_rate": 0.00020403882279273637,
+      "loss": 0.7116,
       "step": 12400
     },
     {
       "epoch": 7.83,
-      "learning_rate": 0.00020298476309747442,
-      "loss": 0.6887,
+      "learning_rate": 0.00020299519933208098,
+      "loss": 0.6899,
       "step": 12500
     },
     {
       "epoch": 7.89,
-      "learning_rate": 0.000201941139636819,
-      "loss": 0.6622,
+      "learning_rate": 0.00020195157587142557,
+      "loss": 0.6892,
       "step": 12600
     },
     {
       "epoch": 7.95,
-      "learning_rate": 0.00020089751617616362,
-      "loss": 0.6992,
+      "learning_rate": 0.00020090795241077016,
+      "loss": 0.6873,
       "step": 12700
     },
     {
       "epoch": 8.0,
-      "eval_accuracy": 0.9032068255369227,
-      "eval_loss": 0.3186171054840088,
-      "eval_runtime": 139.1912,
-      "eval_samples_per_second": 48.839,
-      "eval_steps_per_second": 0.769,
+      "eval_accuracy": 0.902177110914975,
+      "eval_loss": 0.3205910921096802,
+      "eval_runtime": 152.1731,
+      "eval_samples_per_second": 44.673,
+      "eval_steps_per_second": 0.703,
       "step": 12776
     },
     {
       "epoch": 8.02,
       "learning_rate": 0.0001998643289501148,
-      "loss": 0.6696,
+      "loss": 0.6847,
       "step": 12800
     },
     {
       "epoch": 8.08,
       "learning_rate": 0.00019882070548945938,
-      "loss": 0.6972,
+      "loss": 0.6833,
       "step": 12900
     },
     {
       "epoch": 8.14,
       "learning_rate": 0.00019777708202880397,
-      "loss": 0.6782,
+      "loss": 0.6737,
       "step": 13000
     },
     {
       "epoch": 8.2,
       "learning_rate": 0.0001967334585681486,
-      "loss": 0.6566,
+      "loss": 0.6568,
       "step": 13100
     },
     {
       "epoch": 8.27,
       "learning_rate": 0.0001956898351074932,
-      "loss": 0.6823,
+      "loss": 0.7059,
       "step": 13200
     },
     {
       "epoch": 8.33,
       "learning_rate": 0.00019464621164683778,
-      "loss": 0.6772,
+      "loss": 0.6831,
       "step": 13300
     },
     {
       "epoch": 8.39,
       "learning_rate": 0.00019360258818618243,
-      "loss": 0.7382,
+      "loss": 0.7174,
       "step": 13400
     },
     {
       "epoch": 8.45,
-      "learning_rate": 0.000192558964725527,
-      "loss": 0.6941,
+      "learning_rate": 0.00019256940096013358,
+      "loss": 0.6966,
       "step": 13500
     },
     {
       "epoch": 8.52,
-      "learning_rate": 0.0001915153412648716,
-      "loss": 0.7093,
+      "learning_rate": 0.00019152577749947816,
+      "loss": 0.699,
       "step": 13600
     },
     {
       "epoch": 8.58,
-      "learning_rate": 0.00019047171780421624,
-      "loss": 0.6985,
+      "learning_rate": 0.00019048215403882278,
+      "loss": 0.6774,
       "step": 13700
     },
     {
       "epoch": 8.64,
-      "learning_rate": 0.00018942809434356083,
-      "loss": 0.6738,
+      "learning_rate": 0.0001894385305781674,
+      "loss": 0.6848,
       "step": 13800
     },
     {
       "epoch": 8.7,
-      "learning_rate": 0.00018838447088290544,
-      "loss": 0.6886,
+      "learning_rate": 0.00018839490711751198,
+      "loss": 0.6901,
       "step": 13900
     },
     {
       "epoch": 8.77,
-      "learning_rate": 0.00018734084742225005,
-      "loss": 0.7027,
+      "learning_rate": 0.00018735128365685656,
+      "loss": 0.6994,
       "step": 14000
     },
     {
       "epoch": 8.83,
-      "learning_rate": 0.00018629722396159464,
-      "loss": 0.681,
+      "learning_rate": 0.0001863076601962012,
+      "loss": 0.6739,
       "step": 14100
     },
     {
       "epoch": 8.89,
-      "learning_rate": 0.00018525360050093925,
-      "loss": 0.6649,
+      "learning_rate": 0.0001852640367355458,
+      "loss": 0.6854,
       "step": 14200
     },
     {
       "epoch": 8.95,
-      "learning_rate": 0.00018420997704028384,
-      "loss": 0.6543,
+      "learning_rate": 0.00018422041327489038,
+      "loss": 0.6416,
       "step": 14300
     },
     {
       "epoch": 9.0,
-      "eval_accuracy": 0.911738746690203,
-      "eval_loss": 0.3115650713443756,
-      "eval_runtime": 142.7126,
-      "eval_samples_per_second": 47.634,
-      "eval_steps_per_second": 0.75,
+      "eval_accuracy": 0.9057075610473668,
+      "eval_loss": 0.31237688660621643,
+      "eval_runtime": 151.5107,
+      "eval_samples_per_second": 44.868,
+      "eval_steps_per_second": 0.706,
       "step": 14373
     },
     {
       "epoch": 9.02,
-      "learning_rate": 0.00018316635357962845,
-      "loss": 0.679,
+      "learning_rate": 0.00018317678981423502,
+      "loss": 0.6711,
       "step": 14400
     },
     {
       "epoch": 9.08,
-      "learning_rate": 0.00018212273011897307,
-      "loss": 0.6992,
+      "learning_rate": 0.0001821331663535796,
+      "loss": 0.6965,
       "step": 14500
     },
     {
       "epoch": 9.14,
-      "learning_rate": 0.00018107910665831765,
-      "loss": 0.6443,
+      "learning_rate": 0.00018108954289292422,
+      "loss": 0.6896,
       "step": 14600
     },
     {
       "epoch": 9.2,
-      "learning_rate": 0.00018003548319766227,
-      "loss": 0.6952,
+      "learning_rate": 0.00018004591943226883,
+      "loss": 0.6944,
       "step": 14700
     },
     {
       "epoch": 9.27,
       "learning_rate": 0.00017900229597161342,
-      "loss": 0.7096,
+      "loss": 0.6981,
       "step": 14800
     },
     {
       "epoch": 9.33,
       "learning_rate": 0.00017795867251095803,
-      "loss": 0.6808,
+      "loss": 0.6819,
       "step": 14900
     },
     {
       "epoch": 9.39,
       "learning_rate": 0.00017691504905030265,
-      "loss": 0.6456,
+      "loss": 0.6869,
       "step": 15000
     },
     {
       "epoch": 9.46,
       "learning_rate": 0.00017587142558964723,
-      "loss": 0.68,
+      "loss": 0.6974,
       "step": 15100
     },
     {
       "epoch": 9.52,
       "learning_rate": 0.00017482780212899185,
-      "loss": 0.6753,
+      "loss": 0.6621,
       "step": 15200
     },
     {
       "epoch": 9.58,
       "learning_rate": 0.00017378417866833646,
-      "loss": 0.7015,
+      "loss": 0.6732,
       "step": 15300
     },
     {
       "epoch": 9.64,
       "learning_rate": 0.00017274055520768105,
-      "loss": 0.6481,
+      "loss": 0.678,
       "step": 15400
     },
     {
       "epoch": 9.71,
       "learning_rate": 0.00017169693174702566,
-      "loss": 0.6632,
+      "loss": 0.6494,
       "step": 15500
     },
     {
       "epoch": 9.77,
       "learning_rate": 0.00017065330828637025,
-      "loss": 0.684,
+      "loss": 0.6909,
       "step": 15600
     },
     {
       "epoch": 9.83,
       "learning_rate": 0.00016960968482571486,
-      "loss": 0.6991,
+      "loss": 0.687,
       "step": 15700
     },
     {
       "epoch": 9.89,
       "learning_rate": 0.00016856606136505948,
-      "loss": 0.6676,
+      "loss": 0.6705,
       "step": 15800
     },
     {
       "epoch": 9.96,
       "learning_rate": 0.00016752243790440406,
-      "loss": 0.6346,
+      "loss": 0.6698,
       "step": 15900
     },
     {
       "epoch": 10.0,
-      "eval_accuracy": 0.9015887025595763,
-      "eval_loss": 0.317617267370224,
-      "eval_runtime": 141.8277,
-      "eval_samples_per_second": 47.931,
-      "eval_steps_per_second": 0.754,
+      "eval_accuracy": 0.8949691085613416,
+      "eval_loss": 0.3288457989692688,
+      "eval_runtime": 151.6827,
+      "eval_samples_per_second": 44.817,
+      "eval_steps_per_second": 0.705,
       "step": 15970
     },
     {
       "epoch": 10.02,
-      "learning_rate": 0.00016647881444374868,
-      "loss": 0.6998,
+      "learning_rate": 0.00016648925067835524,
+      "loss": 0.6701,
       "step": 16000
     },
     {
       "epoch": 10.08,
-      "learning_rate": 0.0001654351909830933,
-      "loss": 0.6607,
+      "learning_rate": 0.00016544562721769983,
+      "loss": 0.6771,
       "step": 16100
     },
     {
       "epoch": 10.14,
-      "learning_rate": 0.00016439156752243788,
-      "loss": 0.6642,
+      "learning_rate": 0.00016440200375704444,
+      "loss": 0.6877,
       "step": 16200
     },
     {
       "epoch": 10.21,
-      "learning_rate": 0.00016334794406178252,
-      "loss": 0.6505,
+      "learning_rate": 0.00016335838029638906,
+      "loss": 0.6495,
       "step": 16300
     },
     {
       "epoch": 10.27,
-      "learning_rate": 0.0001623043206011271,
-      "loss": 0.6772,
+      "learning_rate": 0.00016231475683573364,
+      "loss": 0.6925,
       "step": 16400
     },
     {
       "epoch": 10.33,
-      "learning_rate": 0.0001612606971404717,
-      "loss": 0.6641,
+      "learning_rate": 0.00016127113337507826,
+      "loss": 0.647,
       "step": 16500
     },
     {
       "epoch": 10.39,
-      "learning_rate": 0.00016021707367981633,
-      "loss": 0.6523,
+      "learning_rate": 0.00016022750991442287,
+      "loss": 0.658,
       "step": 16600
     },
     {
       "epoch": 10.46,
-      "learning_rate": 0.00015917345021916092,
-      "loss": 0.7135,
+      "learning_rate": 0.00015918388645376746,
+      "loss": 0.7033,
       "step": 16700
     },
     {
       "epoch": 10.52,
-      "learning_rate": 0.0001581298267585055,
-      "loss": 0.6796,
+      "learning_rate": 0.00015814026299311207,
+      "loss": 0.6675,
       "step": 16800
     },
     {
       "epoch": 10.58,
-      "learning_rate": 0.00015708620329785015,
-      "loss": 0.665,
+      "learning_rate": 0.00015709663953245666,
+      "loss": 0.6905,
       "step": 16900
     },
     {
       "epoch": 10.64,
-      "learning_rate": 0.00015604257983719473,
-      "loss": 0.6853,
+      "learning_rate": 0.0001560530160718013,
+      "loss": 0.6766,
       "step": 17000
     },
     {
       "epoch": 10.71,
-      "learning_rate": 0.00015499895637653932,
-      "loss": 0.6851,
+      "learning_rate": 0.00015500939261114588,
+      "loss": 0.684,
       "step": 17100
     },
     {
       "epoch": 10.77,
-      "learning_rate": 0.00015395533291588393,
-      "loss": 0.647,
+      "learning_rate": 0.00015396576915049047,
+      "loss": 0.6382,
       "step": 17200
     },
     {
       "epoch": 10.83,
-      "learning_rate": 0.00015291170945522855,
-      "loss": 0.6446,
+      "learning_rate": 0.0001529221456898351,
+      "loss": 0.6737,
       "step": 17300
     },
     {
       "epoch": 10.9,
-      "learning_rate": 0.00015186808599457313,
-      "loss": 0.6798,
+      "learning_rate": 0.0001518785222291797,
+      "loss": 0.691,
       "step": 17400
     },
     {
       "epoch": 10.96,
-      "learning_rate": 0.00015082446253391775,
-      "loss": 0.7189,
+      "learning_rate": 0.00015083489876852429,
+      "loss": 0.716,
       "step": 17500
     },
     {
       "epoch": 11.0,
-      "eval_accuracy": 0.9018829067372757,
-      "eval_loss": 0.3116227686405182,
-      "eval_runtime": 142.1447,
-      "eval_samples_per_second": 47.825,
-      "eval_steps_per_second": 0.753,
+      "eval_accuracy": 0.8998234774933804,
+      "eval_loss": 0.31469690799713135,
+      "eval_runtime": 150.1862,
+      "eval_samples_per_second": 45.264,
+      "eval_steps_per_second": 0.712,
       "step": 17567
     },
     {
       "epoch": 11.02,
-      "learning_rate": 0.00014978083907326236,
-      "loss": 0.7129,
+      "learning_rate": 0.0001497912753078689,
+      "loss": 0.7326,
       "step": 17600
     },
     {
       "epoch": 11.08,
-      "learning_rate": 0.00014873721561260695,
-      "loss": 0.6561,
+      "learning_rate": 0.0001487476518472135,
+      "loss": 0.6747,
       "step": 17700
     },
     {
       "epoch": 11.15,
       "learning_rate": 0.00014770402838655813,
-      "loss": 0.6796,
+      "loss": 0.7075,
       "step": 17800
     },
     {
       "epoch": 11.21,
       "learning_rate": 0.0001466604049259027,
-      "loss": 0.7068,
+      "loss": 0.69,
       "step": 17900
     },
     {
       "epoch": 11.27,
       "learning_rate": 0.00014561678146524733,
-      "loss": 0.6543,
+      "loss": 0.6793,
       "step": 18000
     },
     {
       "epoch": 11.33,
       "learning_rate": 0.00014457315800459194,
-      "loss": 0.6708,
+      "loss": 0.6782,
       "step": 18100
     },
     {
       "epoch": 11.4,
       "learning_rate": 0.00014352953454393655,
-      "loss": 0.6442,
+      "loss": 0.6532,
       "step": 18200
     },
     {
       "epoch": 11.46,
-      "learning_rate": 0.00014248591108328114,
-      "loss": 0.7371,
+      "learning_rate": 0.00014249634731788768,
+      "loss": 0.7053,
       "step": 18300
     },
     {
       "epoch": 11.52,
-      "learning_rate": 0.00014144228762262575,
-      "loss": 0.6647,
+      "learning_rate": 0.0001414527238572323,
+      "loss": 0.6476,
       "step": 18400
     },
     {
       "epoch": 11.58,
-      "learning_rate": 0.00014039866416197034,
-      "loss": 0.6755,
+      "learning_rate": 0.0001404091003965769,
+      "loss": 0.6308,
       "step": 18500
     },
     {
       "epoch": 11.65,
-      "learning_rate": 0.00013935504070131495,
-      "loss": 0.686,
+      "learning_rate": 0.0001393654769359215,
+      "loss": 0.6886,
       "step": 18600
     },
     {
       "epoch": 11.71,
-      "learning_rate": 0.00013831141724065957,
-      "loss": 0.6826,
+      "learning_rate": 0.0001383218534752661,
+      "loss": 0.6631,
       "step": 18700
     },
     {
       "epoch": 11.77,
-      "learning_rate": 0.00013726779378000416,
-      "loss": 0.723,
+      "learning_rate": 0.00013727823001461072,
+      "loss": 0.7056,
       "step": 18800
     },
     {
       "epoch": 11.83,
-      "learning_rate": 0.00013622417031934877,
-      "loss": 0.6481,
+      "learning_rate": 0.00013623460655395533,
+      "loss": 0.6602,
       "step": 18900
     },
     {
       "epoch": 11.9,
-      "learning_rate": 0.00013518054685869336,
-      "loss": 0.6872,
+      "learning_rate": 0.00013519098309329992,
+      "loss": 0.6728,
       "step": 19000
     },
     {
       "epoch": 11.96,
-      "learning_rate": 0.00013413692339803797,
-      "loss": 0.6628,
+      "learning_rate": 0.00013414735963264453,
+      "loss": 0.6514,
       "step": 19100
     },
     {
       "epoch": 12.0,
       "eval_accuracy": 0.9111503383348043,
-      "eval_loss": 0.30716028809547424,
-      "eval_runtime": 141.8819,
-      "eval_samples_per_second": 47.913,
-      "eval_steps_per_second": 0.754,
+      "eval_loss": 0.3034283220767975,
+      "eval_runtime": 150.4249,
+      "eval_samples_per_second": 45.192,
+      "eval_steps_per_second": 0.711,
       "step": 19164
     },
     {
       "epoch": 12.02,
-      "learning_rate": 0.00013309329993738258,
-      "loss": 0.6766,
+      "learning_rate": 0.00013310373617198915,
+      "loss": 0.6567,
       "step": 19200
     },
     {
       "epoch": 12.09,
-      "learning_rate": 0.00013204967647672717,
-      "loss": 0.682,
+      "learning_rate": 0.00013206011271133373,
+      "loss": 0.6882,
       "step": 19300
     },
     {
       "epoch": 12.15,
-      "learning_rate": 0.00013100605301607178,
-      "loss": 0.6623,
+      "learning_rate": 0.00013101648925067835,
+      "loss": 0.6511,
       "step": 19400
     },
     {
       "epoch": 12.21,
-      "learning_rate": 0.0001299624295554164,
-      "loss": 0.6597,
+      "learning_rate": 0.00012997286579002296,
+      "loss": 0.6705,
       "step": 19500
     },
     {
       "epoch": 12.27,
-      "learning_rate": 0.000128918806094761,
-      "loss": 0.669,
+      "learning_rate": 0.00012892924232936755,
+      "loss": 0.6693,
       "step": 19600
     },
     {
       "epoch": 12.34,
-      "learning_rate": 0.0001278751826341056,
-      "loss": 0.6832,
+      "learning_rate": 0.00012788561886871216,
+      "loss": 0.68,
       "step": 19700
     },
     {
       "epoch": 12.4,
-      "learning_rate": 0.0001268315591734502,
-      "loss": 0.6536,
+      "learning_rate": 0.00012684199540805675,
+      "loss": 0.6767,
       "step": 19800
     },
     {
       "epoch": 12.46,
-      "learning_rate": 0.00012578793571279482,
-      "loss": 0.6808,
+      "learning_rate": 0.00012579837194740136,
+      "loss": 0.6768,
       "step": 19900
     },
     {
       "epoch": 12.52,
-      "learning_rate": 0.0001247443122521394,
-      "loss": 0.6783,
+      "learning_rate": 0.00012475474848674598,
+      "loss": 0.6662,
       "step": 20000
     },
     {
       "epoch": 12.59,
-      "learning_rate": 0.00012370068879148403,
-      "loss": 0.6648,
+      "learning_rate": 0.00012371112502609056,
+      "loss": 0.6511,
       "step": 20100
     },
     {
       "epoch": 12.65,
-      "learning_rate": 0.00012265706533082864,
-      "loss": 0.679,
+      "learning_rate": 0.00012266750156543518,
+      "loss": 0.7057,
       "step": 20200
     },
     {
       "epoch": 12.71,
-      "learning_rate": 0.00012161344187017323,
-      "loss": 0.6344,
+      "learning_rate": 0.00012163431433938634,
+      "loss": 0.6699,
       "step": 20300
     },
     {
       "epoch": 12.77,
-      "learning_rate": 0.00012056981840951783,
-      "loss": 0.6441,
+      "learning_rate": 0.00012059069087873094,
+      "loss": 0.6541,
       "step": 20400
     },
     {
       "epoch": 12.84,
-      "learning_rate": 0.00011952619494886244,
-      "loss": 0.6464,
+      "learning_rate": 0.00011954706741807554,
+      "loss": 0.6741,
       "step": 20500
     },
     {
       "epoch": 12.9,
-      "learning_rate": 0.00011848257148820705,
-      "loss": 0.6828,
+      "learning_rate": 0.00011850344395742016,
+      "loss": 0.658,
       "step": 20600
     },
     {
       "epoch": 12.96,
-      "learning_rate": 0.00011743894802755164,
-      "loss": 0.6855,
+      "learning_rate": 0.00011745982049676476,
+      "loss": 0.6513,
       "step": 20700
     },
     {
       "epoch": 13.0,
-      "eval_accuracy": 0.903059723448073,
-      "eval_loss": 0.31616511940956116,
-      "eval_runtime": 141.8344,
-      "eval_samples_per_second": 47.929,
-      "eval_steps_per_second": 0.754,
+      "eval_accuracy": 0.9092380111797588,
+      "eval_loss": 0.30905914306640625,
+      "eval_runtime": 152.1119,
+      "eval_samples_per_second": 44.691,
+      "eval_steps_per_second": 0.703,
       "step": 20761
     },
     {
       "epoch": 13.02,
-      "learning_rate": 0.00011639532456689625,
-      "loss": 0.6492,
+      "learning_rate": 0.00011641619703610936,
+      "loss": 0.6568,
       "step": 20800
     },
     {
       "epoch": 13.09,
-      "learning_rate": 0.00011536213734084742,
-      "loss": 0.6584,
+      "learning_rate": 0.00011537257357545397,
+      "loss": 0.6853,
       "step": 20900
     },
     {
       "epoch": 13.15,
-      "learning_rate": 0.000114318513880192,
-      "loss": 0.6705,
+      "learning_rate": 0.00011432895011479858,
+      "loss": 0.6699,
       "step": 21000
     },
     {
       "epoch": 13.21,
-      "learning_rate": 0.00011327489041953662,
-      "loss": 0.6644,
+      "learning_rate": 0.00011328532665414317,
+      "loss": 0.6494,
       "step": 21100
     },
     {
       "epoch": 13.27,
-      "learning_rate": 0.00011223126695888123,
-      "loss": 0.6988,
+      "learning_rate": 0.00011224170319348778,
+      "loss": 0.7118,
       "step": 21200
     },
     {
       "epoch": 13.34,
-      "learning_rate": 0.00011118764349822583,
-      "loss": 0.6575,
+      "learning_rate": 0.00011119807973283239,
+      "loss": 0.6649,
       "step": 21300
     },
     {
       "epoch": 13.4,
-      "learning_rate": 0.00011014402003757043,
-      "loss": 0.6529,
+      "learning_rate": 0.00011015445627217699,
+      "loss": 0.6646,
       "step": 21400
     },
     {
       "epoch": 13.46,
-      "learning_rate": 0.00010910039657691503,
-      "loss": 0.6453,
+      "learning_rate": 0.0001091108328115216,
+      "loss": 0.6436,
       "step": 21500
     },
     {
       "epoch": 13.53,
-      "learning_rate": 0.00010805677311625965,
-      "loss": 0.6446,
+      "learning_rate": 0.0001080672093508662,
+      "loss": 0.6258,
       "step": 21600
     },
     {
       "epoch": 13.59,
-      "learning_rate": 0.00010701314965560425,
-      "loss": 0.6554,
+      "learning_rate": 0.0001070235858902108,
+      "loss": 0.6754,
       "step": 21700
     },
     {
       "epoch": 13.65,
-      "learning_rate": 0.00010596952619494885,
-      "loss": 0.6584,
+      "learning_rate": 0.0001059799624295554,
+      "loss": 0.6737,
       "step": 21800
     },
     {
       "epoch": 13.71,
-      "learning_rate": 0.00010492590273429346,
-      "loss": 0.7099,
+      "learning_rate": 0.00010493633896890001,
+      "loss": 0.6511,
       "step": 21900
     },
     {
       "epoch": 13.78,
-      "learning_rate": 0.00010388227927363808,
-      "loss": 0.6365,
+      "learning_rate": 0.00010389271550824463,
+      "loss": 0.6472,
       "step": 22000
     },
     {
       "epoch": 13.84,
-      "learning_rate": 0.00010283865581298266,
-      "loss": 0.6561,
+      "learning_rate": 0.00010284909204758921,
+      "loss": 0.6571,
       "step": 22100
     },
     {
       "epoch": 13.9,
-      "learning_rate": 0.00010179503235232728,
-      "loss": 0.6913,
+      "learning_rate": 0.00010180546858693383,
+      "loss": 0.693,
       "step": 22200
     },
     {
       "epoch": 13.96,
-      "learning_rate": 0.00010075140889167188,
-      "loss": 0.6651,
+      "learning_rate": 0.00010076184512627843,
+      "loss": 0.652,
       "step": 22300
     },
     {
       "epoch": 14.0,
-      "eval_accuracy": 0.9111503383348043,
-      "eval_loss": 0.30470559000968933,
-      "eval_runtime": 142.7059,
-      "eval_samples_per_second": 47.636,
-      "eval_steps_per_second": 0.75,
+      "eval_accuracy": 0.909973521624007,
+      "eval_loss": 0.30560359358787537,
+      "eval_runtime": 151.7724,
+      "eval_samples_per_second": 44.791,
+      "eval_steps_per_second": 0.705,
       "step": 22358
     },
     {
       "epoch": 14.03,
-      "learning_rate": 9.970778543101648e-05,
-      "loss": 0.6411,
+      "learning_rate": 9.971822166562303e-05,
+      "loss": 0.6286,
       "step": 22400
     },
     {
       "epoch": 14.09,
-      "learning_rate": 9.866416197036109e-05,
-      "loss": 0.6754,
+      "learning_rate": 9.867459820496764e-05,
+      "loss": 0.6503,
       "step": 22500
     },
     {
       "epoch": 14.15,
-      "learning_rate": 9.762053850970569e-05,
-      "loss": 0.6787,
+      "learning_rate": 9.763097474431224e-05,
+      "loss": 0.6514,
       "step": 22600
     },
     {
       "epoch": 14.21,
-      "learning_rate": 9.65769150490503e-05,
-      "loss": 0.6687,
+      "learning_rate": 9.659778751826341e-05,
+      "loss": 0.6728,
       "step": 22700
     },
     {
       "epoch": 14.28,
-      "learning_rate": 9.553329158839489e-05,
-      "loss": 0.6355,
+      "learning_rate": 9.555416405760801e-05,
+      "loss": 0.6621,
       "step": 22800
     },
     {
       "epoch": 14.34,
-      "learning_rate": 9.44896681277395e-05,
-      "loss": 0.6498,
+      "learning_rate": 9.451054059695261e-05,
+      "loss": 0.6771,
       "step": 22900
     },
     {
       "epoch": 14.4,
-      "learning_rate": 9.344604466708412e-05,
-      "loss": 0.6685,
+      "learning_rate": 9.346691713629722e-05,
+      "loss": 0.6689,
       "step": 23000
     },
     {
       "epoch": 14.46,
-      "learning_rate": 9.24024212064287e-05,
-      "loss": 0.6741,
+      "learning_rate": 9.242329367564181e-05,
+      "loss": 0.6712,
       "step": 23100
     },
     {
       "epoch": 14.53,
-      "learning_rate": 9.135879774577332e-05,
-      "loss": 0.6701,
+      "learning_rate": 9.137967021498642e-05,
+      "loss": 0.6761,
       "step": 23200
     },
     {
       "epoch": 14.59,
-      "learning_rate": 9.032561051972448e-05,
-      "loss": 0.6508,
+      "learning_rate": 9.033604675433104e-05,
+      "loss": 0.6327,
       "step": 23300
     },
     {
       "epoch": 14.65,
-      "learning_rate": 8.928198705906908e-05,
-      "loss": 0.6571,
+      "learning_rate": 8.929242329367564e-05,
+      "loss": 0.6671,
       "step": 23400
     },
     {
       "epoch": 14.72,
-      "learning_rate": 8.823836359841368e-05,
-      "loss": 0.6508,
+      "learning_rate": 8.824879983302024e-05,
+      "loss": 0.6598,
       "step": 23500
     },
     {
       "epoch": 14.78,
-      "learning_rate": 8.719474013775828e-05,
-      "loss": 0.6202,
+      "learning_rate": 8.720517637236485e-05,
+      "loss": 0.6317,
       "step": 23600
     },
     {
       "epoch": 14.84,
-      "learning_rate": 8.61511166771029e-05,
-      "loss": 0.6657,
+      "learning_rate": 8.616155291170945e-05,
+      "loss": 0.6615,
       "step": 23700
     },
     {
       "epoch": 14.9,
-      "learning_rate": 8.51074932164475e-05,
-      "loss": 0.6301,
+      "learning_rate": 8.511792945105405e-05,
+      "loss": 0.6087,
       "step": 23800
     },
     {
       "epoch": 14.97,
-      "learning_rate": 8.40638697557921e-05,
-      "loss": 0.6962,
+      "learning_rate": 8.407430599039865e-05,
+      "loss": 0.7105,
       "step": 23900
     },
     {
       "epoch": 15.0,
-      "eval_accuracy": 0.912621359223301,
-      "eval_loss": 0.30054593086242676,
-      "eval_runtime": 147.8277,
-      "eval_samples_per_second": 45.986,
-      "eval_steps_per_second": 0.724,
+      "eval_accuracy": 0.9149749926448956,
+      "eval_loss": 0.30149412155151367,
+      "eval_runtime": 151.4148,
+      "eval_samples_per_second": 44.897,
+      "eval_steps_per_second": 0.707,
       "step": 23955
     },
     {
       "epoch": 15.03,
-      "learning_rate": 8.302024629513671e-05,
-      "loss": 0.6898,
+      "learning_rate": 8.303068252974326e-05,
+      "loss": 0.6911,
       "step": 24000
     },
     {
       "epoch": 15.09,
-      "learning_rate": 8.19766228344813e-05,
-      "loss": 0.6417,
+      "learning_rate": 8.198705906908788e-05,
+      "loss": 0.6717,
       "step": 24100
     },
     {
       "epoch": 15.15,
-      "learning_rate": 8.093299937382591e-05,
-      "loss": 0.6322,
+      "learning_rate": 8.094343560843246e-05,
+      "loss": 0.6564,
       "step": 24200
     },
     {
       "epoch": 15.22,
-      "learning_rate": 7.988937591317053e-05,
-      "loss": 0.6142,
+      "learning_rate": 7.989981214777708e-05,
+      "loss": 0.6446,
       "step": 24300
     },
     {
       "epoch": 15.28,
-      "learning_rate": 7.884575245251513e-05,
-      "loss": 0.6659,
+      "learning_rate": 7.885618868712169e-05,
+      "loss": 0.6431,
       "step": 24400
     },
     {
       "epoch": 15.34,
-      "learning_rate": 7.780212899185973e-05,
-      "loss": 0.6626,
+      "learning_rate": 7.781256522646628e-05,
+      "loss": 0.6762,
       "step": 24500
     },
     {
       "epoch": 15.4,
-      "learning_rate": 7.675850553120434e-05,
-      "loss": 0.6477,
+      "learning_rate": 7.676894176581089e-05,
+      "loss": 0.6656,
       "step": 24600
     },
     {
       "epoch": 15.47,
-      "learning_rate": 7.571488207054894e-05,
-      "loss": 0.6066,
+      "learning_rate": 7.572531830515549e-05,
+      "loss": 0.6337,
       "step": 24700
     },
     {
       "epoch": 15.53,
-      "learning_rate": 7.467125860989354e-05,
-      "loss": 0.6713,
+      "learning_rate": 7.468169484450009e-05,
+      "loss": 0.6541,
       "step": 24800
     },
     {
       "epoch": 15.59,
-      "learning_rate": 7.362763514923814e-05,
-      "loss": 0.6725,
+      "learning_rate": 7.363807138384469e-05,
+      "loss": 0.6772,
       "step": 24900
     },
     {
       "epoch": 15.65,
-      "learning_rate": 7.258401168858275e-05,
-      "loss": 0.6442,
+      "learning_rate": 7.260488415779586e-05,
+      "loss": 0.629,
       "step": 25000
     },
     {
       "epoch": 15.72,
-      "learning_rate": 7.154038822792735e-05,
-      "loss": 0.7171,
+      "learning_rate": 7.156126069714046e-05,
+      "loss": 0.6998,
       "step": 25100
     },
     {
       "epoch": 15.78,
-      "learning_rate": 7.049676476727197e-05,
-      "loss": 0.6558,
+      "learning_rate": 7.051763723648507e-05,
+      "loss": 0.6686,
       "step": 25200
     },
     {
       "epoch": 15.84,
-      "learning_rate": 6.945314130661657e-05,
-      "loss": 0.6759,
+      "learning_rate": 6.947401377582967e-05,
+      "loss": 0.6822,
       "step": 25300
     },
     {
       "epoch": 15.9,
-      "learning_rate": 6.840951784596117e-05,
-      "loss": 0.6378,
+      "learning_rate": 6.843039031517429e-05,
+      "loss": 0.6143,
       "step": 25400
     },
     {
       "epoch": 15.97,
-      "learning_rate": 6.736589438530578e-05,
-      "loss": 0.6348,
+      "learning_rate": 6.738676685451889e-05,
+      "loss": 0.6337,
       "step": 25500
     },
     {
       "epoch": 16.0,
-      "eval_accuracy": 0.9079140923801118,
-      "eval_loss": 0.31096547842025757,
-      "eval_runtime": 148.4859,
-      "eval_samples_per_second": 45.782,
-      "eval_steps_per_second": 0.721,
+      "eval_accuracy": 0.9090909090909091,
+      "eval_loss": 0.30700910091400146,
+      "eval_runtime": 152.1641,
+      "eval_samples_per_second": 44.675,
+      "eval_steps_per_second": 0.703,
       "step": 25552
     },
     {
       "epoch": 16.03,
-      "learning_rate": 6.632227092465038e-05,
-      "loss": 0.6608,
+      "learning_rate": 6.634314339386349e-05,
+      "loss": 0.6502,
       "step": 25600
     },
     {
       "epoch": 16.09,
-      "learning_rate": 6.527864746399498e-05,
-      "loss": 0.667,
+      "learning_rate": 6.52995199332081e-05,
+      "loss": 0.6448,
       "step": 25700
     },
     {
       "epoch": 16.16,
-      "learning_rate": 6.423502400333958e-05,
-      "loss": 0.6447,
+      "learning_rate": 6.42558964725527e-05,
+      "loss": 0.6272,
       "step": 25800
     },
     {
       "epoch": 16.22,
-      "learning_rate": 6.31914005426842e-05,
-      "loss": 0.649,
+      "learning_rate": 6.32122730118973e-05,
+      "loss": 0.6316,
       "step": 25900
     },
     {
       "epoch": 16.28,
-      "learning_rate": 6.21477770820288e-05,
-      "loss": 0.6438,
+      "learning_rate": 6.21686495512419e-05,
+      "loss": 0.6442,
       "step": 26000
     },
     {
       "epoch": 16.34,
-      "learning_rate": 6.11041536213734e-05,
-      "loss": 0.6727,
+      "learning_rate": 6.11250260905865e-05,
+      "loss": 0.6489,
       "step": 26100
     },
     {
       "epoch": 16.41,
-      "learning_rate": 6.006053016071801e-05,
-      "loss": 0.6467,
+      "learning_rate": 6.0081402629931114e-05,
+      "loss": 0.6603,
       "step": 26200
     },
     {
       "epoch": 16.47,
-      "learning_rate": 5.901690670006261e-05,
-      "loss": 0.6365,
+      "learning_rate": 5.9037779169275714e-05,
+      "loss": 0.6449,
       "step": 26300
     },
     {
       "epoch": 16.53,
-      "learning_rate": 5.797328323940722e-05,
-      "loss": 0.6642,
+      "learning_rate": 5.799415570862033e-05,
+      "loss": 0.661,
       "step": 26400
     },
     {
       "epoch": 16.59,
-      "learning_rate": 5.692965977875182e-05,
-      "loss": 0.6404,
+      "learning_rate": 5.695053224796493e-05,
+      "loss": 0.642,
       "step": 26500
     },
     {
       "epoch": 16.66,
-      "learning_rate": 5.5896472552702984e-05,
-      "loss": 0.6483,
+      "learning_rate": 5.5906908787309535e-05,
+      "loss": 0.6498,
       "step": 26600
     },
     {
       "epoch": 16.72,
-      "learning_rate": 5.4852849092047584e-05,
-      "loss": 0.7041,
+      "learning_rate": 5.4863285326654135e-05,
+      "loss": 0.6902,
       "step": 26700
     },
     {
       "epoch": 16.78,
-      "learning_rate": 5.3809225631392184e-05,
-      "loss": 0.6534,
+      "learning_rate": 5.381966186599874e-05,
+      "loss": 0.6336,
       "step": 26800
     },
     {
       "epoch": 16.84,
-      "learning_rate": 5.27656021707368e-05,
-      "loss": 0.6355,
+      "learning_rate": 5.277603840534335e-05,
+      "loss": 0.6393,
       "step": 26900
     },
     {
       "epoch": 16.91,
-      "learning_rate": 5.17219787100814e-05,
-      "loss": 0.6394,
+      "learning_rate": 5.173241494468795e-05,
+      "loss": 0.6496,
       "step": 27000
     },
     {
       "epoch": 16.97,
-      "learning_rate": 5.0678355249426005e-05,
-      "loss": 0.6595,
+      "learning_rate": 5.0688791484032557e-05,
+      "loss": 0.63,
       "step": 27100
     },
     {
       "epoch": 17.0,
-      "eval_accuracy": 0.9168873197999412,
-      "eval_loss": 0.2988849878311157,
-      "eval_runtime": 147.5031,
-      "eval_samples_per_second": 46.087,
-      "eval_steps_per_second": 0.725,
+      "eval_accuracy": 0.913503971756399,
+      "eval_loss": 0.30175167322158813,
+      "eval_runtime": 146.2393,
+      "eval_samples_per_second": 46.485,
+      "eval_steps_per_second": 0.732,
       "step": 27149
     },
     {
       "epoch": 17.03,
-      "learning_rate": 4.9634731788770605e-05,
-      "loss": 0.6207,
+      "learning_rate": 4.9645168023377163e-05,
+      "loss": 0.6195,
       "step": 27200
     },
     {
       "epoch": 17.09,
-      "learning_rate": 4.8591108328115205e-05,
-      "loss": 0.6216,
+      "learning_rate": 4.861198079732832e-05,
+      "loss": 0.6299,
       "step": 27300
     },
     {
       "epoch": 17.16,
-      "learning_rate": 4.754748486745982e-05,
-      "loss": 0.6647,
+      "learning_rate": 4.756835733667292e-05,
+      "loss": 0.6939,
       "step": 27400
     },
     {
       "epoch": 17.22,
-      "learning_rate": 4.650386140680442e-05,
-      "loss": 0.6152,
+      "learning_rate": 4.652473387601752e-05,
+      "loss": 0.6416,
       "step": 27500
     },
     {
       "epoch": 17.28,
-      "learning_rate": 4.5460237946149026e-05,
-      "loss": 0.6531,
+      "learning_rate": 4.5481110415362136e-05,
+      "loss": 0.6626,
       "step": 27600
     },
     {
       "epoch": 17.35,
-      "learning_rate": 4.4416614485493626e-05,
-      "loss": 0.6108,
+      "learning_rate": 4.4437486954706736e-05,
+      "loss": 0.6444,
       "step": 27700
     },
     {
       "epoch": 17.41,
-      "learning_rate": 4.337299102483824e-05,
-      "loss": 0.6569,
+      "learning_rate": 4.339386349405134e-05,
+      "loss": 0.6238,
       "step": 27800
     },
     {
       "epoch": 17.47,
-      "learning_rate": 4.232936756418284e-05,
-      "loss": 0.666,
+      "learning_rate": 4.2350240033395944e-05,
+      "loss": 0.6505,
       "step": 27900
     },
     {
       "epoch": 17.53,
-      "learning_rate": 4.128574410352744e-05,
-      "loss": 0.6344,
+      "learning_rate": 4.130661657274056e-05,
+      "loss": 0.6346,
       "step": 28000
     },
     {
       "epoch": 17.6,
-      "learning_rate": 4.024212064287205e-05,
-      "loss": 0.6627,
+      "learning_rate": 4.026299311208516e-05,
+      "loss": 0.6319,
       "step": 28100
     },
     {
       "epoch": 17.66,
-      "learning_rate": 3.9198497182216654e-05,
-      "loss": 0.6863,
+      "learning_rate": 3.921936965142976e-05,
+      "loss": 0.6811,
       "step": 28200
     },
     {
       "epoch": 17.72,
-      "learning_rate": 3.815487372156126e-05,
-      "loss": 0.7073,
+      "learning_rate": 3.8175746190774365e-05,
+      "loss": 0.6645,
       "step": 28300
     },
     {
       "epoch": 17.78,
-      "learning_rate": 3.711125026090586e-05,
-      "loss": 0.6618,
+      "learning_rate": 3.713212273011897e-05,
+      "loss": 0.6512,
       "step": 28400
     },
     {
       "epoch": 17.85,
-      "learning_rate": 3.606762680025047e-05,
-      "loss": 0.6535,
+      "learning_rate": 3.608849926946357e-05,
+      "loss": 0.6578,
       "step": 28500
     },
     {
       "epoch": 17.91,
-      "learning_rate": 3.5024003339595075e-05,
-      "loss": 0.6106,
+      "learning_rate": 3.504487580880818e-05,
+      "loss": 0.6358,
       "step": 28600
     },
     {
       "epoch": 17.97,
-      "learning_rate": 3.3980379878939675e-05,
-      "loss": 0.677,
+      "learning_rate": 3.4001252348152786e-05,
+      "loss": 0.6672,
       "step": 28700
     },
     {
       "epoch": 18.0,
-      "eval_accuracy": 0.9098264195351574,
-      "eval_loss": 0.31027764081954956,
-      "eval_runtime": 142.2177,
-      "eval_samples_per_second": 47.8,
-      "eval_steps_per_second": 0.752,
+      "eval_accuracy": 0.9087967049132097,
+      "eval_loss": 0.30836355686187744,
+      "eval_runtime": 148.5759,
+      "eval_samples_per_second": 45.754,
+      "eval_steps_per_second": 0.72,
       "step": 28746
     },
     {
       "epoch": 18.03,
-      "learning_rate": 3.2936756418284276e-05,
-      "loss": 0.6368,
+      "learning_rate": 3.295762888749739e-05,
+      "loss": 0.65,
       "step": 28800
     },
     {
       "epoch": 18.1,
-      "learning_rate": 3.189313295762888e-05,
-      "loss": 0.6739,
+      "learning_rate": 3.191400542684199e-05,
+      "loss": 0.6293,
       "step": 28900
     },
     {
       "epoch": 18.16,
-      "learning_rate": 3.084950949697349e-05,
-      "loss": 0.6187,
+      "learning_rate": 3.087038196618659e-05,
+      "loss": 0.6059,
       "step": 29000
     },
     {
       "epoch": 18.22,
-      "learning_rate": 2.9805886036318093e-05,
-      "loss": 0.6197,
+      "learning_rate": 2.98267585055312e-05,
+      "loss": 0.5975,
       "step": 29100
     },
     {
       "epoch": 18.28,
-      "learning_rate": 2.87622625756627e-05,
-      "loss": 0.6583,
+      "learning_rate": 2.8783135044875807e-05,
+      "loss": 0.6759,
       "step": 29200
     },
     {
       "epoch": 18.35,
-      "learning_rate": 2.7718639115007304e-05,
-      "loss": 0.6446,
+      "learning_rate": 2.773951158422041e-05,
+      "loss": 0.6457,
       "step": 29300
     },
     {
       "epoch": 18.41,
-      "learning_rate": 2.6675015654351907e-05,
-      "loss": 0.6569,
+      "learning_rate": 2.6695888123565017e-05,
+      "loss": 0.6715,
       "step": 29400
     },
     {
       "epoch": 18.47,
-      "learning_rate": 2.563139219369651e-05,
-      "loss": 0.6681,
+      "learning_rate": 2.565226466290962e-05,
+      "loss": 0.6642,
       "step": 29500
     },
     {
       "epoch": 18.53,
-      "learning_rate": 2.4587768733041114e-05,
-      "loss": 0.627,
+      "learning_rate": 2.460864120225422e-05,
+      "loss": 0.6432,
       "step": 29600
     },
     {
       "epoch": 18.6,
-      "learning_rate": 2.354414527238572e-05,
-      "loss": 0.6477,
+      "learning_rate": 2.3565017741598828e-05,
+      "loss": 0.6441,
       "step": 29700
     },
     {
       "epoch": 18.66,
-      "learning_rate": 2.2500521811730325e-05,
-      "loss": 0.6745,
+      "learning_rate": 2.2521394280943432e-05,
+      "loss": 0.6843,
       "step": 29800
     },
     {
       "epoch": 18.72,
-      "learning_rate": 2.1456898351074932e-05,
-      "loss": 0.6365,
+      "learning_rate": 2.147777082028804e-05,
+      "loss": 0.6459,
       "step": 29900
     },
     {
       "epoch": 18.79,
-      "learning_rate": 2.0413274890419532e-05,
-      "loss": 0.6254,
+      "learning_rate": 2.0434147359632642e-05,
+      "loss": 0.6233,
       "step": 30000
     },
     {
       "epoch": 18.85,
-      "learning_rate": 1.936965142976414e-05,
-      "loss": 0.6445,
+      "learning_rate": 1.939052389897725e-05,
+      "loss": 0.6634,
       "step": 30100
     },
     {
       "epoch": 18.91,
-      "learning_rate": 1.8326027969108743e-05,
-      "loss": 0.6744,
+      "learning_rate": 1.8346900438321853e-05,
+      "loss": 0.6701,
       "step": 30200
     },
     {
       "epoch": 18.97,
-      "learning_rate": 1.728240450845335e-05,
-      "loss": 0.6394,
+      "learning_rate": 1.7303276977666456e-05,
+      "loss": 0.6479,
       "step": 30300
     },
     {
       "epoch": 19.0,
-      "eval_accuracy": 0.9085025007355104,
-      "eval_loss": 0.3055914342403412,
-      "eval_runtime": 142.7169,
-      "eval_samples_per_second": 47.633,
-      "eval_steps_per_second": 0.75,
+      "eval_accuracy": 0.9101206237128567,
+      "eval_loss": 0.3060016632080078,
+      "eval_runtime": 145.9275,
+      "eval_samples_per_second": 46.585,
+      "eval_steps_per_second": 0.733,
       "step": 30343
     },
     {
       "epoch": 19.04,
-      "learning_rate": 1.6238781047797953e-05,
-      "loss": 0.6256,
+      "learning_rate": 1.625965351701106e-05,
+      "loss": 0.6155,
       "step": 30400
     },
     {
       "epoch": 19.1,
-      "learning_rate": 1.5195157587142557e-05,
-      "loss": 0.6218,
+      "learning_rate": 1.5216030056355665e-05,
+      "loss": 0.6079,
       "step": 30500
     },
     {
       "epoch": 19.16,
-      "learning_rate": 1.4161970361093715e-05,
-      "loss": 0.6752,
+      "learning_rate": 1.417240659570027e-05,
+      "loss": 0.6709,
       "step": 30600
     },
     {
       "epoch": 19.22,
-      "learning_rate": 1.311834690043832e-05,
-      "loss": 0.6387,
+      "learning_rate": 1.3128783135044874e-05,
+      "loss": 0.6604,
       "step": 30700
     },
     {
       "epoch": 19.29,
-      "learning_rate": 1.2074723439782926e-05,
-      "loss": 0.6291,
+      "learning_rate": 1.208515967438948e-05,
+      "loss": 0.6222,
       "step": 30800
     },
     {
       "epoch": 19.35,
-      "learning_rate": 1.103109997912753e-05,
-      "loss": 0.6009,
+      "learning_rate": 1.1041536213734085e-05,
+      "loss": 0.6428,
       "step": 30900
     },
     {
       "epoch": 19.41,
-      "learning_rate": 9.987476518472135e-06,
-      "loss": 0.6347,
+      "learning_rate": 9.997912753078688e-06,
+      "loss": 0.6664,
       "step": 31000
     },
     {
       "epoch": 19.47,
-      "learning_rate": 8.943853057816738e-06,
-      "loss": 0.6795,
+      "learning_rate": 8.954289292423293e-06,
+      "loss": 0.6489,
       "step": 31100
     },
     {
       "epoch": 19.54,
-      "learning_rate": 7.900229597161344e-06,
-      "loss": 0.6558,
+      "learning_rate": 7.910665831767897e-06,
+      "loss": 0.6414,
       "step": 31200
     },
     {
       "epoch": 19.6,
-      "learning_rate": 6.856606136505948e-06,
-      "loss": 0.6577,
+      "learning_rate": 6.877478605719056e-06,
+      "loss": 0.6599,
       "step": 31300
     },
     {
       "epoch": 19.66,
-      "learning_rate": 5.812982675850552e-06,
-      "loss": 0.6258,
+      "learning_rate": 5.83385514506366e-06,
+      "loss": 0.6433,
       "step": 31400
     },
     {
       "epoch": 19.72,
-      "learning_rate": 4.769359215195158e-06,
-      "loss": 0.6196,
+      "learning_rate": 4.790231684408265e-06,
+      "loss": 0.6325,
       "step": 31500
     },
     {
       "epoch": 19.79,
-      "learning_rate": 3.7257357545397616e-06,
-      "loss": 0.6232,
+      "learning_rate": 3.7466082237528697e-06,
+      "loss": 0.6363,
       "step": 31600
     },
     {
       "epoch": 19.85,
-      "learning_rate": 2.6821122938843665e-06,
-      "loss": 0.6747,
+      "learning_rate": 2.7029847630974745e-06,
+      "loss": 0.6408,
       "step": 31700
     },
     {
       "epoch": 19.91,
-      "learning_rate": 1.638488833228971e-06,
-      "loss": 0.6485,
+      "learning_rate": 1.6593613024420787e-06,
+      "loss": 0.645,
       "step": 31800
     },
     {
       "epoch": 19.97,
-      "learning_rate": 5.948653725735754e-07,
-      "loss": 0.6411,
+      "learning_rate": 6.157378417866834e-07,
+      "loss": 0.6658,
       "step": 31900
     },
     {
       "epoch": 20.0,
       "eval_accuracy": 0.9089438070020595,
-      "eval_loss": 0.30604276061058044,
-      "eval_runtime": 142.2608,
-      "eval_samples_per_second": 47.785,
-      "eval_steps_per_second": 0.752,
+      "eval_loss": 0.3071773946285248,
+      "eval_runtime": 145.97,
+      "eval_samples_per_second": 46.571,
+      "eval_steps_per_second": 0.733,
       "step": 31940
     },
     {
       "epoch": 20.0,
       "step": 31940,
       "total_flos": 0.0,
-      "train_loss": 0.8135456519344858,
-      "train_runtime": 27591.4491,
-      "train_samples_per_second": 37.036,
-      "train_steps_per_second": 1.158
+      "train_loss": 0.8136612295581911,
+      "train_runtime": 28286.4741,
+      "train_samples_per_second": 36.126,
+      "train_steps_per_second": 1.129
     }
   ],
   "max_steps": 31940,