diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,13729 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.131931740614334,
+  "global_step": 65300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.998743993010048e-05,
+      "loss": 5.9677,
+      "step": 50
+    },
+    {
+      "epoch": 0.01,
+      "learning_rate": 4.99737876802097e-05,
+      "loss": 3.506,
+      "step": 100
+    },
+    {
+      "epoch": 0.01,
+      "eval_cer": 0.4539164593605962,
+      "eval_loss": 2.664681911468506,
+      "eval_runtime": 1456.6338,
+      "eval_samples_per_second": 5.158,
+      "eval_steps_per_second": 0.645,
+      "step": 100
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.996013543031892e-05,
+      "loss": 2.7796,
+      "step": 150
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 4.994648318042814e-05,
+      "loss": 2.4129,
+      "step": 200
+    },
+    {
+      "epoch": 0.02,
+      "eval_cer": 0.41749788701790647,
+      "eval_loss": 2.389951705932617,
+      "eval_runtime": 1302.8974,
+      "eval_samples_per_second": 5.766,
+      "eval_steps_per_second": 0.721,
+      "step": 200
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.993283093053736e-05,
+      "loss": 2.2714,
+      "step": 250
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 4.9919178680646574e-05,
+      "loss": 2.2433,
+      "step": 300
+    },
+    {
+      "epoch": 0.03,
+      "eval_cer": 0.3008544317874631,
+      "eval_loss": 1.778554916381836,
+      "eval_runtime": 1143.042,
+      "eval_samples_per_second": 6.573,
+      "eval_steps_per_second": 0.822,
+      "step": 300
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.990552643075579e-05,
+      "loss": 1.9436,
+      "step": 350
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 4.989187418086501e-05,
+      "loss": 1.8459,
+      "step": 400
+    },
+    {
+      "epoch": 0.04,
+      "eval_cer": 0.26927090996814274,
+      "eval_loss": 1.7008424997329712,
+      "eval_runtime": 1063.6345,
+      "eval_samples_per_second": 7.064,
+      "eval_steps_per_second": 0.884,
+      "step": 400
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.987822193097423e-05,
+      "loss": 1.7638,
+      "step": 450
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 4.9864569681083444e-05,
+      "loss": 1.7375,
+      "step": 500
+    },
+    {
+      "epoch": 0.05,
+      "eval_cer": 0.28988061223442296,
+      "eval_loss": 1.690486192703247,
+      "eval_runtime": 1230.2876,
+      "eval_samples_per_second": 6.107,
+      "eval_steps_per_second": 0.764,
+      "step": 500
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 4.985091743119266e-05,
+      "loss": 1.6963,
+      "step": 550
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.9837538226299694e-05,
+      "loss": 1.6133,
+      "step": 600
+    },
+    {
+      "epoch": 0.07,
+      "eval_cer": 0.24955430620617913,
+      "eval_loss": 1.496834397315979,
+      "eval_runtime": 1083.5748,
+      "eval_samples_per_second": 6.934,
+      "eval_steps_per_second": 0.867,
+      "step": 600
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 4.982388597640891e-05,
+      "loss": 1.5559,
+      "step": 650
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.981023372651813e-05,
+      "loss": 1.5477,
+      "step": 700
+    },
+    {
+      "epoch": 0.08,
+      "eval_cer": 0.23770791914892161,
+      "eval_loss": 1.4269628524780273,
+      "eval_runtime": 1176.6532,
+      "eval_samples_per_second": 6.385,
+      "eval_steps_per_second": 0.799,
+      "step": 700
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 4.9796581476627355e-05,
+      "loss": 1.5112,
+      "step": 750
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.978292922673657e-05,
+      "loss": 1.4376,
+      "step": 800
+    },
+    {
+      "epoch": 0.09,
+      "eval_cer": 0.25057743437779095,
+      "eval_loss": 1.388639211654663,
+      "eval_runtime": 1506.6054,
+      "eval_samples_per_second": 4.987,
+      "eval_steps_per_second": 0.624,
+      "step": 800
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 4.976927697684579e-05,
+      "loss": 1.4,
+      "step": 850
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.9755624726955e-05,
+      "loss": 1.2982,
+      "step": 900
+    },
+    {
+      "epoch": 0.1,
+      "eval_cer": 0.27491693499543185,
+      "eval_loss": 1.3276695013046265,
+      "eval_runtime": 1886.1604,
+      "eval_samples_per_second": 3.983,
+      "eval_steps_per_second": 0.498,
+      "step": 900
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 4.9741972477064225e-05,
+      "loss": 1.394,
+      "step": 950
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.972832022717344e-05,
+      "loss": 1.2517,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "eval_cer": 0.19267659226460354,
+      "eval_loss": 1.152963399887085,
+      "eval_runtime": 1240.808,
+      "eval_samples_per_second": 6.055,
+      "eval_steps_per_second": 0.758,
+      "step": 1000
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 4.971466797728266e-05,
+      "loss": 1.2324,
+      "step": 1050
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 4.970101572739187e-05,
+      "loss": 1.2044,
+      "step": 1100
+    },
+    {
+      "epoch": 0.12,
+      "eval_cer": 0.1959649741138307,
+      "eval_loss": 1.1271371841430664,
+      "eval_runtime": 1192.9542,
+      "eval_samples_per_second": 6.298,
+      "eval_steps_per_second": 0.788,
+      "step": 1100
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.9687363477501095e-05,
+      "loss": 1.1874,
+      "step": 1150
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.967371122761031e-05,
+      "loss": 1.1988,
+      "step": 1200
+    },
+    {
+      "epoch": 0.13,
+      "eval_cer": 0.2019942444763055,
+      "eval_loss": 1.1370463371276855,
+      "eval_runtime": 1202.9894,
+      "eval_samples_per_second": 6.245,
+      "eval_steps_per_second": 0.781,
+      "step": 1200
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.9660058977719534e-05,
+      "loss": 1.1858,
+      "step": 1250
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 4.964640672782875e-05,
+      "loss": 1.1685,
+      "step": 1300
+    },
+    {
+      "epoch": 0.14,
+      "eval_cer": 0.1879031347415318,
+      "eval_loss": 1.1137348413467407,
+      "eval_runtime": 1125.3048,
+      "eval_samples_per_second": 6.676,
+      "eval_steps_per_second": 0.835,
+      "step": 1300
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.9632754477937965e-05,
+      "loss": 1.1165,
+      "step": 1350
+    },
+    {
+      "epoch": 0.15,
+      "learning_rate": 4.961910222804718e-05,
+      "loss": 1.135,
+      "step": 1400
+    },
+    {
+      "epoch": 0.15,
+      "eval_cer": 0.18725983007175584,
+      "eval_loss": 1.0930421352386475,
+      "eval_runtime": 1179.9127,
+      "eval_samples_per_second": 6.367,
+      "eval_steps_per_second": 0.797,
+      "step": 1400
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.9605449978156404e-05,
+      "loss": 1.0825,
+      "step": 1450
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 4.959179772826562e-05,
+      "loss": 1.0914,
+      "step": 1500
+    },
+    {
+      "epoch": 0.16,
+      "eval_cer": 0.18079940870719713,
+      "eval_loss": 1.032211184501648,
+      "eval_runtime": 1192.2191,
+      "eval_samples_per_second": 6.302,
+      "eval_steps_per_second": 0.788,
+      "step": 1500
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 4.9578145478374836e-05,
+      "loss": 1.0302,
+      "step": 1550
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 4.956449322848405e-05,
+      "loss": 1.1226,
+      "step": 1600
+    },
+    {
+      "epoch": 0.17,
+      "eval_cer": 0.17277863133509672,
+      "eval_loss": 1.0084491968154907,
+      "eval_runtime": 1179.5417,
+      "eval_samples_per_second": 6.369,
+      "eval_steps_per_second": 0.797,
+      "step": 1600
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 4.9550840978593274e-05,
+      "loss": 1.0392,
+      "step": 1650
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.9537188728702497e-05,
+      "loss": 1.0158,
+      "step": 1700
+    },
+    {
+      "epoch": 0.19,
+      "eval_cer": 0.16822417114641683,
+      "eval_loss": 1.0280424356460571,
+      "eval_runtime": 1116.1285,
+      "eval_samples_per_second": 6.731,
+      "eval_steps_per_second": 0.842,
+      "step": 1700
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.952353647881171e-05,
+      "loss": 1.0508,
+      "step": 1750
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.950988422892093e-05,
+      "loss": 1.0199,
+      "step": 1800
+    },
+    {
+      "epoch": 0.2,
+      "eval_cer": 0.1638339589585308,
+      "eval_loss": 0.9974864721298218,
+      "eval_runtime": 1121.8717,
+      "eval_samples_per_second": 6.697,
+      "eval_steps_per_second": 0.838,
+      "step": 1800
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 4.9496231979030144e-05,
+      "loss": 1.0024,
+      "step": 1850
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.9482579729139367e-05,
+      "loss": 0.9729,
+      "step": 1900
+    },
+    {
+      "epoch": 0.21,
+      "eval_cer": 0.17101638715991255,
+      "eval_loss": 0.9937753081321716,
+      "eval_runtime": 1156.6861,
+      "eval_samples_per_second": 6.495,
+      "eval_steps_per_second": 0.813,
+      "step": 1900
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 4.946892747924858e-05,
+      "loss": 1.0344,
+      "step": 1950
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.94552752293578e-05,
+      "loss": 0.9539,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "eval_cer": 0.15910498526900743,
+      "eval_loss": 0.9161514639854431,
+      "eval_runtime": 1178.9029,
+      "eval_samples_per_second": 6.373,
+      "eval_steps_per_second": 0.797,
+      "step": 2000
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 4.9441622979467014e-05,
+      "loss": 0.9257,
+      "step": 2050
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.9427970729576237e-05,
+      "loss": 1.0588,
+      "step": 2100
+    },
+    {
+      "epoch": 0.23,
+      "eval_cer": 0.15335972707457202,
+      "eval_loss": 0.9163301587104797,
+      "eval_runtime": 1165.5791,
+      "eval_samples_per_second": 6.446,
+      "eval_steps_per_second": 0.806,
+      "step": 2100
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 4.941431847968546e-05,
+      "loss": 0.942,
+      "step": 2150
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 4.9400666229794675e-05,
+      "loss": 1.0196,
+      "step": 2200
+    },
+    {
+      "epoch": 0.24,
+      "eval_cer": 0.16217094795049292,
+      "eval_loss": 0.9203193783760071,
+      "eval_runtime": 1167.9514,
+      "eval_samples_per_second": 6.433,
+      "eval_steps_per_second": 0.805,
+      "step": 2200
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.938701397990389e-05,
+      "loss": 0.9518,
+      "step": 2250
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 4.937336173001311e-05,
+      "loss": 0.9547,
+      "step": 2300
+    },
+    {
+      "epoch": 0.25,
+      "eval_cer": 0.1646825736292991,
+      "eval_loss": 0.8791403770446777,
+      "eval_runtime": 1267.8673,
+      "eval_samples_per_second": 5.926,
+      "eval_steps_per_second": 0.741,
+      "step": 2300
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.935970948012233e-05,
+      "loss": 0.9117,
+      "step": 2350
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.9346057230231545e-05,
+      "loss": 0.9091,
+      "step": 2400
+    },
+    {
+      "epoch": 0.26,
+      "eval_cer": 0.14374437536143114,
+      "eval_loss": 0.8452526330947876,
+      "eval_runtime": 1242.3604,
+      "eval_samples_per_second": 6.047,
+      "eval_steps_per_second": 0.757,
+      "step": 2400
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.933240498034076e-05,
+      "loss": 0.9276,
+      "step": 2450
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 4.931875273044998e-05,
+      "loss": 0.9199,
+      "step": 2500
+    },
+    {
+      "epoch": 0.27,
+      "eval_cer": 0.14217033202048993,
+      "eval_loss": 0.828856348991394,
+      "eval_runtime": 1444.7023,
+      "eval_samples_per_second": 5.2,
+      "eval_steps_per_second": 0.651,
+      "step": 2500
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.93051004805592e-05,
+      "loss": 0.9014,
+      "step": 2550
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 4.929144823066842e-05,
+      "loss": 0.8825,
+      "step": 2600
+    },
+    {
+      "epoch": 0.28,
+      "eval_cer": 0.13830366033513436,
+      "eval_loss": 0.8257022500038147,
+      "eval_runtime": 1455.1873,
+      "eval_samples_per_second": 5.163,
+      "eval_steps_per_second": 0.646,
+      "step": 2600
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.927779598077764e-05,
+      "loss": 0.9311,
+      "step": 2650
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 4.9264143730886854e-05,
+      "loss": 0.8439,
+      "step": 2700
+    },
+    {
+      "epoch": 0.29,
+      "eval_cer": 0.146505794874778,
+      "eval_loss": 0.8311769366264343,
+      "eval_runtime": 1282.7141,
+      "eval_samples_per_second": 5.857,
+      "eval_steps_per_second": 0.733,
+      "step": 2700
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 4.925049148099607e-05,
+      "loss": 0.8646,
+      "step": 2750
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 4.9236839231105285e-05,
+      "loss": 0.8431,
+      "step": 2800
+    },
+    {
+      "epoch": 0.31,
+      "eval_cer": 0.136733038827543,
+      "eval_loss": 0.8049759864807129,
+      "eval_runtime": 1195.0102,
+      "eval_samples_per_second": 6.287,
+      "eval_steps_per_second": 0.787,
+      "step": 2800
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 4.922318698121451e-05,
+      "loss": 0.8711,
+      "step": 2850
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 4.9209534731323724e-05,
+      "loss": 0.8549,
+      "step": 2900
+    },
+    {
+      "epoch": 0.32,
+      "eval_cer": 0.1316276634695337,
+      "eval_loss": 0.7712754011154175,
+      "eval_runtime": 1199.3555,
+      "eval_samples_per_second": 6.264,
+      "eval_steps_per_second": 0.784,
+      "step": 2900
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 4.919588248143294e-05,
+      "loss": 0.8794,
+      "step": 2950
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 4.918223023154216e-05,
+      "loss": 0.8285,
+      "step": 3000
+    },
+    {
+      "epoch": 0.33,
+      "eval_cer": 0.13497079465235884,
+      "eval_loss": 0.7963769435882568,
+      "eval_runtime": 1149.3465,
+      "eval_samples_per_second": 6.537,
+      "eval_steps_per_second": 0.818,
+      "step": 3000
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 4.916857798165138e-05,
+      "loss": 0.8368,
+      "step": 3050
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 4.91549257317606e-05,
+      "loss": 0.7937,
+      "step": 3100
+    },
+    {
+      "epoch": 0.34,
+      "eval_cer": 0.12768571145048094,
+      "eval_loss": 0.7565470933914185,
+      "eval_runtime": 1157.2539,
+      "eval_samples_per_second": 6.492,
+      "eval_steps_per_second": 0.812,
+      "step": 3100
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 4.9141273481869816e-05,
+      "loss": 0.8224,
+      "step": 3150
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 4.912762123197903e-05,
+      "loss": 0.7887,
+      "step": 3200
+    },
+    {
+      "epoch": 0.35,
+      "eval_cer": 0.1362300293251118,
+      "eval_loss": 0.7719568014144897,
+      "eval_runtime": 1229.1352,
+      "eval_samples_per_second": 6.112,
+      "eval_steps_per_second": 0.765,
+      "step": 3200
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 4.911396898208825e-05,
+      "loss": 0.788,
+      "step": 3250
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 4.910031673219747e-05,
+      "loss": 0.8322,
+      "step": 3300
+    },
+    {
+      "epoch": 0.36,
+      "eval_cer": 0.13488182698526216,
+      "eval_loss": 0.7592117190361023,
+      "eval_runtime": 1412.8986,
+      "eval_samples_per_second": 5.317,
+      "eval_steps_per_second": 0.665,
+      "step": 3300
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 4.9086664482306686e-05,
+      "loss": 0.8167,
+      "step": 3350
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 4.90730122324159e-05,
+      "loss": 0.8116,
+      "step": 3400
+    },
+    {
+      "epoch": 0.37,
+      "eval_cer": 0.16377578779158297,
+      "eval_loss": 0.7804229259490967,
+      "eval_runtime": 1491.597,
+      "eval_samples_per_second": 5.037,
+      "eval_steps_per_second": 0.63,
+      "step": 3400
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 4.9059359982525125e-05,
+      "loss": 0.7847,
+      "step": 3450
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 4.904570773263434e-05,
+      "loss": 0.816,
+      "step": 3500
+    },
+    {
+      "epoch": 0.38,
+      "eval_cer": 0.13126152730109739,
+      "eval_loss": 0.7453513145446777,
+      "eval_runtime": 1217.9912,
+      "eval_samples_per_second": 6.168,
+      "eval_steps_per_second": 0.772,
+      "step": 3500
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 4.903205548274356e-05,
+      "loss": 0.781,
+      "step": 3550
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 4.901840323285278e-05,
+      "loss": 0.7941,
+      "step": 3600
+    },
+    {
+      "epoch": 0.39,
+      "eval_cer": 0.13394424464739718,
+      "eval_loss": 0.7644733786582947,
+      "eval_runtime": 1216.9116,
+      "eval_samples_per_second": 6.174,
+      "eval_steps_per_second": 0.772,
+      "step": 3600
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 4.9004750982961995e-05,
+      "loss": 0.7896,
+      "step": 3650
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 4.899109873307121e-05,
+      "loss": 0.7556,
+      "step": 3700
+    },
+    {
+      "epoch": 0.4,
+      "eval_cer": 0.12223130909078465,
+      "eval_loss": 0.7185453772544861,
+      "eval_runtime": 1225.4621,
+      "eval_samples_per_second": 6.131,
+      "eval_steps_per_second": 0.767,
+      "step": 3700
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 4.8977446483180426e-05,
+      "loss": 0.7616,
+      "step": 3750
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 4.896379423328965e-05,
+      "loss": 0.766,
+      "step": 3800
+    },
+    {
+      "epoch": 0.42,
+      "eval_cer": 0.12583792144154995,
+      "eval_loss": 0.7360116839408875,
+      "eval_runtime": 1167.0386,
+      "eval_samples_per_second": 6.438,
+      "eval_steps_per_second": 0.805,
+      "step": 3800
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 4.8950141983398865e-05,
+      "loss": 0.7699,
+      "step": 3850
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 4.893648973350809e-05,
+      "loss": 0.7201,
+      "step": 3900
+    },
+    {
+      "epoch": 0.43,
+      "eval_cer": 0.12545809793971413,
+      "eval_loss": 0.7248350381851196,
+      "eval_runtime": 1214.3633,
+      "eval_samples_per_second": 6.187,
+      "eval_steps_per_second": 0.774,
+      "step": 3900
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 4.89228374836173e-05,
+      "loss": 0.7671,
+      "step": 3950
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 4.890918523372652e-05,
+      "loss": 0.7502,
+      "step": 4000
+    },
+    {
+      "epoch": 0.44,
+      "eval_cer": 0.11814221823768739,
+      "eval_loss": 0.7216148376464844,
+      "eval_runtime": 1179.5688,
+      "eval_samples_per_second": 6.369,
+      "eval_steps_per_second": 0.797,
+      "step": 4000
+    },
+    {
+      "epoch": 0.44,
+      "learning_rate": 4.889553298383574e-05,
+      "loss": 0.7319,
+      "step": 4050
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 4.888188073394496e-05,
+      "loss": 0.7187,
+      "step": 4100
+    },
+    {
+      "epoch": 0.45,
+      "eval_cer": 0.11959649741138308,
+      "eval_loss": 0.7175608277320862,
+      "eval_runtime": 1154.0936,
+      "eval_samples_per_second": 6.51,
+      "eval_steps_per_second": 0.814,
+      "step": 4100
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 4.886822848405417e-05,
+      "loss": 0.725,
+      "step": 4150
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 4.885457623416339e-05,
+      "loss": 0.7606,
+      "step": 4200
+    },
+    {
+      "epoch": 0.46,
+      "eval_cer": 0.1217967362553509,
+      "eval_loss": 0.7174265384674072,
+      "eval_runtime": 1171.3267,
+      "eval_samples_per_second": 6.414,
+      "eval_steps_per_second": 0.803,
+      "step": 4200
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 4.884092398427261e-05,
+      "loss": 0.7281,
+      "step": 4250
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 4.882727173438183e-05,
+      "loss": 0.7184,
+      "step": 4300
+    },
+    {
+      "epoch": 0.47,
+      "eval_cer": 0.11260569187759417,
+      "eval_loss": 0.6905629634857178,
+      "eval_runtime": 1345.0272,
+      "eval_samples_per_second": 5.586,
+      "eval_steps_per_second": 0.699,
+      "step": 4300
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 4.881361948449104e-05,
+      "loss": 0.7349,
+      "step": 4350
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 4.8799967234600266e-05,
+      "loss": 0.7077,
+      "step": 4400
+    },
+    {
+      "epoch": 0.48,
+      "eval_cer": 0.11969915241187924,
+      "eval_loss": 0.6907770037651062,
+      "eval_runtime": 1448.5704,
+      "eval_samples_per_second": 5.186,
+      "eval_steps_per_second": 0.649,
+      "step": 4400
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 4.878631498470948e-05,
+      "loss": 0.6844,
+      "step": 4450
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 4.8772662734818704e-05,
+      "loss": 0.7288,
+      "step": 4500
+    },
+    {
+      "epoch": 0.49,
+      "eval_cer": 0.11132592620474198,
+      "eval_loss": 0.6686322093009949,
+      "eval_runtime": 1442.9609,
+      "eval_samples_per_second": 5.207,
+      "eval_steps_per_second": 0.651,
+      "step": 4500
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.875901048492792e-05,
+      "loss": 0.7375,
+      "step": 4550
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 4.8745358235037136e-05,
+      "loss": 0.7161,
+      "step": 4600
+    },
+    {
+      "epoch": 0.5,
+      "eval_cer": 0.12221077809068542,
+      "eval_loss": 0.6822640895843506,
+      "eval_runtime": 1261.344,
+      "eval_samples_per_second": 5.956,
+      "eval_steps_per_second": 0.745,
+      "step": 4600
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.873170598514635e-05,
+      "loss": 0.6956,
+      "step": 4650
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 4.871805373525557e-05,
+      "loss": 0.6666,
+      "step": 4700
+    },
+    {
+      "epoch": 0.51,
+      "eval_cer": 0.1256976262742052,
+      "eval_loss": 0.673389196395874,
+      "eval_runtime": 1470.5096,
+      "eval_samples_per_second": 5.109,
+      "eval_steps_per_second": 0.639,
+      "step": 4700
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.870440148536479e-05,
+      "loss": 0.7583,
+      "step": 4750
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 4.8690749235474006e-05,
+      "loss": 0.7239,
+      "step": 4800
+    },
+    {
+      "epoch": 0.52,
+      "eval_cer": 0.11146622137208674,
+      "eval_loss": 0.6682428121566772,
+      "eval_runtime": 1393.331,
+      "eval_samples_per_second": 5.392,
+      "eval_steps_per_second": 0.675,
+      "step": 4800
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 4.867709698558323e-05,
+      "loss": 0.7167,
+      "step": 4850
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.8663444735692444e-05,
+      "loss": 0.6723,
+      "step": 4900
+    },
+    {
+      "epoch": 0.54,
+      "eval_cer": 0.11162020387283099,
+      "eval_loss": 0.6625186204910278,
+      "eval_runtime": 1481.6224,
+      "eval_samples_per_second": 5.071,
+      "eval_steps_per_second": 0.634,
+      "step": 4900
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 4.864979248580167e-05,
+      "loss": 0.7173,
+      "step": 4950
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.863614023591088e-05,
+      "loss": 0.683,
+      "step": 5000
+    },
+    {
+      "epoch": 0.55,
+      "eval_cer": 0.11195554354111846,
+      "eval_loss": 0.6574082970619202,
+      "eval_runtime": 1193.9295,
+      "eval_samples_per_second": 6.293,
+      "eval_steps_per_second": 0.787,
+      "step": 5000
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 4.86224879860201e-05,
+      "loss": 0.6623,
+      "step": 5050
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.8608835736129314e-05,
+      "loss": 0.6982,
+      "step": 5100
+    },
+    {
+      "epoch": 0.56,
+      "eval_cer": 0.11072710536851434,
+      "eval_loss": 0.6460687518119812,
+      "eval_runtime": 1243.3246,
+      "eval_samples_per_second": 6.043,
+      "eval_steps_per_second": 0.756,
+      "step": 5100
+    },
+    {
+      "epoch": 0.56,
+      "learning_rate": 4.859518348623853e-05,
+      "loss": 0.6779,
+      "step": 5150
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.858153123634775e-05,
+      "loss": 0.6681,
+      "step": 5200
+    },
+    {
+      "epoch": 0.57,
+      "eval_cer": 0.12255638325902252,
+      "eval_loss": 0.6535465121269226,
+      "eval_runtime": 1269.3739,
+      "eval_samples_per_second": 5.919,
+      "eval_steps_per_second": 0.741,
+      "step": 5200
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 4.856787898645697e-05,
+      "loss": 0.6924,
+      "step": 5250
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.855422673656619e-05,
+      "loss": 0.6701,
+      "step": 5300
+    },
+    {
+      "epoch": 0.58,
+      "eval_cer": 0.10567647934410299,
+      "eval_loss": 0.6478390693664551,
+      "eval_runtime": 1185.9589,
+      "eval_samples_per_second": 6.335,
+      "eval_steps_per_second": 0.793,
+      "step": 5300
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 4.854057448667541e-05,
+      "loss": 0.7025,
+      "step": 5350
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 4.852692223678462e-05,
+      "loss": 0.6928,
+      "step": 5400
+    },
+    {
+      "epoch": 0.59,
+      "eval_cer": 0.10365417583432851,
+      "eval_loss": 0.6242822408676147,
+      "eval_runtime": 1155.8357,
+      "eval_samples_per_second": 6.5,
+      "eval_steps_per_second": 0.813,
+      "step": 5400
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.8513269986893845e-05,
+      "loss": 0.6665,
+      "step": 5450
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 4.849961773700306e-05,
+      "loss": 0.6919,
+      "step": 5500
+    },
+    {
+      "epoch": 0.6,
+      "eval_cer": 0.10333252349944053,
+      "eval_loss": 0.6287544369697571,
+      "eval_runtime": 1139.7191,
+      "eval_samples_per_second": 6.592,
+      "eval_steps_per_second": 0.825,
+      "step": 5500
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 4.848596548711228e-05,
+      "loss": 0.6408,
+      "step": 5550
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 4.847231323722149e-05,
+      "loss": 0.6573,
+      "step": 5600
+    },
+    {
+      "epoch": 0.61,
+      "eval_cer": 0.11020698669933376,
+      "eval_loss": 0.6342067122459412,
+      "eval_runtime": 1184.0803,
+      "eval_samples_per_second": 6.345,
+      "eval_steps_per_second": 0.794,
+      "step": 5600
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 4.845866098733071e-05,
+      "loss": 0.7322,
+      "step": 5650
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 4.844500873743993e-05,
+      "loss": 0.6518,
+      "step": 5700
+    },
+    {
+      "epoch": 0.62,
+      "eval_cer": 0.10025287348455555,
+      "eval_loss": 0.6224654316902161,
+      "eval_runtime": 1109.2466,
+      "eval_samples_per_second": 6.773,
+      "eval_steps_per_second": 0.847,
+      "step": 5700
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 4.8431356487549154e-05,
+      "loss": 0.6683,
+      "step": 5750
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 4.841770423765837e-05,
+      "loss": 0.6502,
+      "step": 5800
+    },
+    {
+      "epoch": 0.63,
+      "eval_cer": 0.09801841630708902,
+      "eval_loss": 0.6061940789222717,
+      "eval_runtime": 1144.048,
+      "eval_samples_per_second": 6.567,
+      "eval_steps_per_second": 0.822,
+      "step": 5800
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 4.8404051987767586e-05,
+      "loss": 0.6692,
+      "step": 5850
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 4.839039973787681e-05,
+      "loss": 0.6802,
+      "step": 5900
+    },
+    {
+      "epoch": 0.64,
+      "eval_cer": 0.10424273117050653,
+      "eval_loss": 0.6255876421928406,
+      "eval_runtime": 1180.3322,
+      "eval_samples_per_second": 6.365,
+      "eval_steps_per_second": 0.796,
+      "step": 5900
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 4.8376747487986024e-05,
+      "loss": 0.6465,
+      "step": 5950
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 4.836309523809524e-05,
+      "loss": 0.6015,
+      "step": 6000
+    },
+    {
+      "epoch": 0.66,
+      "eval_cer": 0.10244969049517351,
+      "eval_loss": 0.6007630825042725,
+      "eval_runtime": 1184.9976,
+      "eval_samples_per_second": 6.34,
+      "eval_steps_per_second": 0.793,
+      "step": 6000
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 4.8349442988204456e-05,
+      "loss": 0.6529,
+      "step": 6050
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 4.833579073831367e-05,
+      "loss": 0.6279,
+      "step": 6100
+    },
+    {
+      "epoch": 0.67,
+      "eval_cer": 0.0989183584781054,
+      "eval_loss": 0.5849039554595947,
+      "eval_runtime": 1147.0099,
+      "eval_samples_per_second": 6.55,
+      "eval_steps_per_second": 0.82,
+      "step": 6100
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 4.8322138488422894e-05,
+      "loss": 0.612,
+      "step": 6150
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 4.8308486238532117e-05,
+      "loss": 0.636,
+      "step": 6200
+    },
+    {
+      "epoch": 0.68,
+      "eval_cer": 0.10187482249239498,
+      "eval_loss": 0.602621853351593,
+      "eval_runtime": 1163.5869,
+      "eval_samples_per_second": 6.457,
+      "eval_steps_per_second": 0.808,
+      "step": 6200
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 4.829483398864133e-05,
+      "loss": 0.6153,
+      "step": 6250
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 4.828118173875055e-05,
+      "loss": 0.5761,
+      "step": 6300
+    },
+    {
+      "epoch": 0.69,
+      "eval_cer": 0.10323329033229424,
+      "eval_loss": 0.6136672496795654,
+      "eval_runtime": 1156.1641,
+      "eval_samples_per_second": 6.498,
+      "eval_steps_per_second": 0.813,
+      "step": 6300
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 4.8267529488859764e-05,
+      "loss": 0.6564,
+      "step": 6350
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 4.825387723896899e-05,
+      "loss": 0.6106,
+      "step": 6400
+    },
+    {
+      "epoch": 0.7,
+      "eval_cer": 0.10424957483720627,
+      "eval_loss": 0.5963040590286255,
+      "eval_runtime": 1209.0706,
+      "eval_samples_per_second": 6.214,
+      "eval_steps_per_second": 0.777,
+      "step": 6400
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 4.82402249890782e-05,
+      "loss": 0.6102,
+      "step": 6450
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 4.822657273918742e-05,
+      "loss": 0.6471,
+      "step": 6500
+    },
+    {
+      "epoch": 0.71,
+      "eval_cer": 0.09800130714033965,
+      "eval_loss": 0.5907582640647888,
+      "eval_runtime": 1159.4797,
+      "eval_samples_per_second": 6.48,
+      "eval_steps_per_second": 0.811,
+      "step": 6500
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 4.8212920489296634e-05,
+      "loss": 0.6466,
+      "step": 6550
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 4.819926823940586e-05,
+      "loss": 0.5745,
+      "step": 6600
+    },
+    {
+      "epoch": 0.72,
+      "eval_cer": 0.0992263234795939,
+      "eval_loss": 0.5862923264503479,
+      "eval_runtime": 1177.95,
+      "eval_samples_per_second": 6.378,
+      "eval_steps_per_second": 0.798,
+      "step": 6600
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 4.818561598951508e-05,
+      "loss": 0.5995,
+      "step": 6650
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 4.8171963739624295e-05,
+      "loss": 0.6058,
+      "step": 6700
+    },
+    {
+      "epoch": 0.73,
+      "eval_cer": 0.09561628929547873,
+      "eval_loss": 0.5826772451400757,
+      "eval_runtime": 1154.6872,
+      "eval_samples_per_second": 6.507,
+      "eval_steps_per_second": 0.814,
+      "step": 6700
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 4.815831148973351e-05,
+      "loss": 0.6223,
+      "step": 6750
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 4.814465923984273e-05,
+      "loss": 0.5968,
+      "step": 6800
+    },
+    {
+      "epoch": 0.74,
+      "eval_cer": 0.09658124630014268,
+      "eval_loss": 0.5713803172111511,
+      "eval_runtime": 1148.4157,
+      "eval_samples_per_second": 6.542,
+      "eval_steps_per_second": 0.819,
+      "step": 6800
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 4.813100698995195e-05,
+      "loss": 0.6244,
+      "step": 6850
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 4.8117354740061165e-05,
+      "loss": 0.6262,
+      "step": 6900
+    },
+    {
+      "epoch": 0.75,
+      "eval_cer": 0.09553074346173193,
+      "eval_loss": 0.5605288147926331,
+      "eval_runtime": 1166.3348,
+      "eval_samples_per_second": 6.442,
+      "eval_steps_per_second": 0.806,
+      "step": 6900
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 4.810370249017038e-05,
+      "loss": 0.5902,
+      "step": 6950
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 4.80900502402796e-05,
+      "loss": 0.566,
+      "step": 7000
+    },
+    {
+      "epoch": 0.76,
+      "eval_cer": 0.11679743773118761,
+      "eval_loss": 0.6162604689598083,
+      "eval_runtime": 1280.6689,
+      "eval_samples_per_second": 5.866,
+      "eval_steps_per_second": 0.734,
+      "step": 7000
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 4.807639799038882e-05,
+      "loss": 0.6245,
+      "step": 7050
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 4.806274574049804e-05,
+      "loss": 0.5711,
+      "step": 7100
+    },
+    {
+      "epoch": 0.78,
+      "eval_cer": 0.0962595939652547,
+      "eval_loss": 0.5639761090278625,
+      "eval_runtime": 1182.4487,
+      "eval_samples_per_second": 6.354,
+      "eval_steps_per_second": 0.795,
+      "step": 7100
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 4.804909349060726e-05,
+      "loss": 0.621,
+      "step": 7150
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 4.803571428571429e-05,
+      "loss": 0.6164,
+      "step": 7200
+    },
+    {
+      "epoch": 0.79,
+      "eval_cer": 0.09956166314788137,
+      "eval_loss": 0.561359703540802,
+      "eval_runtime": 1220.9677,
+      "eval_samples_per_second": 6.153,
+      "eval_steps_per_second": 0.77,
+      "step": 7200
+    },
+    {
+      "epoch": 0.79,
+      "learning_rate": 4.802206203582351e-05,
+      "loss": 0.582,
+      "step": 7250
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 4.8008409785932724e-05,
+      "loss": 0.5574,
+      "step": 7300
+    },
+    {
+      "epoch": 0.8,
+      "eval_cer": 0.09406277695463676,
+      "eval_loss": 0.5611019134521484,
+      "eval_runtime": 1157.6717,
+      "eval_samples_per_second": 6.49,
+      "eval_steps_per_second": 0.812,
+      "step": 7300
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 4.799475753604194e-05,
+      "loss": 0.5702,
+      "step": 7350
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.798110528615116e-05,
+      "loss": 0.5772,
+      "step": 7400
+    },
+    {
+      "epoch": 0.81,
+      "eval_cer": 0.09428861795572832,
+      "eval_loss": 0.5664366483688354,
+      "eval_runtime": 1143.9283,
+      "eval_samples_per_second": 6.568,
+      "eval_steps_per_second": 0.822,
+      "step": 7400
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 4.796745303626038e-05,
+      "loss": 0.5813,
+      "step": 7450
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 4.7953800786369594e-05,
+      "loss": 0.5836,
+      "step": 7500
+    },
+    {
+      "epoch": 0.82,
+      "eval_cer": 0.09800130714033965,
+      "eval_loss": 0.569098174571991,
+      "eval_runtime": 1171.894,
+      "eval_samples_per_second": 6.411,
+      "eval_steps_per_second": 0.802,
+      "step": 7500
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 4.794014853647881e-05,
+      "loss": 0.5426,
+      "step": 7550
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 4.792649628658803e-05,
+      "loss": 0.5729,
+      "step": 7600
+    },
+    {
+      "epoch": 0.83,
+      "eval_cer": 0.09232790744625155,
+      "eval_loss": 0.5581706762313843,
+      "eval_runtime": 1144.9014,
+      "eval_samples_per_second": 6.562,
+      "eval_steps_per_second": 0.821,
+      "step": 7600
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 4.7912844036697255e-05,
+      "loss": 0.5861,
+      "step": 7650
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 4.789919178680647e-05,
+      "loss": 0.5852,
+      "step": 7700
+    },
+    {
+      "epoch": 0.84,
+      "eval_cer": 0.0938848416204434,
+      "eval_loss": 0.5493680834770203,
+      "eval_runtime": 1170.4762,
+      "eval_samples_per_second": 6.419,
+      "eval_steps_per_second": 0.803,
+      "step": 7700
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 4.788553953691569e-05,
+      "loss": 0.5467,
+      "step": 7750
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 4.78718872870249e-05,
+      "loss": 0.5809,
+      "step": 7800
+    },
+    {
+      "epoch": 0.85,
+      "eval_cer": 0.0912260771075927,
+      "eval_loss": 0.5456804037094116,
+      "eval_runtime": 1175.5115,
+      "eval_samples_per_second": 6.391,
+      "eval_steps_per_second": 0.8,
+      "step": 7800
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 4.785823503713412e-05,
+      "loss": 0.5769,
+      "step": 7850
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 4.784458278724334e-05,
+      "loss": 0.6034,
+      "step": 7900
+    },
+    {
+      "epoch": 0.86,
+      "eval_cer": 0.09061014710461571,
+      "eval_loss": 0.544008195400238,
+      "eval_runtime": 1149.8414,
+      "eval_samples_per_second": 6.534,
+      "eval_steps_per_second": 0.818,
+      "step": 7900
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 4.783093053735256e-05,
+      "loss": 0.5827,
+      "step": 7950
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 4.781727828746177e-05,
+      "loss": 0.5553,
+      "step": 8000
+    },
+    {
+      "epoch": 0.87,
+      "eval_cer": 0.0900215917684377,
+      "eval_loss": 0.5491546392440796,
+      "eval_runtime": 1136.1393,
+      "eval_samples_per_second": 6.613,
+      "eval_steps_per_second": 0.827,
+      "step": 8000
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 4.7803626037570995e-05,
+      "loss": 0.6212,
+      "step": 8050
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 4.778997378768021e-05,
+      "loss": 0.5762,
+      "step": 8100
+    },
+    {
+      "epoch": 0.88,
+      "eval_cer": 0.09335103561786334,
+      "eval_loss": 0.5435011386871338,
+      "eval_runtime": 1264.0887,
+      "eval_samples_per_second": 5.943,
+      "eval_steps_per_second": 0.744,
+      "step": 8100
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 4.7776321537789434e-05,
+      "loss": 0.5516,
+      "step": 8150
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.776266928789865e-05,
+      "loss": 0.5587,
+      "step": 8200
+    },
+    {
+      "epoch": 0.9,
+      "eval_cer": 0.09445286595652219,
+      "eval_loss": 0.5462570190429688,
+      "eval_runtime": 1479.3094,
+      "eval_samples_per_second": 5.079,
+      "eval_steps_per_second": 0.635,
+      "step": 8200
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 4.7749017038007865e-05,
+      "loss": 0.5745,
+      "step": 8250
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.773536478811708e-05,
+      "loss": 0.5416,
+      "step": 8300
+    },
+    {
+      "epoch": 0.91,
+      "eval_cer": 0.08941250543216045,
+      "eval_loss": 0.5406492948532104,
+      "eval_runtime": 1459.8897,
+      "eval_samples_per_second": 5.146,
+      "eval_steps_per_second": 0.644,
+      "step": 8300
+    },
+    {
+      "epoch": 0.91,
+      "learning_rate": 4.7721712538226304e-05,
+      "loss": 0.561,
+      "step": 8350
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.770806028833552e-05,
+      "loss": 0.5443,
+      "step": 8400
+    },
+    {
+      "epoch": 0.92,
+      "eval_cer": 0.08896766709667706,
+      "eval_loss": 0.5334160327911377,
+      "eval_runtime": 1344.1845,
+      "eval_samples_per_second": 5.589,
+      "eval_steps_per_second": 0.699,
+      "step": 8400
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 4.7694408038444735e-05,
+      "loss": 0.5721,
+      "step": 8450
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 4.768075578855395e-05,
+      "loss": 0.5798,
+      "step": 8500
+    },
+    {
+      "epoch": 0.93,
+      "eval_cer": 0.09518513829339484,
+      "eval_loss": 0.5398730039596558,
+      "eval_runtime": 1463.1068,
+      "eval_samples_per_second": 5.135,
+      "eval_steps_per_second": 0.642,
+      "step": 8500
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 4.7667103538663174e-05,
+      "loss": 0.5383,
+      "step": 8550
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.7653451288772397e-05,
+      "loss": 0.57,
+      "step": 8600
+    },
+    {
+      "epoch": 0.94,
+      "eval_cer": 0.09988673731611923,
+      "eval_loss": 0.5255935788154602,
+      "eval_runtime": 1549.7656,
+      "eval_samples_per_second": 4.848,
+      "eval_steps_per_second": 0.607,
+      "step": 8600
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.763979903888161e-05,
+      "loss": 0.559,
+      "step": 8650
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 4.762614678899083e-05,
+      "loss": 0.5773,
+      "step": 8700
+    },
+    {
+      "epoch": 0.95,
+      "eval_cer": 0.08790005509151694,
+      "eval_loss": 0.528082549571991,
+      "eval_runtime": 1175.5328,
+      "eval_samples_per_second": 6.391,
+      "eval_steps_per_second": 0.8,
+      "step": 8700
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 4.7612494539100044e-05,
+      "loss": 0.5594,
+      "step": 8750
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 4.7598842289209267e-05,
+      "loss": 0.5252,
+      "step": 8800
+    },
+    {
+      "epoch": 0.96,
+      "eval_cer": 0.08552530274670563,
+      "eval_loss": 0.510725200176239,
+      "eval_runtime": 1194.9094,
+      "eval_samples_per_second": 6.288,
+      "eval_steps_per_second": 0.787,
+      "step": 8800
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 4.758519003931848e-05,
+      "loss": 0.5344,
+      "step": 8850
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 4.75715377894277e-05,
+      "loss": 0.5453,
+      "step": 8900
+    },
+    {
+      "epoch": 0.97,
+      "eval_cer": 0.09173935211007353,
+      "eval_loss": 0.5302219390869141,
+      "eval_runtime": 1258.3889,
+      "eval_samples_per_second": 5.97,
+      "eval_steps_per_second": 0.747,
+      "step": 8900
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 4.7557885539536914e-05,
+      "loss": 0.5709,
+      "step": 8950
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 4.7544233289646137e-05,
+      "loss": 0.5435,
+      "step": 9000
+    },
+    {
+      "epoch": 0.98,
+      "eval_cer": 0.08849887592774457,
+      "eval_loss": 0.5267897248268127,
+      "eval_runtime": 1447.3374,
+      "eval_samples_per_second": 5.191,
+      "eval_steps_per_second": 0.649,
+      "step": 9000
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.753058103975536e-05,
+      "loss": 0.5276,
+      "step": 9050
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 4.7516928789864575e-05,
+      "loss": 0.5448,
+      "step": 9100
+    },
+    {
+      "epoch": 0.99,
+      "eval_cer": 0.09276248028168532,
+      "eval_loss": 0.521221399307251,
+      "eval_runtime": 1481.2749,
+      "eval_samples_per_second": 5.072,
+      "eval_steps_per_second": 0.635,
+      "step": 9100
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.750327653997379e-05,
+      "loss": 0.5232,
+      "step": 9150
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 4.7489897335080826e-05,
+      "loss": 0.4876,
+      "step": 9200
+    },
+    {
+      "epoch": 1.0,
+      "eval_cer": 0.08981628176744536,
+      "eval_loss": 0.5108721256256104,
+      "eval_runtime": 1277.6548,
+      "eval_samples_per_second": 5.88,
+      "eval_steps_per_second": 0.736,
+      "step": 9200
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 4.747624508519004e-05,
+      "loss": 0.4831,
+      "step": 9250
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 4.746259283529926e-05,
+      "loss": 0.4454,
+      "step": 9300
+    },
+    {
+      "epoch": 1.02,
+      "eval_cer": 0.08573403458104784,
+      "eval_loss": 0.5104303359985352,
+      "eval_runtime": 1166.2306,
+      "eval_samples_per_second": 6.442,
+      "eval_steps_per_second": 0.806,
+      "step": 9300
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 4.744894058540848e-05,
+      "loss": 0.492,
+      "step": 9350
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 4.7435288335517696e-05,
+      "loss": 0.4575,
+      "step": 9400
+    },
+    {
+      "epoch": 1.03,
+      "eval_cer": 0.0861822947498811,
+      "eval_loss": 0.5089101195335388,
+      "eval_runtime": 1209.7895,
+      "eval_samples_per_second": 6.21,
+      "eval_steps_per_second": 0.777,
+      "step": 9400
+    },
+    {
+      "epoch": 1.03,
+      "learning_rate": 4.742163608562691e-05,
+      "loss": 0.4754,
+      "step": 9450
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 4.740798383573613e-05,
+      "loss": 0.5084,
+      "step": 9500
+    },
+    {
+      "epoch": 1.04,
+      "eval_cer": 0.08640471391762278,
+      "eval_loss": 0.5144868493080139,
+      "eval_runtime": 1163.3162,
+      "eval_samples_per_second": 6.458,
+      "eval_steps_per_second": 0.808,
+      "step": 9500
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 4.739433158584535e-05,
+      "loss": 0.4832,
+      "step": 9550
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 4.738067933595457e-05,
+      "loss": 0.474,
+      "step": 9600
+    },
+    {
+      "epoch": 1.05,
+      "eval_cer": 0.08605568691593582,
+      "eval_loss": 0.5110841989517212,
+      "eval_runtime": 1149.6213,
+      "eval_samples_per_second": 6.535,
+      "eval_steps_per_second": 0.818,
+      "step": 9600
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 4.736702708606379e-05,
+      "loss": 0.479,
+      "step": 9650
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 4.7353374836173004e-05,
+      "loss": 0.4837,
+      "step": 9700
+    },
+    {
+      "epoch": 1.06,
+      "eval_cer": 0.08458429857549078,
+      "eval_loss": 0.5060144066810608,
+      "eval_runtime": 1137.2697,
+      "eval_samples_per_second": 6.606,
+      "eval_steps_per_second": 0.827,
+      "step": 9700
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 4.733972258628222e-05,
+      "loss": 0.4502,
+      "step": 9750
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 4.7326070336391436e-05,
+      "loss": 0.4898,
+      "step": 9800
+    },
+    {
+      "epoch": 1.07,
+      "eval_cer": 0.08435161390769946,
+      "eval_loss": 0.5079366564750671,
+      "eval_runtime": 1150.9922,
+      "eval_samples_per_second": 6.527,
+      "eval_steps_per_second": 0.817,
+      "step": 9800
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 4.731241808650066e-05,
+      "loss": 0.4827,
+      "step": 9850
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 4.7298765836609874e-05,
+      "loss": 0.47,
+      "step": 9900
+    },
+    {
+      "epoch": 1.08,
+      "eval_cer": 0.08341060973648462,
+      "eval_loss": 0.4975820481777191,
+      "eval_runtime": 1149.8902,
+      "eval_samples_per_second": 6.534,
+      "eval_steps_per_second": 0.817,
+      "step": 9900
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 4.728511358671909e-05,
+      "loss": 0.4675,
+      "step": 9950
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 4.727146133682831e-05,
+      "loss": 0.4593,
+      "step": 10000
+    },
+    {
+      "epoch": 1.09,
+      "eval_cer": 0.08362618523752656,
+      "eval_loss": 0.4920589327812195,
+      "eval_runtime": 1171.3132,
+      "eval_samples_per_second": 6.414,
+      "eval_steps_per_second": 0.803,
+      "step": 10000
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 4.7257809086937535e-05,
+      "loss": 0.4898,
+      "step": 10050
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 4.724415683704675e-05,
+      "loss": 0.4723,
+      "step": 10100
+    },
+    {
+      "epoch": 1.1,
+      "eval_cer": 0.087362827255587,
+      "eval_loss": 0.5022254586219788,
+      "eval_runtime": 1155.2292,
+      "eval_samples_per_second": 6.503,
+      "eval_steps_per_second": 0.814,
+      "step": 10100
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 4.723050458715597e-05,
+      "loss": 0.4627,
+      "step": 10150
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 4.721685233726518e-05,
+      "loss": 0.4562,
+      "step": 10200
+    },
+    {
+      "epoch": 1.11,
+      "eval_cer": 0.08214110956368204,
+      "eval_loss": 0.4981001019477844,
+      "eval_runtime": 1146.0235,
+      "eval_samples_per_second": 6.556,
+      "eval_steps_per_second": 0.82,
+      "step": 10200
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 4.72032000873744e-05,
+      "loss": 0.43,
+      "step": 10250
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 4.718954783748362e-05,
+      "loss": 0.4701,
+      "step": 10300
+    },
+    {
+      "epoch": 1.12,
+      "eval_cer": 0.08535421107921203,
+      "eval_loss": 0.5037410259246826,
+      "eval_runtime": 1149.3397,
+      "eval_samples_per_second": 6.537,
+      "eval_steps_per_second": 0.818,
+      "step": 10300
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 4.717589558759284e-05,
+      "loss": 0.4429,
+      "step": 10350
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 4.716224333770205e-05,
+      "loss": 0.4207,
+      "step": 10400
+    },
+    {
+      "epoch": 1.14,
+      "eval_cer": 0.08262016623266413,
+      "eval_loss": 0.4911841154098511,
+      "eval_runtime": 1168.0546,
+      "eval_samples_per_second": 6.432,
+      "eval_steps_per_second": 0.805,
+      "step": 10400
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 4.7148591087811275e-05,
+      "loss": 0.4737,
+      "step": 10450
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 4.713493883792049e-05,
+      "loss": 0.4413,
+      "step": 10500
+    },
+    {
+      "epoch": 1.15,
+      "eval_cer": 0.08833120609360083,
+      "eval_loss": 0.49298039078712463,
+      "eval_runtime": 1177.0621,
+      "eval_samples_per_second": 6.383,
+      "eval_steps_per_second": 0.799,
+      "step": 10500
+    },
+    {
+      "epoch": 1.15,
+      "learning_rate": 4.7121286588029714e-05,
+      "loss": 0.4406,
+      "step": 10550
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 4.710763433813893e-05,
+      "loss": 0.4868,
+      "step": 10600
+    },
+    {
+      "epoch": 1.16,
+      "eval_cer": 0.08830383142680186,
+      "eval_loss": 0.5031660795211792,
+      "eval_runtime": 1196.51,
+      "eval_samples_per_second": 6.279,
+      "eval_steps_per_second": 0.786,
+      "step": 10600
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 4.7093982088248145e-05,
+      "loss": 0.4756,
+      "step": 10650
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 4.708032983835736e-05,
+      "loss": 0.4392,
+      "step": 10700
+    },
+    {
+      "epoch": 1.17,
+      "eval_cer": 0.09982514431582153,
+      "eval_loss": 0.4886401891708374,
+      "eval_runtime": 1325.7127,
+      "eval_samples_per_second": 5.667,
+      "eval_steps_per_second": 0.709,
+      "step": 10700
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 4.706667758846658e-05,
+      "loss": 0.4553,
+      "step": 10750
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 4.70530253385758e-05,
+      "loss": 0.4832,
+      "step": 10800
+    },
+    {
+      "epoch": 1.18,
+      "eval_cer": 0.08149096122720631,
+      "eval_loss": 0.4746004343032837,
+      "eval_runtime": 1165.4679,
+      "eval_samples_per_second": 6.446,
+      "eval_steps_per_second": 0.807,
+      "step": 10800
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 4.7039373088685015e-05,
+      "loss": 0.4311,
+      "step": 10850
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 4.702572083879424e-05,
+      "loss": 0.4427,
+      "step": 10900
+    },
+    {
+      "epoch": 1.19,
+      "eval_cer": 0.08554583374680487,
+      "eval_loss": 0.4992435574531555,
+      "eval_runtime": 1168.4072,
+      "eval_samples_per_second": 6.43,
+      "eval_steps_per_second": 0.805,
+      "step": 10900
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 4.7012068588903454e-05,
+      "loss": 0.4384,
+      "step": 10950
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 4.6998416339012676e-05,
+      "loss": 0.4192,
+      "step": 11000
+    },
+    {
+      "epoch": 1.2,
+      "eval_cer": 0.08168258389479915,
+      "eval_loss": 0.4806784689426422,
+      "eval_runtime": 1157.026,
+      "eval_samples_per_second": 6.493,
+      "eval_steps_per_second": 0.812,
+      "step": 11000
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 4.698476408912189e-05,
+      "loss": 0.4645,
+      "step": 11050
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 4.697111183923111e-05,
+      "loss": 0.4342,
+      "step": 11100
+    },
+    {
+      "epoch": 1.21,
+      "eval_cer": 0.08099479539147485,
+      "eval_loss": 0.48337244987487793,
+      "eval_runtime": 1152.0989,
+      "eval_samples_per_second": 6.521,
+      "eval_steps_per_second": 0.816,
+      "step": 11100
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 4.6957459589340324e-05,
+      "loss": 0.4268,
+      "step": 11150
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 4.694380733944954e-05,
+      "loss": 0.4439,
+      "step": 11200
+    },
+    {
+      "epoch": 1.22,
+      "eval_cer": 0.0805328478892421,
+      "eval_loss": 0.47774767875671387,
+      "eval_runtime": 1151.292,
+      "eval_samples_per_second": 6.526,
+      "eval_steps_per_second": 0.816,
+      "step": 11200
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 4.693015508955876e-05,
+      "loss": 0.4213,
+      "step": 11250
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 4.69167758846658e-05,
+      "loss": 0.4614,
+      "step": 11300
+    },
+    {
+      "epoch": 1.23,
+      "eval_cer": 0.08316423773529381,
+      "eval_loss": 0.47571513056755066,
+      "eval_runtime": 1198.3379,
+      "eval_samples_per_second": 6.27,
+      "eval_steps_per_second": 0.784,
+      "step": 11300
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 4.690312363477501e-05,
+      "loss": 0.4417,
+      "step": 11350
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 4.688947138488423e-05,
+      "loss": 0.4773,
+      "step": 11400
+    },
+    {
+      "epoch": 1.25,
+      "eval_cer": 0.0813438223931618,
+      "eval_loss": 0.4786456525325775,
+      "eval_runtime": 1170.4782,
+      "eval_samples_per_second": 6.419,
+      "eval_steps_per_second": 0.803,
+      "step": 11400
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 4.687581913499345e-05,
+      "loss": 0.4111,
+      "step": 11450
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 4.686216688510267e-05,
+      "loss": 0.4483,
+      "step": 11500
+    },
+    {
+      "epoch": 1.26,
+      "eval_cer": 0.08518311941171841,
+      "eval_loss": 0.4720217287540436,
+      "eval_runtime": 1220.5252,
+      "eval_samples_per_second": 6.156,
+      "eval_steps_per_second": 0.77,
+      "step": 11500
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 4.684851463521189e-05,
+      "loss": 0.467,
+      "step": 11550
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 4.6834862385321106e-05,
+      "loss": 0.4479,
+      "step": 11600
+    },
+    {
+      "epoch": 1.27,
+      "eval_cer": 0.08453981474194244,
+      "eval_loss": 0.470368891954422,
+      "eval_runtime": 1179.3323,
+      "eval_samples_per_second": 6.371,
+      "eval_steps_per_second": 0.797,
+      "step": 11600
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 4.682121013543032e-05,
+      "loss": 0.435,
+      "step": 11650
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 4.680755788553954e-05,
+      "loss": 0.4549,
+      "step": 11700
+    },
+    {
+      "epoch": 1.28,
+      "eval_cer": 0.0796705458850743,
+      "eval_loss": 0.47497525811195374,
+      "eval_runtime": 1142.2374,
+      "eval_samples_per_second": 6.577,
+      "eval_steps_per_second": 0.823,
+      "step": 11700
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 4.679390563564875e-05,
+      "loss": 0.4084,
+      "step": 11750
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 4.6780253385757976e-05,
+      "loss": 0.4918,
+      "step": 11800
+    },
+    {
+      "epoch": 1.29,
+      "eval_cer": 0.07762771137520061,
+      "eval_loss": 0.46993914246559143,
+      "eval_runtime": 1152.0072,
+      "eval_samples_per_second": 6.522,
+      "eval_steps_per_second": 0.816,
+      "step": 11800
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 4.676660113586719e-05,
+      "loss": 0.4552,
+      "step": 11850
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 4.6752948885976414e-05,
+      "loss": 0.471,
+      "step": 11900
+    },
+    {
+      "epoch": 1.3,
+      "eval_cer": 0.07532139569738674,
+      "eval_loss": 0.4603424072265625,
+      "eval_runtime": 1141.6643,
+      "eval_samples_per_second": 6.581,
+      "eval_steps_per_second": 0.823,
+      "step": 11900
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 4.673929663608563e-05,
+      "loss": 0.3913,
+      "step": 11950
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 4.6725644386194846e-05,
+      "loss": 0.476,
+      "step": 12000
+    },
+    {
+      "epoch": 1.31,
+      "eval_cer": 0.07731974637371211,
+      "eval_loss": 0.46096158027648926,
+      "eval_runtime": 1151.1862,
+      "eval_samples_per_second": 6.526,
+      "eval_steps_per_second": 0.817,
+      "step": 12000
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 4.671199213630407e-05,
+      "loss": 0.4183,
+      "step": 12050
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 4.6698339886413284e-05,
+      "loss": 0.4305,
+      "step": 12100
+    },
+    {
+      "epoch": 1.32,
+      "eval_cer": 0.07767903887544869,
+      "eval_loss": 0.46618959307670593,
+      "eval_runtime": 1147.4361,
+      "eval_samples_per_second": 6.548,
+      "eval_steps_per_second": 0.819,
+      "step": 12100
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 4.66846876365225e-05,
+      "loss": 0.4616,
+      "step": 12150
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 4.6671035386631716e-05,
+      "loss": 0.4678,
+      "step": 12200
+    },
+    {
+      "epoch": 1.33,
+      "eval_cer": 0.08326689273578998,
+      "eval_loss": 0.4704437553882599,
+      "eval_runtime": 1193.667,
+      "eval_samples_per_second": 6.294,
+      "eval_steps_per_second": 0.787,
+      "step": 12200
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 4.665738313674094e-05,
+      "loss": 0.4114,
+      "step": 12250
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 4.6643730886850154e-05,
+      "loss": 0.4238,
+      "step": 12300
+    },
+    {
+      "epoch": 1.34,
+      "eval_cer": 0.07919491104944207,
+      "eval_loss": 0.4706832468509674,
+      "eval_runtime": 1155.3699,
+      "eval_samples_per_second": 6.503,
+      "eval_steps_per_second": 0.814,
+      "step": 12300
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 4.663007863695938e-05,
+      "loss": 0.4057,
+      "step": 12350
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 4.661642638706859e-05,
+      "loss": 0.4633,
+      "step": 12400
+    },
+    {
+      "epoch": 1.35,
+      "eval_cer": 0.07833260904527428,
+      "eval_loss": 0.46324703097343445,
+      "eval_runtime": 1156.5395,
+      "eval_samples_per_second": 6.496,
+      "eval_steps_per_second": 0.813,
+      "step": 12400
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 4.660277413717781e-05,
+      "loss": 0.4525,
+      "step": 12450
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 4.658912188728703e-05,
+      "loss": 0.4248,
+      "step": 12500
+    },
+    {
+      "epoch": 1.37,
+      "eval_cer": 0.07947207955078171,
+      "eval_loss": 0.4599738121032715,
+      "eval_runtime": 1174.5926,
+      "eval_samples_per_second": 6.396,
+      "eval_steps_per_second": 0.8,
+      "step": 12500
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 4.657546963739625e-05,
+      "loss": 0.4218,
+      "step": 12550
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 4.656181738750546e-05,
+      "loss": 0.4595,
+      "step": 12600
+    },
+    {
+      "epoch": 1.38,
+      "eval_cer": 0.07839762387892185,
+      "eval_loss": 0.455277681350708,
+      "eval_runtime": 1183.3222,
+      "eval_samples_per_second": 6.349,
+      "eval_steps_per_second": 0.794,
+      "step": 12600
+    },
+    {
+      "epoch": 1.38,
+      "learning_rate": 4.654816513761468e-05,
+      "loss": 0.462,
+      "step": 12650
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 4.6534512887723894e-05,
+      "loss": 0.4352,
+      "step": 12700
+    },
+    {
+      "epoch": 1.39,
+      "eval_cer": 0.07753874370810393,
+      "eval_loss": 0.45391160249710083,
+      "eval_runtime": 1160.3415,
+      "eval_samples_per_second": 6.475,
+      "eval_steps_per_second": 0.81,
+      "step": 12700
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 4.652086063783312e-05,
+      "loss": 0.4388,
+      "step": 12750
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 4.650720838794233e-05,
+      "loss": 0.4238,
+      "step": 12800
+    },
+    {
+      "epoch": 1.4,
+      "eval_cer": 0.07942075205053363,
+      "eval_loss": 0.4653577208518982,
+      "eval_runtime": 1157.7661,
+      "eval_samples_per_second": 6.489,
+      "eval_steps_per_second": 0.812,
+      "step": 12800
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 4.6493556138051555e-05,
+      "loss": 0.4501,
+      "step": 12850
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 4.647990388816077e-05,
+      "loss": 0.4153,
+      "step": 12900
+    },
+    {
+      "epoch": 1.41,
+      "eval_cer": 0.07721709137321595,
+      "eval_loss": 0.45755788683891296,
+      "eval_runtime": 1151.6369,
+      "eval_samples_per_second": 6.524,
+      "eval_steps_per_second": 0.816,
+      "step": 12900
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 4.646625163826999e-05,
+      "loss": 0.4314,
+      "step": 12950
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 4.645259938837921e-05,
+      "loss": 0.4354,
+      "step": 13000
+    },
+    {
+      "epoch": 1.42,
+      "eval_cer": 0.0735659951889023,
+      "eval_loss": 0.44789427518844604,
+      "eval_runtime": 1142.8652,
+      "eval_samples_per_second": 6.574,
+      "eval_steps_per_second": 0.822,
+      "step": 13000
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 4.6438947138488425e-05,
+      "loss": 0.3863,
+      "step": 13050
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 4.642529488859764e-05,
+      "loss": 0.3837,
+      "step": 13100
+    },
+    {
+      "epoch": 1.43,
+      "eval_cer": 0.07796647287683796,
+      "eval_loss": 0.4503418803215027,
+      "eval_runtime": 1162.223,
+      "eval_samples_per_second": 6.464,
+      "eval_steps_per_second": 0.809,
+      "step": 13100
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 4.641164263870686e-05,
+      "loss": 0.4048,
+      "step": 13150
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 4.639799038881608e-05,
+      "loss": 0.4325,
+      "step": 13200
+    },
+    {
+      "epoch": 1.44,
+      "eval_cer": 0.07542062886453305,
+      "eval_loss": 0.4409065842628479,
+      "eval_runtime": 1164.1583,
+      "eval_samples_per_second": 6.454,
+      "eval_steps_per_second": 0.807,
+      "step": 13200
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 4.6384338138925295e-05,
+      "loss": 0.4185,
+      "step": 13250
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 4.637068588903452e-05,
+      "loss": 0.3799,
+      "step": 13300
+    },
+    {
+      "epoch": 1.45,
+      "eval_cer": 0.07549590919823022,
+      "eval_loss": 0.44226187467575073,
+      "eval_runtime": 1167.1145,
+      "eval_samples_per_second": 6.437,
+      "eval_steps_per_second": 0.805,
+      "step": 13300
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 4.6357033639143734e-05,
+      "loss": 0.4415,
+      "step": 13350
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 4.634338138925295e-05,
+      "loss": 0.4073,
+      "step": 13400
+    },
+    {
+      "epoch": 1.46,
+      "eval_cer": 0.07687490803822872,
+      "eval_loss": 0.4516383111476898,
+      "eval_runtime": 1152.5579,
+      "eval_samples_per_second": 6.519,
+      "eval_steps_per_second": 0.816,
+      "step": 13400
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 4.632972913936217e-05,
+      "loss": 0.4191,
+      "step": 13450
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 4.631607688947139e-05,
+      "loss": 0.4051,
+      "step": 13500
+    },
+    {
+      "epoch": 1.47,
+      "eval_cer": 0.0723854626831964,
+      "eval_loss": 0.44055524468421936,
+      "eval_runtime": 1134.572,
+      "eval_samples_per_second": 6.622,
+      "eval_steps_per_second": 0.829,
+      "step": 13500
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 4.6302424639580604e-05,
+      "loss": 0.4252,
+      "step": 13550
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 4.628877238968982e-05,
+      "loss": 0.4494,
+      "step": 13600
+    },
+    {
+      "epoch": 1.49,
+      "eval_cer": 0.07588942003346553,
+      "eval_loss": 0.4374730885028839,
+      "eval_runtime": 1179.3142,
+      "eval_samples_per_second": 6.371,
+      "eval_steps_per_second": 0.797,
+      "step": 13600
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 4.6275120139799035e-05,
+      "loss": 0.4612,
+      "step": 13650
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 4.626146788990826e-05,
+      "loss": 0.4158,
+      "step": 13700
+    },
+    {
+      "epoch": 1.5,
+      "eval_cer": 0.07792541087663948,
+      "eval_loss": 0.44300341606140137,
+      "eval_runtime": 1180.7216,
+      "eval_samples_per_second": 6.363,
+      "eval_steps_per_second": 0.796,
+      "step": 13700
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 4.624781564001748e-05,
+      "loss": 0.4,
+      "step": 13750
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 4.6234163390126696e-05,
+      "loss": 0.424,
+      "step": 13800
+    },
+    {
+      "epoch": 1.51,
+      "eval_cer": 0.07315879702026752,
+      "eval_loss": 0.4424927234649658,
+      "eval_runtime": 1179.1114,
+      "eval_samples_per_second": 6.372,
+      "eval_steps_per_second": 0.797,
+      "step": 13800
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 4.622051114023591e-05,
+      "loss": 0.4021,
+      "step": 13850
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 4.6206858890345135e-05,
+      "loss": 0.4351,
+      "step": 13900
+    },
+    {
+      "epoch": 1.52,
+      "eval_cer": 0.07417165969182969,
+      "eval_loss": 0.44169941544532776,
+      "eval_runtime": 1443.9806,
+      "eval_samples_per_second": 5.203,
+      "eval_steps_per_second": 0.651,
+      "step": 13900
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 4.619320664045435e-05,
+      "loss": 0.4413,
+      "step": 13950
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 4.6179554390563567e-05,
+      "loss": 0.4298,
+      "step": 14000
+    },
+    {
+      "epoch": 1.53,
+      "eval_cer": 0.08301025523454957,
+      "eval_loss": 0.44149938225746155,
+      "eval_runtime": 1516.9826,
+      "eval_samples_per_second": 4.953,
+      "eval_steps_per_second": 0.62,
+      "step": 14000
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 4.616590214067278e-05,
+      "loss": 0.4379,
+      "step": 14050
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 4.6152249890782e-05,
+      "loss": 0.409,
+      "step": 14100
+    },
+    {
+      "epoch": 1.54,
+      "eval_cer": 0.07659773953688907,
+      "eval_loss": 0.45608383417129517,
+      "eval_runtime": 1213.6853,
+      "eval_samples_per_second": 6.19,
+      "eval_steps_per_second": 0.775,
+      "step": 14100
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 4.613859764089122e-05,
+      "loss": 0.3977,
+      "step": 14150
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 4.612494539100044e-05,
+      "loss": 0.4253,
+      "step": 14200
+    },
+    {
+      "epoch": 1.55,
+      "eval_cer": 0.07670381637073512,
+      "eval_loss": 0.4387846887111664,
+      "eval_runtime": 1160.8358,
+      "eval_samples_per_second": 6.472,
+      "eval_steps_per_second": 0.81,
+      "step": 14200
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 4.611129314110966e-05,
+      "loss": 0.4107,
+      "step": 14250
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 4.6097640891218875e-05,
+      "loss": 0.4413,
+      "step": 14300
+    },
+    {
+      "epoch": 1.56,
+      "eval_cer": 0.07185165668061634,
+      "eval_loss": 0.42884722352027893,
+      "eval_runtime": 1159.6073,
+      "eval_samples_per_second": 6.479,
+      "eval_steps_per_second": 0.811,
+      "step": 14300
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 4.608398864132809e-05,
+      "loss": 0.4288,
+      "step": 14350
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 4.607033639143731e-05,
+      "loss": 0.4253,
+      "step": 14400
+    },
+    {
+      "epoch": 1.57,
+      "eval_cer": 0.07265236568448644,
+      "eval_loss": 0.4315667152404785,
+      "eval_runtime": 1149.883,
+      "eval_samples_per_second": 6.534,
+      "eval_steps_per_second": 0.817,
+      "step": 14400
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 4.605668414154653e-05,
+      "loss": 0.396,
+      "step": 14450
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 4.6043031891655745e-05,
+      "loss": 0.4063,
+      "step": 14500
+    },
+    {
+      "epoch": 1.58,
+      "eval_cer": 0.07520847519684096,
+      "eval_loss": 0.42946287989616394,
+      "eval_runtime": 1172.1069,
+      "eval_samples_per_second": 6.41,
+      "eval_steps_per_second": 0.802,
+      "step": 14500
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 4.602937964176496e-05,
+      "loss": 0.4058,
+      "step": 14550
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 4.6015727391874183e-05,
+      "loss": 0.4054,
+      "step": 14600
+    },
+    {
+      "epoch": 1.59,
+      "eval_cer": 0.07184823484726646,
+      "eval_loss": 0.43230387568473816,
+      "eval_runtime": 1137.0608,
+      "eval_samples_per_second": 6.607,
+      "eval_steps_per_second": 0.827,
+      "step": 14600
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 4.6002075141983406e-05,
+      "loss": 0.4349,
+      "step": 14650
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 4.598842289209262e-05,
+      "loss": 0.4117,
+      "step": 14700
+    },
+    {
+      "epoch": 1.61,
+      "eval_cer": 0.0719064060142143,
+      "eval_loss": 0.4254385530948639,
+      "eval_runtime": 1163.8583,
+      "eval_samples_per_second": 6.455,
+      "eval_steps_per_second": 0.808,
+      "step": 14700
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 4.597477064220184e-05,
+      "loss": 0.3811,
+      "step": 14750
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 4.5961118392311053e-05,
+      "loss": 0.3963,
+      "step": 14800
+    },
+    {
+      "epoch": 1.62,
+      "eval_cer": 0.07202274834810996,
+      "eval_loss": 0.42654383182525635,
+      "eval_runtime": 1155.0915,
+      "eval_samples_per_second": 6.504,
+      "eval_steps_per_second": 0.814,
+      "step": 14800
+    },
+    {
+      "epoch": 1.62,
+      "learning_rate": 4.5947466142420276e-05,
+      "loss": 0.3831,
+      "step": 14850
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 4.593381389252949e-05,
+      "loss": 0.4148,
+      "step": 14900
+    },
+    {
+      "epoch": 1.63,
+      "eval_cer": 0.07073271717520813,
+      "eval_loss": 0.42668265104293823,
+      "eval_runtime": 1139.0726,
+      "eval_samples_per_second": 6.596,
+      "eval_steps_per_second": 0.825,
+      "step": 14900
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 4.592016164263871e-05,
+      "loss": 0.4053,
+      "step": 14950
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 4.5906509392747924e-05,
+      "loss": 0.4056,
+      "step": 15000
+    },
+    {
+      "epoch": 1.64,
+      "eval_cer": 0.07124257034433909,
+      "eval_loss": 0.41616320610046387,
+      "eval_runtime": 1406.7221,
+      "eval_samples_per_second": 5.341,
+      "eval_steps_per_second": 0.668,
+      "step": 15000
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 4.5892857142857146e-05,
+      "loss": 0.3748,
+      "step": 15050
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 4.587920489296637e-05,
+      "loss": 0.3842,
+      "step": 15100
+    },
+    {
+      "epoch": 1.65,
+      "eval_cer": 0.06995253917143728,
+      "eval_loss": 0.41423478722572327,
+      "eval_runtime": 1316.6097,
+      "eval_samples_per_second": 5.706,
+      "eval_steps_per_second": 0.714,
+      "step": 15100
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 4.5865552643075585e-05,
+      "loss": 0.3705,
+      "step": 15150
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 4.58519003931848e-05,
+      "loss": 0.4091,
+      "step": 15200
+    },
+    {
+      "epoch": 1.66,
+      "eval_cer": 0.06825530982990066,
+      "eval_loss": 0.4133824408054352,
+      "eval_runtime": 1388.2867,
+      "eval_samples_per_second": 5.412,
+      "eval_steps_per_second": 0.677,
+      "step": 15200
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 4.5838248143294016e-05,
+      "loss": 0.3925,
+      "step": 15250
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 4.582459589340323e-05,
+      "loss": 0.3844,
+      "step": 15300
+    },
+    {
+      "epoch": 1.67,
+      "eval_cer": 0.07246416485024346,
+      "eval_loss": 0.42426353693008423,
+      "eval_runtime": 1364.96,
+      "eval_samples_per_second": 5.504,
+      "eval_steps_per_second": 0.689,
+      "step": 15300
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 4.5810943643512455e-05,
+      "loss": 0.4009,
+      "step": 15350
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 4.579729139362167e-05,
+      "loss": 0.4136,
+      "step": 15400
+    },
+    {
+      "epoch": 1.68,
+      "eval_cer": 0.07132811617808589,
+      "eval_loss": 0.4231643080711365,
+      "eval_runtime": 1408.2259,
+      "eval_samples_per_second": 5.335,
+      "eval_steps_per_second": 0.668,
+      "step": 15400
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 4.5783639143730886e-05,
+      "loss": 0.4087,
+      "step": 15450
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 4.576998689384011e-05,
+      "loss": 0.4216,
+      "step": 15500
+    },
+    {
+      "epoch": 1.69,
+      "eval_cer": 0.07285425385212889,
+      "eval_loss": 0.4176812171936035,
+      "eval_runtime": 1387.2566,
+      "eval_samples_per_second": 5.416,
+      "eval_steps_per_second": 0.678,
+      "step": 15500
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 4.5756334643949325e-05,
+      "loss": 0.4044,
+      "step": 15550
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 4.574268239405855e-05,
+      "loss": 0.4197,
+      "step": 15600
+    },
+    {
+      "epoch": 1.7,
+      "eval_cer": 0.07007230333868279,
+      "eval_loss": 0.4150015711784363,
+      "eval_runtime": 1151.9702,
+      "eval_samples_per_second": 6.522,
+      "eval_steps_per_second": 0.816,
+      "step": 15600
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 4.572903014416776e-05,
+      "loss": 0.3793,
+      "step": 15650
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 4.571537789427698e-05,
+      "loss": 0.3867,
+      "step": 15700
+    },
+    {
+      "epoch": 1.71,
+      "eval_cer": 0.06928185983486232,
+      "eval_loss": 0.4184330999851227,
+      "eval_runtime": 1160.7464,
+      "eval_samples_per_second": 6.473,
+      "eval_steps_per_second": 0.81,
+      "step": 15700
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 4.5701725644386195e-05,
+      "loss": 0.4238,
+      "step": 15750
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 4.568807339449542e-05,
+      "loss": 0.4149,
+      "step": 15800
+    },
+    {
+      "epoch": 1.73,
+      "eval_cer": 0.07171820517997132,
+      "eval_loss": 0.41364967823028564,
+      "eval_runtime": 1161.925,
+      "eval_samples_per_second": 6.466,
+      "eval_steps_per_second": 0.809,
+      "step": 15800
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 4.567442114460463e-05,
+      "loss": 0.4028,
+      "step": 15850
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 4.566076889471385e-05,
+      "loss": 0.4122,
+      "step": 15900
+    },
+    {
+      "epoch": 1.74,
+      "eval_cer": 0.07077035734205672,
+      "eval_loss": 0.4160652160644531,
+      "eval_runtime": 1143.2321,
+      "eval_samples_per_second": 6.572,
+      "eval_steps_per_second": 0.822,
+      "step": 15900
+    },
+    {
+      "epoch": 1.74,
+      "learning_rate": 4.5647116644823065e-05,
+      "loss": 0.4263,
+      "step": 15950
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 4.56337374399301e-05,
+      "loss": 0.424,
+      "step": 16000
+    },
+    {
+      "epoch": 1.75,
+      "eval_cer": 0.07280634818523068,
+      "eval_loss": 0.4145605266094208,
+      "eval_runtime": 1166.7151,
+      "eval_samples_per_second": 6.439,
+      "eval_steps_per_second": 0.806,
+      "step": 16000
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 4.562008519003932e-05,
+      "loss": 0.4219,
+      "step": 16050
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 4.560643294014854e-05,
+      "loss": 0.4096,
+      "step": 16100
+    },
+    {
+      "epoch": 1.76,
+      "eval_cer": 0.07225201118255138,
+      "eval_loss": 0.41958579421043396,
+      "eval_runtime": 1161.2372,
+      "eval_samples_per_second": 6.47,
+      "eval_steps_per_second": 0.809,
+      "step": 16100
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 4.559278069025776e-05,
+      "loss": 0.3963,
+      "step": 16150
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 4.5579128440366976e-05,
+      "loss": 0.3692,
+      "step": 16200
+    },
+    {
+      "epoch": 1.77,
+      "eval_cer": 0.07155737901252733,
+      "eval_loss": 0.4091661870479584,
+      "eval_runtime": 1183.6151,
+      "eval_samples_per_second": 6.348,
+      "eval_steps_per_second": 0.794,
+      "step": 16200
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 4.556547619047619e-05,
+      "loss": 0.3938,
+      "step": 16250
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 4.555182394058541e-05,
+      "loss": 0.4094,
+      "step": 16300
+    },
+    {
+      "epoch": 1.78,
+      "eval_cer": 0.07196115534781225,
+      "eval_loss": 0.40848201513290405,
+      "eval_runtime": 1175.7709,
+      "eval_samples_per_second": 6.39,
+      "eval_steps_per_second": 0.799,
+      "step": 16300
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 4.553817169069463e-05,
+      "loss": 0.4047,
+      "step": 16350
+    },
+    {
+      "epoch": 1.79,
+      "learning_rate": 4.5524519440803846e-05,
+      "loss": 0.3869,
+      "step": 16400
+    },
+    {
+      "epoch": 1.79,
+      "eval_cer": 0.06926132883476309,
+      "eval_loss": 0.40593230724334717,
+      "eval_runtime": 1163.6552,
+      "eval_samples_per_second": 6.456,
+      "eval_steps_per_second": 0.808,
+      "step": 16400
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 4.551086719091306e-05,
+      "loss": 0.4197,
+      "step": 16450
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 4.549721494102228e-05,
+      "loss": 0.3814,
+      "step": 16500
+    },
+    {
+      "epoch": 1.8,
+      "eval_cer": 0.07038711200687105,
+      "eval_loss": 0.40712347626686096,
+      "eval_runtime": 1163.3519,
+      "eval_samples_per_second": 6.458,
+      "eval_steps_per_second": 0.808,
+      "step": 16500
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 4.54835626911315e-05,
+      "loss": 0.3532,
+      "step": 16550
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 4.546991044124072e-05,
+      "loss": 0.4102,
+      "step": 16600
+    },
+    {
+      "epoch": 1.81,
+      "eval_cer": 0.0705513600076649,
+      "eval_loss": 0.4008554518222809,
+      "eval_runtime": 1174.5887,
+      "eval_samples_per_second": 6.396,
+      "eval_steps_per_second": 0.8,
+      "step": 16600
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 4.545625819134994e-05,
+      "loss": 0.3598,
+      "step": 16650
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 4.5442605941459155e-05,
+      "loss": 0.4015,
+      "step": 16700
+    },
+    {
+      "epoch": 1.82,
+      "eval_cer": 0.06728350915853697,
+      "eval_loss": 0.3957868218421936,
+      "eval_runtime": 1148.3376,
+      "eval_samples_per_second": 6.543,
+      "eval_steps_per_second": 0.819,
+      "step": 16700
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 4.542895369156837e-05,
+      "loss": 0.3946,
+      "step": 16750
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 4.5415574486675406e-05,
+      "loss": 0.3765,
+      "step": 16800
+    },
+    {
+      "epoch": 1.83,
+      "eval_cer": 0.06861802416498712,
+      "eval_loss": 0.4007869362831116,
+      "eval_runtime": 1150.6892,
+      "eval_samples_per_second": 6.529,
+      "eval_steps_per_second": 0.817,
+      "step": 16800
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 4.540192223678462e-05,
+      "loss": 0.359,
+      "step": 16850
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 4.5388269986893844e-05,
+      "loss": 0.4047,
+      "step": 16900
+    },
+    {
+      "epoch": 1.85,
+      "eval_cer": 0.07087985600925263,
+      "eval_loss": 0.40284430980682373,
+      "eval_runtime": 1179.5455,
+      "eval_samples_per_second": 6.369,
+      "eval_steps_per_second": 0.797,
+      "step": 16900
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 4.537461773700306e-05,
+      "loss": 0.3911,
+      "step": 16950
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 4.5360965487112276e-05,
+      "loss": 0.3915,
+      "step": 17000
+    },
+    {
+      "epoch": 1.86,
+      "eval_cer": 0.06871725733213341,
+      "eval_loss": 0.3993259072303772,
+      "eval_runtime": 1144.6731,
+      "eval_samples_per_second": 6.563,
+      "eval_steps_per_second": 0.821,
+      "step": 17000
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 4.534731323722149e-05,
+      "loss": 0.3652,
+      "step": 17050
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 4.5333660987330714e-05,
+      "loss": 0.3939,
+      "step": 17100
+    },
+    {
+      "epoch": 1.87,
+      "eval_cer": 0.06720138515814003,
+      "eval_loss": 0.395829975605011,
+      "eval_runtime": 1148.798,
+      "eval_samples_per_second": 6.54,
+      "eval_steps_per_second": 0.818,
+      "step": 17100
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 4.532000873743994e-05,
+      "loss": 0.3864,
+      "step": 17150
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 4.530635648754915e-05,
+      "loss": 0.3617,
+      "step": 17200
+    },
+    {
+      "epoch": 1.88,
+      "eval_cer": 0.07021259850602755,
+      "eval_loss": 0.4057924449443817,
+      "eval_runtime": 1153.1488,
+      "eval_samples_per_second": 6.515,
+      "eval_steps_per_second": 0.815,
+      "step": 17200
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 4.529270423765837e-05,
+      "loss": 0.3855,
+      "step": 17250
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 4.5279051987767584e-05,
+      "loss": 0.3741,
+      "step": 17300
+    },
+    {
+      "epoch": 1.89,
+      "eval_cer": 0.07238888451654628,
+      "eval_loss": 0.41401273012161255,
+      "eval_runtime": 1185.6581,
+      "eval_samples_per_second": 6.337,
+      "eval_steps_per_second": 0.793,
+      "step": 17300
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 4.526539973787681e-05,
+      "loss": 0.3724,
+      "step": 17350
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 4.525174748798602e-05,
+      "loss": 0.408,
+      "step": 17400
+    },
+    {
+      "epoch": 1.9,
+      "eval_cer": 0.06712610482444284,
+      "eval_loss": 0.398795485496521,
+      "eval_runtime": 1152.8077,
+      "eval_samples_per_second": 6.517,
+      "eval_steps_per_second": 0.815,
+      "step": 17400
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 4.523809523809524e-05,
+      "loss": 0.3974,
+      "step": 17450
+    },
+    {
+      "epoch": 1.91,
+      "learning_rate": 4.5224442988204454e-05,
+      "loss": 0.3812,
+      "step": 17500
+    },
+    {
+      "epoch": 1.91,
+      "eval_cer": 0.07113991534384292,
+      "eval_loss": 0.41947415471076965,
+      "eval_runtime": 1123.4148,
+      "eval_samples_per_second": 6.688,
+      "eval_steps_per_second": 0.837,
+      "step": 17500
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 4.521079073831368e-05,
+      "loss": 0.3823,
+      "step": 17550
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 4.51971384884229e-05,
+      "loss": 0.3848,
+      "step": 17600
+    },
+    {
+      "epoch": 1.92,
+      "eval_cer": 0.0706882333416598,
+      "eval_loss": 0.401214599609375,
+      "eval_runtime": 1163.9479,
+      "eval_samples_per_second": 6.455,
+      "eval_steps_per_second": 0.808,
+      "step": 17600
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 4.5183486238532115e-05,
+      "loss": 0.3645,
+      "step": 17650
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 4.516983398864133e-05,
+      "loss": 0.3587,
+      "step": 17700
+    },
+    {
+      "epoch": 1.93,
+      "eval_cer": 0.0772342005399653,
+      "eval_loss": 0.40040454268455505,
+      "eval_runtime": 1206.7666,
+      "eval_samples_per_second": 6.226,
+      "eval_steps_per_second": 0.779,
+      "step": 17700
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 4.515618173875055e-05,
+      "loss": 0.4196,
+      "step": 17750
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 4.514252948885976e-05,
+      "loss": 0.364,
+      "step": 17800
+    },
+    {
+      "epoch": 1.94,
+      "eval_cer": 0.07031525350652372,
+      "eval_loss": 0.3970361649990082,
+      "eval_runtime": 1161.1563,
+      "eval_samples_per_second": 6.47,
+      "eval_steps_per_second": 0.81,
+      "step": 17800
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 4.5128877238968985e-05,
+      "loss": 0.4055,
+      "step": 17850
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 4.51152249890782e-05,
+      "loss": 0.3667,
+      "step": 17900
+    },
+    {
+      "epoch": 1.95,
+      "eval_cer": 0.07077035734205672,
+      "eval_loss": 0.3838033676147461,
+      "eval_runtime": 1185.1264,
+      "eval_samples_per_second": 6.339,
+      "eval_steps_per_second": 0.793,
+      "step": 17900
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 4.510157273918742e-05,
+      "loss": 0.3816,
+      "step": 17950
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 4.508792048929664e-05,
+      "loss": 0.3619,
+      "step": 18000
+    },
+    {
+      "epoch": 1.97,
+      "eval_cer": 0.07000728850503522,
+      "eval_loss": 0.39164894819259644,
+      "eval_runtime": 1179.3108,
+      "eval_samples_per_second": 6.371,
+      "eval_steps_per_second": 0.797,
+      "step": 18000
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 4.5074268239405855e-05,
+      "loss": 0.3989,
+      "step": 18050
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 4.506061598951508e-05,
+      "loss": 0.4249,
+      "step": 18100
+    },
+    {
+      "epoch": 1.98,
+      "eval_cer": 0.06841613599734465,
+      "eval_loss": 0.3852728307247162,
+      "eval_runtime": 1156.9882,
+      "eval_samples_per_second": 6.494,
+      "eval_steps_per_second": 0.812,
+      "step": 18100
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 4.5046963739624294e-05,
+      "loss": 0.3723,
+      "step": 18150
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 4.503331148973351e-05,
+      "loss": 0.3633,
+      "step": 18200
+    },
+    {
+      "epoch": 1.99,
+      "eval_cer": 0.06889519266632677,
+      "eval_loss": 0.387567400932312,
+      "eval_runtime": 1174.6487,
+      "eval_samples_per_second": 6.396,
+      "eval_steps_per_second": 0.8,
+      "step": 18200
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 4.5019659239842725e-05,
+      "loss": 0.3559,
+      "step": 18250
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 4.500600698995195e-05,
+      "loss": 0.4151,
+      "step": 18300
+    },
+    {
+      "epoch": 2.0,
+      "eval_cer": 0.06865224249848584,
+      "eval_loss": 0.38729149103164673,
+      "eval_runtime": 1181.6644,
+      "eval_samples_per_second": 6.358,
+      "eval_steps_per_second": 0.795,
+      "step": 18300
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 4.4992354740061164e-05,
+      "loss": 0.317,
+      "step": 18350
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 4.497870249017038e-05,
+      "loss": 0.2872,
+      "step": 18400
+    },
+    {
+      "epoch": 2.01,
+      "eval_cer": 0.06717058865799118,
+      "eval_loss": 0.392042875289917,
+      "eval_runtime": 1137.7306,
+      "eval_samples_per_second": 6.603,
+      "eval_steps_per_second": 0.826,
+      "step": 18400
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 4.49650502402796e-05,
+      "loss": 0.2926,
+      "step": 18450
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 4.495139799038882e-05,
+      "loss": 0.3,
+      "step": 18500
+    },
+    {
+      "epoch": 2.02,
+      "eval_cer": 0.06902864416697178,
+      "eval_loss": 0.38552695512771606,
+      "eval_runtime": 1176.6053,
+      "eval_samples_per_second": 6.385,
+      "eval_steps_per_second": 0.799,
+      "step": 18500
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 4.493774574049804e-05,
+      "loss": 0.3154,
+      "step": 18550
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 4.4924093490607256e-05,
+      "loss": 0.3287,
+      "step": 18600
+    },
+    {
+      "epoch": 2.03,
+      "eval_cer": 0.06579158981799269,
+      "eval_loss": 0.39401352405548096,
+      "eval_runtime": 1146.5226,
+      "eval_samples_per_second": 6.553,
+      "eval_steps_per_second": 0.82,
+      "step": 18600
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 4.491044124071647e-05,
+      "loss": 0.3288,
+      "step": 18650
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 4.489678899082569e-05,
+      "loss": 0.3124,
+      "step": 18700
+    },
+    {
+      "epoch": 2.04,
+      "eval_cer": 0.06543571914960597,
+      "eval_loss": 0.3829745948314667,
+      "eval_runtime": 1170.1361,
+      "eval_samples_per_second": 6.421,
+      "eval_steps_per_second": 0.803,
+      "step": 18700
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 4.4883136740934904e-05,
+      "loss": 0.3132,
+      "step": 18750
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 4.4869484491044126e-05,
+      "loss": 0.2994,
+      "step": 18800
+    },
+    {
+      "epoch": 2.05,
+      "eval_cer": 0.06607560198603207,
+      "eval_loss": 0.3876435458660126,
+      "eval_runtime": 1143.5794,
+      "eval_samples_per_second": 6.57,
+      "eval_steps_per_second": 0.822,
+      "step": 18800
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 4.485583224115334e-05,
+      "loss": 0.3342,
+      "step": 18850
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 4.4842179991262565e-05,
+      "loss": 0.2923,
+      "step": 18900
+    },
+    {
+      "epoch": 2.06,
+      "eval_cer": 0.06948032616915491,
+      "eval_loss": 0.3895299434661865,
+      "eval_runtime": 1153.5637,
+      "eval_samples_per_second": 6.513,
+      "eval_steps_per_second": 0.815,
+      "step": 18900
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 4.482852774137178e-05,
+      "loss": 0.3297,
+      "step": 18950
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 4.4814875491481e-05,
+      "loss": 0.3179,
+      "step": 19000
+    },
+    {
+      "epoch": 2.08,
+      "eval_cer": 0.06705766815744539,
+      "eval_loss": 0.38019657135009766,
+      "eval_runtime": 1179.7174,
+      "eval_samples_per_second": 6.368,
+      "eval_steps_per_second": 0.797,
+      "step": 19000
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 4.480122324159022e-05,
+      "loss": 0.2929,
+      "step": 19050
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 4.4787570991699435e-05,
+      "loss": 0.3521,
+      "step": 19100
+    },
+    {
+      "epoch": 2.09,
+      "eval_cer": 0.06731772749203568,
+      "eval_loss": 0.37924471497535706,
+      "eval_runtime": 1165.4502,
+      "eval_samples_per_second": 6.446,
+      "eval_steps_per_second": 0.807,
+      "step": 19100
+    },
+    {
+      "epoch": 2.09,
+      "learning_rate": 4.477391874180865e-05,
+      "loss": 0.3111,
+      "step": 19150
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 4.4760266491917866e-05,
+      "loss": 0.3007,
+      "step": 19200
+    },
+    {
+      "epoch": 2.1,
+      "eval_cer": 0.06791997016161319,
+      "eval_loss": 0.3855767548084259,
+      "eval_runtime": 1158.659,
+      "eval_samples_per_second": 6.484,
+      "eval_steps_per_second": 0.811,
+      "step": 19200
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 4.474661424202709e-05,
+      "loss": 0.3176,
+      "step": 19250
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 4.4732961992136305e-05,
+      "loss": 0.294,
+      "step": 19300
+    },
+    {
+      "epoch": 2.11,
+      "eval_cer": 0.06327654230583662,
+      "eval_loss": 0.37883859872817993,
+      "eval_runtime": 1146.7655,
+      "eval_samples_per_second": 6.551,
+      "eval_steps_per_second": 0.82,
+      "step": 19300
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 4.471930974224553e-05,
+      "loss": 0.3376,
+      "step": 19350
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 4.470565749235474e-05,
+      "loss": 0.3348,
+      "step": 19400
+    },
+    {
+      "epoch": 2.12,
+      "eval_cer": 0.06263323763606064,
+      "eval_loss": 0.37968873977661133,
+      "eval_runtime": 1136.9087,
+      "eval_samples_per_second": 6.608,
+      "eval_steps_per_second": 0.827,
+      "step": 19400
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 4.469200524246396e-05,
+      "loss": 0.2803,
+      "step": 19450
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 4.467835299257318e-05,
+      "loss": 0.2794,
+      "step": 19500
+    },
+    {
+      "epoch": 2.13,
+      "eval_cer": 0.06554179598345201,
+      "eval_loss": 0.37674739956855774,
+      "eval_runtime": 1167.3995,
+      "eval_samples_per_second": 6.436,
+      "eval_steps_per_second": 0.805,
+      "step": 19500
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 4.46647007426824e-05,
+      "loss": 0.3175,
+      "step": 19550
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 4.465104849279161e-05,
+      "loss": 0.3098,
+      "step": 19600
+    },
+    {
+      "epoch": 2.14,
+      "eval_cer": 0.06358108547397524,
+      "eval_loss": 0.3796171247959137,
+      "eval_runtime": 1151.1497,
+      "eval_samples_per_second": 6.527,
+      "eval_steps_per_second": 0.817,
+      "step": 19600
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 4.463739624290083e-05,
+      "loss": 0.3399,
+      "step": 19650
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 4.4623743993010045e-05,
+      "loss": 0.289,
+      "step": 19700
+    },
+    {
+      "epoch": 2.15,
+      "eval_cer": 0.06953165366940299,
+      "eval_loss": 0.39493194222450256,
+      "eval_runtime": 1365.0679,
+      "eval_samples_per_second": 5.504,
+      "eval_steps_per_second": 0.689,
+      "step": 19700
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 4.461009174311927e-05,
+      "loss": 0.3148,
+      "step": 19750
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 4.459643949322849e-05,
+      "loss": 0.3033,
+      "step": 19800
+    },
+    {
+      "epoch": 2.16,
+      "eval_cer": 0.06477530531308064,
+      "eval_loss": 0.3851621747016907,
+      "eval_runtime": 1450.2983,
+      "eval_samples_per_second": 5.18,
+      "eval_steps_per_second": 0.648,
+      "step": 19800
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 4.458306028833552e-05,
+      "loss": 0.3174,
+      "step": 19850
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 4.456940803844474e-05,
+      "loss": 0.3354,
+      "step": 19900
+    },
+    {
+      "epoch": 2.17,
+      "eval_cer": 0.06344763397333023,
+      "eval_loss": 0.38615599274635315,
+      "eval_runtime": 1300.2394,
+      "eval_samples_per_second": 5.778,
+      "eval_steps_per_second": 0.723,
+      "step": 19900
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 4.455575578855396e-05,
+      "loss": 0.3462,
+      "step": 19950
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 4.454210353866317e-05,
+      "loss": 0.3373,
+      "step": 20000
+    },
+    {
+      "epoch": 2.18,
+      "eval_cer": 0.06389247230881362,
+      "eval_loss": 0.3779688775539398,
+      "eval_runtime": 1247.3861,
+      "eval_samples_per_second": 6.023,
+      "eval_steps_per_second": 0.754,
+      "step": 20000
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 4.4528451288772395e-05,
+      "loss": 0.3024,
+      "step": 20050
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 4.451479903888161e-05,
+      "loss": 0.2955,
+      "step": 20100
+    },
+    {
+      "epoch": 2.2,
+      "eval_cer": 0.06926817250146283,
+      "eval_loss": 0.3931354880332947,
+      "eval_runtime": 1363.3975,
+      "eval_samples_per_second": 5.51,
+      "eval_steps_per_second": 0.689,
+      "step": 20100
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 4.450114678899083e-05,
+      "loss": 0.3167,
+      "step": 20150
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 4.448749453910004e-05,
+      "loss": 0.3228,
+      "step": 20200
+    },
+    {
+      "epoch": 2.21,
+      "eval_cer": 0.06707135549084489,
+      "eval_loss": 0.37632882595062256,
+      "eval_runtime": 1493.5017,
+      "eval_samples_per_second": 5.03,
+      "eval_steps_per_second": 0.629,
+      "step": 20200
+    },
+    {
+      "epoch": 2.21,
+      "learning_rate": 4.4473842289209265e-05,
+      "loss": 0.316,
+      "step": 20250
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 4.446019003931848e-05,
+      "loss": 0.3392,
+      "step": 20300
+    },
+    {
+      "epoch": 2.22,
+      "eval_cer": 0.06558970165035022,
+      "eval_loss": 0.38044553995132446,
+      "eval_runtime": 1382.6368,
+      "eval_samples_per_second": 5.434,
+      "eval_steps_per_second": 0.68,
+      "step": 20300
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 4.4446537789427704e-05,
+      "loss": 0.2981,
+      "step": 20350
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 4.443288553953692e-05,
+      "loss": 0.3149,
+      "step": 20400
+    },
+    {
+      "epoch": 2.23,
+      "eval_cer": 0.0647547743129814,
+      "eval_loss": 0.3692227602005005,
+      "eval_runtime": 1474.8703,
+      "eval_samples_per_second": 5.094,
+      "eval_steps_per_second": 0.637,
+      "step": 20400
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 4.4419233289646135e-05,
+      "loss": 0.311,
+      "step": 20450
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 4.440558103975536e-05,
+      "loss": 0.2676,
+      "step": 20500
+    },
+    {
+      "epoch": 2.24,
+      "eval_cer": 0.06711583932439322,
+      "eval_loss": 0.3739226758480072,
+      "eval_runtime": 1465.1736,
+      "eval_samples_per_second": 5.128,
+      "eval_steps_per_second": 0.642,
+      "step": 20500
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 4.4391928789864574e-05,
+      "loss": 0.2736,
+      "step": 20550
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 4.437827653997379e-05,
+      "loss": 0.3111,
+      "step": 20600
+    },
+    {
+      "epoch": 2.25,
+      "eval_cer": 0.0664417381544684,
+      "eval_loss": 0.3745150864124298,
+      "eval_runtime": 1170.2827,
+      "eval_samples_per_second": 6.42,
+      "eval_steps_per_second": 0.803,
+      "step": 20600
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 4.4364624290083005e-05,
+      "loss": 0.2983,
+      "step": 20650
+    },
+    {
+      "epoch": 2.26,
+      "learning_rate": 4.435097204019222e-05,
+      "loss": 0.3049,
+      "step": 20700
+    },
+    {
+      "epoch": 2.26,
+      "eval_cer": 0.06717401049134106,
+      "eval_loss": 0.3762642443180084,
+      "eval_runtime": 1206.4849,
+      "eval_samples_per_second": 6.227,
+      "eval_steps_per_second": 0.779,
+      "step": 20700
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 4.4337319790301444e-05,
+      "loss": 0.3181,
+      "step": 20750
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 4.432366754041066e-05,
+      "loss": 0.2593,
+      "step": 20800
+    },
+    {
+      "epoch": 2.27,
+      "eval_cer": 0.06381034830841668,
+      "eval_loss": 0.37320199608802795,
+      "eval_runtime": 1196.0544,
+      "eval_samples_per_second": 6.281,
+      "eval_steps_per_second": 0.786,
+      "step": 20800
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 4.431001529051988e-05,
+      "loss": 0.3106,
+      "step": 20850
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 4.42963630406291e-05,
+      "loss": 0.2765,
+      "step": 20900
+    },
+    {
+      "epoch": 2.28,
+      "eval_cer": 0.06492586598047502,
+      "eval_loss": 0.37560248374938965,
+      "eval_runtime": 1186.1802,
+      "eval_samples_per_second": 6.334,
+      "eval_steps_per_second": 0.792,
+      "step": 20900
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 4.4282710790738314e-05,
+      "loss": 0.2789,
+      "step": 20950
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 4.4269058540847536e-05,
+      "loss": 0.3202,
+      "step": 21000
+    },
+    {
+      "epoch": 2.29,
+      "eval_cer": 0.06184279413224017,
+      "eval_loss": 0.3614885210990906,
+      "eval_runtime": 1146.8682,
+      "eval_samples_per_second": 6.551,
+      "eval_steps_per_second": 0.82,
+      "step": 21000
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 4.425540629095675e-05,
+      "loss": 0.3247,
+      "step": 21050
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 4.424175404106597e-05,
+      "loss": 0.3105,
+      "step": 21100
+    },
+    {
+      "epoch": 2.3,
+      "eval_cer": 0.0640669858096571,
+      "eval_loss": 0.3699856996536255,
+      "eval_runtime": 1154.5375,
+      "eval_samples_per_second": 6.507,
+      "eval_steps_per_second": 0.814,
+      "step": 21100
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 4.4228101791175184e-05,
+      "loss": 0.2951,
+      "step": 21150
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 4.4214449541284406e-05,
+      "loss": 0.3252,
+      "step": 21200
+    },
+    {
+      "epoch": 2.32,
+      "eval_cer": 0.06651017482146585,
+      "eval_loss": 0.367857426404953,
+      "eval_runtime": 1184.7517,
+      "eval_samples_per_second": 6.341,
+      "eval_steps_per_second": 0.793,
+      "step": 21200
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 4.420079729139362e-05,
+      "loss": 0.2952,
+      "step": 21250
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 4.4187145041502845e-05,
+      "loss": 0.3112,
+      "step": 21300
+    },
+    {
+      "epoch": 2.33,
+      "eval_cer": 0.06518934714841518,
+      "eval_loss": 0.3664465844631195,
+      "eval_runtime": 1166.6886,
+      "eval_samples_per_second": 6.44,
+      "eval_steps_per_second": 0.806,
+      "step": 21300
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 4.417349279161206e-05,
+      "loss": 0.3146,
+      "step": 21350
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 4.4159840541721276e-05,
+      "loss": 0.3088,
+      "step": 21400
+    },
+    {
+      "epoch": 2.34,
+      "eval_cer": 0.06262981580271078,
+      "eval_loss": 0.3656906187534332,
+      "eval_runtime": 1164.8946,
+      "eval_samples_per_second": 6.45,
+      "eval_steps_per_second": 0.807,
+      "step": 21400
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 4.41461882918305e-05,
+      "loss": 0.3172,
+      "step": 21450
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 4.4132536041939715e-05,
+      "loss": 0.3134,
+      "step": 21500
+    },
+    {
+      "epoch": 2.35,
+      "eval_cer": 0.06641094165431956,
+      "eval_loss": 0.37143638730049133,
+      "eval_runtime": 1189.7181,
+      "eval_samples_per_second": 6.315,
+      "eval_steps_per_second": 0.79,
+      "step": 21500
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 4.411888379204893e-05,
+      "loss": 0.312,
+      "step": 21550
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 4.4105231542158146e-05,
+      "loss": 0.2935,
+      "step": 21600
+    },
+    {
+      "epoch": 2.36,
+      "eval_cer": 0.060775182127080045,
+      "eval_loss": 0.36119917035102844,
+      "eval_runtime": 1143.708,
+      "eval_samples_per_second": 6.569,
+      "eval_steps_per_second": 0.822,
+      "step": 21600
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 4.409157929226736e-05,
+      "loss": 0.3247,
+      "step": 21650
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 4.4077927042376585e-05,
+      "loss": 0.3451,
+      "step": 21700
+    },
+    {
+      "epoch": 2.37,
+      "eval_cer": 0.061541672797451416,
+      "eval_loss": 0.3566524386405945,
+      "eval_runtime": 1165.536,
+      "eval_samples_per_second": 6.446,
+      "eval_steps_per_second": 0.806,
+      "step": 21700
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 4.406427479248581e-05,
+      "loss": 0.3164,
+      "step": 21750
+    },
+    {
+      "epoch": 2.38,
+      "learning_rate": 4.405062254259502e-05,
+      "loss": 0.3113,
+      "step": 21800
+    },
+    {
+      "epoch": 2.38,
+      "eval_cer": 0.06236291280142074,
+      "eval_loss": 0.3675330579280853,
+      "eval_runtime": 1149.7199,
+      "eval_samples_per_second": 6.535,
+      "eval_steps_per_second": 0.818,
+      "step": 21800
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 4.403697029270424e-05,
+      "loss": 0.3079,
+      "step": 21850
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 4.4023318042813455e-05,
+      "loss": 0.2795,
+      "step": 21900
+    },
+    {
+      "epoch": 2.39,
+      "eval_cer": 0.062013885799733784,
+      "eval_loss": 0.3630739152431488,
+      "eval_runtime": 1162.4946,
+      "eval_samples_per_second": 6.463,
+      "eval_steps_per_second": 0.809,
+      "step": 21900
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 4.400966579292268e-05,
+      "loss": 0.3273,
+      "step": 21950
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 4.399601354303189e-05,
+      "loss": 0.3165,
+      "step": 22000
+    },
+    {
+      "epoch": 2.4,
+      "eval_cer": 0.06382403564181617,
+      "eval_loss": 0.3620857298374176,
+      "eval_runtime": 1163.7577,
+      "eval_samples_per_second": 6.456,
+      "eval_steps_per_second": 0.808,
+      "step": 22000
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 4.398236129314111e-05,
+      "loss": 0.3265,
+      "step": 22050
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 4.3968709043250325e-05,
+      "loss": 0.293,
+      "step": 22100
+    },
+    {
+      "epoch": 2.41,
+      "eval_cer": 0.06423123381045097,
+      "eval_loss": 0.36310431361198425,
+      "eval_runtime": 1165.8222,
+      "eval_samples_per_second": 6.444,
+      "eval_steps_per_second": 0.806,
+      "step": 22100
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 4.395505679335955e-05,
+      "loss": 0.3111,
+      "step": 22150
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 4.394140454346877e-05,
+      "loss": 0.3052,
+      "step": 22200
+    },
+    {
+      "epoch": 2.42,
+      "eval_cer": 0.06236291280142074,
+      "eval_loss": 0.3586342930793762,
+      "eval_runtime": 1151.2813,
+      "eval_samples_per_second": 6.526,
+      "eval_steps_per_second": 0.816,
+      "step": 22200
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 4.3927752293577986e-05,
+      "loss": 0.3033,
+      "step": 22250
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 4.39141000436872e-05,
+      "loss": 0.3183,
+      "step": 22300
+    },
+    {
+      "epoch": 2.44,
+      "eval_cer": 0.060504857292440144,
+      "eval_loss": 0.3589232861995697,
+      "eval_runtime": 1143.1903,
+      "eval_samples_per_second": 6.572,
+      "eval_steps_per_second": 0.822,
+      "step": 22300
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 4.390044779379642e-05,
+      "loss": 0.299,
+      "step": 22350
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 4.388679554390564e-05,
+      "loss": 0.3395,
+      "step": 22400
+    },
+    {
+      "epoch": 2.45,
+      "eval_cer": 0.06262297213601103,
+      "eval_loss": 0.3562231659889221,
+      "eval_runtime": 1171.0509,
+      "eval_samples_per_second": 6.416,
+      "eval_steps_per_second": 0.803,
+      "step": 22400
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 4.3873143294014856e-05,
+      "loss": 0.2849,
+      "step": 22450
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 4.385949104412407e-05,
+      "loss": 0.2904,
+      "step": 22500
+    },
+    {
+      "epoch": 2.46,
+      "eval_cer": 0.06267429963625912,
+      "eval_loss": 0.3656773865222931,
+      "eval_runtime": 1135.83,
+      "eval_samples_per_second": 6.615,
+      "eval_steps_per_second": 0.828,
+      "step": 22500
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 4.384583879423329e-05,
+      "loss": 0.2895,
+      "step": 22550
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 4.383218654434251e-05,
+      "loss": 0.3042,
+      "step": 22600
+    },
+    {
+      "epoch": 2.47,
+      "eval_cer": 0.06159642213104937,
+      "eval_loss": 0.35998255014419556,
+      "eval_runtime": 1154.6766,
+      "eval_samples_per_second": 6.507,
+      "eval_steps_per_second": 0.814,
+      "step": 22600
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 4.381853429445173e-05,
+      "loss": 0.2958,
+      "step": 22650
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 4.380488204456095e-05,
+      "loss": 0.2925,
+      "step": 22700
+    },
+    {
+      "epoch": 2.48,
+      "eval_cer": 0.0625408481356141,
+      "eval_loss": 0.35356906056404114,
+      "eval_runtime": 1163.7126,
+      "eval_samples_per_second": 6.456,
+      "eval_steps_per_second": 0.808,
+      "step": 22700
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 4.3791229794670164e-05,
+      "loss": 0.3009,
+      "step": 22750
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 4.377757754477938e-05,
+      "loss": 0.2875,
+      "step": 22800
+    },
+    {
+      "epoch": 2.49,
+      "eval_cer": 0.06025506345789947,
+      "eval_loss": 0.3555847704410553,
+      "eval_runtime": 1146.3487,
+      "eval_samples_per_second": 6.554,
+      "eval_steps_per_second": 0.82,
+      "step": 22800
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 4.37639252948886e-05,
+      "loss": 0.3123,
+      "step": 22850
+    },
+    {
+      "epoch": 2.5,
+      "learning_rate": 4.375027304499782e-05,
+      "loss": 0.3042,
+      "step": 22900
+    },
+    {
+      "epoch": 2.5,
+      "eval_cer": 0.06097022662802276,
+      "eval_loss": 0.3487367331981659,
+      "eval_runtime": 1161.3152,
+      "eval_samples_per_second": 6.469,
+      "eval_steps_per_second": 0.809,
+      "step": 22900
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 4.3736620795107034e-05,
+      "loss": 0.2855,
+      "step": 22950
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 4.372296854521625e-05,
+      "loss": 0.277,
+      "step": 23000
+    },
+    {
+      "epoch": 2.51,
+      "eval_cer": 0.060617777792985925,
+      "eval_loss": 0.34876227378845215,
+      "eval_runtime": 1159.4584,
+      "eval_samples_per_second": 6.48,
+      "eval_steps_per_second": 0.811,
+      "step": 23000
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 4.370931629532547e-05,
+      "loss": 0.2976,
+      "step": 23050
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 4.3695664045434695e-05,
+      "loss": 0.3092,
+      "step": 23100
+    },
+    {
+      "epoch": 2.52,
+      "eval_cer": 0.059519369287676954,
+      "eval_loss": 0.3475951552391052,
+      "eval_runtime": 1144.1568,
+      "eval_samples_per_second": 6.566,
+      "eval_steps_per_second": 0.822,
+      "step": 23100
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 4.368201179554391e-05,
+      "loss": 0.2982,
+      "step": 23150
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 4.366835954565313e-05,
+      "loss": 0.3148,
+      "step": 23200
+    },
+    {
+      "epoch": 2.53,
+      "eval_cer": 0.0602892817913982,
+      "eval_loss": 0.3553316295146942,
+      "eval_runtime": 1154.791,
+      "eval_samples_per_second": 6.506,
+      "eval_steps_per_second": 0.814,
+      "step": 23200
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 4.365470729576234e-05,
+      "loss": 0.3058,
+      "step": 23250
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 4.364105504587156e-05,
+      "loss": 0.3245,
+      "step": 23300
+    },
+    {
+      "epoch": 2.54,
+      "eval_cer": 0.07005177233858356,
+      "eval_loss": 0.3529204726219177,
+      "eval_runtime": 1195.1336,
+      "eval_samples_per_second": 6.286,
+      "eval_steps_per_second": 0.787,
+      "step": 23300
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 4.362740279598078e-05,
+      "loss": 0.3058,
+      "step": 23350
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 4.361375054609e-05,
+      "loss": 0.2828,
+      "step": 23400
+    },
+    {
+      "epoch": 2.56,
+      "eval_cer": 0.062260257800924576,
+      "eval_loss": 0.35099321603775024,
+      "eval_runtime": 1152.0634,
+      "eval_samples_per_second": 6.521,
+      "eval_steps_per_second": 0.816,
+      "step": 23400
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 4.360009829619921e-05,
+      "loss": 0.3004,
+      "step": 23450
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 4.3586446046308436e-05,
+      "loss": 0.3078,
+      "step": 23500
+    },
+    {
+      "epoch": 2.57,
+      "eval_cer": 0.06309176330494352,
+      "eval_loss": 0.3564070463180542,
+      "eval_runtime": 1168.3273,
+      "eval_samples_per_second": 6.431,
+      "eval_steps_per_second": 0.805,
+      "step": 23500
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 4.357279379641765e-05,
+      "loss": 0.3201,
+      "step": 23550
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 4.3559141546526874e-05,
+      "loss": 0.297,
+      "step": 23600
+    },
+    {
+      "epoch": 2.58,
+      "eval_cer": 0.0617161862982949,
+      "eval_loss": 0.3490002155303955,
+      "eval_runtime": 1170.8594,
+      "eval_samples_per_second": 6.417,
+      "eval_steps_per_second": 0.803,
+      "step": 23600
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 4.354548929663609e-05,
+      "loss": 0.2804,
+      "step": 23650
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 4.3531837046745306e-05,
+      "loss": 0.2812,
+      "step": 23700
+    },
+    {
+      "epoch": 2.59,
+      "eval_cer": 0.05973836662206877,
+      "eval_loss": 0.34518638253211975,
+      "eval_runtime": 1156.4633,
+      "eval_samples_per_second": 6.497,
+      "eval_steps_per_second": 0.813,
+      "step": 23700
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 4.351818479685452e-05,
+      "loss": 0.2962,
+      "step": 23750
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 4.3504532546963744e-05,
+      "loss": 0.2938,
+      "step": 23800
+    },
+    {
+      "epoch": 2.6,
+      "eval_cer": 0.06105577246176957,
+      "eval_loss": 0.3440614342689514,
+      "eval_runtime": 1166.2989,
+      "eval_samples_per_second": 6.442,
+      "eval_steps_per_second": 0.806,
+      "step": 23800
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 4.349115334207077e-05,
+      "loss": 0.2741,
+      "step": 23850
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 4.3477501092179995e-05,
+      "loss": 0.2992,
+      "step": 23900
+    },
+    {
+      "epoch": 2.61,
+      "eval_cer": 0.06274273630325655,
+      "eval_loss": 0.34478849172592163,
+      "eval_runtime": 1168.87,
+      "eval_samples_per_second": 6.428,
+      "eval_steps_per_second": 0.804,
+      "step": 23900
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 4.346384884228921e-05,
+      "loss": 0.3067,
+      "step": 23950
+    },
+    {
+      "epoch": 2.62,
+      "learning_rate": 4.3450196592398426e-05,
+      "loss": 0.3091,
+      "step": 24000
+    },
+    {
+      "epoch": 2.62,
+      "eval_cer": 0.062486098802016145,
+      "eval_loss": 0.3409504294395447,
+      "eval_runtime": 1187.759,
+      "eval_samples_per_second": 6.325,
+      "eval_steps_per_second": 0.791,
+      "step": 24000
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 4.343654434250765e-05,
+      "loss": 0.2917,
+      "step": 24050
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 4.3422892092616865e-05,
+      "loss": 0.2831,
+      "step": 24100
+    },
+    {
+      "epoch": 2.63,
+      "eval_cer": 0.061654593297997204,
+      "eval_loss": 0.34767550230026245,
+      "eval_runtime": 1162.5382,
+      "eval_samples_per_second": 6.463,
+      "eval_steps_per_second": 0.809,
+      "step": 24100
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 4.340923984272609e-05,
+      "loss": 0.2899,
+      "step": 24150
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 4.33955875928353e-05,
+      "loss": 0.316,
+      "step": 24200
+    },
+    {
+      "epoch": 2.64,
+      "eval_cer": 0.06590793215188834,
+      "eval_loss": 0.3456661105155945,
+      "eval_runtime": 1228.1783,
+      "eval_samples_per_second": 6.117,
+      "eval_steps_per_second": 0.765,
+      "step": 24200
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 4.338193534294452e-05,
+      "loss": 0.2818,
+      "step": 24250
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 4.3368283093053735e-05,
+      "loss": 0.3216,
+      "step": 24300
+    },
+    {
+      "epoch": 2.65,
+      "eval_cer": 0.06014214295735369,
+      "eval_loss": 0.3410555124282837,
+      "eval_runtime": 1339.8538,
+      "eval_samples_per_second": 5.607,
+      "eval_steps_per_second": 0.702,
+      "step": 24300
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 4.335463084316296e-05,
+      "loss": 0.3092,
+      "step": 24350
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 4.334097859327217e-05,
+      "loss": 0.3041,
+      "step": 24400
+    },
+    {
+      "epoch": 2.66,
+      "eval_cer": 0.06188727796578851,
+      "eval_loss": 0.345905065536499,
+      "eval_runtime": 1188.7161,
+      "eval_samples_per_second": 6.32,
+      "eval_steps_per_second": 0.791,
+      "step": 24400
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 4.332732634338139e-05,
+      "loss": 0.2921,
+      "step": 24450
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 4.331367409349061e-05,
+      "loss": 0.2946,
+      "step": 24500
+    },
+    {
+      "epoch": 2.68,
+      "eval_cer": 0.061298722629610496,
+      "eval_loss": 0.34493544697761536,
+      "eval_runtime": 1156.8097,
+      "eval_samples_per_second": 6.495,
+      "eval_steps_per_second": 0.813,
+      "step": 24500
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 4.330002184359983e-05,
+      "loss": 0.2961,
+      "step": 24550
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 4.328636959370905e-05,
+      "loss": 0.2969,
+      "step": 24600
+    },
+    {
+      "epoch": 2.69,
+      "eval_cer": 0.06008739362375574,
+      "eval_loss": 0.341984361410141,
+      "eval_runtime": 1170.1423,
+      "eval_samples_per_second": 6.421,
+      "eval_steps_per_second": 0.803,
+      "step": 24600
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 4.3272717343818266e-05,
+      "loss": 0.2912,
+      "step": 24650
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 4.325906509392748e-05,
+      "loss": 0.3111,
+      "step": 24700
+    },
+    {
+      "epoch": 2.7,
+      "eval_cer": 0.060039487956857526,
+      "eval_loss": 0.3431969881057739,
+      "eval_runtime": 1162.683,
+      "eval_samples_per_second": 6.462,
+      "eval_steps_per_second": 0.808,
+      "step": 24700
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 4.32454128440367e-05,
+      "loss": 0.3057,
+      "step": 24750
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 4.323176059414591e-05,
+      "loss": 0.2951,
+      "step": 24800
+    },
+    {
+      "epoch": 2.71,
+      "eval_cer": 0.05896503228499766,
+      "eval_loss": 0.3386366367340088,
+      "eval_runtime": 1155.7752,
+      "eval_samples_per_second": 6.5,
+      "eval_steps_per_second": 0.813,
+      "step": 24800
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 4.3218108344255136e-05,
+      "loss": 0.2803,
+      "step": 24850
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 4.320445609436435e-05,
+      "loss": 0.2848,
+      "step": 24900
+    },
+    {
+      "epoch": 2.72,
+      "eval_cer": 0.06460079181223716,
+      "eval_loss": 0.34501972794532776,
+      "eval_runtime": 1185.2116,
+      "eval_samples_per_second": 6.339,
+      "eval_steps_per_second": 0.793,
+      "step": 24900
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 4.319080384447357e-05,
+      "loss": 0.2823,
+      "step": 24950
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 4.317715159458279e-05,
+      "loss": 0.3066,
+      "step": 25000
+    },
+    {
+      "epoch": 2.73,
+      "eval_cer": 0.06447076214494202,
+      "eval_loss": 0.3437393605709076,
+      "eval_runtime": 1201.6699,
+      "eval_samples_per_second": 6.252,
+      "eval_steps_per_second": 0.782,
+      "step": 25000
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 4.316349934469201e-05,
+      "loss": 0.2575,
+      "step": 25050
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 4.314984709480123e-05,
+      "loss": 0.2893,
+      "step": 25100
+    },
+    {
+      "epoch": 2.74,
+      "eval_cer": 0.06085046246077724,
+      "eval_loss": 0.3420575261116028,
+      "eval_runtime": 1164.3677,
+      "eval_samples_per_second": 6.452,
+      "eval_steps_per_second": 0.807,
+      "step": 25100
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 4.3136194844910444e-05,
+      "loss": 0.3092,
+      "step": 25150
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 4.312254259501966e-05,
+      "loss": 0.2743,
+      "step": 25200
+    },
+    {
+      "epoch": 2.75,
+      "eval_cer": 0.05815747961442782,
+      "eval_loss": 0.331033319234848,
+      "eval_runtime": 1172.6397,
+      "eval_samples_per_second": 6.407,
+      "eval_steps_per_second": 0.802,
+      "step": 25200
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 4.3108890345128876e-05,
+      "loss": 0.2971,
+      "step": 25250
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 4.30952380952381e-05,
+      "loss": 0.2997,
+      "step": 25300
+    },
+    {
+      "epoch": 2.76,
+      "eval_cer": 0.05869812928370762,
+      "eval_loss": 0.3358995318412781,
+      "eval_runtime": 1146.9262,
+      "eval_samples_per_second": 6.551,
+      "eval_steps_per_second": 0.82,
+      "step": 25300
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 4.3081585845347314e-05,
+      "loss": 0.2666,
+      "step": 25350
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 4.306793359545653e-05,
+      "loss": 0.2939,
+      "step": 25400
+    },
+    {
+      "epoch": 2.77,
+      "eval_cer": 0.058930813951498934,
+      "eval_loss": 0.33529844880104065,
+      "eval_runtime": 1187.2055,
+      "eval_samples_per_second": 6.328,
+      "eval_steps_per_second": 0.792,
+      "step": 25400
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 4.305428134556575e-05,
+      "loss": 0.289,
+      "step": 25450
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 4.304062909567497e-05,
+      "loss": 0.3051,
+      "step": 25500
+    },
+    {
+      "epoch": 2.79,
+      "eval_cer": 0.05896503228499766,
+      "eval_loss": 0.3372917175292969,
+      "eval_runtime": 1419.1461,
+      "eval_samples_per_second": 5.294,
+      "eval_steps_per_second": 0.662,
+      "step": 25500
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 4.302697684578419e-05,
+      "loss": 0.2801,
+      "step": 25550
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 4.301332459589341e-05,
+      "loss": 0.3016,
+      "step": 25600
+    },
+    {
+      "epoch": 2.8,
+      "eval_cer": 0.05905742178544421,
+      "eval_loss": 0.3363926410675049,
+      "eval_runtime": 1464.9044,
+      "eval_samples_per_second": 5.129,
+      "eval_steps_per_second": 0.642,
+      "step": 25600
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 4.299967234600262e-05,
+      "loss": 0.271,
+      "step": 25650
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 4.298602009611184e-05,
+      "loss": 0.3148,
+      "step": 25700
+    },
+    {
+      "epoch": 2.81,
+      "eval_cer": 0.05901635978524574,
+      "eval_loss": 0.33722782135009766,
+      "eval_runtime": 1460.2748,
+      "eval_samples_per_second": 5.145,
+      "eval_steps_per_second": 0.644,
+      "step": 25700
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 4.2972367846221054e-05,
+      "loss": 0.2866,
+      "step": 25750
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 4.295871559633028e-05,
+      "loss": 0.3225,
+      "step": 25800
+    },
+    {
+      "epoch": 2.82,
+      "eval_cer": 0.057562080611550055,
+      "eval_loss": 0.33164745569229126,
+      "eval_runtime": 1430.5344,
+      "eval_samples_per_second": 5.252,
+      "eval_steps_per_second": 0.657,
+      "step": 25800
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 4.294506334643949e-05,
+      "loss": 0.2966,
+      "step": 25850
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 4.293168414154653e-05,
+      "loss": 0.2998,
+      "step": 25900
+    },
+    {
+      "epoch": 2.83,
+      "eval_cer": 0.057589455278349036,
+      "eval_loss": 0.33457526564598083,
+      "eval_runtime": 1421.5781,
+      "eval_samples_per_second": 5.285,
+      "eval_steps_per_second": 0.661,
+      "step": 25900
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 4.2918031891655744e-05,
+      "loss": 0.2741,
+      "step": 25950
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 4.2904379641764966e-05,
+      "loss": 0.3025,
+      "step": 26000
+    },
+    {
+      "epoch": 2.84,
+      "eval_cer": 0.05718567894306412,
+      "eval_loss": 0.33198389410972595,
+      "eval_runtime": 1194.7499,
+      "eval_samples_per_second": 6.288,
+      "eval_steps_per_second": 0.787,
+      "step": 26000
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 4.289072739187418e-05,
+      "loss": 0.2841,
+      "step": 26050
+    },
+    {
+      "epoch": 2.85,
+      "learning_rate": 4.2877075141983405e-05,
+      "loss": 0.2882,
+      "step": 26100
+    },
+    {
+      "epoch": 2.85,
+      "eval_cer": 0.05745942561105389,
+      "eval_loss": 0.3311343789100647,
+      "eval_runtime": 1280.2736,
+      "eval_samples_per_second": 5.868,
+      "eval_steps_per_second": 0.734,
+      "step": 26100
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 4.286342289209262e-05,
+      "loss": 0.2725,
+      "step": 26150
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 4.2849770642201836e-05,
+      "loss": 0.3311,
+      "step": 26200
+    },
+    {
+      "epoch": 2.86,
+      "eval_cer": 0.05959464962137414,
+      "eval_loss": 0.33679836988449097,
+      "eval_runtime": 1337.0011,
+      "eval_samples_per_second": 5.619,
+      "eval_steps_per_second": 0.703,
+      "step": 26200
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 4.283611839231105e-05,
+      "loss": 0.2945,
+      "step": 26250
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 4.2822466142420275e-05,
+      "loss": 0.2976,
+      "step": 26300
+    },
+    {
+      "epoch": 2.87,
+      "eval_cer": 0.05775712511249277,
+      "eval_loss": 0.328110933303833,
+      "eval_runtime": 1396.3907,
+      "eval_samples_per_second": 5.38,
+      "eval_steps_per_second": 0.673,
+      "step": 26300
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 4.280881389252949e-05,
+      "loss": 0.311,
+      "step": 26350
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 4.2795161642638706e-05,
+      "loss": 0.2873,
+      "step": 26400
+    },
+    {
+      "epoch": 2.88,
+      "eval_cer": 0.0635263361403773,
+      "eval_loss": 0.32722070813179016,
+      "eval_runtime": 1462.697,
+      "eval_samples_per_second": 5.136,
+      "eval_steps_per_second": 0.643,
+      "step": 26400
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 4.278150939274793e-05,
+      "loss": 0.2689,
+      "step": 26450
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 4.2767857142857145e-05,
+      "loss": 0.2823,
+      "step": 26500
+    },
+    {
+      "epoch": 2.89,
+      "eval_cer": 0.06995938283813702,
+      "eval_loss": 0.34899574518203735,
+      "eval_runtime": 1723.4358,
+      "eval_samples_per_second": 4.359,
+      "eval_steps_per_second": 0.545,
+      "step": 26500
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 4.275420489296637e-05,
+      "loss": 0.2491,
+      "step": 26550
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 4.274055264307558e-05,
+      "loss": 0.2775,
+      "step": 26600
+    },
+    {
+      "epoch": 2.91,
+      "eval_cer": 0.05953647845442631,
+      "eval_loss": 0.33239951729774475,
+      "eval_runtime": 1151.2678,
+      "eval_samples_per_second": 6.526,
+      "eval_steps_per_second": 0.816,
+      "step": 26600
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 4.27269003931848e-05,
+      "loss": 0.2867,
+      "step": 26650
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 4.2713248143294015e-05,
+      "loss": 0.2721,
+      "step": 26700
+    },
+    {
+      "epoch": 2.92,
+      "eval_cer": 0.059091640118942926,
+      "eval_loss": 0.32586294412612915,
+      "eval_runtime": 1182.0981,
+      "eval_samples_per_second": 6.356,
+      "eval_steps_per_second": 0.795,
+      "step": 26700
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 4.269959589340323e-05,
+      "loss": 0.3009,
+      "step": 26750
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 4.268594364351245e-05,
+      "loss": 0.2994,
+      "step": 26800
+    },
+    {
+      "epoch": 2.93,
+      "eval_cer": 0.0615690474642504,
+      "eval_loss": 0.33627235889434814,
+      "eval_runtime": 1216.0666,
+      "eval_samples_per_second": 6.178,
+      "eval_steps_per_second": 0.773,
+      "step": 26800
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 4.267229139362167e-05,
+      "loss": 0.3008,
+      "step": 26850
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 4.265863914373089e-05,
+      "loss": 0.3001,
+      "step": 26900
+    },
+    {
+      "epoch": 2.94,
+      "eval_cer": 0.06160668763109899,
+      "eval_loss": 0.33180370926856995,
+      "eval_runtime": 1284.8633,
+      "eval_samples_per_second": 5.847,
+      "eval_steps_per_second": 0.732,
+      "step": 26900
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 4.264498689384011e-05,
+      "loss": 0.2861,
+      "step": 26950
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 4.263133464394932e-05,
+      "loss": 0.2527,
+      "step": 27000
+    },
+    {
+      "epoch": 2.95,
+      "eval_cer": 0.06202757313313327,
+      "eval_loss": 0.32616278529167175,
+      "eval_runtime": 1198.1406,
+      "eval_samples_per_second": 6.271,
+      "eval_steps_per_second": 0.785,
+      "step": 27000
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 4.2617682394058546e-05,
+      "loss": 0.2836,
+      "step": 27050
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 4.260403014416776e-05,
+      "loss": 0.2746,
+      "step": 27100
+    },
+    {
+      "epoch": 2.96,
+      "eval_cer": 0.058434648115767464,
+      "eval_loss": 0.32697415351867676,
+      "eval_runtime": 1148.8419,
+      "eval_samples_per_second": 6.54,
+      "eval_steps_per_second": 0.818,
+      "step": 27100
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 4.259037789427698e-05,
+      "loss": 0.2861,
+      "step": 27150
+    },
+    {
+      "epoch": 2.97,
+      "learning_rate": 4.257672564438619e-05,
+      "loss": 0.277,
+      "step": 27200
+    },
+    {
+      "epoch": 2.97,
+      "eval_cer": 0.057876889279738296,
+      "eval_loss": 0.32712793350219727,
+      "eval_runtime": 1150.7522,
+      "eval_samples_per_second": 6.529,
+      "eval_steps_per_second": 0.817,
+      "step": 27200
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 4.2563073394495416e-05,
+      "loss": 0.2656,
+      "step": 27250
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 4.254942114460463e-05,
+      "loss": 0.3059,
+      "step": 27300
+    },
+    {
+      "epoch": 2.98,
+      "eval_cer": 0.05807193378068101,
+      "eval_loss": 0.32142749428749084,
+      "eval_runtime": 1151.3245,
+      "eval_samples_per_second": 6.526,
+      "eval_steps_per_second": 0.816,
+      "step": 27300
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 4.2535768894713854e-05,
+      "loss": 0.2706,
+      "step": 27350
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 4.252211664482307e-05,
+      "loss": 0.2603,
+      "step": 27400
+    },
+    {
+      "epoch": 2.99,
+      "eval_cer": 0.05766815744539609,
+      "eval_loss": 0.3238297700881958,
+      "eval_runtime": 1147.7577,
+      "eval_samples_per_second": 6.546,
+      "eval_steps_per_second": 0.819,
+      "step": 27400
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 4.2508464394932286e-05,
+      "loss": 0.2459,
+      "step": 27450
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 4.249481214504151e-05,
+      "loss": 0.2803,
+      "step": 27500
+    },
+    {
+      "epoch": 3.0,
+      "eval_cer": 0.05597777177055923,
+      "eval_loss": 0.32824379205703735,
+      "eval_runtime": 1138.6321,
+      "eval_samples_per_second": 6.598,
+      "eval_steps_per_second": 0.826,
+      "step": 27500
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 4.2481159895150724e-05,
+      "loss": 0.2304,
+      "step": 27550
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 4.246750764525994e-05,
+      "loss": 0.233,
+      "step": 27600
+    },
+    {
+      "epoch": 3.01,
+      "eval_cer": 0.060460373458891804,
+      "eval_loss": 0.32600587606430054,
+      "eval_runtime": 1173.6339,
+      "eval_samples_per_second": 6.401,
+      "eval_steps_per_second": 0.801,
+      "step": 27600
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 4.2453855395369156e-05,
+      "loss": 0.2192,
+      "step": 27650
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 4.244020314547837e-05,
+      "loss": 0.245,
+      "step": 27700
+    },
+    {
+      "epoch": 3.03,
+      "eval_cer": 0.05650815593978942,
+      "eval_loss": 0.3225695788860321,
+      "eval_runtime": 1140.3828,
+      "eval_samples_per_second": 6.588,
+      "eval_steps_per_second": 0.824,
+      "step": 27700
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 4.2426550895587594e-05,
+      "loss": 0.2309,
+      "step": 27750
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 4.241289864569682e-05,
+      "loss": 0.241,
+      "step": 27800
+    },
+    {
+      "epoch": 3.04,
+      "eval_cer": 0.05897529778504727,
+      "eval_loss": 0.3347548544406891,
+      "eval_runtime": 1129.3341,
+      "eval_samples_per_second": 6.653,
+      "eval_steps_per_second": 0.832,
+      "step": 27800
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 4.239924639580603e-05,
+      "loss": 0.2422,
+      "step": 27850
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 4.238559414591525e-05,
+      "loss": 0.2394,
+      "step": 27900
+    },
+    {
+      "epoch": 3.05,
+      "eval_cer": 0.05539606010108096,
+      "eval_loss": 0.32074490189552307,
+      "eval_runtime": 1144.36,
+      "eval_samples_per_second": 6.565,
+      "eval_steps_per_second": 0.821,
+      "step": 27900
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 4.237194189602447e-05,
+      "loss": 0.2108,
+      "step": 27950
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 4.235828964613369e-05,
+      "loss": 0.2254,
+      "step": 28000
+    },
+    {
+      "epoch": 3.06,
+      "eval_cer": 0.05462956943070959,
+      "eval_loss": 0.318286269903183,
+      "eval_runtime": 1136.2023,
+      "eval_samples_per_second": 6.612,
+      "eval_steps_per_second": 0.827,
+      "step": 28000
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 4.23446373962429e-05,
+      "loss": 0.2219,
+      "step": 28050
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 4.233098514635212e-05,
+      "loss": 0.2416,
+      "step": 28100
+    },
+    {
+      "epoch": 3.07,
+      "eval_cer": 0.056121488771253863,
+      "eval_loss": 0.322205513715744,
+      "eval_runtime": 1151.0902,
+      "eval_samples_per_second": 6.527,
+      "eval_steps_per_second": 0.817,
+      "step": 28100
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 4.2317332896461334e-05,
+      "loss": 0.2157,
+      "step": 28150
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 4.230368064657056e-05,
+      "loss": 0.2262,
+      "step": 28200
+    },
+    {
+      "epoch": 3.08,
+      "eval_cer": 0.05716172610961501,
+      "eval_loss": 0.32194942235946655,
+      "eval_runtime": 1150.2928,
+      "eval_samples_per_second": 6.531,
+      "eval_steps_per_second": 0.817,
+      "step": 28200
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 4.229002839667977e-05,
+      "loss": 0.2428,
+      "step": 28250
+    },
+    {
+      "epoch": 3.09,
+      "learning_rate": 4.2276376146788995e-05,
+      "loss": 0.2193,
+      "step": 28300
+    },
+    {
+      "epoch": 3.09,
+      "eval_cer": 0.05849281928271529,
+      "eval_loss": 0.32632672786712646,
+      "eval_runtime": 1147.7549,
+      "eval_samples_per_second": 6.546,
+      "eval_steps_per_second": 0.819,
+      "step": 28300
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 4.226272389689821e-05,
+      "loss": 0.2357,
+      "step": 28350
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 4.224907164700743e-05,
+      "loss": 0.248,
+      "step": 28400
+    },
+    {
+      "epoch": 3.1,
+      "eval_cer": 0.05890001745135008,
+      "eval_loss": 0.320720911026001,
+      "eval_runtime": 1158.2635,
+      "eval_samples_per_second": 6.486,
+      "eval_steps_per_second": 0.812,
+      "step": 28400
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 4.223541939711665e-05,
+      "loss": 0.2138,
+      "step": 28450
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 4.2221767147225865e-05,
+      "loss": 0.2215,
+      "step": 28500
+    },
+    {
+      "epoch": 3.11,
+      "eval_cer": 0.05739441077740632,
+      "eval_loss": 0.32517117261886597,
+      "eval_runtime": 1159.9181,
+      "eval_samples_per_second": 6.477,
+      "eval_steps_per_second": 0.81,
+      "step": 28500
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 4.22083879423329e-05,
+      "loss": 0.2348,
+      "step": 28550
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 4.2194735692442116e-05,
+      "loss": 0.2149,
+      "step": 28600
+    },
+    {
+      "epoch": 3.12,
+      "eval_cer": 0.05660738910693571,
+      "eval_loss": 0.3221997618675232,
+      "eval_runtime": 1165.8197,
+      "eval_samples_per_second": 6.444,
+      "eval_steps_per_second": 0.806,
+      "step": 28600
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 4.218108344255133e-05,
+      "loss": 0.237,
+      "step": 28650
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 4.216743119266055e-05,
+      "loss": 0.2135,
+      "step": 28700
+    },
+    {
+      "epoch": 3.13,
+      "eval_cer": 0.057018009108920374,
+      "eval_loss": 0.32162022590637207,
+      "eval_runtime": 1320.2861,
+      "eval_samples_per_second": 5.69,
+      "eval_steps_per_second": 0.712,
+      "step": 28700
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 4.215377894276977e-05,
+      "loss": 0.2187,
+      "step": 28750
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 4.2140126692878986e-05,
+      "loss": 0.2358,
+      "step": 28800
+    },
+    {
+      "epoch": 3.15,
+      "eval_cer": 0.05756550244489993,
+      "eval_loss": 0.3216935396194458,
+      "eval_runtime": 1263.617,
+      "eval_samples_per_second": 5.946,
+      "eval_steps_per_second": 0.744,
+      "step": 28800
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 4.212647444298821e-05,
+      "loss": 0.2094,
+      "step": 28850
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 4.2112822193097425e-05,
+      "loss": 0.2131,
+      "step": 28900
+    },
+    {
+      "epoch": 3.16,
+      "eval_cer": 0.05689824494167485,
+      "eval_loss": 0.3266179859638214,
+      "eval_runtime": 1161.2332,
+      "eval_samples_per_second": 6.47,
+      "eval_steps_per_second": 0.809,
+      "step": 28900
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 4.209916994320664e-05,
+      "loss": 0.2061,
+      "step": 28950
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 4.208551769331586e-05,
+      "loss": 0.2162,
+      "step": 29000
+    },
+    {
+      "epoch": 3.17,
+      "eval_cer": 0.05746284744440376,
+      "eval_loss": 0.32129406929016113,
+      "eval_runtime": 1211.1118,
+      "eval_samples_per_second": 6.203,
+      "eval_steps_per_second": 0.776,
+      "step": 29000
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 4.207186544342508e-05,
+      "loss": 0.2245,
+      "step": 29050
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 4.2058213193534295e-05,
+      "loss": 0.221,
+      "step": 29100
+    },
+    {
+      "epoch": 3.18,
+      "eval_cer": 0.05628573677204773,
+      "eval_loss": 0.3239326477050781,
+      "eval_runtime": 1182.7128,
+      "eval_samples_per_second": 6.352,
+      "eval_steps_per_second": 0.795,
+      "step": 29100
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 4.204456094364351e-05,
+      "loss": 0.2451,
+      "step": 29150
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 4.203090869375273e-05,
+      "loss": 0.2629,
+      "step": 29200
+    },
+    {
+      "epoch": 3.19,
+      "eval_cer": 0.056980368942071785,
+      "eval_loss": 0.3187606930732727,
+      "eval_runtime": 1203.5217,
+      "eval_samples_per_second": 6.243,
+      "eval_steps_per_second": 0.781,
+      "step": 29200
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 4.201725644386195e-05,
+      "loss": 0.2233,
+      "step": 29250
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 4.200360419397117e-05,
+      "loss": 0.2185,
+      "step": 29300
+    },
+    {
+      "epoch": 3.2,
+      "eval_cer": 0.05524207760033671,
+      "eval_loss": 0.31554532051086426,
+      "eval_runtime": 1201.7452,
+      "eval_samples_per_second": 6.252,
+      "eval_steps_per_second": 0.782,
+      "step": 29300
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 4.198995194408039e-05,
+      "loss": 0.2282,
+      "step": 29350
+    },
+    {
+      "epoch": 3.21,
+      "learning_rate": 4.19762996941896e-05,
+      "loss": 0.2115,
+      "step": 29400
+    },
+    {
+      "epoch": 3.21,
+      "eval_cer": 0.05779818711269124,
+      "eval_loss": 0.31857335567474365,
+      "eval_runtime": 1213.1112,
+      "eval_samples_per_second": 6.193,
+      "eval_steps_per_second": 0.775,
+      "step": 29400
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 4.1962647444298826e-05,
+      "loss": 0.2139,
+      "step": 29450
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 4.194899519440804e-05,
+      "loss": 0.2145,
+      "step": 29500
+    },
+    {
+      "epoch": 3.22,
+      "eval_cer": 0.06090863362772506,
+      "eval_loss": 0.3194052278995514,
+      "eval_runtime": 1182.218,
+      "eval_samples_per_second": 6.355,
+      "eval_steps_per_second": 0.795,
+      "step": 29500
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 4.193534294451726e-05,
+      "loss": 0.2111,
+      "step": 29550
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 4.192169069462647e-05,
+      "loss": 0.2189,
+      "step": 29600
+    },
+    {
+      "epoch": 3.23,
+      "eval_cer": 0.057295177610260026,
+      "eval_loss": 0.3254971206188202,
+      "eval_runtime": 1158.4532,
+      "eval_samples_per_second": 6.485,
+      "eval_steps_per_second": 0.811,
+      "step": 29600
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 4.190803844473569e-05,
+      "loss": 0.2112,
+      "step": 29650
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 4.189438619484491e-05,
+      "loss": 0.2175,
+      "step": 29700
+    },
+    {
+      "epoch": 3.24,
+      "eval_cer": 0.05672373144083137,
+      "eval_loss": 0.3198707401752472,
+      "eval_runtime": 1157.0972,
+      "eval_samples_per_second": 6.493,
+      "eval_steps_per_second": 0.812,
+      "step": 29700
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 4.1880733944954134e-05,
+      "loss": 0.2352,
+      "step": 29750
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 4.186708169506335e-05,
+      "loss": 0.2431,
+      "step": 29800
+    },
+    {
+      "epoch": 3.25,
+      "eval_cer": 0.05510862609969169,
+      "eval_loss": 0.3135843276977539,
+      "eval_runtime": 1163.9894,
+      "eval_samples_per_second": 6.455,
+      "eval_steps_per_second": 0.808,
+      "step": 29800
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 4.1853429445172566e-05,
+      "loss": 0.216,
+      "step": 29850
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 4.183977719528178e-05,
+      "loss": 0.2095,
+      "step": 29900
+    },
+    {
+      "epoch": 3.27,
+      "eval_cer": 0.05986839628936391,
+      "eval_loss": 0.316039502620697,
+      "eval_runtime": 1182.2319,
+      "eval_samples_per_second": 6.355,
+      "eval_steps_per_second": 0.795,
+      "step": 29900
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 4.1826124945391004e-05,
+      "loss": 0.218,
+      "step": 29950
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 4.181247269550022e-05,
+      "loss": 0.2449,
+      "step": 30000
+    },
+    {
+      "epoch": 3.28,
+      "eval_cer": 0.05585116393661396,
+      "eval_loss": 0.31165117025375366,
+      "eval_runtime": 1175.7213,
+      "eval_samples_per_second": 6.39,
+      "eval_steps_per_second": 0.8,
+      "step": 30000
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 4.1798820445609436e-05,
+      "loss": 0.2124,
+      "step": 30050
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 4.178516819571865e-05,
+      "loss": 0.2007,
+      "step": 30100
+    },
+    {
+      "epoch": 3.29,
+      "eval_cer": 0.05651842143983904,
+      "eval_loss": 0.31462639570236206,
+      "eval_runtime": 1169.6031,
+      "eval_samples_per_second": 6.424,
+      "eval_steps_per_second": 0.804,
+      "step": 30100
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 4.1771515945827874e-05,
+      "loss": 0.2019,
+      "step": 30150
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 4.17578636959371e-05,
+      "loss": 0.2395,
+      "step": 30200
+    },
+    {
+      "epoch": 3.3,
+      "eval_cer": 0.05408549792807991,
+      "eval_loss": 0.31565436720848083,
+      "eval_runtime": 1163.1123,
+      "eval_samples_per_second": 6.459,
+      "eval_steps_per_second": 0.808,
+      "step": 30200
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 4.174421144604631e-05,
+      "loss": 0.2252,
+      "step": 30250
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 4.173055919615553e-05,
+      "loss": 0.2166,
+      "step": 30300
+    },
+    {
+      "epoch": 3.31,
+      "eval_cer": 0.05582721110316485,
+      "eval_loss": 0.3155282437801361,
+      "eval_runtime": 1216.7838,
+      "eval_samples_per_second": 6.174,
+      "eval_steps_per_second": 0.773,
+      "step": 30300
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 4.1716906946264744e-05,
+      "loss": 0.224,
+      "step": 30350
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 4.170325469637397e-05,
+      "loss": 0.2512,
+      "step": 30400
+    },
+    {
+      "epoch": 3.32,
+      "eval_cer": 0.055180484600039006,
+      "eval_loss": 0.31320416927337646,
+      "eval_runtime": 1208.865,
+      "eval_samples_per_second": 6.215,
+      "eval_steps_per_second": 0.778,
+      "step": 30400
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 4.168960244648318e-05,
+      "loss": 0.2261,
+      "step": 30450
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 4.16759501965924e-05,
+      "loss": 0.2319,
+      "step": 30500
+    },
+    {
+      "epoch": 3.33,
+      "eval_cer": 0.055782727269616514,
+      "eval_loss": 0.3205634355545044,
+      "eval_runtime": 1175.9742,
+      "eval_samples_per_second": 6.389,
+      "eval_steps_per_second": 0.799,
+      "step": 30500
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 4.1662297946701614e-05,
+      "loss": 0.222,
+      "step": 30550
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 4.164864569681084e-05,
+      "loss": 0.2333,
+      "step": 30600
+    },
+    {
+      "epoch": 3.34,
+      "eval_cer": 0.05437977559616892,
+      "eval_loss": 0.31233859062194824,
+      "eval_runtime": 1187.2819,
+      "eval_samples_per_second": 6.328,
+      "eval_steps_per_second": 0.792,
+      "step": 30600
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 4.163499344692006e-05,
+      "loss": 0.2463,
+      "step": 30650
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 4.1621341197029275e-05,
+      "loss": 0.2376,
+      "step": 30700
+    },
+    {
+      "epoch": 3.35,
+      "eval_cer": 0.05478355193145384,
+      "eval_loss": 0.31242722272872925,
+      "eval_runtime": 1189.2518,
+      "eval_samples_per_second": 6.317,
+      "eval_steps_per_second": 0.79,
+      "step": 30700
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 4.160768894713849e-05,
+      "loss": 0.2144,
+      "step": 30750
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 4.159403669724771e-05,
+      "loss": 0.2149,
+      "step": 30800
+    },
+    {
+      "epoch": 3.36,
+      "eval_cer": 0.05400337392768297,
+      "eval_loss": 0.31088629364967346,
+      "eval_runtime": 1204.1106,
+      "eval_samples_per_second": 6.239,
+      "eval_steps_per_second": 0.781,
+      "step": 30800
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 4.158038444735692e-05,
+      "loss": 0.2245,
+      "step": 30850
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 4.1566732197466145e-05,
+      "loss": 0.2346,
+      "step": 30900
+    },
+    {
+      "epoch": 3.37,
+      "eval_cer": 0.053336116424457895,
+      "eval_loss": 0.3114325702190399,
+      "eval_runtime": 1199.4806,
+      "eval_samples_per_second": 6.264,
+      "eval_steps_per_second": 0.784,
+      "step": 30900
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 4.155307994757536e-05,
+      "loss": 0.2346,
+      "step": 30950
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 4.153942769768458e-05,
+      "loss": 0.22,
+      "step": 31000
+    },
+    {
+      "epoch": 3.39,
+      "eval_cer": 0.05588196043676281,
+      "eval_loss": 0.3127199113368988,
+      "eval_runtime": 1171.8911,
+      "eval_samples_per_second": 6.411,
+      "eval_steps_per_second": 0.802,
+      "step": 31000
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 4.15257754477938e-05,
+      "loss": 0.2366,
+      "step": 31050
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 4.151212319790302e-05,
+      "loss": 0.2589,
+      "step": 31100
+    },
+    {
+      "epoch": 3.4,
+      "eval_cer": 0.0532129304238625,
+      "eval_loss": 0.3078201115131378,
+      "eval_runtime": 1179.1501,
+      "eval_samples_per_second": 6.372,
+      "eval_steps_per_second": 0.797,
+      "step": 31100
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 4.149847094801224e-05,
+      "loss": 0.2256,
+      "step": 31150
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 4.1484818698121454e-05,
+      "loss": 0.2331,
+      "step": 31200
+    },
+    {
+      "epoch": 3.41,
+      "eval_cer": 0.055495293268227254,
+      "eval_loss": 0.30701497197151184,
+      "eval_runtime": 1363.5561,
+      "eval_samples_per_second": 5.51,
+      "eval_steps_per_second": 0.689,
+      "step": 31200
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 4.147116644823067e-05,
+      "loss": 0.2491,
+      "step": 31250
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 4.1457514198339885e-05,
+      "loss": 0.1912,
+      "step": 31300
+    },
+    {
+      "epoch": 3.42,
+      "eval_cer": 0.054999127432495784,
+      "eval_loss": 0.31193986535072327,
+      "eval_runtime": 1470.9109,
+      "eval_samples_per_second": 5.108,
+      "eval_steps_per_second": 0.639,
+      "step": 31300
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 4.144386194844911e-05,
+      "loss": 0.2294,
+      "step": 31350
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 4.1430209698558324e-05,
+      "loss": 0.2178,
+      "step": 31400
+    },
+    {
+      "epoch": 3.43,
+      "eval_cer": 0.055721134269318816,
+      "eval_loss": 0.31223970651626587,
+      "eval_runtime": 1504.6961,
+      "eval_samples_per_second": 4.993,
+      "eval_steps_per_second": 0.625,
+      "step": 31400
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 4.141655744866754e-05,
+      "loss": 0.227,
+      "step": 31450
+    },
+    {
+      "epoch": 3.44,
+      "learning_rate": 4.140290519877676e-05,
+      "loss": 0.2341,
+      "step": 31500
+    },
+    {
+      "epoch": 3.44,
+      "eval_cer": 0.05363723775924665,
+      "eval_loss": 0.3130796253681183,
+      "eval_runtime": 1438.0987,
+      "eval_samples_per_second": 5.224,
+      "eval_steps_per_second": 0.654,
+      "step": 31500
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 4.138925294888598e-05,
+      "loss": 0.2199,
+      "step": 31550
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 4.13756006989952e-05,
+      "loss": 0.227,
+      "step": 31600
+    },
+    {
+      "epoch": 3.45,
+      "eval_cer": 0.052867325255525406,
+      "eval_loss": 0.3069811761379242,
+      "eval_runtime": 1423.702,
+      "eval_samples_per_second": 5.277,
+      "eval_steps_per_second": 0.66,
+      "step": 31600
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 4.1361948449104417e-05,
+      "loss": 0.2467,
+      "step": 31650
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 4.134856924421145e-05,
+      "loss": 0.2168,
+      "step": 31700
+    },
+    {
+      "epoch": 3.46,
+      "eval_cer": 0.057274646610160795,
+      "eval_loss": 0.30950018763542175,
+      "eval_runtime": 1485.941,
+      "eval_samples_per_second": 5.056,
+      "eval_steps_per_second": 0.633,
+      "step": 31700
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 4.133491699432067e-05,
+      "loss": 0.2426,
+      "step": 31750
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 4.132126474442988e-05,
+      "loss": 0.2088,
+      "step": 31800
+    },
+    {
+      "epoch": 3.47,
+      "eval_cer": 0.05539263826773109,
+      "eval_loss": 0.3096722960472107,
+      "eval_runtime": 1520.0393,
+      "eval_samples_per_second": 4.943,
+      "eval_steps_per_second": 0.618,
+      "step": 31800
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 4.13076124945391e-05,
+      "loss": 0.2269,
+      "step": 31850
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 4.129396024464832e-05,
+      "loss": 0.2332,
+      "step": 31900
+    },
+    {
+      "epoch": 3.48,
+      "eval_cer": 0.05160124691607269,
+      "eval_loss": 0.3041413426399231,
+      "eval_runtime": 1479.0494,
+      "eval_samples_per_second": 5.08,
+      "eval_steps_per_second": 0.636,
+      "step": 31900
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 4.128030799475754e-05,
+      "loss": 0.2293,
+      "step": 31950
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 4.126665574486675e-05,
+      "loss": 0.2501,
+      "step": 32000
+    },
+    {
+      "epoch": 3.49,
+      "eval_cer": 0.05571429060261907,
+      "eval_loss": 0.3081224262714386,
+      "eval_runtime": 1376.0064,
+      "eval_samples_per_second": 5.46,
+      "eval_steps_per_second": 0.683,
+      "step": 32000
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 4.1253003494975976e-05,
+      "loss": 0.2398,
+      "step": 32050
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 4.123935124508519e-05,
+      "loss": 0.2375,
+      "step": 32100
+    },
+    {
+      "epoch": 3.51,
+      "eval_cer": 0.053113697256716205,
+      "eval_loss": 0.3007121980190277,
+      "eval_runtime": 1448.1367,
+      "eval_samples_per_second": 5.188,
+      "eval_steps_per_second": 0.649,
+      "step": 32100
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 4.1225698995194414e-05,
+      "loss": 0.2178,
+      "step": 32150
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 4.121204674530363e-05,
+      "loss": 0.213,
+      "step": 32200
+    },
+    {
+      "epoch": 3.52,
+      "eval_cer": 0.05521470293353773,
+      "eval_loss": 0.3040596842765808,
+      "eval_runtime": 1193.1522,
+      "eval_samples_per_second": 6.297,
+      "eval_steps_per_second": 0.788,
+      "step": 32200
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 4.1198394495412846e-05,
+      "loss": 0.1994,
+      "step": 32250
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 4.118474224552206e-05,
+      "loss": 0.2116,
+      "step": 32300
+    },
+    {
+      "epoch": 3.53,
+      "eval_cer": 0.05221717691904969,
+      "eval_loss": 0.3014806807041168,
+      "eval_runtime": 1146.7847,
+      "eval_samples_per_second": 6.551,
+      "eval_steps_per_second": 0.82,
+      "step": 32300
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 4.1171089995631284e-05,
+      "loss": 0.2131,
+      "step": 32350
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 4.11574377457405e-05,
+      "loss": 0.2425,
+      "step": 32400
+    },
+    {
+      "epoch": 3.54,
+      "eval_cer": 0.058444913615817086,
+      "eval_loss": 0.3058461546897888,
+      "eval_runtime": 1212.9428,
+      "eval_samples_per_second": 6.194,
+      "eval_steps_per_second": 0.775,
+      "step": 32400
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 4.1143785495849716e-05,
+      "loss": 0.2267,
+      "step": 32450
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 4.113013324595894e-05,
+      "loss": 0.2195,
+      "step": 32500
+    },
+    {
+      "epoch": 3.55,
+      "eval_cer": 0.05577588360291677,
+      "eval_loss": 0.302491158246994,
+      "eval_runtime": 1260.9141,
+      "eval_samples_per_second": 5.958,
+      "eval_steps_per_second": 0.745,
+      "step": 32500
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 4.1116480996068154e-05,
+      "loss": 0.2363,
+      "step": 32550
+    },
+    {
+      "epoch": 3.56,
+      "learning_rate": 4.110282874617738e-05,
+      "loss": 0.217,
+      "step": 32600
+    },
+    {
+      "epoch": 3.56,
+      "eval_cer": 0.05627547127199811,
+      "eval_loss": 0.30593034625053406,
+      "eval_runtime": 1316.3658,
+      "eval_samples_per_second": 5.707,
+      "eval_steps_per_second": 0.714,
+      "step": 32600
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 4.108917649628659e-05,
+      "loss": 0.2104,
+      "step": 32650
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 4.107552424639581e-05,
+      "loss": 0.226,
+      "step": 32700
+    },
+    {
+      "epoch": 3.57,
+      "eval_cer": 0.05334638192450751,
+      "eval_loss": 0.30213332176208496,
+      "eval_runtime": 1175.6285,
+      "eval_samples_per_second": 6.391,
+      "eval_steps_per_second": 0.8,
+      "step": 32700
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 4.1061871996505024e-05,
+      "loss": 0.2309,
+      "step": 32750
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 4.104821974661424e-05,
+      "loss": 0.245,
+      "step": 32800
+    },
+    {
+      "epoch": 3.58,
+      "eval_cer": 0.05487251959855051,
+      "eval_loss": 0.30643877387046814,
+      "eval_runtime": 1166.4446,
+      "eval_samples_per_second": 6.441,
+      "eval_steps_per_second": 0.806,
+      "step": 32800
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 4.103456749672346e-05,
+      "loss": 0.2258,
+      "step": 32850
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 4.102091524683268e-05,
+      "loss": 0.2279,
+      "step": 32900
+    },
+    {
+      "epoch": 3.59,
+      "eval_cer": 0.053449036925003676,
+      "eval_loss": 0.3022903501987457,
+      "eval_runtime": 1152.2154,
+      "eval_samples_per_second": 6.52,
+      "eval_steps_per_second": 0.816,
+      "step": 32900
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 4.1007262996941894e-05,
+      "loss": 0.2067,
+      "step": 32950
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 4.099361074705112e-05,
+      "loss": 0.2144,
+      "step": 33000
+    },
+    {
+      "epoch": 3.6,
+      "eval_cer": 0.052997354922820546,
+      "eval_loss": 0.298502653837204,
+      "eval_runtime": 1158.0782,
+      "eval_samples_per_second": 6.487,
+      "eval_steps_per_second": 0.812,
+      "step": 33000
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 4.097995849716034e-05,
+      "loss": 0.2372,
+      "step": 33050
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 4.0966306247269555e-05,
+      "loss": 0.254,
+      "step": 33100
+    },
+    {
+      "epoch": 3.62,
+      "eval_cer": 0.05426685509562313,
+      "eval_loss": 0.30310696363449097,
+      "eval_runtime": 1161.4263,
+      "eval_samples_per_second": 6.469,
+      "eval_steps_per_second": 0.809,
+      "step": 33100
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 4.095265399737877e-05,
+      "loss": 0.2082,
+      "step": 33150
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 4.093900174748799e-05,
+      "loss": 0.2498,
+      "step": 33200
+    },
+    {
+      "epoch": 3.63,
+      "eval_cer": 0.05420526209532543,
+      "eval_loss": 0.29893621802330017,
+      "eval_runtime": 1174.6966,
+      "eval_samples_per_second": 6.396,
+      "eval_steps_per_second": 0.8,
+      "step": 33200
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 4.09253494975972e-05,
+      "loss": 0.2326,
+      "step": 33250
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 4.0911697247706425e-05,
+      "loss": 0.2402,
+      "step": 33300
+    },
+    {
+      "epoch": 3.64,
+      "eval_cer": 0.057024852775620125,
+      "eval_loss": 0.30172908306121826,
+      "eval_runtime": 1168.0386,
+      "eval_samples_per_second": 6.432,
+      "eval_steps_per_second": 0.805,
+      "step": 33300
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 4.089804499781564e-05,
+      "loss": 0.2413,
+      "step": 33350
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 4.088439274792486e-05,
+      "loss": 0.2352,
+      "step": 33400
+    },
+    {
+      "epoch": 3.65,
+      "eval_cer": 0.05724042827666207,
+      "eval_loss": 0.29973098635673523,
+      "eval_runtime": 1177.8105,
+      "eval_samples_per_second": 6.379,
+      "eval_steps_per_second": 0.798,
+      "step": 33400
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 4.087074049803408e-05,
+      "loss": 0.2159,
+      "step": 33450
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 4.0857088248143295e-05,
+      "loss": 0.2342,
+      "step": 33500
+    },
+    {
+      "epoch": 3.66,
+      "eval_cer": 0.05494095626554796,
+      "eval_loss": 0.3017541170120239,
+      "eval_runtime": 1163.4535,
+      "eval_samples_per_second": 6.457,
+      "eval_steps_per_second": 0.808,
+      "step": 33500
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 4.084343599825252e-05,
+      "loss": 0.2366,
+      "step": 33550
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 4.0829783748361734e-05,
+      "loss": 0.2155,
+      "step": 33600
+    },
+    {
+      "epoch": 3.67,
+      "eval_cer": 0.055628744768872265,
+      "eval_loss": 0.3011651635169983,
+      "eval_runtime": 1171.9956,
+      "eval_samples_per_second": 6.41,
+      "eval_steps_per_second": 0.802,
+      "step": 33600
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 4.081613149847095e-05,
+      "loss": 0.2168,
+      "step": 33650
+    },
+    {
+      "epoch": 3.68,
+      "learning_rate": 4.0802479248580165e-05,
+      "loss": 0.219,
+      "step": 33700
+    },
+    {
+      "epoch": 3.68,
+      "eval_cer": 0.05231298825284611,
+      "eval_loss": 0.2997644245624542,
+      "eval_runtime": 1153.1499,
+      "eval_samples_per_second": 6.515,
+      "eval_steps_per_second": 0.815,
+      "step": 33700
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 4.078882699868938e-05,
+      "loss": 0.2006,
+      "step": 33750
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 4.0775174748798604e-05,
+      "loss": 0.2,
+      "step": 33800
+    },
+    {
+      "epoch": 3.69,
+      "eval_cer": 0.05348667709185227,
+      "eval_loss": 0.2957291901111603,
+      "eval_runtime": 1161.6858,
+      "eval_samples_per_second": 6.467,
+      "eval_steps_per_second": 0.809,
+      "step": 33800
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 4.076152249890782e-05,
+      "loss": 0.2528,
+      "step": 33850
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 4.074787024901704e-05,
+      "loss": 0.2247,
+      "step": 33900
+    },
+    {
+      "epoch": 3.7,
+      "eval_cer": 0.05367145609274537,
+      "eval_loss": 0.29320961236953735,
+      "eval_runtime": 1170.6891,
+      "eval_samples_per_second": 6.418,
+      "eval_steps_per_second": 0.803,
+      "step": 33900
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 4.073421799912626e-05,
+      "loss": 0.2122,
+      "step": 33950
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 4.072056574923548e-05,
+      "loss": 0.2386,
+      "step": 34000
+    },
+    {
+      "epoch": 3.71,
+      "eval_cer": 0.05359617575904818,
+      "eval_loss": 0.29429566860198975,
+      "eval_runtime": 1174.7513,
+      "eval_samples_per_second": 6.395,
+      "eval_steps_per_second": 0.8,
+      "step": 34000
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 4.0706913499344696e-05,
+      "loss": 0.2547,
+      "step": 34050
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 4.069326124945391e-05,
+      "loss": 0.2235,
+      "step": 34100
+    },
+    {
+      "epoch": 3.72,
+      "eval_cer": 0.05282284142197707,
+      "eval_loss": 0.293638676404953,
+      "eval_runtime": 1171.6751,
+      "eval_samples_per_second": 6.412,
+      "eval_steps_per_second": 0.802,
+      "step": 34100
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 4.067960899956313e-05,
+      "loss": 0.2224,
+      "step": 34150
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 4.0665956749672344e-05,
+      "loss": 0.2245,
+      "step": 34200
+    },
+    {
+      "epoch": 3.74,
+      "eval_cer": 0.052275348085997514,
+      "eval_loss": 0.30025115609169006,
+      "eval_runtime": 1146.2881,
+      "eval_samples_per_second": 6.554,
+      "eval_steps_per_second": 0.82,
+      "step": 34200
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 4.0652304499781567e-05,
+      "loss": 0.2471,
+      "step": 34250
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 4.063865224989078e-05,
+      "loss": 0.2363,
+      "step": 34300
+    },
+    {
+      "epoch": 3.75,
+      "eval_cer": 0.053336116424457895,
+      "eval_loss": 0.2921861410140991,
+      "eval_runtime": 1188.9193,
+      "eval_samples_per_second": 6.319,
+      "eval_steps_per_second": 0.791,
+      "step": 34300
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 4.0625000000000005e-05,
+      "loss": 0.2122,
+      "step": 34350
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 4.061134775010922e-05,
+      "loss": 0.2613,
+      "step": 34400
+    },
+    {
+      "epoch": 3.76,
+      "eval_cer": 0.05454060176361291,
+      "eval_loss": 0.29321932792663574,
+      "eval_runtime": 1219.9866,
+      "eval_samples_per_second": 6.158,
+      "eval_steps_per_second": 0.771,
+      "step": 34400
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 4.0597695500218437e-05,
+      "loss": 0.2628,
+      "step": 34450
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 4.058404325032766e-05,
+      "loss": 0.2226,
+      "step": 34500
+    },
+    {
+      "epoch": 3.77,
+      "eval_cer": 0.051833931583864006,
+      "eval_loss": 0.29715830087661743,
+      "eval_runtime": 1206.9387,
+      "eval_samples_per_second": 6.225,
+      "eval_steps_per_second": 0.779,
+      "step": 34500
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 4.0570391000436875e-05,
+      "loss": 0.1874,
+      "step": 34550
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 4.055673875054609e-05,
+      "loss": 0.244,
+      "step": 34600
+    },
+    {
+      "epoch": 3.78,
+      "eval_cer": 0.051714167416618474,
+      "eval_loss": 0.29057419300079346,
+      "eval_runtime": 1202.0051,
+      "eval_samples_per_second": 6.25,
+      "eval_steps_per_second": 0.782,
+      "step": 34600
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 4.0543086500655307e-05,
+      "loss": 0.2246,
+      "step": 34650
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 4.052943425076452e-05,
+      "loss": 0.2198,
+      "step": 34700
+    },
+    {
+      "epoch": 3.79,
+      "eval_cer": 0.05181340058376477,
+      "eval_loss": 0.2926875650882721,
+      "eval_runtime": 1199.7282,
+      "eval_samples_per_second": 6.262,
+      "eval_steps_per_second": 0.784,
+      "step": 34700
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 4.0515782000873745e-05,
+      "loss": 0.2338,
+      "step": 34750
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 4.050212975098297e-05,
+      "loss": 0.2304,
+      "step": 34800
+    },
+    {
+      "epoch": 3.8,
+      "eval_cer": 0.05333953825780777,
+      "eval_loss": 0.2972395420074463,
+      "eval_runtime": 1220.1743,
+      "eval_samples_per_second": 6.157,
+      "eval_steps_per_second": 0.77,
+      "step": 34800
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 4.0488750546089996e-05,
+      "loss": 0.2541,
+      "step": 34850
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 4.047509829619922e-05,
+      "loss": 0.2478,
+      "step": 34900
+    },
+    {
+      "epoch": 3.81,
+      "eval_cer": 0.052597000420885505,
+      "eval_loss": 0.2926911413669586,
+      "eval_runtime": 1192.6912,
+      "eval_samples_per_second": 6.299,
+      "eval_steps_per_second": 0.788,
+      "step": 34900
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 4.0461446046308434e-05,
+      "loss": 0.224,
+      "step": 34950
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 4.044779379641765e-05,
+      "loss": 0.2234,
+      "step": 35000
+    },
+    {
+      "epoch": 3.82,
+      "eval_cer": 0.05634048610564568,
+      "eval_loss": 0.29346540570259094,
+      "eval_runtime": 1213.2584,
+      "eval_samples_per_second": 6.192,
+      "eval_steps_per_second": 0.775,
+      "step": 35000
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 4.043414154652687e-05,
+      "loss": 0.2257,
+      "step": 35050
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 4.042048929663609e-05,
+      "loss": 0.2041,
+      "step": 35100
+    },
+    {
+      "epoch": 3.83,
+      "eval_cer": 0.05821907261472552,
+      "eval_loss": 0.2906344532966614,
+      "eval_runtime": 1243.2388,
+      "eval_samples_per_second": 6.043,
+      "eval_steps_per_second": 0.756,
+      "step": 35100
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 4.0406837046745304e-05,
+      "loss": 0.2152,
+      "step": 35150
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 4.039318479685452e-05,
+      "loss": 0.2172,
+      "step": 35200
+    },
+    {
+      "epoch": 3.84,
+      "eval_cer": 0.05438661926286866,
+      "eval_loss": 0.29344433546066284,
+      "eval_runtime": 1198.4162,
+      "eval_samples_per_second": 6.269,
+      "eval_steps_per_second": 0.784,
+      "step": 35200
+    },
+    {
+      "epoch": 3.85,
+      "learning_rate": 4.037953254696374e-05,
+      "loss": 0.2172,
+      "step": 35250
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 4.036588029707296e-05,
+      "loss": 0.2321,
+      "step": 35300
+    },
+    {
+      "epoch": 3.86,
+      "eval_cer": 0.05137198408163126,
+      "eval_loss": 0.29439395666122437,
+      "eval_runtime": 1168.9807,
+      "eval_samples_per_second": 6.427,
+      "eval_steps_per_second": 0.804,
+      "step": 35300
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 4.035222804718218e-05,
+      "loss": 0.2222,
+      "step": 35350
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 4.03385757972914e-05,
+      "loss": 0.2207,
+      "step": 35400
+    },
+    {
+      "epoch": 3.87,
+      "eval_cer": 0.05017092057582612,
+      "eval_loss": 0.2857843339443207,
+      "eval_runtime": 1150.2318,
+      "eval_samples_per_second": 6.532,
+      "eval_steps_per_second": 0.817,
+      "step": 35400
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 4.032492354740061e-05,
+      "loss": 0.2155,
+      "step": 35450
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 4.0311271297509835e-05,
+      "loss": 0.2405,
+      "step": 35500
+    },
+    {
+      "epoch": 3.88,
+      "eval_cer": 0.051833931583864006,
+      "eval_loss": 0.28927794098854065,
+      "eval_runtime": 1156.4708,
+      "eval_samples_per_second": 6.496,
+      "eval_steps_per_second": 0.813,
+      "step": 35500
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 4.029761904761905e-05,
+      "loss": 0.2211,
+      "step": 35550
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 4.028396679772827e-05,
+      "loss": 0.2579,
+      "step": 35600
+    },
+    {
+      "epoch": 3.89,
+      "eval_cer": 0.05231641008619598,
+      "eval_loss": 0.28790414333343506,
+      "eval_runtime": 1154.5522,
+      "eval_samples_per_second": 6.507,
+      "eval_steps_per_second": 0.814,
+      "step": 35600
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 4.027031454783748e-05,
+      "loss": 0.2331,
+      "step": 35650
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 4.02566622979467e-05,
+      "loss": 0.2188,
+      "step": 35700
+    },
+    {
+      "epoch": 3.9,
+      "eval_cer": 0.05226850441929777,
+      "eval_loss": 0.2905963063240051,
+      "eval_runtime": 1151.8907,
+      "eval_samples_per_second": 6.522,
+      "eval_steps_per_second": 0.816,
+      "step": 35700
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 4.024301004805592e-05,
+      "loss": 0.2364,
+      "step": 35750
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 4.0229357798165144e-05,
+      "loss": 0.205,
+      "step": 35800
+    },
+    {
+      "epoch": 3.91,
+      "eval_cer": 0.053407974924805215,
+      "eval_loss": 0.2937009036540985,
+      "eval_runtime": 1188.6581,
+      "eval_samples_per_second": 6.321,
+      "eval_steps_per_second": 0.791,
+      "step": 35800
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 4.021570554827436e-05,
+      "loss": 0.2184,
+      "step": 35850
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 4.0202053298383575e-05,
+      "loss": 0.2432,
+      "step": 35900
+    },
+    {
+      "epoch": 3.92,
+      "eval_cer": 0.051584137749323335,
+      "eval_loss": 0.287141352891922,
+      "eval_runtime": 1147.8604,
+      "eval_samples_per_second": 6.545,
+      "eval_steps_per_second": 0.819,
+      "step": 35900
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 4.018840104849279e-05,
+      "loss": 0.2394,
+      "step": 35950
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 4.0174748798602014e-05,
+      "loss": 0.2324,
+      "step": 36000
+    },
+    {
+      "epoch": 3.93,
+      "eval_cer": 0.050954520412946847,
+      "eval_loss": 0.2883908152580261,
+      "eval_runtime": 1153.0185,
+      "eval_samples_per_second": 6.516,
+      "eval_steps_per_second": 0.815,
+      "step": 36000
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 4.016109654871123e-05,
+      "loss": 0.2182,
+      "step": 36050
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 4.0147444298820445e-05,
+      "loss": 0.22,
+      "step": 36100
+    },
+    {
+      "epoch": 3.94,
+      "eval_cer": 0.0522650825859479,
+      "eval_loss": 0.2854180932044983,
+      "eval_runtime": 1185.1981,
+      "eval_samples_per_second": 6.339,
+      "eval_steps_per_second": 0.793,
+      "step": 36100
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 4.013379204892966e-05,
+      "loss": 0.2301,
+      "step": 36150
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 4.0120139799038884e-05,
+      "loss": 0.2026,
+      "step": 36200
+    },
+    {
+      "epoch": 3.95,
+      "eval_cer": 0.051697058249869116,
+      "eval_loss": 0.2849681079387665,
+      "eval_runtime": 1167.8292,
+      "eval_samples_per_second": 6.433,
+      "eval_steps_per_second": 0.805,
+      "step": 36200
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 4.01064875491481e-05,
+      "loss": 0.2268,
+      "step": 36250
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 4.009283529925732e-05,
+      "loss": 0.2164,
+      "step": 36300
+    },
+    {
+      "epoch": 3.96,
+      "eval_cer": 0.050372808743468576,
+      "eval_loss": 0.28152966499328613,
+      "eval_runtime": 1156.0778,
+      "eval_samples_per_second": 6.499,
+      "eval_steps_per_second": 0.813,
+      "step": 36300
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 4.007918304936654e-05,
+      "loss": 0.2254,
+      "step": 36350
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 4.0065530799475754e-05,
+      "loss": 0.2153,
+      "step": 36400
+    },
+    {
+      "epoch": 3.98,
+      "eval_cer": 0.05203924158485634,
+      "eval_loss": 0.2812017798423767,
+      "eval_runtime": 1170.4653,
+      "eval_samples_per_second": 6.419,
+      "eval_steps_per_second": 0.803,
+      "step": 36400
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 4.0051878549584976e-05,
+      "loss": 0.2031,
+      "step": 36450
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 4.003822629969419e-05,
+      "loss": 0.2128,
+      "step": 36500
+    },
+    {
+      "epoch": 3.99,
+      "eval_cer": 0.05301788592291978,
+      "eval_loss": 0.2868877649307251,
+      "eval_runtime": 1161.9761,
+      "eval_samples_per_second": 6.466,
+      "eval_steps_per_second": 0.809,
+      "step": 36500
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 4.002457404980341e-05,
+      "loss": 0.2131,
+      "step": 36550
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 4.0010921799912624e-05,
+      "loss": 0.2209,
+      "step": 36600
+    },
+    {
+      "epoch": 4.0,
+      "eval_cer": 0.050047734575230714,
+      "eval_loss": 0.28464749455451965,
+      "eval_runtime": 1152.6846,
+      "eval_samples_per_second": 6.518,
+      "eval_steps_per_second": 0.815,
+      "step": 36600
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 3.9997269550021846e-05,
+      "loss": 0.1963,
+      "step": 36650
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 3.998361730013106e-05,
+      "loss": 0.1744,
+      "step": 36700
+    },
+    {
+      "epoch": 4.01,
+      "eval_cer": 0.055201015600138244,
+      "eval_loss": 0.29001498222351074,
+      "eval_runtime": 1191.6495,
+      "eval_samples_per_second": 6.305,
+      "eval_steps_per_second": 0.789,
+      "step": 36700
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 3.9969965050240285e-05,
+      "loss": 0.1677,
+      "step": 36750
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 3.99563128003495e-05,
+      "loss": 0.1774,
+      "step": 36800
+    },
+    {
+      "epoch": 4.02,
+      "eval_cer": 0.05026331007627267,
+      "eval_loss": 0.2934856712818146,
+      "eval_runtime": 1168.1862,
+      "eval_samples_per_second": 6.431,
+      "eval_steps_per_second": 0.805,
+      "step": 36800
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 3.9942660550458716e-05,
+      "loss": 0.1708,
+      "step": 36850
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 3.992900830056793e-05,
+      "loss": 0.1749,
+      "step": 36900
+    },
+    {
+      "epoch": 4.03,
+      "eval_cer": 0.05236431575309419,
+      "eval_loss": 0.2882940471172333,
+      "eval_runtime": 1173.1669,
+      "eval_samples_per_second": 6.404,
+      "eval_steps_per_second": 0.801,
+      "step": 36900
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 3.9915356050677155e-05,
+      "loss": 0.159,
+      "step": 36950
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 3.990170380078637e-05,
+      "loss": 0.166,
+      "step": 37000
+    },
+    {
+      "epoch": 4.04,
+      "eval_cer": 0.05076974141205375,
+      "eval_loss": 0.28926318883895874,
+      "eval_runtime": 1465.1032,
+      "eval_samples_per_second": 5.128,
+      "eval_steps_per_second": 0.642,
+      "step": 37000
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 3.9888051550895587e-05,
+      "loss": 0.1893,
+      "step": 37050
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 3.98743993010048e-05,
+      "loss": 0.184,
+      "step": 37100
+    },
+    {
+      "epoch": 4.05,
+      "eval_cer": 0.050547322244312055,
+      "eval_loss": 0.28999829292297363,
+      "eval_runtime": 1279.6549,
+      "eval_samples_per_second": 5.871,
+      "eval_steps_per_second": 0.735,
+      "step": 37100
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 3.9860747051114025e-05,
+      "loss": 0.1672,
+      "step": 37150
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 3.984709480122325e-05,
+      "loss": 0.1614,
+      "step": 37200
+    },
+    {
+      "epoch": 4.06,
+      "eval_cer": 0.05136856224828138,
+      "eval_loss": 0.28663578629493713,
+      "eval_runtime": 1454.1636,
+      "eval_samples_per_second": 5.167,
+      "eval_steps_per_second": 0.646,
+      "step": 37200
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 3.983344255133246e-05,
+      "loss": 0.1651,
+      "step": 37250
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 3.981979030144168e-05,
+      "loss": 0.1842,
+      "step": 37300
+    },
+    {
+      "epoch": 4.07,
+      "eval_cer": 0.05069788291170643,
+      "eval_loss": 0.29018309712409973,
+      "eval_runtime": 1179.9051,
+      "eval_samples_per_second": 6.367,
+      "eval_steps_per_second": 0.797,
+      "step": 37300
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 3.9806138051550895e-05,
+      "loss": 0.1817,
+      "step": 37350
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 3.979248580166012e-05,
+      "loss": 0.1681,
+      "step": 37400
+    },
+    {
+      "epoch": 4.08,
+      "eval_cer": 0.0527201864214809,
+      "eval_loss": 0.29232168197631836,
+      "eval_runtime": 1152.2145,
+      "eval_samples_per_second": 6.52,
+      "eval_steps_per_second": 0.816,
+      "step": 37400
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 3.9778833551769333e-05,
+      "loss": 0.1687,
+      "step": 37450
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 3.976518130187855e-05,
+      "loss": 0.1613,
+      "step": 37500
+    },
+    {
+      "epoch": 4.1,
+      "eval_cer": 0.05114272124718982,
+      "eval_loss": 0.28755250573158264,
+      "eval_runtime": 1157.0153,
+      "eval_samples_per_second": 6.493,
+      "eval_steps_per_second": 0.812,
+      "step": 37500
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 3.9751529051987765e-05,
+      "loss": 0.183,
+      "step": 37550
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 3.973787680209699e-05,
+      "loss": 0.1678,
+      "step": 37600
+    },
+    {
+      "epoch": 4.11,
+      "eval_cer": 0.051245376247685985,
+      "eval_loss": 0.2812196612358093,
+      "eval_runtime": 1157.3725,
+      "eval_samples_per_second": 6.491,
+      "eval_steps_per_second": 0.812,
+      "step": 37600
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 3.972422455220621e-05,
+      "loss": 0.171,
+      "step": 37650
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 3.9710572302315426e-05,
+      "loss": 0.1763,
+      "step": 37700
+    },
+    {
+      "epoch": 4.12,
+      "eval_cer": 0.050886083745949405,
+      "eval_loss": 0.2909301817417145,
+      "eval_runtime": 1146.6592,
+      "eval_samples_per_second": 6.552,
+      "eval_steps_per_second": 0.82,
+      "step": 37700
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 3.969692005242464e-05,
+      "loss": 0.1461,
+      "step": 37750
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 3.968326780253386e-05,
+      "loss": 0.1714,
+      "step": 37800
+    },
+    {
+      "epoch": 4.13,
+      "eval_cer": 0.05107770641354225,
+      "eval_loss": 0.28945988416671753,
+      "eval_runtime": 1206.247,
+      "eval_samples_per_second": 6.228,
+      "eval_steps_per_second": 0.779,
+      "step": 37800
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 3.966961555264308e-05,
+      "loss": 0.1786,
+      "step": 37850
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 3.9655963302752296e-05,
+      "loss": 0.1665,
+      "step": 37900
+    },
+    {
+      "epoch": 4.14,
+      "eval_cer": 0.04967817657344452,
+      "eval_loss": 0.2878923714160919,
+      "eval_runtime": 1367.4992,
+      "eval_samples_per_second": 5.494,
+      "eval_steps_per_second": 0.687,
+      "step": 37900
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 3.964231105286151e-05,
+      "loss": 0.1629,
+      "step": 37950
+    },
+    {
+      "epoch": 4.15,
+      "learning_rate": 3.962865880297073e-05,
+      "loss": 0.1744,
+      "step": 38000
+    },
+    {
+      "epoch": 4.15,
+      "eval_cer": 0.05413340359497812,
+      "eval_loss": 0.28816238045692444,
+      "eval_runtime": 1362.8508,
+      "eval_samples_per_second": 5.513,
+      "eval_steps_per_second": 0.69,
+      "step": 38000
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 3.961500655307995e-05,
+      "loss": 0.1571,
+      "step": 38050
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 3.960135430318917e-05,
+      "loss": 0.1769,
+      "step": 38100
+    },
+    {
+      "epoch": 4.16,
+      "eval_cer": 0.05594697527041038,
+      "eval_loss": 0.28923317790031433,
+      "eval_runtime": 1450.036,
+      "eval_samples_per_second": 5.181,
+      "eval_steps_per_second": 0.648,
+      "step": 38100
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 3.958770205329839e-05,
+      "loss": 0.1781,
+      "step": 38150
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 3.9574049803407605e-05,
+      "loss": 0.1769,
+      "step": 38200
+    },
+    {
+      "epoch": 4.17,
+      "eval_cer": 0.05179629141701541,
+      "eval_loss": 0.28494036197662354,
+      "eval_runtime": 1181.7013,
+      "eval_samples_per_second": 6.358,
+      "eval_steps_per_second": 0.795,
+      "step": 38200
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 3.956039755351682e-05,
+      "loss": 0.18,
+      "step": 38250
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 3.9546745303626036e-05,
+      "loss": 0.1846,
+      "step": 38300
+    },
+    {
+      "epoch": 4.18,
+      "eval_cer": 0.0502462009095233,
+      "eval_loss": 0.2893485128879547,
+      "eval_runtime": 1168.764,
+      "eval_samples_per_second": 6.428,
+      "eval_steps_per_second": 0.804,
+      "step": 38300
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 3.953309305373526e-05,
+      "loss": 0.1811,
+      "step": 38350
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 3.9519440803844475e-05,
+      "loss": 0.1944,
+      "step": 38400
+    },
+    {
+      "epoch": 4.19,
+      "eval_cer": 0.05107428458019238,
+      "eval_loss": 0.28696149587631226,
+      "eval_runtime": 1172.9327,
+      "eval_samples_per_second": 6.405,
+      "eval_steps_per_second": 0.801,
+      "step": 38400
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 3.950578855395369e-05,
+      "loss": 0.1608,
+      "step": 38450
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 3.949213630406291e-05,
+      "loss": 0.1626,
+      "step": 38500
+    },
+    {
+      "epoch": 4.2,
+      "eval_cer": 0.049335993238457304,
+      "eval_loss": 0.2923007607460022,
+      "eval_runtime": 1166.3295,
+      "eval_samples_per_second": 6.442,
+      "eval_steps_per_second": 0.806,
+      "step": 38500
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 3.947848405417213e-05,
+      "loss": 0.1567,
+      "step": 38550
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 3.946483180428135e-05,
+      "loss": 0.1635,
+      "step": 38600
+    },
+    {
+      "epoch": 4.22,
+      "eval_cer": 0.056754527940980216,
+      "eval_loss": 0.29253220558166504,
+      "eval_runtime": 1320.7091,
+      "eval_samples_per_second": 5.689,
+      "eval_steps_per_second": 0.712,
+      "step": 38600
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 3.945117955439057e-05,
+      "loss": 0.1848,
+      "step": 38650
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 3.943752730449978e-05,
+      "loss": 0.1633,
+      "step": 38700
+    },
+    {
+      "epoch": 4.23,
+      "eval_cer": 0.04997245424153353,
+      "eval_loss": 0.28573524951934814,
+      "eval_runtime": 1459.4135,
+      "eval_samples_per_second": 5.148,
+      "eval_steps_per_second": 0.644,
+      "step": 38700
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 3.9423875054609e-05,
+      "loss": 0.175,
+      "step": 38750
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 3.941022280471822e-05,
+      "loss": 0.1895,
+      "step": 38800
+    },
+    {
+      "epoch": 4.24,
+      "eval_cer": 0.0490040754035197,
+      "eval_loss": 0.2857818901538849,
+      "eval_runtime": 1374.6623,
+      "eval_samples_per_second": 5.465,
+      "eval_steps_per_second": 0.684,
+      "step": 38800
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 3.939657055482744e-05,
+      "loss": 0.156,
+      "step": 38850
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 3.938291830493665e-05,
+      "loss": 0.1762,
+      "step": 38900
+    },
+    {
+      "epoch": 4.25,
+      "eval_cer": 0.04956525607289874,
+      "eval_loss": 0.28907591104507446,
+      "eval_runtime": 1452.6364,
+      "eval_samples_per_second": 5.172,
+      "eval_steps_per_second": 0.647,
+      "step": 38900
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 3.9369266055045876e-05,
+      "loss": 0.1567,
+      "step": 38950
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 3.935561380515509e-05,
+      "loss": 0.1567,
+      "step": 39000
+    },
+    {
+      "epoch": 4.26,
+      "eval_cer": 0.049517350406000527,
+      "eval_loss": 0.28758588433265686,
+      "eval_runtime": 1139.6582,
+      "eval_samples_per_second": 6.592,
+      "eval_steps_per_second": 0.825,
+      "step": 39000
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 3.9341961555264314e-05,
+      "loss": 0.1706,
+      "step": 39050
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 3.932830930537353e-05,
+      "loss": 0.1542,
+      "step": 39100
+    },
+    {
+      "epoch": 4.27,
+      "eval_cer": 0.05102980074664404,
+      "eval_loss": 0.2869040369987488,
+      "eval_runtime": 1158.6876,
+      "eval_samples_per_second": 6.484,
+      "eval_steps_per_second": 0.811,
+      "step": 39100
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 3.9314657055482746e-05,
+      "loss": 0.1741,
+      "step": 39150
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 3.930100480559196e-05,
+      "loss": 0.178,
+      "step": 39200
+    },
+    {
+      "epoch": 4.28,
+      "eval_cer": 0.049780831573940684,
+      "eval_loss": 0.28901004791259766,
+      "eval_runtime": 1161.4194,
+      "eval_samples_per_second": 6.469,
+      "eval_steps_per_second": 0.809,
+      "step": 39200
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 3.928735255570118e-05,
+      "loss": 0.1832,
+      "step": 39250
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 3.92737003058104e-05,
+      "loss": 0.1613,
+      "step": 39300
+    },
+    {
+      "epoch": 4.29,
+      "eval_cer": 0.05036254324341896,
+      "eval_loss": 0.2918330132961273,
+      "eval_runtime": 1156.9299,
+      "eval_samples_per_second": 6.494,
+      "eval_steps_per_second": 0.812,
+      "step": 39300
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 3.9260048055919616e-05,
+      "loss": 0.1793,
+      "step": 39350
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 3.924639580602884e-05,
+      "loss": 0.1801,
+      "step": 39400
+    },
+    {
+      "epoch": 4.3,
+      "eval_cer": 0.050280419243022026,
+      "eval_loss": 0.28874126076698303,
+      "eval_runtime": 1174.6787,
+      "eval_samples_per_second": 6.396,
+      "eval_steps_per_second": 0.8,
+      "step": 39400
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 3.9232743556138054e-05,
+      "loss": 0.1789,
+      "step": 39450
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 3.921909130624727e-05,
+      "loss": 0.1885,
+      "step": 39500
+    },
+    {
+      "epoch": 4.31,
+      "eval_cer": 0.05452691443021342,
+      "eval_loss": 0.282078355550766,
+      "eval_runtime": 1194.6585,
+      "eval_samples_per_second": 6.289,
+      "eval_steps_per_second": 0.787,
+      "step": 39500
+    },
+    {
+      "epoch": 4.32,
+      "learning_rate": 3.920543905635649e-05,
+      "loss": 0.1593,
+      "step": 39550
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 3.919178680646571e-05,
+      "loss": 0.1773,
+      "step": 39600
+    },
+    {
+      "epoch": 4.33,
+      "eval_cer": 0.05213163108530288,
+      "eval_loss": 0.2819925844669342,
+      "eval_runtime": 1165.2004,
+      "eval_samples_per_second": 6.448,
+      "eval_steps_per_second": 0.807,
+      "step": 39600
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 3.9178134556574924e-05,
+      "loss": 0.1718,
+      "step": 39650
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 3.916448230668414e-05,
+      "loss": 0.1718,
+      "step": 39700
+    },
+    {
+      "epoch": 4.34,
+      "eval_cer": 0.049561834239548866,
+      "eval_loss": 0.2793809175491333,
+      "eval_runtime": 1157.9372,
+      "eval_samples_per_second": 6.488,
+      "eval_steps_per_second": 0.812,
+      "step": 39700
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 3.915083005679336e-05,
+      "loss": 0.1726,
+      "step": 39750
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 3.913717780690258e-05,
+      "loss": 0.1756,
+      "step": 39800
+    },
+    {
+      "epoch": 4.35,
+      "eval_cer": 0.04887746756957442,
+      "eval_loss": 0.2796994745731354,
+      "eval_runtime": 1151.5087,
+      "eval_samples_per_second": 6.524,
+      "eval_steps_per_second": 0.816,
+      "step": 39800
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 3.9123525557011794e-05,
+      "loss": 0.1565,
+      "step": 39850
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 3.910987330712102e-05,
+      "loss": 0.1785,
+      "step": 39900
+    },
+    {
+      "epoch": 4.36,
+      "eval_cer": 0.052774935755078856,
+      "eval_loss": 0.28161171078681946,
+      "eval_runtime": 1287.3124,
+      "eval_samples_per_second": 5.836,
+      "eval_steps_per_second": 0.73,
+      "step": 39900
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 3.909622105723023e-05,
+      "loss": 0.1682,
+      "step": 39950
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 3.9082568807339455e-05,
+      "loss": 0.1748,
+      "step": 40000
+    },
+    {
+      "epoch": 4.37,
+      "eval_cer": 0.05262437508768448,
+      "eval_loss": 0.2795943319797516,
+      "eval_runtime": 1472.9079,
+      "eval_samples_per_second": 5.101,
+      "eval_steps_per_second": 0.638,
+      "step": 40000
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 3.906891655744867e-05,
+      "loss": 0.1823,
+      "step": 40050
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 3.905526430755789e-05,
+      "loss": 0.1526,
+      "step": 40100
+    },
+    {
+      "epoch": 4.38,
+      "eval_cer": 0.048610564568284394,
+      "eval_loss": 0.2830022871494293,
+      "eval_runtime": 1231.8361,
+      "eval_samples_per_second": 6.099,
+      "eval_steps_per_second": 0.763,
+      "step": 40100
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 3.90416120576671e-05,
+      "loss": 0.1638,
+      "step": 40150
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 3.902795980777632e-05,
+      "loss": 0.1873,
+      "step": 40200
+    },
+    {
+      "epoch": 4.39,
+      "eval_cer": 0.053401131258105465,
+      "eval_loss": 0.2826146185398102,
+      "eval_runtime": 1183.8819,
+      "eval_samples_per_second": 6.346,
+      "eval_steps_per_second": 0.794,
+      "step": 40200
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 3.901430755788554e-05,
+      "loss": 0.1605,
+      "step": 40250
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 3.900065530799476e-05,
+      "loss": 0.1576,
+      "step": 40300
+    },
+    {
+      "epoch": 4.4,
+      "eval_cer": 0.04860714273493452,
+      "eval_loss": 0.28039631247520447,
+      "eval_runtime": 1139.9713,
+      "eval_samples_per_second": 6.591,
+      "eval_steps_per_second": 0.825,
+      "step": 40300
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 3.898700305810398e-05,
+      "loss": 0.1778,
+      "step": 40350
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 3.8973350808213195e-05,
+      "loss": 0.1715,
+      "step": 40400
+    },
+    {
+      "epoch": 4.41,
+      "eval_cer": 0.049062246570467524,
+      "eval_loss": 0.27709659934043884,
+      "eval_runtime": 1152.7021,
+      "eval_samples_per_second": 6.518,
+      "eval_steps_per_second": 0.815,
+      "step": 40400
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 3.895969855832242e-05,
+      "loss": 0.1774,
+      "step": 40450
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 3.8946046308431634e-05,
+      "loss": 0.1675,
+      "step": 40500
+    },
+    {
+      "epoch": 4.42,
+      "eval_cer": 0.047919354231610214,
+      "eval_loss": 0.280836284160614,
+      "eval_runtime": 1141.4777,
+      "eval_samples_per_second": 6.582,
+      "eval_steps_per_second": 0.823,
+      "step": 40500
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 3.893239405854085e-05,
+      "loss": 0.1776,
+      "step": 40550
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 3.8918741808650065e-05,
+      "loss": 0.1735,
+      "step": 40600
+    },
+    {
+      "epoch": 4.43,
+      "eval_cer": 0.050119593075578034,
+      "eval_loss": 0.28058889508247375,
+      "eval_runtime": 1152.5493,
+      "eval_samples_per_second": 6.519,
+      "eval_steps_per_second": 0.816,
+      "step": 40600
+    },
+    {
+      "epoch": 4.44,
+      "learning_rate": 3.890508955875928e-05,
+      "loss": 0.1555,
+      "step": 40650
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 3.8891437308868504e-05,
+      "loss": 0.1462,
+      "step": 40700
+    },
+    {
+      "epoch": 4.45,
+      "eval_cer": 0.050150389575726886,
+      "eval_loss": 0.2807774245738983,
+      "eval_runtime": 1157.0666,
+      "eval_samples_per_second": 6.493,
+      "eval_steps_per_second": 0.812,
+      "step": 40700
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 3.887778505897772e-05,
+      "loss": 0.1823,
+      "step": 40750
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 3.886413280908694e-05,
+      "loss": 0.1817,
+      "step": 40800
+    },
+    {
+      "epoch": 4.46,
+      "eval_cer": 0.04801516556540664,
+      "eval_loss": 0.27960580587387085,
+      "eval_runtime": 1144.0954,
+      "eval_samples_per_second": 6.567,
+      "eval_steps_per_second": 0.822,
+      "step": 40800
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 3.885048055919616e-05,
+      "loss": 0.1831,
+      "step": 40850
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 3.8836828309305374e-05,
+      "loss": 0.1561,
+      "step": 40900
+    },
+    {
+      "epoch": 4.47,
+      "eval_cer": 0.04895959156997136,
+      "eval_loss": 0.2814878821372986,
+      "eval_runtime": 1156.0109,
+      "eval_samples_per_second": 6.499,
+      "eval_steps_per_second": 0.813,
+      "step": 40900
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 3.882344910441241e-05,
+      "loss": 0.1673,
+      "step": 40950
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 3.880979685452163e-05,
+      "loss": 0.2072,
+      "step": 41000
+    },
+    {
+      "epoch": 4.48,
+      "eval_cer": 0.05174496391676733,
+      "eval_loss": 0.28528130054473877,
+      "eval_runtime": 1279.9842,
+      "eval_samples_per_second": 5.87,
+      "eval_steps_per_second": 0.734,
+      "step": 41000
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 3.879614460463085e-05,
+      "loss": 0.1719,
+      "step": 41050
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 3.878249235474006e-05,
+      "loss": 0.1675,
+      "step": 41100
+    },
+    {
+      "epoch": 4.49,
+      "eval_cer": 0.04981162807408954,
+      "eval_loss": 0.2832522988319397,
+      "eval_runtime": 1157.4874,
+      "eval_samples_per_second": 6.491,
+      "eval_steps_per_second": 0.812,
+      "step": 41100
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 3.876884010484928e-05,
+      "loss": 0.1789,
+      "step": 41150
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 3.8755187854958495e-05,
+      "loss": 0.2017,
+      "step": 41200
+    },
+    {
+      "epoch": 4.5,
+      "eval_cer": 0.04828549040004654,
+      "eval_loss": 0.28346341848373413,
+      "eval_runtime": 1137.5974,
+      "eval_samples_per_second": 6.604,
+      "eval_steps_per_second": 0.826,
+      "step": 41200
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 3.874153560506772e-05,
+      "loss": 0.1719,
+      "step": 41250
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 3.872788335517693e-05,
+      "loss": 0.2023,
+      "step": 41300
+    },
+    {
+      "epoch": 4.51,
+      "eval_cer": 0.05051310391081334,
+      "eval_loss": 0.27435365319252014,
+      "eval_runtime": 1287.7747,
+      "eval_samples_per_second": 5.834,
+      "eval_steps_per_second": 0.73,
+      "step": 41300
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 3.8714231105286156e-05,
+      "loss": 0.1624,
+      "step": 41350
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 3.870057885539537e-05,
+      "loss": 0.1696,
+      "step": 41400
+    },
+    {
+      "epoch": 4.52,
+      "eval_cer": 0.047361595395581046,
+      "eval_loss": 0.2742946147918701,
+      "eval_runtime": 1143.0851,
+      "eval_samples_per_second": 6.573,
+      "eval_steps_per_second": 0.822,
+      "step": 41400
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 3.868692660550459e-05,
+      "loss": 0.1732,
+      "step": 41450
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 3.867327435561381e-05,
+      "loss": 0.1709,
+      "step": 41500
+    },
+    {
+      "epoch": 4.53,
+      "eval_cer": 0.04716655089463833,
+      "eval_loss": 0.27990689873695374,
+      "eval_runtime": 1145.7372,
+      "eval_samples_per_second": 6.557,
+      "eval_steps_per_second": 0.82,
+      "step": 41500
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 3.8659622105723026e-05,
+      "loss": 0.1567,
+      "step": 41550
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 3.864596985583224e-05,
+      "loss": 0.1785,
+      "step": 41600
+    },
+    {
+      "epoch": 4.54,
+      "eval_cer": 0.04891168590307315,
+      "eval_loss": 0.2770211398601532,
+      "eval_runtime": 1150.6726,
+      "eval_samples_per_second": 6.529,
+      "eval_steps_per_second": 0.817,
+      "step": 41600
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 3.863231760594146e-05,
+      "loss": 0.1767,
+      "step": 41650
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 3.861866535605068e-05,
+      "loss": 0.1923,
+      "step": 41700
+    },
+    {
+      "epoch": 4.55,
+      "eval_cer": 0.04813492973265216,
+      "eval_loss": 0.27829426527023315,
+      "eval_runtime": 1159.8988,
+      "eval_samples_per_second": 6.477,
+      "eval_steps_per_second": 0.81,
+      "step": 41700
+    },
+    {
+      "epoch": 4.56,
+      "learning_rate": 3.8605013106159896e-05,
+      "loss": 0.1827,
+      "step": 41750
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 3.859136085626912e-05,
+      "loss": 0.2008,
+      "step": 41800
+    },
+    {
+      "epoch": 4.57,
+      "eval_cer": 0.04740950106247926,
+      "eval_loss": 0.27811819314956665,
+      "eval_runtime": 1139.7714,
+      "eval_samples_per_second": 6.592,
+      "eval_steps_per_second": 0.825,
+      "step": 41800
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 3.8577708606378334e-05,
+      "loss": 0.1752,
+      "step": 41850
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 3.856405635648755e-05,
+      "loss": 0.1746,
+      "step": 41900
+    },
+    {
+      "epoch": 4.58,
+      "eval_cer": 0.047378704562330405,
+      "eval_loss": 0.2777942717075348,
+      "eval_runtime": 1136.9039,
+      "eval_samples_per_second": 6.608,
+      "eval_steps_per_second": 0.827,
+      "step": 41900
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 3.855040410659677e-05,
+      "loss": 0.1668,
+      "step": 41950
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 3.853675185670599e-05,
+      "loss": 0.1732,
+      "step": 42000
+    },
+    {
+      "epoch": 4.59,
+      "eval_cer": 0.05069446107835656,
+      "eval_loss": 0.2736971378326416,
+      "eval_runtime": 1150.4296,
+      "eval_samples_per_second": 6.531,
+      "eval_steps_per_second": 0.817,
+      "step": 42000
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 3.8523099606815204e-05,
+      "loss": 0.1847,
+      "step": 42050
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 3.850944735692442e-05,
+      "loss": 0.1883,
+      "step": 42100
+    },
+    {
+      "epoch": 4.6,
+      "eval_cer": 0.05216242758545173,
+      "eval_loss": 0.2768714427947998,
+      "eval_runtime": 1164.9599,
+      "eval_samples_per_second": 6.449,
+      "eval_steps_per_second": 0.807,
+      "step": 42100
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 3.8495795107033636e-05,
+      "loss": 0.1605,
+      "step": 42150
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 3.848214285714286e-05,
+      "loss": 0.1462,
+      "step": 42200
+    },
+    {
+      "epoch": 4.61,
+      "eval_cer": 0.05187841541741234,
+      "eval_loss": 0.27714163064956665,
+      "eval_runtime": 1164.401,
+      "eval_samples_per_second": 6.452,
+      "eval_steps_per_second": 0.807,
+      "step": 42200
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 3.846849060725208e-05,
+      "loss": 0.1724,
+      "step": 42250
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 3.84548383573613e-05,
+      "loss": 0.1813,
+      "step": 42300
+    },
+    {
+      "epoch": 4.62,
+      "eval_cer": 0.048521596901187715,
+      "eval_loss": 0.27737104892730713,
+      "eval_runtime": 1153.5255,
+      "eval_samples_per_second": 6.513,
+      "eval_steps_per_second": 0.815,
+      "step": 42300
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 3.844118610747051e-05,
+      "loss": 0.1676,
+      "step": 42350
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 3.842753385757973e-05,
+      "loss": 0.2156,
+      "step": 42400
+    },
+    {
+      "epoch": 4.63,
+      "eval_cer": 0.04726236222843475,
+      "eval_loss": 0.2811882495880127,
+      "eval_runtime": 1139.084,
+      "eval_samples_per_second": 6.596,
+      "eval_steps_per_second": 0.825,
+      "step": 42400
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 3.841388160768895e-05,
+      "loss": 0.175,
+      "step": 42450
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 3.840022935779817e-05,
+      "loss": 0.1854,
+      "step": 42500
+    },
+    {
+      "epoch": 4.64,
+      "eval_cer": 0.04832997423359488,
+      "eval_loss": 0.27360132336616516,
+      "eval_runtime": 1176.5116,
+      "eval_samples_per_second": 6.386,
+      "eval_steps_per_second": 0.799,
+      "step": 42500
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 3.838657710790738e-05,
+      "loss": 0.1697,
+      "step": 42550
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 3.83729248580166e-05,
+      "loss": 0.1728,
+      "step": 42600
+    },
+    {
+      "epoch": 4.65,
+      "eval_cer": 0.04801174373205676,
+      "eval_loss": 0.2755264639854431,
+      "eval_runtime": 1183.4193,
+      "eval_samples_per_second": 6.349,
+      "eval_steps_per_second": 0.794,
+      "step": 42600
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 3.835927260812582e-05,
+      "loss": 0.1757,
+      "step": 42650
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 3.8345620358235044e-05,
+      "loss": 0.1784,
+      "step": 42700
+    },
+    {
+      "epoch": 4.66,
+      "eval_cer": 0.04826838123329718,
+      "eval_loss": 0.27424073219299316,
+      "eval_runtime": 1313.6948,
+      "eval_samples_per_second": 5.719,
+      "eval_steps_per_second": 0.716,
+      "step": 42700
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 3.833196810834426e-05,
+      "loss": 0.1756,
+      "step": 42750
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 3.8318315858453475e-05,
+      "loss": 0.1549,
+      "step": 42800
+    },
+    {
+      "epoch": 4.67,
+      "eval_cer": 0.05006826557532995,
+      "eval_loss": 0.2791966199874878,
+      "eval_runtime": 1518.3443,
+      "eval_samples_per_second": 4.948,
+      "eval_steps_per_second": 0.619,
+      "step": 42800
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 3.830466360856269e-05,
+      "loss": 0.1886,
+      "step": 42850
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 3.8291284403669726e-05,
+      "loss": 0.178,
+      "step": 42900
+    },
+    {
+      "epoch": 4.69,
+      "eval_cer": 0.048815874569276725,
+      "eval_loss": 0.2769632041454315,
+      "eval_runtime": 1583.845,
+      "eval_samples_per_second": 4.744,
+      "eval_steps_per_second": 0.593,
+      "step": 42900
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 3.827763215377895e-05,
+      "loss": 0.1778,
+      "step": 42950
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 3.8263979903888164e-05,
+      "loss": 0.1683,
+      "step": 43000
+    },
+    {
+      "epoch": 4.7,
+      "eval_cer": 0.048702954068730944,
+      "eval_loss": 0.2739656865596771,
+      "eval_runtime": 1397.9506,
+      "eval_samples_per_second": 5.374,
+      "eval_steps_per_second": 0.672,
+      "step": 43000
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 3.825032765399738e-05,
+      "loss": 0.1738,
+      "step": 43050
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 3.8236675404106596e-05,
+      "loss": 0.183,
+      "step": 43100
+    },
+    {
+      "epoch": 4.71,
+      "eval_cer": 0.04755663989652376,
+      "eval_loss": 0.2752831280231476,
+      "eval_runtime": 1171.1241,
+      "eval_samples_per_second": 6.415,
+      "eval_steps_per_second": 0.803,
+      "step": 43100
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 3.822302315421581e-05,
+      "loss": 0.1802,
+      "step": 43150
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 3.8209370904325034e-05,
+      "loss": 0.1727,
+      "step": 43200
+    },
+    {
+      "epoch": 4.72,
+      "eval_cer": 0.05161151241612231,
+      "eval_loss": 0.2738773226737976,
+      "eval_runtime": 1208.5144,
+      "eval_samples_per_second": 6.217,
+      "eval_steps_per_second": 0.778,
+      "step": 43200
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 3.819571865443426e-05,
+      "loss": 0.1818,
+      "step": 43250
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 3.818206640454347e-05,
+      "loss": 0.1866,
+      "step": 43300
+    },
+    {
+      "epoch": 4.73,
+      "eval_cer": 0.049089621237266505,
+      "eval_loss": 0.2753521502017975,
+      "eval_runtime": 1160.1441,
+      "eval_samples_per_second": 6.476,
+      "eval_steps_per_second": 0.81,
+      "step": 43300
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 3.816841415465269e-05,
+      "loss": 0.1413,
+      "step": 43350
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 3.8154761904761904e-05,
+      "loss": 0.1639,
+      "step": 43400
+    },
+    {
+      "epoch": 4.74,
+      "eval_cer": 0.04883298373602609,
+      "eval_loss": 0.27695029973983765,
+      "eval_runtime": 1168.439,
+      "eval_samples_per_second": 6.43,
+      "eval_steps_per_second": 0.804,
+      "step": 43400
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 3.814110965487113e-05,
+      "loss": 0.1736,
+      "step": 43450
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 3.812745740498034e-05,
+      "loss": 0.1608,
+      "step": 43500
+    },
+    {
+      "epoch": 4.75,
+      "eval_cer": 0.047128910727789734,
+      "eval_loss": 0.271191269159317,
+      "eval_runtime": 1202.4574,
+      "eval_samples_per_second": 6.248,
+      "eval_steps_per_second": 0.782,
+      "step": 43500
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 3.811380515508956e-05,
+      "loss": 0.1753,
+      "step": 43550
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 3.8100152905198775e-05,
+      "loss": 0.1793,
+      "step": 43600
+    },
+    {
+      "epoch": 4.76,
+      "eval_cer": 0.04675593089265367,
+      "eval_loss": 0.27248117327690125,
+      "eval_runtime": 1138.4082,
+      "eval_samples_per_second": 6.6,
+      "eval_steps_per_second": 0.826,
+      "step": 43600
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 3.8086500655308e-05,
+      "loss": 0.1798,
+      "step": 43650
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 3.807284840541721e-05,
+      "loss": 0.1765,
+      "step": 43700
+    },
+    {
+      "epoch": 4.77,
+      "eval_cer": 0.04771062239726801,
+      "eval_loss": 0.2712666094303131,
+      "eval_runtime": 1149.9923,
+      "eval_samples_per_second": 6.533,
+      "eval_steps_per_second": 0.817,
+      "step": 43700
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 3.8059196155526436e-05,
+      "loss": 0.1543,
+      "step": 43750
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 3.804554390563565e-05,
+      "loss": 0.197,
+      "step": 43800
+    },
+    {
+      "epoch": 4.78,
+      "eval_cer": 0.047895401398161105,
+      "eval_loss": 0.27246350049972534,
+      "eval_runtime": 1136.3564,
+      "eval_samples_per_second": 6.611,
+      "eval_steps_per_second": 0.827,
+      "step": 43800
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 3.803189165574487e-05,
+      "loss": 0.1743,
+      "step": 43850
+    },
+    {
+      "epoch": 4.79,
+      "learning_rate": 3.801823940585409e-05,
+      "loss": 0.1795,
+      "step": 43900
+    },
+    {
+      "epoch": 4.79,
+      "eval_cer": 0.04925729107141024,
+      "eval_loss": 0.26492729783058167,
+      "eval_runtime": 1162.7543,
+      "eval_samples_per_second": 6.461,
+      "eval_steps_per_second": 0.808,
+      "step": 43900
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 3.8004587155963306e-05,
+      "loss": 0.2045,
+      "step": 43950
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 3.799093490607252e-05,
+      "loss": 0.1855,
+      "step": 44000
+    },
+    {
+      "epoch": 4.81,
+      "eval_cer": 0.048767968902378514,
+      "eval_loss": 0.2690161168575287,
+      "eval_runtime": 1151.9666,
+      "eval_samples_per_second": 6.522,
+      "eval_steps_per_second": 0.816,
+      "step": 44000
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 3.797728265618174e-05,
+      "loss": 0.1865,
+      "step": 44050
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 3.796363040629096e-05,
+      "loss": 0.1731,
+      "step": 44100
+    },
+    {
+      "epoch": 4.82,
+      "eval_cer": 0.04666354139220712,
+      "eval_loss": 0.26857924461364746,
+      "eval_runtime": 1138.8024,
+      "eval_samples_per_second": 6.597,
+      "eval_steps_per_second": 0.825,
+      "step": 44100
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 3.7949978156400176e-05,
+      "loss": 0.1811,
+      "step": 44150
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 3.79363259065094e-05,
+      "loss": 0.1909,
+      "step": 44200
+    },
+    {
+      "epoch": 4.83,
+      "eval_cer": 0.048347083400344236,
+      "eval_loss": 0.26807311177253723,
+      "eval_runtime": 1136.7106,
+      "eval_samples_per_second": 6.609,
+      "eval_steps_per_second": 0.827,
+      "step": 44200
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 3.7922673656618614e-05,
+      "loss": 0.1673,
+      "step": 44250
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 3.790902140672783e-05,
+      "loss": 0.1722,
+      "step": 44300
+    },
+    {
+      "epoch": 4.84,
+      "eval_cer": 0.04721787839488641,
+      "eval_loss": 0.2638927400112152,
+      "eval_runtime": 1151.5301,
+      "eval_samples_per_second": 6.524,
+      "eval_steps_per_second": 0.816,
+      "step": 44300
+    },
+    {
+      "epoch": 4.84,
+      "learning_rate": 3.7895369156837046e-05,
+      "loss": 0.1785,
+      "step": 44350
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 3.788171690694627e-05,
+      "loss": 0.1691,
+      "step": 44400
+    },
+    {
+      "epoch": 4.85,
+      "eval_cer": 0.04767640406376929,
+      "eval_loss": 0.2721834182739258,
+      "eval_runtime": 1200.0834,
+      "eval_samples_per_second": 6.26,
+      "eval_steps_per_second": 0.783,
+      "step": 44400
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 3.7868064657055484e-05,
+      "loss": 0.1949,
+      "step": 44450
+    },
+    {
+      "epoch": 4.86,
+      "learning_rate": 3.78544124071647e-05,
+      "loss": 0.1563,
+      "step": 44500
+    },
+    {
+      "epoch": 4.86,
+      "eval_cer": 0.04779274639766494,
+      "eval_loss": 0.26372846961021423,
+      "eval_runtime": 1394.4107,
+      "eval_samples_per_second": 5.388,
+      "eval_steps_per_second": 0.674,
+      "step": 44500
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 3.7840760157273916e-05,
+      "loss": 0.1728,
+      "step": 44550
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 3.782710790738314e-05,
+      "loss": 0.159,
+      "step": 44600
+    },
+    {
+      "epoch": 4.87,
+      "eval_cer": 0.04847369123428951,
+      "eval_loss": 0.26685479283332825,
+      "eval_runtime": 1409.3734,
+      "eval_samples_per_second": 5.331,
+      "eval_steps_per_second": 0.667,
+      "step": 44600
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 3.781345565749236e-05,
+      "loss": 0.1585,
+      "step": 44650
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 3.779980340760158e-05,
+      "loss": 0.1902,
+      "step": 44700
+    },
+    {
+      "epoch": 4.88,
+      "eval_cer": 0.04875428156897903,
+      "eval_loss": 0.26479294896125793,
+      "eval_runtime": 1474.9654,
+      "eval_samples_per_second": 5.094,
+      "eval_steps_per_second": 0.637,
+      "step": 44700
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 3.778615115771079e-05,
+      "loss": 0.1986,
+      "step": 44750
+    },
+    {
+      "epoch": 4.89,
+      "learning_rate": 3.777249890782001e-05,
+      "loss": 0.1975,
+      "step": 44800
+    },
+    {
+      "epoch": 4.89,
+      "eval_cer": 0.04979451890734018,
+      "eval_loss": 0.2669293284416199,
+      "eval_runtime": 1193.7755,
+      "eval_samples_per_second": 6.293,
+      "eval_steps_per_second": 0.787,
+      "step": 44800
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 3.775884665792923e-05,
+      "loss": 0.1921,
+      "step": 44850
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 3.774519440803845e-05,
+      "loss": 0.1757,
+      "step": 44900
+    },
+    {
+      "epoch": 4.9,
+      "eval_cer": 0.046643010392107886,
+      "eval_loss": 0.26375335454940796,
+      "eval_runtime": 1407.5693,
+      "eval_samples_per_second": 5.338,
+      "eval_steps_per_second": 0.668,
+      "step": 44900
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 3.773154215814766e-05,
+      "loss": 0.1806,
+      "step": 44950
+    },
+    {
+      "epoch": 4.91,
+      "learning_rate": 3.771788990825688e-05,
+      "loss": 0.1499,
+      "step": 45000
+    },
+    {
+      "epoch": 4.91,
+      "eval_cer": 0.04707758322754165,
+      "eval_loss": 0.26409247517585754,
+      "eval_runtime": 1611.3579,
+      "eval_samples_per_second": 4.663,
+      "eval_steps_per_second": 0.583,
+      "step": 45000
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 3.77042376583661e-05,
+      "loss": 0.1803,
+      "step": 45050
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 3.7690585408475324e-05,
+      "loss": 0.1808,
+      "step": 45100
+    },
+    {
+      "epoch": 4.93,
+      "eval_cer": 0.04849764406773861,
+      "eval_loss": 0.267391562461853,
+      "eval_runtime": 1404.0846,
+      "eval_samples_per_second": 5.351,
+      "eval_steps_per_second": 0.669,
+      "step": 45100
+    },
+    {
+      "epoch": 4.93,
+      "learning_rate": 3.767693315858454e-05,
+      "loss": 0.1713,
+      "step": 45150
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 3.7663280908693755e-05,
+      "loss": 0.1783,
+      "step": 45200
+    },
+    {
+      "epoch": 4.94,
+      "eval_cer": 0.047553218063173884,
+      "eval_loss": 0.2665572762489319,
+      "eval_runtime": 1430.7712,
+      "eval_samples_per_second": 5.251,
+      "eval_steps_per_second": 0.657,
+      "step": 45200
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 3.764962865880297e-05,
+      "loss": 0.1873,
+      "step": 45250
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 3.763597640891219e-05,
+      "loss": 0.1676,
+      "step": 45300
+    },
+    {
+      "epoch": 4.95,
+      "eval_cer": 0.04647191872461427,
+      "eval_loss": 0.26687777042388916,
+      "eval_runtime": 1165.6382,
+      "eval_samples_per_second": 6.445,
+      "eval_steps_per_second": 0.806,
+      "step": 45300
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 3.762232415902141e-05,
+      "loss": 0.1789,
+      "step": 45350
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 3.7608671909130625e-05,
+      "loss": 0.1869,
+      "step": 45400
+    },
+    {
+      "epoch": 4.96,
+      "eval_cer": 0.04635899822406849,
+      "eval_loss": 0.2622717320919037,
+      "eval_runtime": 1313.8343,
+      "eval_samples_per_second": 5.718,
+      "eval_steps_per_second": 0.715,
+      "step": 45400
+    },
+    {
+      "epoch": 4.96,
+      "learning_rate": 3.759501965923984e-05,
+      "loss": 0.1605,
+      "step": 45450
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 3.7581367409349064e-05,
+      "loss": 0.1633,
+      "step": 45500
+    },
+    {
+      "epoch": 4.97,
+      "eval_cer": 0.047361595395581046,
+      "eval_loss": 0.26969292759895325,
+      "eval_runtime": 1475.6758,
+      "eval_samples_per_second": 5.091,
+      "eval_steps_per_second": 0.637,
+      "step": 45500
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 3.7567715159458286e-05,
+      "loss": 0.1649,
+      "step": 45550
+    },
+    {
+      "epoch": 4.98,
+      "learning_rate": 3.75540629095675e-05,
+      "loss": 0.1488,
+      "step": 45600
+    },
+    {
+      "epoch": 4.98,
+      "eval_cer": 0.047204191061486926,
+      "eval_loss": 0.26253628730773926,
+      "eval_runtime": 1213.3298,
+      "eval_samples_per_second": 6.192,
+      "eval_steps_per_second": 0.775,
+      "step": 45600
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 3.754041065967672e-05,
+      "loss": 0.1776,
+      "step": 45650
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 3.7526758409785934e-05,
+      "loss": 0.1415,
+      "step": 45700
+    },
+    {
+      "epoch": 4.99,
+      "eval_cer": 0.05035912141006909,
+      "eval_loss": 0.2769641876220703,
+      "eval_runtime": 1194.9242,
+      "eval_samples_per_second": 6.287,
+      "eval_steps_per_second": 0.787,
+      "step": 45700
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 3.751310615989515e-05,
+      "loss": 0.1828,
+      "step": 45750
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 3.749945391000437e-05,
+      "loss": 0.1522,
+      "step": 45800
+    },
+    {
+      "epoch": 5.0,
+      "eval_cer": 0.0480562275656051,
+      "eval_loss": 0.2658143639564514,
+      "eval_runtime": 1166.1118,
+      "eval_samples_per_second": 6.443,
+      "eval_steps_per_second": 0.806,
+      "step": 45800
+    },
+    {
+      "epoch": 5.01,
+      "learning_rate": 3.748580166011359e-05,
+      "loss": 0.1204,
+      "step": 45850
+    },
+    {
+      "epoch": 5.01,
+      "learning_rate": 3.7472149410222804e-05,
+      "loss": 0.1203,
+      "step": 45900
+    },
+    {
+      "epoch": 5.01,
+      "eval_cer": 0.04960973990644708,
+      "eval_loss": 0.272051066160202,
+      "eval_runtime": 1159.3999,
+      "eval_samples_per_second": 6.48,
+      "eval_steps_per_second": 0.811,
+      "step": 45900
+    },
+    {
+      "epoch": 5.02,
+      "learning_rate": 3.7458497160332026e-05,
+      "loss": 0.1304,
+      "step": 45950
+    },
+    {
+      "epoch": 5.02,
+      "learning_rate": 3.744484491044124e-05,
+      "loss": 0.1256,
+      "step": 46000
+    },
+    {
+      "epoch": 5.02,
+      "eval_cer": 0.04871321956878056,
+      "eval_loss": 0.267743319272995,
+      "eval_runtime": 1167.536,
+      "eval_samples_per_second": 6.435,
+      "eval_steps_per_second": 0.805,
+      "step": 46000
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 3.7431192660550465e-05,
+      "loss": 0.1462,
+      "step": 46050
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 3.741754041065968e-05,
+      "loss": 0.1306,
+      "step": 46100
+    },
+    {
+      "epoch": 5.03,
+      "eval_cer": 0.04901434090356931,
+      "eval_loss": 0.2696629464626312,
+      "eval_runtime": 1172.6718,
+      "eval_samples_per_second": 6.407,
+      "eval_steps_per_second": 0.802,
+      "step": 46100
+    },
+    {
+      "epoch": 5.04,
+      "learning_rate": 3.7403888160768896e-05,
+      "loss": 0.1378,
+      "step": 46150
+    },
+    {
+      "epoch": 5.05,
+      "learning_rate": 3.739023591087811e-05,
+      "loss": 0.1373,
+      "step": 46200
+    },
+    {
+      "epoch": 5.05,
+      "eval_cer": 0.04708100506089152,
+      "eval_loss": 0.26910731196403503,
+      "eval_runtime": 1223.1444,
+      "eval_samples_per_second": 6.142,
+      "eval_steps_per_second": 0.769,
+      "step": 46200
+    },
+    {
+      "epoch": 5.05,
+      "learning_rate": 3.737658366098733e-05,
+      "loss": 0.1354,
+      "step": 46250
+    },
+    {
+      "epoch": 5.06,
+      "learning_rate": 3.736293141109655e-05,
+      "loss": 0.1421,
+      "step": 46300
+    },
+    {
+      "epoch": 5.06,
+      "eval_cer": 0.04830944323349564,
+      "eval_loss": 0.2727450728416443,
+      "eval_runtime": 1245.9075,
+      "eval_samples_per_second": 6.03,
+      "eval_steps_per_second": 0.754,
+      "step": 46300
+    },
+    {
+      "epoch": 5.06,
+      "learning_rate": 3.7349279161205766e-05,
+      "loss": 0.115,
+      "step": 46350
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 3.733562691131499e-05,
+      "loss": 0.1281,
+      "step": 46400
+    },
+    {
+      "epoch": 5.07,
+      "eval_cer": 0.050030625408481355,
+      "eval_loss": 0.2738962173461914,
+      "eval_runtime": 1175.6177,
+      "eval_samples_per_second": 6.391,
+      "eval_steps_per_second": 0.8,
+      "step": 46400
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 3.7321974661424205e-05,
+      "loss": 0.1344,
+      "step": 46450
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 3.730832241153343e-05,
+      "loss": 0.1257,
+      "step": 46500
+    },
+    {
+      "epoch": 5.08,
+      "eval_cer": 0.050684195578306945,
+      "eval_loss": 0.26944053173065186,
+      "eval_runtime": 1196.4417,
+      "eval_samples_per_second": 6.279,
+      "eval_steps_per_second": 0.786,
+      "step": 46500
+    },
+    {
+      "epoch": 5.08,
+      "learning_rate": 3.729467016164264e-05,
+      "loss": 0.1408,
+      "step": 46550
+    },
+    {
+      "epoch": 5.09,
+      "learning_rate": 3.728101791175186e-05,
+      "loss": 0.1329,
+      "step": 46600
+    },
+    {
+      "epoch": 5.09,
+      "eval_cer": 0.04847711306763938,
+      "eval_loss": 0.27407070994377136,
+      "eval_runtime": 1178.5414,
+      "eval_samples_per_second": 6.375,
+      "eval_steps_per_second": 0.798,
+      "step": 46600
+    },
+    {
+      "epoch": 5.1,
+      "learning_rate": 3.7267365661861075e-05,
+      "loss": 0.1386,
+      "step": 46650
+    },
+    {
+      "epoch": 5.1,
+      "learning_rate": 3.725371341197029e-05,
+      "loss": 0.1181,
+      "step": 46700
+    },
+    {
+      "epoch": 5.1,
+      "eval_cer": 0.048323130566895134,
+      "eval_loss": 0.27152273058891296,
+      "eval_runtime": 1196.4525,
+      "eval_samples_per_second": 6.279,
+      "eval_steps_per_second": 0.786,
+      "step": 46700
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 3.724006116207951e-05,
+      "loss": 0.1269,
+      "step": 46750
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 3.722640891218873e-05,
+      "loss": 0.1485,
+      "step": 46800
+    },
+    {
+      "epoch": 5.11,
+      "eval_cer": 0.04719050372808743,
+      "eval_loss": 0.272051602602005,
+      "eval_runtime": 1208.2115,
+      "eval_samples_per_second": 6.218,
+      "eval_steps_per_second": 0.778,
+      "step": 46800
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 3.721275666229795e-05,
+      "loss": 0.1299,
+      "step": 46850
+    },
+    {
+      "epoch": 5.12,
+      "learning_rate": 3.719910441240717e-05,
+      "loss": 0.1208,
+      "step": 46900
+    },
+    {
+      "epoch": 5.12,
+      "eval_cer": 0.046954397226946255,
+      "eval_loss": 0.2721458971500397,
+      "eval_runtime": 1204.2184,
+      "eval_samples_per_second": 6.239,
+      "eval_steps_per_second": 0.781,
+      "step": 46900
+    },
+    {
+      "epoch": 5.13,
+      "learning_rate": 3.718545216251638e-05,
+      "loss": 0.1464,
+      "step": 46950
+    },
+    {
+      "epoch": 5.13,
+      "learning_rate": 3.7171799912625606e-05,
+      "loss": 0.1208,
+      "step": 47000
+    },
+    {
+      "epoch": 5.13,
+      "eval_cer": 0.04879192173582762,
+      "eval_loss": 0.27154362201690674,
+      "eval_runtime": 1204.4392,
+      "eval_samples_per_second": 6.238,
+      "eval_steps_per_second": 0.78,
+      "step": 47000
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 3.715869375273045e-05,
+      "loss": 0.1329,
+      "step": 47050
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 3.714504150283967e-05,
+      "loss": 0.1369,
+      "step": 47100
+    },
+    {
+      "epoch": 5.14,
+      "eval_cer": 0.048282068566696666,
+      "eval_loss": 0.27170222997665405,
+      "eval_runtime": 1194.6083,
+      "eval_samples_per_second": 6.289,
+      "eval_steps_per_second": 0.787,
+      "step": 47100
+    },
+    {
+      "epoch": 5.15,
+      "learning_rate": 3.713138925294889e-05,
+      "loss": 0.1339,
+      "step": 47150
+    },
+    {
+      "epoch": 5.16,
+      "learning_rate": 3.711773700305811e-05,
+      "loss": 0.1421,
+      "step": 47200
+    },
+    {
+      "epoch": 5.16,
+      "eval_cer": 0.04744714122932785,
+      "eval_loss": 0.26895907521247864,
+      "eval_runtime": 1185.6835,
+      "eval_samples_per_second": 6.336,
+      "eval_steps_per_second": 0.793,
+      "step": 47200
+    },
+    {
+      "epoch": 5.16,
+      "learning_rate": 3.710408475316732e-05,
+      "loss": 0.1211,
+      "step": 47250
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 3.709043250327654e-05,
+      "loss": 0.1335,
+      "step": 47300
+    },
+    {
+      "epoch": 5.17,
+      "eval_cer": 0.04832997423359488,
+      "eval_loss": 0.26967930793762207,
+      "eval_runtime": 1179.959,
+      "eval_samples_per_second": 6.367,
+      "eval_steps_per_second": 0.797,
+      "step": 47300
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 3.707678025338576e-05,
+      "loss": 0.1212,
+      "step": 47350
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 3.706312800349498e-05,
+      "loss": 0.1288,
+      "step": 47400
+    },
+    {
+      "epoch": 5.18,
+      "eval_cer": 0.04718023822803782,
+      "eval_loss": 0.2701777219772339,
+      "eval_runtime": 1172.1235,
+      "eval_samples_per_second": 6.41,
+      "eval_steps_per_second": 0.802,
+      "step": 47400
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 3.704947575360419e-05,
+      "loss": 0.1317,
+      "step": 47450
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 3.7035823503713416e-05,
+      "loss": 0.1178,
+      "step": 47500
+    },
+    {
+      "epoch": 5.19,
+      "eval_cer": 0.04687569505989919,
+      "eval_loss": 0.2715287208557129,
+      "eval_runtime": 1159.6419,
+      "eval_samples_per_second": 6.479,
+      "eval_steps_per_second": 0.811,
+      "step": 47500
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 3.702217125382263e-05,
+      "loss": 0.1387,
+      "step": 47550
+    },
+    {
+      "epoch": 5.2,
+      "learning_rate": 3.7008519003931854e-05,
+      "loss": 0.1401,
+      "step": 47600
+    },
+    {
+      "epoch": 5.2,
+      "eval_cer": 0.0451887312184122,
+      "eval_loss": 0.2689366936683655,
+      "eval_runtime": 1153.2799,
+      "eval_samples_per_second": 6.514,
+      "eval_steps_per_second": 0.815,
+      "step": 47600
+    },
+    {
+      "epoch": 5.2,
+      "learning_rate": 3.699486675404107e-05,
+      "loss": 0.1537,
+      "step": 47650
+    },
+    {
+      "epoch": 5.21,
+      "learning_rate": 3.6981214504150286e-05,
+      "loss": 0.131,
+      "step": 47700
+    },
+    {
+      "epoch": 5.21,
+      "eval_cer": 0.04627003055697181,
+      "eval_loss": 0.27360963821411133,
+      "eval_runtime": 1144.9881,
+      "eval_samples_per_second": 6.562,
+      "eval_steps_per_second": 0.821,
+      "step": 47700
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 3.69675622542595e-05,
+      "loss": 0.1171,
+      "step": 47750
+    },
+    {
+      "epoch": 5.22,
+      "learning_rate": 3.695391000436872e-05,
+      "loss": 0.1227,
+      "step": 47800
+    },
+    {
+      "epoch": 5.22,
+      "eval_cer": 0.04878849990247775,
+      "eval_loss": 0.2697356641292572,
+      "eval_runtime": 1167.8301,
+      "eval_samples_per_second": 6.433,
+      "eval_steps_per_second": 0.805,
+      "step": 47800
+    },
+    {
+      "epoch": 5.23,
+      "learning_rate": 3.694025775447794e-05,
+      "loss": 0.1356,
+      "step": 47850
+    },
+    {
+      "epoch": 5.23,
+      "learning_rate": 3.6926605504587156e-05,
+      "loss": 0.1377,
+      "step": 47900
+    },
+    {
+      "epoch": 5.23,
+      "eval_cer": 0.05130696924798368,
+      "eval_loss": 0.27117544412612915,
+      "eval_runtime": 1196.2902,
+      "eval_samples_per_second": 6.28,
+      "eval_steps_per_second": 0.786,
+      "step": 47900
+    },
+    {
+      "epoch": 5.24,
+      "learning_rate": 3.691295325469638e-05,
+      "loss": 0.1426,
+      "step": 47950
+    },
+    {
+      "epoch": 5.24,
+      "learning_rate": 3.6899301004805594e-05,
+      "loss": 0.1354,
+      "step": 48000
+    },
+    {
+      "epoch": 5.24,
+      "eval_cer": 0.047423188395878745,
+      "eval_loss": 0.2683400809764862,
+      "eval_runtime": 1161.2453,
+      "eval_samples_per_second": 6.47,
+      "eval_steps_per_second": 0.809,
+      "step": 48000
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 3.688564875491482e-05,
+      "loss": 0.1168,
+      "step": 48050
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 3.687199650502403e-05,
+      "loss": 0.1334,
+      "step": 48100
+    },
+    {
+      "epoch": 5.25,
+      "eval_cer": 0.04782012106446392,
+      "eval_loss": 0.2731897532939911,
+      "eval_runtime": 1159.6513,
+      "eval_samples_per_second": 6.479,
+      "eval_steps_per_second": 0.811,
+      "step": 48100
+    },
+    {
+      "epoch": 5.26,
+      "learning_rate": 3.685834425513325e-05,
+      "loss": 0.1383,
+      "step": 48150
+    },
+    {
+      "epoch": 5.26,
+      "learning_rate": 3.6844692005242464e-05,
+      "loss": 0.1405,
+      "step": 48200
+    },
+    {
+      "epoch": 5.26,
+      "eval_cer": 0.045941534555384085,
+      "eval_loss": 0.27034494280815125,
+      "eval_runtime": 1268.9769,
+      "eval_samples_per_second": 5.921,
+      "eval_steps_per_second": 0.741,
+      "step": 48200
+    },
+    {
+      "epoch": 5.27,
+      "learning_rate": 3.683103975535168e-05,
+      "loss": 0.1294,
+      "step": 48250
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 3.68173875054609e-05,
+      "loss": 0.116,
+      "step": 48300
+    },
+    {
+      "epoch": 5.28,
+      "eval_cer": 0.04806649306565472,
+      "eval_loss": 0.27247393131256104,
+      "eval_runtime": 1271.3216,
+      "eval_samples_per_second": 5.91,
+      "eval_steps_per_second": 0.739,
+      "step": 48300
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 3.680373525557012e-05,
+      "loss": 0.1136,
+      "step": 48350
+    },
+    {
+      "epoch": 5.29,
+      "learning_rate": 3.6790083005679334e-05,
+      "loss": 0.1479,
+      "step": 48400
+    },
+    {
+      "epoch": 5.29,
+      "eval_cer": 0.044699409049380474,
+      "eval_loss": 0.27049344778060913,
+      "eval_runtime": 1251.951,
+      "eval_samples_per_second": 6.001,
+      "eval_steps_per_second": 0.751,
+      "step": 48400
+    },
+    {
+      "epoch": 5.29,
+      "learning_rate": 3.677643075578856e-05,
+      "loss": 0.1308,
+      "step": 48450
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 3.676277850589777e-05,
+      "loss": 0.1352,
+      "step": 48500
+    },
+    {
+      "epoch": 5.3,
+      "eval_cer": 0.04500737405086897,
+      "eval_loss": 0.26786038279533386,
+      "eval_runtime": 1245.5699,
+      "eval_samples_per_second": 6.032,
+      "eval_steps_per_second": 0.755,
+      "step": 48500
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 3.6749126256006995e-05,
+      "loss": 0.1188,
+      "step": 48550
+    },
+    {
+      "epoch": 5.31,
+      "learning_rate": 3.673547400611621e-05,
+      "loss": 0.1272,
+      "step": 48600
+    },
+    {
+      "epoch": 5.31,
+      "eval_cer": 0.045086076217916035,
+      "eval_loss": 0.27316638827323914,
+      "eval_runtime": 1239.7313,
+      "eval_samples_per_second": 6.06,
+      "eval_steps_per_second": 0.758,
+      "step": 48600
+    },
+    {
+      "epoch": 5.31,
+      "learning_rate": 3.672182175622543e-05,
+      "loss": 0.1235,
+      "step": 48650
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 3.670816950633464e-05,
+      "loss": 0.1682,
+      "step": 48700
+    },
+    {
+      "epoch": 5.32,
+      "eval_cer": 0.04581834855478868,
+      "eval_loss": 0.2646952271461487,
+      "eval_runtime": 1268.735,
+      "eval_samples_per_second": 5.922,
+      "eval_steps_per_second": 0.741,
+      "step": 48700
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 3.669451725644386e-05,
+      "loss": 0.133,
+      "step": 48750
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 3.668086500655308e-05,
+      "loss": 0.1172,
+      "step": 48800
+    },
+    {
+      "epoch": 5.33,
+      "eval_cer": 0.04660537022525929,
+      "eval_loss": 0.269662082195282,
+      "eval_runtime": 1258.7176,
+      "eval_samples_per_second": 5.969,
+      "eval_steps_per_second": 0.747,
+      "step": 48800
+    },
+    {
+      "epoch": 5.34,
+      "learning_rate": 3.66672127566623e-05,
+      "loss": 0.1423,
+      "step": 48850
+    },
+    {
+      "epoch": 5.34,
+      "learning_rate": 3.665356050677152e-05,
+      "loss": 0.1353,
+      "step": 48900
+    },
+    {
+      "epoch": 5.34,
+      "eval_cer": 0.04585941055498715,
+      "eval_loss": 0.26135268807411194,
+      "eval_runtime": 1262.3154,
+      "eval_samples_per_second": 5.952,
+      "eval_steps_per_second": 0.745,
+      "step": 48900
+    },
+    {
+      "epoch": 5.35,
+      "learning_rate": 3.6639908256880735e-05,
+      "loss": 0.1148,
+      "step": 48950
+    },
+    {
+      "epoch": 5.35,
+      "learning_rate": 3.662625600698996e-05,
+      "loss": 0.1316,
+      "step": 49000
+    },
+    {
+      "epoch": 5.35,
+      "eval_cer": 0.045784130221289965,
+      "eval_loss": 0.26784536242485046,
+      "eval_runtime": 1276.1191,
+      "eval_samples_per_second": 5.887,
+      "eval_steps_per_second": 0.737,
+      "step": 49000
+    },
+    {
+      "epoch": 5.36,
+      "learning_rate": 3.6612603757099174e-05,
+      "loss": 0.1386,
+      "step": 49050
+    },
+    {
+      "epoch": 5.36,
+      "learning_rate": 3.659895150720839e-05,
+      "loss": 0.1637,
+      "step": 49100
+    },
+    {
+      "epoch": 5.36,
+      "eval_cer": 0.04674224355925418,
+      "eval_loss": 0.26811158657073975,
+      "eval_runtime": 1235.9712,
+      "eval_samples_per_second": 6.079,
+      "eval_steps_per_second": 0.761,
+      "step": 49100
+    },
+    {
+      "epoch": 5.37,
+      "learning_rate": 3.6585299257317606e-05,
+      "loss": 0.122,
+      "step": 49150
+    },
+    {
+      "epoch": 5.37,
+      "learning_rate": 3.657164700742682e-05,
+      "loss": 0.1155,
+      "step": 49200
+    },
+    {
+      "epoch": 5.37,
+      "eval_cer": 0.04763876389692069,
+      "eval_loss": 0.2680990397930145,
+      "eval_runtime": 1264.4978,
+      "eval_samples_per_second": 5.941,
+      "eval_steps_per_second": 0.743,
+      "step": 49200
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 3.6557994757536044e-05,
+      "loss": 0.139,
+      "step": 49250
+    },
+    {
+      "epoch": 5.38,
+      "learning_rate": 3.654434250764526e-05,
+      "loss": 0.118,
+      "step": 49300
+    },
+    {
+      "epoch": 5.38,
+      "eval_cer": 0.04588336338843626,
+      "eval_loss": 0.2674948275089264,
+      "eval_runtime": 1261.4278,
+      "eval_samples_per_second": 5.956,
+      "eval_steps_per_second": 0.745,
+      "step": 49300
+    },
+    {
+      "epoch": 5.39,
+      "learning_rate": 3.653069025775448e-05,
+      "loss": 0.1548,
+      "step": 49350
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 3.65170380078637e-05,
+      "loss": 0.1212,
+      "step": 49400
+    },
+    {
+      "epoch": 5.4,
+      "eval_cer": 0.04661905755865878,
+      "eval_loss": 0.2663722634315491,
+      "eval_runtime": 1258.5871,
+      "eval_samples_per_second": 5.969,
+      "eval_steps_per_second": 0.747,
+      "step": 49400
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 3.6503385757972914e-05,
+      "loss": 0.1128,
+      "step": 49450
+    },
+    {
+      "epoch": 5.41,
+      "learning_rate": 3.6489733508082137e-05,
+      "loss": 0.1252,
+      "step": 49500
+    },
+    {
+      "epoch": 5.41,
+      "eval_cer": 0.04388159087876102,
+      "eval_loss": 0.26354527473449707,
+      "eval_runtime": 1249.6738,
+      "eval_samples_per_second": 6.012,
+      "eval_steps_per_second": 0.752,
+      "step": 49500
+    },
+    {
+      "epoch": 5.41,
+      "learning_rate": 3.647608125819135e-05,
+      "loss": 0.1367,
+      "step": 49550
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 3.646242900830057e-05,
+      "loss": 0.1254,
+      "step": 49600
+    },
+    {
+      "epoch": 5.42,
+      "eval_cer": 0.044230617880447985,
+      "eval_loss": 0.26343604922294617,
+      "eval_runtime": 1249.6982,
+      "eval_samples_per_second": 6.012,
+      "eval_steps_per_second": 0.752,
+      "step": 49600
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 3.6448776758409784e-05,
+      "loss": 0.115,
+      "step": 49650
+    },
+    {
+      "epoch": 5.43,
+      "learning_rate": 3.6435124508519e-05,
+      "loss": 0.1205,
+      "step": 49700
+    },
+    {
+      "epoch": 5.43,
+      "eval_cer": 0.04457280121543521,
+      "eval_loss": 0.2652714252471924,
+      "eval_runtime": 1248.8867,
+      "eval_samples_per_second": 6.016,
+      "eval_steps_per_second": 0.753,
+      "step": 49700
+    },
+    {
+      "epoch": 5.43,
+      "learning_rate": 3.6421745303626035e-05,
+      "loss": 0.1693,
+      "step": 49750
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 3.640809305373526e-05,
+      "loss": 0.1395,
+      "step": 49800
+    },
+    {
+      "epoch": 5.44,
+      "eval_cer": 0.04441881871469096,
+      "eval_loss": 0.26012542843818665,
+      "eval_runtime": 1252.5778,
+      "eval_samples_per_second": 5.998,
+      "eval_steps_per_second": 0.75,
+      "step": 49800
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 3.639444080384447e-05,
+      "loss": 0.1174,
+      "step": 49850
+    },
+    {
+      "epoch": 5.45,
+      "learning_rate": 3.6380788553953696e-05,
+      "loss": 0.1248,
+      "step": 49900
+    },
+    {
+      "epoch": 5.45,
+      "eval_cer": 0.04295427404094566,
+      "eval_loss": 0.2604806423187256,
+      "eval_runtime": 1250.2865,
+      "eval_samples_per_second": 6.009,
+      "eval_steps_per_second": 0.752,
+      "step": 49900
+    },
+    {
+      "epoch": 5.46,
+      "learning_rate": 3.636713630406291e-05,
+      "loss": 0.1504,
+      "step": 49950
+    },
+    {
+      "epoch": 5.46,
+      "learning_rate": 3.635348405417213e-05,
+      "loss": 0.1354,
+      "step": 50000
+    },
+    {
+      "epoch": 5.46,
+      "eval_cer": 0.044569379382085335,
+      "eval_loss": 0.25881579518318176,
+      "eval_runtime": 1274.8628,
+      "eval_samples_per_second": 5.893,
+      "eval_steps_per_second": 0.737,
+      "step": 50000
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 3.633983180428135e-05,
+      "loss": 0.1333,
+      "step": 50050
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 3.6326179554390566e-05,
+      "loss": 0.1274,
+      "step": 50100
+    },
+    {
+      "epoch": 5.47,
+      "eval_cer": 0.04327934820918351,
+      "eval_loss": 0.26810508966445923,
+      "eval_runtime": 1234.1932,
+      "eval_samples_per_second": 6.087,
+      "eval_steps_per_second": 0.762,
+      "step": 50100
+    },
+    {
+      "epoch": 5.48,
+      "learning_rate": 3.631252730449978e-05,
+      "loss": 0.1529,
+      "step": 50150
+    },
+    {
+      "epoch": 5.48,
+      "learning_rate": 3.6298875054609e-05,
+      "loss": 0.1358,
+      "step": 50200
+    },
+    {
+      "epoch": 5.48,
+      "eval_cer": 0.04586625422168689,
+      "eval_loss": 0.2661205530166626,
+      "eval_runtime": 1253.0382,
+      "eval_samples_per_second": 5.996,
+      "eval_steps_per_second": 0.75,
+      "step": 50200
+    },
+    {
+      "epoch": 5.49,
+      "learning_rate": 3.628522280471822e-05,
+      "loss": 0.1378,
+      "step": 50250
+    },
+    {
+      "epoch": 5.49,
+      "learning_rate": 3.6271570554827436e-05,
+      "loss": 0.1197,
+      "step": 50300
+    },
+    {
+      "epoch": 5.49,
+      "eval_cer": 0.047036521227343184,
+      "eval_loss": 0.26336920261383057,
+      "eval_runtime": 1281.5933,
+      "eval_samples_per_second": 5.862,
+      "eval_steps_per_second": 0.733,
+      "step": 50300
+    },
+    {
+      "epoch": 5.5,
+      "learning_rate": 3.625791830493666e-05,
+      "loss": 0.1307,
+      "step": 50350
+    },
+    {
+      "epoch": 5.5,
+      "learning_rate": 3.6244266055045874e-05,
+      "loss": 0.1206,
+      "step": 50400
+    },
+    {
+      "epoch": 5.5,
+      "eval_cer": 0.04744714122932785,
+      "eval_loss": 0.2674347162246704,
+      "eval_runtime": 1268.2958,
+      "eval_samples_per_second": 5.924,
+      "eval_steps_per_second": 0.741,
+      "step": 50400
+    },
+    {
+      "epoch": 5.51,
+      "learning_rate": 3.623061380515509e-05,
+      "loss": 0.1378,
+      "step": 50450
+    },
+    {
+      "epoch": 5.52,
+      "learning_rate": 3.621696155526431e-05,
+      "loss": 0.1356,
+      "step": 50500
+    },
+    {
+      "epoch": 5.52,
+      "eval_cer": 0.04557197655359789,
+      "eval_loss": 0.2662409842014313,
+      "eval_runtime": 1262.0384,
+      "eval_samples_per_second": 5.953,
+      "eval_steps_per_second": 0.745,
+      "step": 50500
+    },
+    {
+      "epoch": 5.52,
+      "learning_rate": 3.620330930537353e-05,
+      "loss": 0.1419,
+      "step": 50550
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 3.6189657055482744e-05,
+      "loss": 0.1365,
+      "step": 50600
+    },
+    {
+      "epoch": 5.53,
+      "eval_cer": 0.0467148688924552,
+      "eval_loss": 0.2625742554664612,
+      "eval_runtime": 1264.3349,
+      "eval_samples_per_second": 5.942,
+      "eval_steps_per_second": 0.743,
+      "step": 50600
+    },
+    {
+      "epoch": 5.53,
+      "learning_rate": 3.617600480559196e-05,
+      "loss": 0.1267,
+      "step": 50650
+    },
+    {
+      "epoch": 5.54,
+      "learning_rate": 3.6162352555701176e-05,
+      "loss": 0.1394,
+      "step": 50700
+    },
+    {
+      "epoch": 5.54,
+      "eval_cer": 0.05017776424252586,
+      "eval_loss": 0.26222458481788635,
+      "eval_runtime": 1284.6728,
+      "eval_samples_per_second": 5.848,
+      "eval_steps_per_second": 0.732,
+      "step": 50700
+    },
+    {
+      "epoch": 5.54,
+      "learning_rate": 3.61487003058104e-05,
+      "loss": 0.1296,
+      "step": 50750
+    },
+    {
+      "epoch": 5.55,
+      "learning_rate": 3.613504805591962e-05,
+      "loss": 0.1269,
+      "step": 50800
+    },
+    {
+      "epoch": 5.55,
+      "eval_cer": 0.04537008838595543,
+      "eval_loss": 0.26166456937789917,
+      "eval_runtime": 1264.8055,
+      "eval_samples_per_second": 5.94,
+      "eval_steps_per_second": 0.743,
+      "step": 50800
+    },
+    {
+      "epoch": 5.55,
+      "learning_rate": 3.612139580602884e-05,
+      "loss": 0.123,
+      "step": 50850
+    },
+    {
+      "epoch": 5.56,
+      "learning_rate": 3.610774355613805e-05,
+      "loss": 0.1518,
+      "step": 50900
+    },
+    {
+      "epoch": 5.56,
+      "eval_cer": 0.045038170551017824,
+      "eval_loss": 0.2633424401283264,
+      "eval_runtime": 1261.6473,
+      "eval_samples_per_second": 5.955,
+      "eval_steps_per_second": 0.745,
+      "step": 50900
+    },
+    {
+      "epoch": 5.56,
+      "learning_rate": 3.609409130624727e-05,
+      "loss": 0.1285,
+      "step": 50950
+    },
+    {
+      "epoch": 5.57,
+      "learning_rate": 3.608043905635649e-05,
+      "loss": 0.1507,
+      "step": 51000
+    },
+    {
+      "epoch": 5.57,
+      "eval_cer": 0.0446720343825815,
+      "eval_loss": 0.26199015974998474,
+      "eval_runtime": 1265.2148,
+      "eval_samples_per_second": 5.938,
+      "eval_steps_per_second": 0.743,
+      "step": 51000
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 3.606678680646571e-05,
+      "loss": 0.147,
+      "step": 51050
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 3.605313455657492e-05,
+      "loss": 0.1235,
+      "step": 51100
+    },
+    {
+      "epoch": 5.58,
+      "eval_cer": 0.04783380839786341,
+      "eval_loss": 0.26737144589424133,
+      "eval_runtime": 1274.337,
+      "eval_samples_per_second": 5.896,
+      "eval_steps_per_second": 0.738,
+      "step": 51100
+    },
+    {
+      "epoch": 5.59,
+      "learning_rate": 3.603948230668414e-05,
+      "loss": 0.1562,
+      "step": 51150
+    },
+    {
+      "epoch": 5.59,
+      "learning_rate": 3.602583005679336e-05,
+      "loss": 0.1505,
+      "step": 51200
+    },
+    {
+      "epoch": 5.59,
+      "eval_cer": 0.045335870052456706,
+      "eval_loss": 0.2606879472732544,
+      "eval_runtime": 1267.847,
+      "eval_samples_per_second": 5.926,
+      "eval_steps_per_second": 0.741,
+      "step": 51200
+    },
+    {
+      "epoch": 5.6,
+      "learning_rate": 3.6012177806902584e-05,
+      "loss": 0.1231,
+      "step": 51250
+    },
+    {
+      "epoch": 5.6,
+      "learning_rate": 3.59985255570118e-05,
+      "loss": 0.1292,
+      "step": 51300
+    },
+    {
+      "epoch": 5.6,
+      "eval_cer": 0.04468914354933086,
+      "eval_loss": 0.26606661081314087,
+      "eval_runtime": 1268.994,
+      "eval_samples_per_second": 5.92,
+      "eval_steps_per_second": 0.741,
+      "step": 51300
+    },
+    {
+      "epoch": 5.61,
+      "learning_rate": 3.5984873307121015e-05,
+      "loss": 0.1312,
+      "step": 51350
+    },
+    {
+      "epoch": 5.61,
+      "learning_rate": 3.597122105723023e-05,
+      "loss": 0.1643,
+      "step": 51400
+    },
+    {
+      "epoch": 5.61,
+      "eval_cer": 0.04536324471925568,
+      "eval_loss": 0.2628079056739807,
+      "eval_runtime": 1244.4761,
+      "eval_samples_per_second": 6.037,
+      "eval_steps_per_second": 0.755,
+      "step": 51400
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 3.5957568807339454e-05,
+      "loss": 0.1213,
+      "step": 51450
+    },
+    {
+      "epoch": 5.62,
+      "learning_rate": 3.594391655744867e-05,
+      "loss": 0.132,
+      "step": 51500
+    },
+    {
+      "epoch": 5.62,
+      "eval_cer": 0.04487734438357383,
+      "eval_loss": 0.26130521297454834,
+      "eval_runtime": 1244.9819,
+      "eval_samples_per_second": 6.035,
+      "eval_steps_per_second": 0.755,
+      "step": 51500
+    },
+    {
+      "epoch": 5.63,
+      "learning_rate": 3.5930264307557885e-05,
+      "loss": 0.14,
+      "step": 51550
+    },
+    {
+      "epoch": 5.64,
+      "learning_rate": 3.59166120576671e-05,
+      "loss": 0.1372,
+      "step": 51600
+    },
+    {
+      "epoch": 5.64,
+      "eval_cer": 0.04443592788144032,
+      "eval_loss": 0.26238834857940674,
+      "eval_runtime": 1252.4851,
+      "eval_samples_per_second": 5.998,
+      "eval_steps_per_second": 0.751,
+      "step": 51600
+    },
+    {
+      "epoch": 5.64,
+      "learning_rate": 3.5902959807776324e-05,
+      "loss": 0.1436,
+      "step": 51650
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 3.5889307557885546e-05,
+      "loss": 0.1445,
+      "step": 51700
+    },
+    {
+      "epoch": 5.65,
+      "eval_cer": 0.04557539838694776,
+      "eval_loss": 0.2595983147621155,
+      "eval_runtime": 1271.1092,
+      "eval_samples_per_second": 5.911,
+      "eval_steps_per_second": 0.74,
+      "step": 51700
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 3.587565530799476e-05,
+      "loss": 0.1417,
+      "step": 51750
+    },
+    {
+      "epoch": 5.66,
+      "learning_rate": 3.586200305810398e-05,
+      "loss": 0.1249,
+      "step": 51800
+    },
+    {
+      "epoch": 5.66,
+      "eval_cer": 0.04567805338744393,
+      "eval_loss": 0.2596184015274048,
+      "eval_runtime": 1269.3848,
+      "eval_samples_per_second": 5.919,
+      "eval_steps_per_second": 0.741,
+      "step": 51800
+    },
+    {
+      "epoch": 5.66,
+      "learning_rate": 3.5848350808213194e-05,
+      "loss": 0.1367,
+      "step": 51850
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 3.5834698558322417e-05,
+      "loss": 0.1467,
+      "step": 51900
+    },
+    {
+      "epoch": 5.67,
+      "eval_cer": 0.04477811121642754,
+      "eval_loss": 0.2584436237812042,
+      "eval_runtime": 1255.3233,
+      "eval_samples_per_second": 5.985,
+      "eval_steps_per_second": 0.749,
+      "step": 51900
+    },
+    {
+      "epoch": 5.67,
+      "learning_rate": 3.582104630843163e-05,
+      "loss": 0.1345,
+      "step": 51950
+    },
+    {
+      "epoch": 5.68,
+      "learning_rate": 3.580739405854085e-05,
+      "loss": 0.1216,
+      "step": 52000
+    },
+    {
+      "epoch": 5.68,
+      "eval_cer": 0.044542004715286354,
+      "eval_loss": 0.26097363233566284,
+      "eval_runtime": 1262.6332,
+      "eval_samples_per_second": 5.95,
+      "eval_steps_per_second": 0.744,
+      "step": 52000
+    },
+    {
+      "epoch": 5.68,
+      "learning_rate": 3.5793741808650064e-05,
+      "loss": 0.1363,
+      "step": 52050
+    },
+    {
+      "epoch": 5.69,
+      "learning_rate": 3.5780089558759287e-05,
+      "loss": 0.132,
+      "step": 52100
+    },
+    {
+      "epoch": 5.69,
+      "eval_cer": 0.04648560605801376,
+      "eval_loss": 0.2578682601451874,
+      "eval_runtime": 1261.5333,
+      "eval_samples_per_second": 5.955,
+      "eval_steps_per_second": 0.745,
+      "step": 52100
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 3.57664373088685e-05,
+      "loss": 0.1387,
+      "step": 52150
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 3.5752785058977725e-05,
+      "loss": 0.1331,
+      "step": 52200
+    },
+    {
+      "epoch": 5.7,
+      "eval_cer": 0.04765929489701993,
+      "eval_loss": 0.25392112135887146,
+      "eval_runtime": 1269.6731,
+      "eval_samples_per_second": 5.917,
+      "eval_steps_per_second": 0.74,
+      "step": 52200
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 3.573913280908694e-05,
+      "loss": 0.1492,
+      "step": 52250
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 3.5725480559196157e-05,
+      "loss": 0.1219,
+      "step": 52300
+    },
+    {
+      "epoch": 5.71,
+      "eval_cer": 0.044743892882928814,
+      "eval_loss": 0.2626784145832062,
+      "eval_runtime": 1270.185,
+      "eval_samples_per_second": 5.915,
+      "eval_steps_per_second": 0.74,
+      "step": 52300
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 3.571182830930537e-05,
+      "loss": 0.1599,
+      "step": 52350
+    },
+    {
+      "epoch": 5.72,
+      "learning_rate": 3.5698176059414595e-05,
+      "loss": 0.1252,
+      "step": 52400
+    },
+    {
+      "epoch": 5.72,
+      "eval_cer": 0.0451271382181145,
+      "eval_loss": 0.2560093402862549,
+      "eval_runtime": 1269.5004,
+      "eval_samples_per_second": 5.918,
+      "eval_steps_per_second": 0.74,
+      "step": 52400
+    },
+    {
+      "epoch": 5.73,
+      "learning_rate": 3.568452380952381e-05,
+      "loss": 0.1291,
+      "step": 52450
+    },
+    {
+      "epoch": 5.73,
+      "learning_rate": 3.567087155963303e-05,
+      "loss": 0.1374,
+      "step": 52500
+    },
+    {
+      "epoch": 5.73,
+      "eval_cer": 0.044836282383375364,
+      "eval_loss": 0.25710663199424744,
+      "eval_runtime": 1269.5647,
+      "eval_samples_per_second": 5.918,
+      "eval_steps_per_second": 0.74,
+      "step": 52500
+    },
+    {
+      "epoch": 5.74,
+      "learning_rate": 3.565721930974224e-05,
+      "loss": 0.1405,
+      "step": 52550
+    },
+    {
+      "epoch": 5.74,
+      "learning_rate": 3.5643567059851465e-05,
+      "loss": 0.1355,
+      "step": 52600
+    },
+    {
+      "epoch": 5.74,
+      "eval_cer": 0.04604761138923012,
+      "eval_loss": 0.2563580572605133,
+      "eval_runtime": 1264.2998,
+      "eval_samples_per_second": 5.942,
+      "eval_steps_per_second": 0.743,
+      "step": 52600
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 3.562991480996069e-05,
+      "loss": 0.1349,
+      "step": 52650
+    },
+    {
+      "epoch": 5.76,
+      "learning_rate": 3.5616262560069903e-05,
+      "loss": 0.1225,
+      "step": 52700
+    },
+    {
+      "epoch": 5.76,
+      "eval_cer": 0.04685858589314983,
+      "eval_loss": 0.2595473527908325,
+      "eval_runtime": 1271.6988,
+      "eval_samples_per_second": 5.908,
+      "eval_steps_per_second": 0.739,
+      "step": 52700
+    },
+    {
+      "epoch": 5.76,
+      "learning_rate": 3.560261031017912e-05,
+      "loss": 0.1259,
+      "step": 52750
+    },
+    {
+      "epoch": 5.77,
+      "learning_rate": 3.5588958060288335e-05,
+      "loss": 0.1449,
+      "step": 52800
+    },
+    {
+      "epoch": 5.77,
+      "eval_cer": 0.044747314716278686,
+      "eval_loss": 0.2548329830169678,
+      "eval_runtime": 1266.1674,
+      "eval_samples_per_second": 5.934,
+      "eval_steps_per_second": 0.742,
+      "step": 52800
+    },
+    {
+      "epoch": 5.77,
+      "learning_rate": 3.557530581039756e-05,
+      "loss": 0.1331,
+      "step": 52850
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 3.5561653560506774e-05,
+      "loss": 0.1509,
+      "step": 52900
+    },
+    {
+      "epoch": 5.78,
+      "eval_cer": 0.04509976355131552,
+      "eval_loss": 0.2525275647640228,
+      "eval_runtime": 1264.0478,
+      "eval_samples_per_second": 5.944,
+      "eval_steps_per_second": 0.744,
+      "step": 52900
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 3.554800131061599e-05,
+      "loss": 0.1198,
+      "step": 52950
+    },
+    {
+      "epoch": 5.79,
+      "learning_rate": 3.5534349060725205e-05,
+      "loss": 0.1418,
+      "step": 53000
+    },
+    {
+      "epoch": 5.79,
+      "eval_cer": 0.04541457221950376,
+      "eval_loss": 0.2532413899898529,
+      "eval_runtime": 1261.684,
+      "eval_samples_per_second": 5.955,
+      "eval_steps_per_second": 0.745,
+      "step": 53000
+    },
+    {
+      "epoch": 5.79,
+      "learning_rate": 3.552069681083443e-05,
+      "loss": 0.1342,
+      "step": 53050
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 3.550704456094365e-05,
+      "loss": 0.1365,
+      "step": 53100
+    },
+    {
+      "epoch": 5.8,
+      "eval_cer": 0.04493551555052166,
+      "eval_loss": 0.2554800808429718,
+      "eval_runtime": 1248.7086,
+      "eval_samples_per_second": 6.017,
+      "eval_steps_per_second": 0.753,
+      "step": 53100
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 3.5493392311052866e-05,
+      "loss": 0.115,
+      "step": 53150
+    },
+    {
+      "epoch": 5.81,
+      "learning_rate": 3.547974006116208e-05,
+      "loss": 0.1405,
+      "step": 53200
+    },
+    {
+      "epoch": 5.81,
+      "eval_cer": 0.044367491214442875,
+      "eval_loss": 0.2578386068344116,
+      "eval_runtime": 1259.4446,
+      "eval_samples_per_second": 5.965,
+      "eval_steps_per_second": 0.746,
+      "step": 53200
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 3.54660878112713e-05,
+      "loss": 0.1349,
+      "step": 53250
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 3.5452435561380514e-05,
+      "loss": 0.1415,
+      "step": 53300
+    },
+    {
+      "epoch": 5.82,
+      "eval_cer": 0.048206788232999474,
+      "eval_loss": 0.2606782019138336,
+      "eval_runtime": 1285.9663,
+      "eval_samples_per_second": 5.842,
+      "eval_steps_per_second": 0.731,
+      "step": 53300
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 3.5438783311489736e-05,
+      "loss": 0.1643,
+      "step": 53350
+    },
+    {
+      "epoch": 5.83,
+      "learning_rate": 3.542513106159895e-05,
+      "loss": 0.122,
+      "step": 53400
+    },
+    {
+      "epoch": 5.83,
+      "eval_cer": 0.04651640255816261,
+      "eval_loss": 0.2521412670612335,
+      "eval_runtime": 1275.8742,
+      "eval_samples_per_second": 5.889,
+      "eval_steps_per_second": 0.737,
+      "step": 53400
+    },
+    {
+      "epoch": 5.84,
+      "learning_rate": 3.541147881170817e-05,
+      "loss": 0.1179,
+      "step": 53450
+    },
+    {
+      "epoch": 5.84,
+      "learning_rate": 3.539782656181739e-05,
+      "loss": 0.136,
+      "step": 53500
+    },
+    {
+      "epoch": 5.84,
+      "eval_cer": 0.045017639550918594,
+      "eval_loss": 0.2541681230068207,
+      "eval_runtime": 1144.0813,
+      "eval_samples_per_second": 6.567,
+      "eval_steps_per_second": 0.822,
+      "step": 53500
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 3.538444735692442e-05,
+      "loss": 0.1342,
+      "step": 53550
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 3.537079510703364e-05,
+      "loss": 0.1477,
+      "step": 53600
+    },
+    {
+      "epoch": 5.85,
+      "eval_cer": 0.045428259552903257,
+      "eval_loss": 0.2599349021911621,
+      "eval_runtime": 1156.9156,
+      "eval_samples_per_second": 6.494,
+      "eval_steps_per_second": 0.813,
+      "step": 53600
+    },
+    {
+      "epoch": 5.86,
+      "learning_rate": 3.5357142857142864e-05,
+      "loss": 0.141,
+      "step": 53650
+    },
+    {
+      "epoch": 5.86,
+      "learning_rate": 3.534349060725208e-05,
+      "loss": 0.1443,
+      "step": 53700
+    },
+    {
+      "epoch": 5.86,
+      "eval_cer": 0.044945781050571273,
+      "eval_loss": 0.2558267116546631,
+      "eval_runtime": 1247.3648,
+      "eval_samples_per_second": 6.023,
+      "eval_steps_per_second": 0.754,
+      "step": 53700
+    },
+    {
+      "epoch": 5.87,
+      "learning_rate": 3.5329838357361295e-05,
+      "loss": 0.1352,
+      "step": 53750
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 3.531618610747051e-05,
+      "loss": 0.1668,
+      "step": 53800
+    },
+    {
+      "epoch": 5.88,
+      "eval_cer": 0.04712548889443986,
+      "eval_loss": 0.25672096014022827,
+      "eval_runtime": 1282.4517,
+      "eval_samples_per_second": 5.858,
+      "eval_steps_per_second": 0.733,
+      "step": 53800
+    },
+    {
+      "epoch": 5.88,
+      "learning_rate": 3.530253385757973e-05,
+      "loss": 0.1336,
+      "step": 53850
+    },
+    {
+      "epoch": 5.89,
+      "learning_rate": 3.528888160768895e-05,
+      "loss": 0.1148,
+      "step": 53900
+    },
+    {
+      "epoch": 5.89,
+      "eval_cer": 0.044613863215633674,
+      "eval_loss": 0.25303828716278076,
+      "eval_runtime": 1256.3406,
+      "eval_samples_per_second": 5.98,
+      "eval_steps_per_second": 0.748,
+      "step": 53900
+    },
+    {
+      "epoch": 5.89,
+      "learning_rate": 3.5275229357798165e-05,
+      "loss": 0.1592,
+      "step": 53950
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 3.526157710790738e-05,
+      "loss": 0.1355,
+      "step": 54000
+    },
+    {
+      "epoch": 5.9,
+      "eval_cer": 0.043782357711614726,
+      "eval_loss": 0.25274816155433655,
+      "eval_runtime": 1245.1946,
+      "eval_samples_per_second": 6.034,
+      "eval_steps_per_second": 0.755,
+      "step": 54000
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 3.5247924858016604e-05,
+      "loss": 0.1342,
+      "step": 54050
+    },
+    {
+      "epoch": 5.91,
+      "learning_rate": 3.5234272608125826e-05,
+      "loss": 0.13,
+      "step": 54100
+    },
+    {
+      "epoch": 5.91,
+      "eval_cer": 0.04428878904739581,
+      "eval_loss": 0.25448983907699585,
+      "eval_runtime": 1257.6849,
+      "eval_samples_per_second": 5.974,
+      "eval_steps_per_second": 0.747,
+      "step": 54100
+    },
+    {
+      "epoch": 5.91,
+      "learning_rate": 3.522062035823504e-05,
+      "loss": 0.1447,
+      "step": 54150
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 3.520696810834426e-05,
+      "loss": 0.1317,
+      "step": 54200
+    },
+    {
+      "epoch": 5.92,
+      "eval_cer": 0.04340595604312879,
+      "eval_loss": 0.253248006105423,
+      "eval_runtime": 1252.06,
+      "eval_samples_per_second": 6.001,
+      "eval_steps_per_second": 0.751,
+      "step": 54200
+    },
+    {
+      "epoch": 5.93,
+      "learning_rate": 3.5193315858453474e-05,
+      "loss": 0.1265,
+      "step": 54250
+    },
+    {
+      "epoch": 5.93,
+      "learning_rate": 3.517966360856269e-05,
+      "loss": 0.1338,
+      "step": 54300
+    },
+    {
+      "epoch": 5.93,
+      "eval_cer": 0.04605787688927974,
+      "eval_loss": 0.2523036301136017,
+      "eval_runtime": 1270.608,
+      "eval_samples_per_second": 5.913,
+      "eval_steps_per_second": 0.74,
+      "step": 54300
+    },
+    {
+      "epoch": 5.94,
+      "learning_rate": 3.516601135867191e-05,
+      "loss": 0.1246,
+      "step": 54350
+    },
+    {
+      "epoch": 5.94,
+      "learning_rate": 3.515235910878113e-05,
+      "loss": 0.1266,
+      "step": 54400
+    },
+    {
+      "epoch": 5.94,
+      "eval_cer": 0.04314247487518863,
+      "eval_loss": 0.2512058913707733,
+      "eval_runtime": 1257.0108,
+      "eval_samples_per_second": 5.977,
+      "eval_steps_per_second": 0.748,
+      "step": 54400
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 3.5138706858890344e-05,
+      "loss": 0.1275,
+      "step": 54450
+    },
+    {
+      "epoch": 5.95,
+      "learning_rate": 3.5125054608999566e-05,
+      "loss": 0.1268,
+      "step": 54500
+    },
+    {
+      "epoch": 5.95,
+      "eval_cer": 0.04472336188282958,
+      "eval_loss": 0.25320523977279663,
+      "eval_runtime": 1262.1566,
+      "eval_samples_per_second": 5.953,
+      "eval_steps_per_second": 0.745,
+      "step": 54500
+    },
+    {
+      "epoch": 5.96,
+      "learning_rate": 3.511140235910878e-05,
+      "loss": 0.1611,
+      "step": 54550
+    },
+    {
+      "epoch": 5.96,
+      "learning_rate": 3.5097750109218005e-05,
+      "loss": 0.1215,
+      "step": 54600
+    },
+    {
+      "epoch": 5.96,
+      "eval_cer": 0.044791798549827025,
+      "eval_loss": 0.2544041574001312,
+      "eval_runtime": 1258.3474,
+      "eval_samples_per_second": 5.971,
+      "eval_steps_per_second": 0.747,
+      "step": 54600
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 3.508409785932722e-05,
+      "loss": 0.1343,
+      "step": 54650
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 3.5070445609436437e-05,
+      "loss": 0.1416,
+      "step": 54700
+    },
+    {
+      "epoch": 5.97,
+      "eval_cer": 0.04583887955488792,
+      "eval_loss": 0.24952231347560883,
+      "eval_runtime": 1266.7565,
+      "eval_samples_per_second": 5.931,
+      "eval_steps_per_second": 0.742,
+      "step": 54700
+    },
+    {
+      "epoch": 5.98,
+      "learning_rate": 3.505679335954565e-05,
+      "loss": 0.1357,
+      "step": 54750
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 3.504314110965487e-05,
+      "loss": 0.1256,
+      "step": 54800
+    },
+    {
+      "epoch": 5.99,
+      "eval_cer": 0.04646507505791453,
+      "eval_loss": 0.2494717687368393,
+      "eval_runtime": 1299.795,
+      "eval_samples_per_second": 5.78,
+      "eval_steps_per_second": 0.723,
+      "step": 54800
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 3.502948885976409e-05,
+      "loss": 0.1354,
+      "step": 54850
+    },
+    {
+      "epoch": 6.0,
+      "learning_rate": 3.5015836609873307e-05,
+      "loss": 0.1463,
+      "step": 54900
+    },
+    {
+      "epoch": 6.0,
+      "eval_cer": 0.04498342121741987,
+      "eval_loss": 0.2538948655128479,
+      "eval_runtime": 1272.2595,
+      "eval_samples_per_second": 5.905,
+      "eval_steps_per_second": 0.739,
+      "step": 54900
+    },
+    {
+      "epoch": 6.0,
+      "learning_rate": 3.500218435998253e-05,
+      "loss": 0.1226,
+      "step": 54950
+    },
+    {
+      "epoch": 6.01,
+      "learning_rate": 3.4988532110091745e-05,
+      "loss": 0.0819,
+      "step": 55000
+    },
+    {
+      "epoch": 6.01,
+      "eval_cer": 0.04370365554456767,
+      "eval_loss": 0.2585710883140564,
+      "eval_runtime": 1254.3721,
+      "eval_samples_per_second": 5.989,
+      "eval_steps_per_second": 0.749,
+      "step": 55000
+    },
+    {
+      "epoch": 6.01,
+      "learning_rate": 3.497487986020097e-05,
+      "loss": 0.0823,
+      "step": 55050
+    },
+    {
+      "epoch": 6.02,
+      "learning_rate": 3.4961227610310183e-05,
+      "loss": 0.1003,
+      "step": 55100
+    },
+    {
+      "epoch": 6.02,
+      "eval_cer": 0.044815751383276134,
+      "eval_loss": 0.2569202780723572,
+      "eval_runtime": 1267.7723,
+      "eval_samples_per_second": 5.926,
+      "eval_steps_per_second": 0.741,
+      "step": 55100
+    },
+    {
+      "epoch": 6.02,
+      "learning_rate": 3.49475753604194e-05,
+      "loss": 0.1029,
+      "step": 55150
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 3.4933923110528615e-05,
+      "loss": 0.0956,
+      "step": 55200
+    },
+    {
+      "epoch": 6.03,
+      "eval_cer": 0.04651640255816261,
+      "eval_loss": 0.2601701319217682,
+      "eval_runtime": 1273.0175,
+      "eval_samples_per_second": 5.902,
+      "eval_steps_per_second": 0.738,
+      "step": 55200
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 3.492027086063783e-05,
+      "loss": 0.0996,
+      "step": 55250
+    },
+    {
+      "epoch": 6.04,
+      "learning_rate": 3.4906618610747053e-05,
+      "loss": 0.089,
+      "step": 55300
+    },
+    {
+      "epoch": 6.04,
+      "eval_cer": 0.04504843605106744,
+      "eval_loss": 0.2632923126220703,
+      "eval_runtime": 1249.6925,
+      "eval_samples_per_second": 6.012,
+      "eval_steps_per_second": 0.752,
+      "step": 55300
+    },
+    {
+      "epoch": 6.05,
+      "learning_rate": 3.489296636085627e-05,
+      "loss": 0.0868,
+      "step": 55350
+    },
+    {
+      "epoch": 6.05,
+      "learning_rate": 3.487931411096549e-05,
+      "loss": 0.0898,
+      "step": 55400
+    },
+    {
+      "epoch": 6.05,
+      "eval_cer": 0.04392265287895949,
+      "eval_loss": 0.258096843957901,
+      "eval_runtime": 1258.7235,
+      "eval_samples_per_second": 5.969,
+      "eval_steps_per_second": 0.747,
+      "step": 55400
+    },
+    {
+      "epoch": 6.06,
+      "learning_rate": 3.486566186107471e-05,
+      "loss": 0.0907,
+      "step": 55450
+    },
+    {
+      "epoch": 6.06,
+      "learning_rate": 3.4852009611183923e-05,
+      "loss": 0.1034,
+      "step": 55500
+    },
+    {
+      "epoch": 6.06,
+      "eval_cer": 0.04375156121146588,
+      "eval_loss": 0.25894665718078613,
+      "eval_runtime": 1253.9835,
+      "eval_samples_per_second": 5.991,
+      "eval_steps_per_second": 0.75,
+      "step": 55500
+    },
+    {
+      "epoch": 6.07,
+      "learning_rate": 3.4838357361293146e-05,
+      "loss": 0.0812,
+      "step": 55550
+    },
+    {
+      "epoch": 6.07,
+      "learning_rate": 3.482470511140236e-05,
+      "loss": 0.098,
+      "step": 55600
+    },
+    {
+      "epoch": 6.07,
+      "eval_cer": 0.044367491214442875,
+      "eval_loss": 0.2576400339603424,
+      "eval_runtime": 1270.1032,
+      "eval_samples_per_second": 5.915,
+      "eval_steps_per_second": 0.74,
+      "step": 55600
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 3.481105286151158e-05,
+      "loss": 0.0796,
+      "step": 55650
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 3.4797400611620794e-05,
+      "loss": 0.1013,
+      "step": 55700
+    },
+    {
+      "epoch": 6.08,
+      "eval_cer": 0.0437857795449646,
+      "eval_loss": 0.261007159948349,
+      "eval_runtime": 1243.8076,
+      "eval_samples_per_second": 6.04,
+      "eval_steps_per_second": 0.756,
+      "step": 55700
+    },
+    {
+      "epoch": 6.09,
+      "learning_rate": 3.4783748361730016e-05,
+      "loss": 0.0766,
+      "step": 55750
+    },
+    {
+      "epoch": 6.09,
+      "learning_rate": 3.477009611183923e-05,
+      "loss": 0.0943,
+      "step": 55800
+    },
+    {
+      "epoch": 6.09,
+      "eval_cer": 0.04406979171300399,
+      "eval_loss": 0.2620924413204193,
+      "eval_runtime": 1266.0942,
+      "eval_samples_per_second": 5.934,
+      "eval_steps_per_second": 0.742,
+      "step": 55800
+    },
+    {
+      "epoch": 6.1,
+      "learning_rate": 3.475644386194845e-05,
+      "loss": 0.0942,
+      "step": 55850
+    },
+    {
+      "epoch": 6.11,
+      "learning_rate": 3.474279161205767e-05,
+      "loss": 0.0958,
+      "step": 55900
+    },
+    {
+      "epoch": 6.11,
+      "eval_cer": 0.043104834708340034,
+      "eval_loss": 0.2605457901954651,
+      "eval_runtime": 1243.8093,
+      "eval_samples_per_second": 6.04,
+      "eval_steps_per_second": 0.756,
+      "step": 55900
+    },
+    {
+      "epoch": 6.11,
+      "learning_rate": 3.4729139362166886e-05,
+      "loss": 0.1016,
+      "step": 55950
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 3.471548711227611e-05,
+      "loss": 0.0975,
+      "step": 56000
+    },
+    {
+      "epoch": 6.12,
+      "eval_cer": 0.04476100204967818,
+      "eval_loss": 0.261498361825943,
+      "eval_runtime": 1275.5024,
+      "eval_samples_per_second": 5.89,
+      "eval_steps_per_second": 0.737,
+      "step": 56000
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 3.4701834862385325e-05,
+      "loss": 0.1053,
+      "step": 56050
+    },
+    {
+      "epoch": 6.13,
+      "learning_rate": 3.468818261249454e-05,
+      "loss": 0.1016,
+      "step": 56100
+    },
+    {
+      "epoch": 6.13,
+      "eval_cer": 0.04587994155508639,
+      "eval_loss": 0.2610916197299957,
+      "eval_runtime": 1273.2615,
+      "eval_samples_per_second": 5.901,
+      "eval_steps_per_second": 0.738,
+      "step": 56100
+    },
+    {
+      "epoch": 6.13,
+      "learning_rate": 3.4674530362603756e-05,
+      "loss": 0.1116,
+      "step": 56150
+    },
+    {
+      "epoch": 6.14,
+      "learning_rate": 3.466087811271297e-05,
+      "loss": 0.0968,
+      "step": 56200
+    },
+    {
+      "epoch": 6.14,
+      "eval_cer": 0.044709674549430097,
+      "eval_loss": 0.2603790760040283,
+      "eval_runtime": 1265.8881,
+      "eval_samples_per_second": 5.935,
+      "eval_steps_per_second": 0.743,
+      "step": 56200
+    },
+    {
+      "epoch": 6.14,
+      "learning_rate": 3.4647225862822195e-05,
+      "loss": 0.1051,
+      "step": 56250
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 3.463357361293141e-05,
+      "loss": 0.0935,
+      "step": 56300
+    },
+    {
+      "epoch": 6.15,
+      "eval_cer": 0.04556513288689814,
+      "eval_loss": 0.2579694092273712,
+      "eval_runtime": 1263.4804,
+      "eval_samples_per_second": 5.946,
+      "eval_steps_per_second": 0.744,
+      "step": 56300
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 3.461992136304063e-05,
+      "loss": 0.0914,
+      "step": 56350
+    },
+    {
+      "epoch": 6.16,
+      "learning_rate": 3.460626911314985e-05,
+      "loss": 0.0953,
+      "step": 56400
+    },
+    {
+      "epoch": 6.16,
+      "eval_cer": 0.047361595395581046,
+      "eval_loss": 0.2568546533584595,
+      "eval_runtime": 1283.114,
+      "eval_samples_per_second": 5.855,
+      "eval_steps_per_second": 0.733,
+      "step": 56400
+    },
+    {
+      "epoch": 6.17,
+      "learning_rate": 3.4592616863259065e-05,
+      "loss": 0.0879,
+      "step": 56450
+    },
+    {
+      "epoch": 6.17,
+      "learning_rate": 3.457896461336829e-05,
+      "loss": 0.0939,
+      "step": 56500
+    },
+    {
+      "epoch": 6.17,
+      "eval_cer": 0.04408005721305361,
+      "eval_loss": 0.2597881853580475,
+      "eval_runtime": 1254.7553,
+      "eval_samples_per_second": 5.988,
+      "eval_steps_per_second": 0.749,
+      "step": 56500
+    },
+    {
+      "epoch": 6.18,
+      "learning_rate": 3.45653123634775e-05,
+      "loss": 0.1095,
+      "step": 56550
+    },
+    {
+      "epoch": 6.18,
+      "learning_rate": 3.455193315858454e-05,
+      "loss": 0.0884,
+      "step": 56600
+    },
+    {
+      "epoch": 6.18,
+      "eval_cer": 0.044097166379802974,
+      "eval_loss": 0.2602521479129791,
+      "eval_runtime": 1251.1413,
+      "eval_samples_per_second": 6.005,
+      "eval_steps_per_second": 0.751,
+      "step": 56600
+    },
+    {
+      "epoch": 6.19,
+      "learning_rate": 3.4538280908693754e-05,
+      "loss": 0.0902,
+      "step": 56650
+    },
+    {
+      "epoch": 6.19,
+      "learning_rate": 3.452462865880297e-05,
+      "loss": 0.0902,
+      "step": 56700
+    },
+    {
+      "epoch": 6.19,
+      "eval_cer": 0.043761826711515496,
+      "eval_loss": 0.26123595237731934,
+      "eval_runtime": 1261.7321,
+      "eval_samples_per_second": 5.955,
+      "eval_steps_per_second": 0.745,
+      "step": 56700
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 3.4510976408912185e-05,
+      "loss": 0.0991,
+      "step": 56750
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 3.449732415902141e-05,
+      "loss": 0.0936,
+      "step": 56800
+    },
+    {
+      "epoch": 6.2,
+      "eval_cer": 0.04275922954000294,
+      "eval_loss": 0.2600439190864563,
+      "eval_runtime": 1244.3219,
+      "eval_samples_per_second": 6.038,
+      "eval_steps_per_second": 0.755,
+      "step": 56800
+    },
+    {
+      "epoch": 6.21,
+      "learning_rate": 3.4483671909130624e-05,
+      "loss": 0.0899,
+      "step": 56850
+    },
+    {
+      "epoch": 6.21,
+      "learning_rate": 3.4470019659239846e-05,
+      "loss": 0.0966,
+      "step": 56900
+    },
+    {
+      "epoch": 6.21,
+      "eval_cer": 0.042122768536926716,
+      "eval_loss": 0.25872135162353516,
+      "eval_runtime": 1250.8633,
+      "eval_samples_per_second": 6.006,
+      "eval_steps_per_second": 0.751,
+      "step": 56900
+    },
+    {
+      "epoch": 6.22,
+      "learning_rate": 3.445636740934906e-05,
+      "loss": 0.0958,
+      "step": 56950
+    },
+    {
+      "epoch": 6.23,
+      "learning_rate": 3.4442715159458285e-05,
+      "loss": 0.0898,
+      "step": 57000
+    },
+    {
+      "epoch": 6.23,
+      "eval_cer": 0.04389185637881064,
+      "eval_loss": 0.2622995972633362,
+      "eval_runtime": 1258.0312,
+      "eval_samples_per_second": 5.972,
+      "eval_steps_per_second": 0.747,
+      "step": 57000
+    },
+    {
+      "epoch": 6.23,
+      "learning_rate": 3.44290629095675e-05,
+      "loss": 0.1116,
+      "step": 57050
+    },
+    {
+      "epoch": 6.24,
+      "learning_rate": 3.4415410659676716e-05,
+      "loss": 0.0959,
+      "step": 57100
+    },
+    {
+      "epoch": 6.24,
+      "eval_cer": 0.042683949206305756,
+      "eval_loss": 0.2615799307823181,
+      "eval_runtime": 1264.8249,
+      "eval_samples_per_second": 5.94,
+      "eval_steps_per_second": 0.743,
+      "step": 57100
+    },
+    {
+      "epoch": 6.24,
+      "learning_rate": 3.440175840978593e-05,
+      "loss": 0.0954,
+      "step": 57150
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 3.438810615989515e-05,
+      "loss": 0.1009,
+      "step": 57200
+    },
+    {
+      "epoch": 6.25,
+      "eval_cer": 0.04379604504501422,
+      "eval_loss": 0.25624653697013855,
+      "eval_runtime": 1276.2697,
+      "eval_samples_per_second": 5.887,
+      "eval_steps_per_second": 0.737,
+      "step": 57200
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 3.437445391000437e-05,
+      "loss": 0.0968,
+      "step": 57250
+    },
+    {
+      "epoch": 6.26,
+      "learning_rate": 3.4360801660113587e-05,
+      "loss": 0.0981,
+      "step": 57300
+    },
+    {
+      "epoch": 6.26,
+      "eval_cer": 0.041849021868936936,
+      "eval_loss": 0.25614550709724426,
+      "eval_runtime": 1247.1429,
+      "eval_samples_per_second": 6.024,
+      "eval_steps_per_second": 0.754,
+      "step": 57300
+    },
+    {
+      "epoch": 6.26,
+      "learning_rate": 3.434714941022281e-05,
+      "loss": 0.1056,
+      "step": 57350
+    },
+    {
+      "epoch": 6.27,
+      "learning_rate": 3.4333497160332025e-05,
+      "loss": 0.0938,
+      "step": 57400
+    },
+    {
+      "epoch": 6.27,
+      "eval_cer": 0.04287215004054872,
+      "eval_loss": 0.2652567923069,
+      "eval_runtime": 1242.8949,
+      "eval_samples_per_second": 6.045,
+      "eval_steps_per_second": 0.756,
+      "step": 57400
+    },
+    {
+      "epoch": 6.27,
+      "learning_rate": 3.431984491044124e-05,
+      "loss": 0.1049,
+      "step": 57450
+    },
+    {
+      "epoch": 6.28,
+      "learning_rate": 3.430619266055046e-05,
+      "loss": 0.1063,
+      "step": 57500
+    },
+    {
+      "epoch": 6.28,
+      "eval_cer": 0.04191403670258451,
+      "eval_loss": 0.2614375054836273,
+      "eval_runtime": 1265.454,
+      "eval_samples_per_second": 5.937,
+      "eval_steps_per_second": 0.743,
+      "step": 57500
+    },
+    {
+      "epoch": 6.29,
+      "learning_rate": 3.429254041065968e-05,
+      "loss": 0.0941,
+      "step": 57550
+    },
+    {
+      "epoch": 6.29,
+      "learning_rate": 3.4278888160768895e-05,
+      "loss": 0.1112,
+      "step": 57600
+    },
+    {
+      "epoch": 6.29,
+      "eval_cer": 0.04245810820521419,
+      "eval_loss": 0.2571166455745697,
+      "eval_runtime": 1277.9712,
+      "eval_samples_per_second": 5.879,
+      "eval_steps_per_second": 0.736,
+      "step": 57600
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 3.426523591087811e-05,
+      "loss": 0.0947,
+      "step": 57650
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 3.4251583660987327e-05,
+      "loss": 0.0998,
+      "step": 57700
+    },
+    {
+      "epoch": 6.3,
+      "eval_cer": 0.041513682200649465,
+      "eval_loss": 0.26077941060066223,
+      "eval_runtime": 1245.1251,
+      "eval_samples_per_second": 6.034,
+      "eval_steps_per_second": 0.755,
+      "step": 57700
+    },
+    {
+      "epoch": 6.31,
+      "learning_rate": 3.423793141109655e-05,
+      "loss": 0.1019,
+      "step": 57750
+    },
+    {
+      "epoch": 6.31,
+      "learning_rate": 3.422427916120577e-05,
+      "loss": 0.1096,
+      "step": 57800
+    },
+    {
+      "epoch": 6.31,
+      "eval_cer": 0.04350518921027508,
+      "eval_loss": 0.25821661949157715,
+      "eval_runtime": 1268.0984,
+      "eval_samples_per_second": 5.925,
+      "eval_steps_per_second": 0.741,
+      "step": 57800
+    },
+    {
+      "epoch": 6.32,
+      "learning_rate": 3.421062691131499e-05,
+      "loss": 0.0893,
+      "step": 57850
+    },
+    {
+      "epoch": 6.32,
+      "learning_rate": 3.4196974661424203e-05,
+      "loss": 0.0944,
+      "step": 57900
+    },
+    {
+      "epoch": 6.32,
+      "eval_cer": 0.04171214853494205,
+      "eval_loss": 0.2577536404132843,
+      "eval_runtime": 1243.3201,
+      "eval_samples_per_second": 6.043,
+      "eval_steps_per_second": 0.756,
+      "step": 57900
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 3.4183322411533426e-05,
+      "loss": 0.0963,
+      "step": 57950
+    },
+    {
+      "epoch": 6.33,
+      "learning_rate": 3.416967016164264e-05,
+      "loss": 0.0984,
+      "step": 58000
+    },
+    {
+      "epoch": 6.33,
+      "eval_cer": 0.041961942369482724,
+      "eval_loss": 0.2591804265975952,
+      "eval_runtime": 1279.5484,
+      "eval_samples_per_second": 5.872,
+      "eval_steps_per_second": 0.735,
+      "step": 58000
+    },
+    {
+      "epoch": 6.34,
+      "learning_rate": 3.415601791175186e-05,
+      "loss": 0.097,
+      "step": 58050
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 3.4142365661861073e-05,
+      "loss": 0.0939,
+      "step": 58100
+    },
+    {
+      "epoch": 6.35,
+      "eval_cer": 0.042122768536926716,
+      "eval_loss": 0.2615301012992859,
+      "eval_runtime": 1263.1636,
+      "eval_samples_per_second": 5.948,
+      "eval_steps_per_second": 0.744,
+      "step": 58100
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 3.412871341197029e-05,
+      "loss": 0.0986,
+      "step": 58150
+    },
+    {
+      "epoch": 6.36,
+      "learning_rate": 3.411506116207951e-05,
+      "loss": 0.1071,
+      "step": 58200
+    },
+    {
+      "epoch": 6.36,
+      "eval_cer": 0.04117834253236199,
+      "eval_loss": 0.2618873119354248,
+      "eval_runtime": 1248.1411,
+      "eval_samples_per_second": 6.019,
+      "eval_steps_per_second": 0.753,
+      "step": 58200
+    },
+    {
+      "epoch": 6.36,
+      "learning_rate": 3.4101408912188734e-05,
+      "loss": 0.1074,
+      "step": 58250
+    },
+    {
+      "epoch": 6.37,
+      "learning_rate": 3.408775666229795e-05,
+      "loss": 0.1063,
+      "step": 58300
+    },
+    {
+      "epoch": 6.37,
+      "eval_cer": 0.04217067420382493,
+      "eval_loss": 0.26021838188171387,
+      "eval_runtime": 1253.6554,
+      "eval_samples_per_second": 5.993,
+      "eval_steps_per_second": 0.75,
+      "step": 58300
+    },
+    {
+      "epoch": 6.37,
+      "learning_rate": 3.4074104412407166e-05,
+      "loss": 0.1066,
+      "step": 58350
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 3.406045216251638e-05,
+      "loss": 0.1085,
+      "step": 58400
+    },
+    {
+      "epoch": 6.38,
+      "eval_cer": 0.04305692904144182,
+      "eval_loss": 0.25534170866012573,
+      "eval_runtime": 1268.3522,
+      "eval_samples_per_second": 5.923,
+      "eval_steps_per_second": 0.741,
+      "step": 58400
+    },
+    {
+      "epoch": 6.38,
+      "learning_rate": 3.4046799912625605e-05,
+      "loss": 0.0962,
+      "step": 58450
+    },
+    {
+      "epoch": 6.39,
+      "learning_rate": 3.403314766273482e-05,
+      "loss": 0.111,
+      "step": 58500
+    },
+    {
+      "epoch": 6.39,
+      "eval_cer": 0.04086353386417375,
+      "eval_loss": 0.2595382034778595,
+      "eval_runtime": 1225.4425,
+      "eval_samples_per_second": 6.131,
+      "eval_steps_per_second": 0.767,
+      "step": 58500
+    },
+    {
+      "epoch": 6.39,
+      "learning_rate": 3.4019495412844036e-05,
+      "loss": 0.0946,
+      "step": 58550
+    },
+    {
+      "epoch": 6.4,
+      "learning_rate": 3.400611620795107e-05,
+      "loss": 0.1098,
+      "step": 58600
+    },
+    {
+      "epoch": 6.4,
+      "eval_cer": 0.04325539537573441,
+      "eval_loss": 0.25475308299064636,
+      "eval_runtime": 1267.9837,
+      "eval_samples_per_second": 5.925,
+      "eval_steps_per_second": 0.741,
+      "step": 58600
+    },
+    {
+      "epoch": 6.41,
+      "learning_rate": 3.399246395806029e-05,
+      "loss": 0.1111,
+      "step": 58650
+    },
+    {
+      "epoch": 6.41,
+      "learning_rate": 3.39788117081695e-05,
+      "loss": 0.1086,
+      "step": 58700
+    },
+    {
+      "epoch": 6.41,
+      "eval_cer": 0.0416573992013441,
+      "eval_loss": 0.2619175314903259,
+      "eval_runtime": 1229.6208,
+      "eval_samples_per_second": 6.11,
+      "eval_steps_per_second": 0.764,
+      "step": 58700
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 3.3965159458278725e-05,
+      "loss": 0.1006,
+      "step": 58750
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 3.395150720838795e-05,
+      "loss": 0.1019,
+      "step": 58800
+    },
+    {
+      "epoch": 6.42,
+      "eval_cer": 0.04108937486526531,
+      "eval_loss": 0.25999510288238525,
+      "eval_runtime": 1248.8363,
+      "eval_samples_per_second": 6.016,
+      "eval_steps_per_second": 0.753,
+      "step": 58800
+    },
+    {
+      "epoch": 6.43,
+      "learning_rate": 3.3937854958497164e-05,
+      "loss": 0.0999,
+      "step": 58850
+    },
+    {
+      "epoch": 6.43,
+      "learning_rate": 3.392420270860638e-05,
+      "loss": 0.1032,
+      "step": 58900
+    },
+    {
+      "epoch": 6.43,
+      "eval_cer": 0.041397339866753806,
+      "eval_loss": 0.2621907591819763,
+      "eval_runtime": 1264.5165,
+      "eval_samples_per_second": 5.941,
+      "eval_steps_per_second": 0.743,
+      "step": 58900
+    },
+    {
+      "epoch": 6.44,
+      "learning_rate": 3.3910550458715595e-05,
+      "loss": 0.1118,
+      "step": 58950
+    },
+    {
+      "epoch": 6.44,
+      "learning_rate": 3.389689820882482e-05,
+      "loss": 0.0951,
+      "step": 59000
+    },
+    {
+      "epoch": 6.44,
+      "eval_cer": 0.04185244370228681,
+      "eval_loss": 0.2569536566734314,
+      "eval_runtime": 1232.8244,
+      "eval_samples_per_second": 6.094,
+      "eval_steps_per_second": 0.762,
+      "step": 59000
+    },
+    {
+      "epoch": 6.45,
+      "learning_rate": 3.3883245958934034e-05,
+      "loss": 0.0999,
+      "step": 59050
+    },
+    {
+      "epoch": 6.45,
+      "learning_rate": 3.386959370904325e-05,
+      "loss": 0.0831,
+      "step": 59100
+    },
+    {
+      "epoch": 6.45,
+      "eval_cer": 0.04179427253533898,
+      "eval_loss": 0.2573639154434204,
+      "eval_runtime": 1143.636,
+      "eval_samples_per_second": 6.569,
+      "eval_steps_per_second": 0.822,
+      "step": 59100
+    },
+    {
+      "epoch": 6.46,
+      "learning_rate": 3.3855941459152465e-05,
+      "loss": 0.1098,
+      "step": 59150
+    },
+    {
+      "epoch": 6.47,
+      "learning_rate": 3.384228920926169e-05,
+      "loss": 0.0976,
+      "step": 59200
+    },
+    {
+      "epoch": 6.47,
+      "eval_cer": 0.04181480353543822,
+      "eval_loss": 0.257639616727829,
+      "eval_runtime": 1202.3685,
+      "eval_samples_per_second": 6.249,
+      "eval_steps_per_second": 0.782,
+      "step": 59200
+    },
+    {
+      "epoch": 6.47,
+      "learning_rate": 3.382863695937091e-05,
+      "loss": 0.0973,
+      "step": 59250
+    },
+    {
+      "epoch": 6.48,
+      "learning_rate": 3.3814984709480126e-05,
+      "loss": 0.0994,
+      "step": 59300
+    },
+    {
+      "epoch": 6.48,
+      "eval_cer": 0.04163002453454512,
+      "eval_loss": 0.2602895498275757,
+      "eval_runtime": 1241.769,
+      "eval_samples_per_second": 6.05,
+      "eval_steps_per_second": 0.757,
+      "step": 59300
+    },
+    {
+      "epoch": 6.48,
+      "learning_rate": 3.380133245958934e-05,
+      "loss": 0.112,
+      "step": 59350
+    },
+    {
+      "epoch": 6.49,
+      "learning_rate": 3.378768020969856e-05,
+      "loss": 0.1099,
+      "step": 59400
+    },
+    {
+      "epoch": 6.49,
+      "eval_cer": 0.042813978873600896,
+      "eval_loss": 0.2561176121234894,
+      "eval_runtime": 1262.5271,
+      "eval_samples_per_second": 5.951,
+      "eval_steps_per_second": 0.745,
+      "step": 59400
+    },
+    {
+      "epoch": 6.49,
+      "learning_rate": 3.377402795980778e-05,
+      "loss": 0.1025,
+      "step": 59450
+    },
+    {
+      "epoch": 6.5,
+      "learning_rate": 3.3760375709916996e-05,
+      "loss": 0.1198,
+      "step": 59500
+    },
+    {
+      "epoch": 6.5,
+      "eval_cer": 0.04231096937116968,
+      "eval_loss": 0.254476934671402,
+      "eval_runtime": 1262.5413,
+      "eval_samples_per_second": 5.951,
+      "eval_steps_per_second": 0.745,
+      "step": 59500
+    },
+    {
+      "epoch": 6.5,
+      "learning_rate": 3.374672346002621e-05,
+      "loss": 0.0967,
+      "step": 59550
+    },
+    {
+      "epoch": 6.51,
+      "learning_rate": 3.373307121013543e-05,
+      "loss": 0.0976,
+      "step": 59600
+    },
+    {
+      "epoch": 6.51,
+      "eval_cer": 0.041267310199458666,
+      "eval_loss": 0.2605424225330353,
+      "eval_runtime": 1248.7159,
+      "eval_samples_per_second": 6.017,
+      "eval_steps_per_second": 0.753,
+      "step": 59600
+    },
+    {
+      "epoch": 6.51,
+      "learning_rate": 3.371941896024465e-05,
+      "loss": 0.0928,
+      "step": 59650
+    },
+    {
+      "epoch": 6.52,
+      "learning_rate": 3.370576671035387e-05,
+      "loss": 0.0977,
+      "step": 59700
+    },
+    {
+      "epoch": 6.52,
+      "eval_cer": 0.04313905304183876,
+      "eval_loss": 0.2529686689376831,
+      "eval_runtime": 1261.516,
+      "eval_samples_per_second": 5.956,
+      "eval_steps_per_second": 0.745,
+      "step": 59700
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 3.369211446046309e-05,
+      "loss": 0.104,
+      "step": 59750
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 3.3678462210572305e-05,
+      "loss": 0.1016,
+      "step": 59800
+    },
+    {
+      "epoch": 6.53,
+      "eval_cer": 0.0412878411995579,
+      "eval_loss": 0.25390204787254333,
+      "eval_runtime": 1246.3223,
+      "eval_samples_per_second": 6.028,
+      "eval_steps_per_second": 0.754,
+      "step": 59800
+    },
+    {
+      "epoch": 6.54,
+      "learning_rate": 3.366480996068152e-05,
+      "loss": 0.098,
+      "step": 59850
+    },
+    {
+      "epoch": 6.54,
+      "learning_rate": 3.3651157710790736e-05,
+      "loss": 0.1009,
+      "step": 59900
+    },
+    {
+      "epoch": 6.54,
+      "eval_cer": 0.0418661310356863,
+      "eval_loss": 0.25637781620025635,
+      "eval_runtime": 1245.1762,
+      "eval_samples_per_second": 6.034,
+      "eval_steps_per_second": 0.755,
+      "step": 59900
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 3.363750546089996e-05,
+      "loss": 0.1131,
+      "step": 59950
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 3.3623853211009175e-05,
+      "loss": 0.1037,
+      "step": 60000
+    },
+    {
+      "epoch": 6.55,
+      "eval_cer": 0.04255391953901061,
+      "eval_loss": 0.25451895594596863,
+      "eval_runtime": 1255.9082,
+      "eval_samples_per_second": 5.982,
+      "eval_steps_per_second": 0.748,
+      "step": 60000
+    },
+    {
+      "epoch": 6.56,
+      "learning_rate": 3.361020096111839e-05,
+      "loss": 0.1005,
+      "step": 60050
+    },
+    {
+      "epoch": 6.56,
+      "learning_rate": 3.359654871122761e-05,
+      "loss": 0.0934,
+      "step": 60100
+    },
+    {
+      "epoch": 6.56,
+      "eval_cer": 0.041825069035487834,
+      "eval_loss": 0.25655898451805115,
+      "eval_runtime": 1247.2834,
+      "eval_samples_per_second": 6.023,
+      "eval_steps_per_second": 0.754,
+      "step": 60100
+    },
+    {
+      "epoch": 6.57,
+      "learning_rate": 3.358289646133683e-05,
+      "loss": 0.1122,
+      "step": 60150
+    },
+    {
+      "epoch": 6.57,
+      "learning_rate": 3.356924421144605e-05,
+      "loss": 0.1072,
+      "step": 60200
+    },
+    {
+      "epoch": 6.57,
+      "eval_cer": 0.044210086880348755,
+      "eval_loss": 0.2543743848800659,
+      "eval_runtime": 1267.4241,
+      "eval_samples_per_second": 5.928,
+      "eval_steps_per_second": 0.742,
+      "step": 60200
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 3.355559196155527e-05,
+      "loss": 0.0921,
+      "step": 60250
+    },
+    {
+      "epoch": 6.59,
+      "learning_rate": 3.354193971166448e-05,
+      "loss": 0.1002,
+      "step": 60300
+    },
+    {
+      "epoch": 6.59,
+      "eval_cer": 0.041893505702485276,
+      "eval_loss": 0.2538849711418152,
+      "eval_runtime": 1265.4609,
+      "eval_samples_per_second": 5.937,
+      "eval_steps_per_second": 0.743,
+      "step": 60300
+    },
+    {
+      "epoch": 6.59,
+      "learning_rate": 3.35282874617737e-05,
+      "loss": 0.1134,
+      "step": 60350
+    },
+    {
+      "epoch": 6.6,
+      "learning_rate": 3.351463521188292e-05,
+      "loss": 0.0978,
+      "step": 60400
+    },
+    {
+      "epoch": 6.6,
+      "eval_cer": 0.041623180867845375,
+      "eval_loss": 0.2523655593395233,
+      "eval_runtime": 1254.741,
+      "eval_samples_per_second": 5.988,
+      "eval_steps_per_second": 0.749,
+      "step": 60400
+    },
+    {
+      "epoch": 6.6,
+      "learning_rate": 3.350098296199214e-05,
+      "loss": 0.1185,
+      "step": 60450
+    },
+    {
+      "epoch": 6.61,
+      "learning_rate": 3.3487330712101353e-05,
+      "loss": 0.0937,
+      "step": 60500
+    },
+    {
+      "epoch": 6.61,
+      "eval_cer": 0.04146235470040138,
+      "eval_loss": 0.25337615609169006,
+      "eval_runtime": 1271.1459,
+      "eval_samples_per_second": 5.91,
+      "eval_steps_per_second": 0.739,
+      "step": 60500
+    },
+    {
+      "epoch": 6.61,
+      "learning_rate": 3.347367846221057e-05,
+      "loss": 0.1064,
+      "step": 60550
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 3.346002621231979e-05,
+      "loss": 0.0945,
+      "step": 60600
+    },
+    {
+      "epoch": 6.62,
+      "eval_cer": 0.041797694368688854,
+      "eval_loss": 0.25419652462005615,
+      "eval_runtime": 1257.8593,
+      "eval_samples_per_second": 5.973,
+      "eval_steps_per_second": 0.747,
+      "step": 60600
+    },
+    {
+      "epoch": 6.62,
+      "learning_rate": 3.3446373962429014e-05,
+      "loss": 0.1099,
+      "step": 60650
+    },
+    {
+      "epoch": 6.63,
+      "learning_rate": 3.343272171253823e-05,
+      "loss": 0.1081,
+      "step": 60700
+    },
+    {
+      "epoch": 6.63,
+      "eval_cer": 0.04154790053414818,
+      "eval_loss": 0.2513173520565033,
+      "eval_runtime": 1254.1003,
+      "eval_samples_per_second": 5.991,
+      "eval_steps_per_second": 0.75,
+      "step": 60700
+    },
+    {
+      "epoch": 6.63,
+      "learning_rate": 3.3419069462647446e-05,
+      "loss": 0.1266,
+      "step": 60750
+    },
+    {
+      "epoch": 6.64,
+      "learning_rate": 3.340541721275666e-05,
+      "loss": 0.1107,
+      "step": 60800
+    },
+    {
+      "epoch": 6.64,
+      "eval_cer": 0.04119545169911135,
+      "eval_loss": 0.257914274930954,
+      "eval_runtime": 1252.8755,
+      "eval_samples_per_second": 5.997,
+      "eval_steps_per_second": 0.75,
+      "step": 60800
+    },
+    {
+      "epoch": 6.65,
+      "learning_rate": 3.3391764962865884e-05,
+      "loss": 0.115,
+      "step": 60850
+    },
+    {
+      "epoch": 6.65,
+      "learning_rate": 3.33781127129751e-05,
+      "loss": 0.0888,
+      "step": 60900
+    },
+    {
+      "epoch": 6.65,
+      "eval_cer": 0.042701058373055115,
+      "eval_loss": 0.2549211382865906,
+      "eval_runtime": 1286.7653,
+      "eval_samples_per_second": 5.839,
+      "eval_steps_per_second": 0.731,
+      "step": 60900
+    },
+    {
+      "epoch": 6.66,
+      "learning_rate": 3.3364460463084316e-05,
+      "loss": 0.1262,
+      "step": 60950
+    },
+    {
+      "epoch": 6.66,
+      "learning_rate": 3.335080821319353e-05,
+      "loss": 0.0977,
+      "step": 61000
+    },
+    {
+      "epoch": 6.66,
+      "eval_cer": 0.04167793020144333,
+      "eval_loss": 0.25344541668891907,
+      "eval_runtime": 1267.7275,
+      "eval_samples_per_second": 5.926,
+      "eval_steps_per_second": 0.741,
+      "step": 61000
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 3.3337155963302755e-05,
+      "loss": 0.111,
+      "step": 61050
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 3.332350371341198e-05,
+      "loss": 0.1139,
+      "step": 61100
+    },
+    {
+      "epoch": 6.67,
+      "eval_cer": 0.040428961028739975,
+      "eval_loss": 0.24946226179599762,
+      "eval_runtime": 1235.1051,
+      "eval_samples_per_second": 6.083,
+      "eval_steps_per_second": 0.761,
+      "step": 61100
+    },
+    {
+      "epoch": 6.68,
+      "learning_rate": 3.330985146352119e-05,
+      "loss": 0.1222,
+      "step": 61150
+    },
+    {
+      "epoch": 6.68,
+      "learning_rate": 3.329619921363041e-05,
+      "loss": 0.1013,
+      "step": 61200
+    },
+    {
+      "epoch": 6.68,
+      "eval_cer": 0.04048713219568781,
+      "eval_loss": 0.25472667813301086,
+      "eval_runtime": 1244.3934,
+      "eval_samples_per_second": 6.037,
+      "eval_steps_per_second": 0.755,
+      "step": 61200
+    },
+    {
+      "epoch": 6.69,
+      "learning_rate": 3.3282546963739625e-05,
+      "loss": 0.1026,
+      "step": 61250
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 3.326889471384884e-05,
+      "loss": 0.0975,
+      "step": 61300
+    },
+    {
+      "epoch": 6.7,
+      "eval_cer": 0.04055556886268525,
+      "eval_loss": 0.25184065103530884,
+      "eval_runtime": 1215.7918,
+      "eval_samples_per_second": 6.18,
+      "eval_steps_per_second": 0.773,
+      "step": 61300
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 3.325524246395806e-05,
+      "loss": 0.1175,
+      "step": 61350
+    },
+    {
+      "epoch": 6.71,
+      "learning_rate": 3.324159021406728e-05,
+      "loss": 0.0887,
+      "step": 61400
+    },
+    {
+      "epoch": 6.71,
+      "eval_cer": 0.04122282636591033,
+      "eval_loss": 0.252297580242157,
+      "eval_runtime": 1438.8019,
+      "eval_samples_per_second": 5.222,
+      "eval_steps_per_second": 0.653,
+      "step": 61400
+    },
+    {
+      "epoch": 6.71,
+      "learning_rate": 3.3227937964176495e-05,
+      "loss": 0.1122,
+      "step": 61450
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 3.321428571428572e-05,
+      "loss": 0.1085,
+      "step": 61500
+    },
+    {
+      "epoch": 6.72,
+      "eval_cer": 0.039946482526408,
+      "eval_loss": 0.24887217581272125,
+      "eval_runtime": 1450.663,
+      "eval_samples_per_second": 5.179,
+      "eval_steps_per_second": 0.648,
+      "step": 61500
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 3.320063346439493e-05,
+      "loss": 0.106,
+      "step": 61550
+    },
+    {
+      "epoch": 6.73,
+      "learning_rate": 3.3186981214504156e-05,
+      "loss": 0.0979,
+      "step": 61600
+    },
+    {
+      "epoch": 6.73,
+      "eval_cer": 0.04282082254030064,
+      "eval_loss": 0.24803991615772247,
+      "eval_runtime": 1237.0461,
+      "eval_samples_per_second": 6.073,
+      "eval_steps_per_second": 0.76,
+      "step": 61600
+    },
+    {
+      "epoch": 6.73,
+      "learning_rate": 3.317332896461337e-05,
+      "loss": 0.1033,
+      "step": 61650
+    },
+    {
+      "epoch": 6.74,
+      "learning_rate": 3.315967671472259e-05,
+      "loss": 0.1077,
+      "step": 61700
+    },
+    {
+      "epoch": 6.74,
+      "eval_cer": 0.04303297620799272,
+      "eval_loss": 0.2528606653213501,
+      "eval_runtime": 1154.6782,
+      "eval_samples_per_second": 6.507,
+      "eval_steps_per_second": 0.814,
+      "step": 61700
+    },
+    {
+      "epoch": 6.74,
+      "learning_rate": 3.31460244648318e-05,
+      "loss": 0.0988,
+      "step": 61750
+    },
+    {
+      "epoch": 6.75,
+      "learning_rate": 3.3132372214941026e-05,
+      "loss": 0.1144,
+      "step": 61800
+    },
+    {
+      "epoch": 6.75,
+      "eval_cer": 0.04078140986377681,
+      "eval_loss": 0.25002339482307434,
+      "eval_runtime": 1141.3931,
+      "eval_samples_per_second": 6.582,
+      "eval_steps_per_second": 0.824,
+      "step": 61800
+    },
+    {
+      "epoch": 6.76,
+      "learning_rate": 3.311871996505024e-05,
+      "loss": 0.1088,
+      "step": 61850
+    },
+    {
+      "epoch": 6.76,
+      "learning_rate": 3.310506771515946e-05,
+      "loss": 0.1093,
+      "step": 61900
+    },
+    {
+      "epoch": 6.76,
+      "eval_cer": 0.04480548588322651,
+      "eval_loss": 0.255926251411438,
+      "eval_runtime": 1166.6291,
+      "eval_samples_per_second": 6.44,
+      "eval_steps_per_second": 0.806,
+      "step": 61900
+    },
+    {
+      "epoch": 6.77,
+      "learning_rate": 3.309141546526868e-05,
+      "loss": 0.096,
+      "step": 61950
+    },
+    {
+      "epoch": 6.77,
+      "learning_rate": 3.3077763215377896e-05,
+      "loss": 0.0998,
+      "step": 62000
+    },
+    {
+      "epoch": 6.77,
+      "eval_cer": 0.042680527372955884,
+      "eval_loss": 0.2501034438610077,
+      "eval_runtime": 1157.0123,
+      "eval_samples_per_second": 6.493,
+      "eval_steps_per_second": 0.812,
+      "step": 62000
+    },
+    {
+      "epoch": 6.78,
+      "learning_rate": 3.306411096548712e-05,
+      "loss": 0.1018,
+      "step": 62050
+    },
+    {
+      "epoch": 6.78,
+      "learning_rate": 3.3050458715596334e-05,
+      "loss": 0.1106,
+      "step": 62100
+    },
+    {
+      "epoch": 6.78,
+      "eval_cer": 0.04183533453553745,
+      "eval_loss": 0.2515909969806671,
+      "eval_runtime": 1422.3584,
+      "eval_samples_per_second": 5.282,
+      "eval_steps_per_second": 0.661,
+      "step": 62100
+    },
+    {
+      "epoch": 6.79,
+      "learning_rate": 3.303680646570555e-05,
+      "loss": 0.1262,
+      "step": 62150
+    },
+    {
+      "epoch": 6.79,
+      "learning_rate": 3.3023154215814766e-05,
+      "loss": 0.094,
+      "step": 62200
+    },
+    {
+      "epoch": 6.79,
+      "eval_cer": 0.042208314370673516,
+      "eval_loss": 0.25220343470573425,
+      "eval_runtime": 1461.0595,
+      "eval_samples_per_second": 5.142,
+      "eval_steps_per_second": 0.643,
+      "step": 62200
+    },
+    {
+      "epoch": 6.8,
+      "learning_rate": 3.300950196592398e-05,
+      "loss": 0.1151,
+      "step": 62250
+    },
+    {
+      "epoch": 6.8,
+      "learning_rate": 3.2995849716033204e-05,
+      "loss": 0.1205,
+      "step": 62300
+    },
+    {
+      "epoch": 6.8,
+      "eval_cer": 0.040049137526904165,
+      "eval_loss": 0.2456672489643097,
+      "eval_runtime": 1314.9991,
+      "eval_samples_per_second": 5.713,
+      "eval_steps_per_second": 0.715,
+      "step": 62300
+    },
+    {
+      "epoch": 6.81,
+      "learning_rate": 3.298219746614242e-05,
+      "loss": 0.1051,
+      "step": 62350
+    },
+    {
+      "epoch": 6.82,
+      "learning_rate": 3.296854521625164e-05,
+      "loss": 0.1036,
+      "step": 62400
+    },
+    {
+      "epoch": 6.82,
+      "eval_cer": 0.042499170205412655,
+      "eval_loss": 0.2505302429199219,
+      "eval_runtime": 1428.5485,
+      "eval_samples_per_second": 5.259,
+      "eval_steps_per_second": 0.658,
+      "step": 62400
+    },
+    {
+      "epoch": 6.82,
+      "learning_rate": 3.295489296636086e-05,
+      "loss": 0.0972,
+      "step": 62450
+    },
+    {
+      "epoch": 6.83,
+      "learning_rate": 3.2941240716470074e-05,
+      "loss": 0.1272,
+      "step": 62500
+    },
+    {
+      "epoch": 6.83,
+      "eval_cer": 0.042047488203229524,
+      "eval_loss": 0.2512877285480499,
+      "eval_runtime": 1355.077,
+      "eval_samples_per_second": 5.544,
+      "eval_steps_per_second": 0.694,
+      "step": 62500
+    },
+    {
+      "epoch": 6.83,
+      "learning_rate": 3.29275884665793e-05,
+      "loss": 0.0955,
+      "step": 62550
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 3.291393621668851e-05,
+      "loss": 0.1016,
+      "step": 62600
+    },
+    {
+      "epoch": 6.84,
+      "eval_cer": 0.040839581030724645,
+      "eval_loss": 0.25099560618400574,
+      "eval_runtime": 1146.6199,
+      "eval_samples_per_second": 6.552,
+      "eval_steps_per_second": 0.82,
+      "step": 62600
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 3.290028396679773e-05,
+      "loss": 0.106,
+      "step": 62650
+    },
+    {
+      "epoch": 6.85,
+      "learning_rate": 3.2886631716906944e-05,
+      "loss": 0.099,
+      "step": 62700
+    },
+    {
+      "epoch": 6.85,
+      "eval_cer": 0.043180115042037226,
+      "eval_loss": 0.24449311196804047,
+      "eval_runtime": 1175.3376,
+      "eval_samples_per_second": 6.392,
+      "eval_steps_per_second": 0.8,
+      "step": 62700
+    },
+    {
+      "epoch": 6.85,
+      "learning_rate": 3.287297946701617e-05,
+      "loss": 0.1034,
+      "step": 62750
+    },
+    {
+      "epoch": 6.86,
+      "learning_rate": 3.2859600262123195e-05,
+      "loss": 0.1019,
+      "step": 62800
+    },
+    {
+      "epoch": 6.86,
+      "eval_cer": 0.041034625531667354,
+      "eval_loss": 0.2486434280872345,
+      "eval_runtime": 1303.8827,
+      "eval_samples_per_second": 5.762,
+      "eval_steps_per_second": 0.721,
+      "step": 62800
+    },
+    {
+      "epoch": 6.86,
+      "learning_rate": 3.284594801223242e-05,
+      "loss": 0.1046,
+      "step": 62850
+    },
+    {
+      "epoch": 6.87,
+      "learning_rate": 3.283229576234163e-05,
+      "loss": 0.1003,
+      "step": 62900
+    },
+    {
+      "epoch": 6.87,
+      "eval_cer": 0.040189432694248926,
+      "eval_loss": 0.24836403131484985,
+      "eval_runtime": 1369.5448,
+      "eval_samples_per_second": 5.486,
+      "eval_steps_per_second": 0.686,
+      "step": 62900
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 3.2818643512450856e-05,
+      "loss": 0.1131,
+      "step": 62950
+    },
+    {
+      "epoch": 6.88,
+      "learning_rate": 3.280499126256007e-05,
+      "loss": 0.1166,
+      "step": 63000
+    },
+    {
+      "epoch": 6.88,
+      "eval_cer": 0.04154105686744844,
+      "eval_loss": 0.24888207018375397,
+      "eval_runtime": 1440.5402,
+      "eval_samples_per_second": 5.215,
+      "eval_steps_per_second": 0.653,
+      "step": 63000
+    },
+    {
+      "epoch": 6.89,
+      "learning_rate": 3.2791339012669294e-05,
+      "loss": 0.1196,
+      "step": 63050
+    },
+    {
+      "epoch": 6.89,
+      "learning_rate": 3.277768676277851e-05,
+      "loss": 0.1125,
+      "step": 63100
+    },
+    {
+      "epoch": 6.89,
+      "eval_cer": 0.04208170653672825,
+      "eval_loss": 0.24433410167694092,
+      "eval_runtime": 1218.0166,
+      "eval_samples_per_second": 6.168,
+      "eval_steps_per_second": 0.772,
+      "step": 63100
+    },
+    {
+      "epoch": 6.9,
+      "learning_rate": 3.2764034512887726e-05,
+      "loss": 0.1071,
+      "step": 63150
+    },
+    {
+      "epoch": 6.9,
+      "learning_rate": 3.275038226299694e-05,
+      "loss": 0.1019,
+      "step": 63200
+    },
+    {
+      "epoch": 6.9,
+      "eval_cer": 0.04413138471330169,
+      "eval_loss": 0.2512649595737457,
+      "eval_runtime": 1158.8136,
+      "eval_samples_per_second": 6.483,
+      "eval_steps_per_second": 0.811,
+      "step": 63200
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 3.273673001310616e-05,
+      "loss": 0.1049,
+      "step": 63250
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 3.272307776321538e-05,
+      "loss": 0.1092,
+      "step": 63300
+    },
+    {
+      "epoch": 6.91,
+      "eval_cer": 0.04346070537672674,
+      "eval_loss": 0.25279322266578674,
+      "eval_runtime": 1145.6556,
+      "eval_samples_per_second": 6.558,
+      "eval_steps_per_second": 0.82,
+      "step": 63300
+    },
+    {
+      "epoch": 6.92,
+      "learning_rate": 3.2709425513324596e-05,
+      "loss": 0.1084,
+      "step": 63350
+    },
+    {
+      "epoch": 6.92,
+      "learning_rate": 3.269577326343382e-05,
+      "loss": 0.1058,
+      "step": 63400
+    },
+    {
+      "epoch": 6.92,
+      "eval_cer": 0.04239309337156662,
+      "eval_loss": 0.24921201169490814,
+      "eval_runtime": 1147.0743,
+      "eval_samples_per_second": 6.55,
+      "eval_steps_per_second": 0.819,
+      "step": 63400
+    },
+    {
+      "epoch": 6.93,
+      "learning_rate": 3.2682121013543034e-05,
+      "loss": 0.1045,
+      "step": 63450
+    },
+    {
+      "epoch": 6.94,
+      "learning_rate": 3.266846876365225e-05,
+      "loss": 0.1019,
+      "step": 63500
+    },
+    {
+      "epoch": 6.94,
+      "eval_cer": 0.04116123336561263,
+      "eval_loss": 0.24884535372257233,
+      "eval_runtime": 1153.8324,
+      "eval_samples_per_second": 6.511,
+      "eval_steps_per_second": 0.815,
+      "step": 63500
+    },
+    {
+      "epoch": 6.94,
+      "learning_rate": 3.265481651376147e-05,
+      "loss": 0.1261,
+      "step": 63550
+    },
+    {
+      "epoch": 6.95,
+      "learning_rate": 3.264116426387069e-05,
+      "loss": 0.0912,
+      "step": 63600
+    },
+    {
+      "epoch": 6.95,
+      "eval_cer": 0.04406636987965412,
+      "eval_loss": 0.24965716898441315,
+      "eval_runtime": 1168.6161,
+      "eval_samples_per_second": 6.429,
+      "eval_steps_per_second": 0.804,
+      "step": 63600
+    },
+    {
+      "epoch": 6.95,
+      "learning_rate": 3.2627512013979904e-05,
+      "loss": 0.1142,
+      "step": 63650
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 3.261385976408912e-05,
+      "loss": 0.1138,
+      "step": 63700
+    },
+    {
+      "epoch": 6.96,
+      "eval_cer": 0.04194483320273336,
+      "eval_loss": 0.24717257916927338,
+      "eval_runtime": 1271.4905,
+      "eval_samples_per_second": 5.909,
+      "eval_steps_per_second": 0.739,
+      "step": 63700
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 3.2600207514198336e-05,
+      "loss": 0.1051,
+      "step": 63750
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 3.258655526430756e-05,
+      "loss": 0.1056,
+      "step": 63800
+    },
+    {
+      "epoch": 6.97,
+      "eval_cer": 0.040675333029930774,
+      "eval_loss": 0.24858665466308594,
+      "eval_runtime": 1147.8949,
+      "eval_samples_per_second": 6.545,
+      "eval_steps_per_second": 0.819,
+      "step": 63800
+    },
+    {
+      "epoch": 6.97,
+      "learning_rate": 3.257290301441678e-05,
+      "loss": 0.1035,
+      "step": 63850
+    },
+    {
+      "epoch": 6.98,
+      "learning_rate": 3.2559250764526e-05,
+      "loss": 0.1128,
+      "step": 63900
+    },
+    {
+      "epoch": 6.98,
+      "eval_cer": 0.04135285603320547,
+      "eval_loss": 0.2538721561431885,
+      "eval_runtime": 1172.9039,
+      "eval_samples_per_second": 6.405,
+      "eval_steps_per_second": 0.801,
+      "step": 63900
+    },
+    {
+      "epoch": 6.98,
+      "learning_rate": 3.254559851463521e-05,
+      "loss": 0.1054,
+      "step": 63950
+    },
+    {
+      "epoch": 6.99,
+      "learning_rate": 3.2531946264744436e-05,
+      "loss": 0.1068,
+      "step": 64000
+    },
+    {
+      "epoch": 6.99,
+      "eval_cer": 0.043919231045609616,
+      "eval_loss": 0.24928919970989227,
+      "eval_runtime": 1456.3353,
+      "eval_samples_per_second": 5.159,
+      "eval_steps_per_second": 0.645,
+      "step": 64000
+    },
+    {
+      "epoch": 7.0,
+      "learning_rate": 3.251829401485365e-05,
+      "loss": 0.1057,
+      "step": 64050
+    },
+    {
+      "epoch": 7.0,
+      "learning_rate": 3.250464176496287e-05,
+      "loss": 0.0955,
+      "step": 64100
+    },
+    {
+      "epoch": 7.0,
+      "eval_cer": 0.042163830537125184,
+      "eval_loss": 0.25061729550361633,
+      "eval_runtime": 1154.3718,
+      "eval_samples_per_second": 6.508,
+      "eval_steps_per_second": 0.814,
+      "step": 64100
+    },
+    {
+      "epoch": 7.01,
+      "learning_rate": 3.249098951507208e-05,
+      "loss": 0.0608,
+      "step": 64150
+    },
+    {
+      "epoch": 7.01,
+      "learning_rate": 3.24773372651813e-05,
+      "loss": 0.0623,
+      "step": 64200
+    },
+    {
+      "epoch": 7.01,
+      "eval_cer": 0.04060347452958346,
+      "eval_loss": 0.26202353835105896,
+      "eval_runtime": 1185.2912,
+      "eval_samples_per_second": 6.339,
+      "eval_steps_per_second": 0.793,
+      "step": 64200
+    },
+    {
+      "epoch": 7.02,
+      "learning_rate": 3.246368501529052e-05,
+      "loss": 0.0705,
+      "step": 64250
+    },
+    {
+      "epoch": 7.02,
+      "learning_rate": 3.245003276539974e-05,
+      "loss": 0.0741,
+      "step": 64300
+    },
+    {
+      "epoch": 7.02,
+      "eval_cer": 0.0432074897088362,
+      "eval_loss": 0.2552080750465393,
+      "eval_runtime": 1154.0385,
+      "eval_samples_per_second": 6.51,
+      "eval_steps_per_second": 0.815,
+      "step": 64300
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 3.243638051550896e-05,
+      "loss": 0.0684,
+      "step": 64350
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 3.2422728265618176e-05,
+      "loss": 0.0807,
+      "step": 64400
+    },
+    {
+      "epoch": 7.03,
+      "eval_cer": 0.04506212338446693,
+      "eval_loss": 0.2651134431362152,
+      "eval_runtime": 1150.1682,
+      "eval_samples_per_second": 6.532,
+      "eval_steps_per_second": 0.817,
+      "step": 64400
+    },
+    {
+      "epoch": 7.04,
+      "learning_rate": 3.240907601572739e-05,
+      "loss": 0.0703,
+      "step": 64450
+    },
+    {
+      "epoch": 7.04,
+      "learning_rate": 3.2395423765836614e-05,
+      "loss": 0.0725,
+      "step": 64500
+    },
+    {
+      "epoch": 7.04,
+      "eval_cer": 0.041239935532659686,
+      "eval_loss": 0.25625482201576233,
+      "eval_runtime": 1149.5117,
+      "eval_samples_per_second": 6.536,
+      "eval_steps_per_second": 0.818,
+      "step": 64500
+    },
+    {
+      "epoch": 7.05,
+      "learning_rate": 3.238177151594583e-05,
+      "loss": 0.0605,
+      "step": 64550
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 3.2368119266055046e-05,
+      "loss": 0.076,
+      "step": 64600
+    },
+    {
+      "epoch": 7.06,
+      "eval_cer": 0.03970695419191694,
+      "eval_loss": 0.2562285363674164,
+      "eval_runtime": 1137.9033,
+      "eval_samples_per_second": 6.602,
+      "eval_steps_per_second": 0.826,
+      "step": 64600
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 3.235446701616426e-05,
+      "loss": 0.067,
+      "step": 64650
+    },
+    {
+      "epoch": 7.07,
+      "learning_rate": 3.2340814766273484e-05,
+      "loss": 0.0735,
+      "step": 64700
+    },
+    {
+      "epoch": 7.07,
+      "eval_cer": 0.040117574193901606,
+      "eval_loss": 0.25367167592048645,
+      "eval_runtime": 1145.7217,
+      "eval_samples_per_second": 6.557,
+      "eval_steps_per_second": 0.82,
+      "step": 64700
+    },
+    {
+      "epoch": 7.07,
+      "learning_rate": 3.23271625163827e-05,
+      "loss": 0.0742,
+      "step": 64750
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 3.231351026649192e-05,
+      "loss": 0.0679,
+      "step": 64800
+    },
+    {
+      "epoch": 7.08,
+      "eval_cer": 0.03999438819330621,
+      "eval_loss": 0.25834035873413086,
+      "eval_runtime": 1164.3956,
+      "eval_samples_per_second": 6.452,
+      "eval_steps_per_second": 0.807,
+      "step": 64800
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 3.229985801660114e-05,
+      "loss": 0.0664,
+      "step": 64850
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 3.2286205766710354e-05,
+      "loss": 0.0609,
+      "step": 64900
+    },
+    {
+      "epoch": 7.09,
+      "eval_cer": 0.0412022953658111,
+      "eval_loss": 0.2590734362602234,
+      "eval_runtime": 1161.4118,
+      "eval_samples_per_second": 6.469,
+      "eval_steps_per_second": 0.809,
+      "step": 64900
+    },
+    {
+      "epoch": 7.09,
+      "learning_rate": 3.227255351681958e-05,
+      "loss": 0.0931,
+      "step": 64950
+    },
+    {
+      "epoch": 7.1,
+      "learning_rate": 3.225890126692879e-05,
+      "loss": 0.0559,
+      "step": 65000
+    },
+    {
+      "epoch": 7.1,
+      "eval_cer": 0.04062058369633282,
+      "eval_loss": 0.25673454999923706,
+      "eval_runtime": 1213.4156,
+      "eval_samples_per_second": 6.192,
+      "eval_steps_per_second": 0.775,
+      "step": 65000
+    },
+    {
+      "epoch": 7.1,
+      "learning_rate": 3.224524901703801e-05,
+      "loss": 0.0676,
+      "step": 65050
+    },
+    {
+      "epoch": 7.11,
+      "learning_rate": 3.2231596767147224e-05,
+      "loss": 0.0704,
+      "step": 65100
+    },
+    {
+      "epoch": 7.11,
+      "eval_cer": 0.04068217669663052,
+      "eval_loss": 0.25640252232551575,
+      "eval_runtime": 1150.6735,
+      "eval_samples_per_second": 6.529,
+      "eval_steps_per_second": 0.817,
+      "step": 65100
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 3.221794451725644e-05,
+      "loss": 0.0661,
+      "step": 65150
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 3.220429226736566e-05,
+      "loss": 0.0771,
+      "step": 65200
+    },
+    {
+      "epoch": 7.12,
+      "eval_cer": 0.04037078986179215,
+      "eval_loss": 0.25912144780158997,
+      "eval_runtime": 1469.693,
+      "eval_samples_per_second": 5.112,
+      "eval_steps_per_second": 0.64,
+      "step": 65200
+    },
+    {
+      "epoch": 7.13,
+      "learning_rate": 3.2190640017474885e-05,
+      "loss": 0.0632,
+      "step": 65250
+    },
+    {
+      "epoch": 7.13,
+      "learning_rate": 3.21769877675841e-05,
+      "loss": 0.0656,
+      "step": 65300
+    },
+    {
+      "epoch": 7.13,
+      "eval_cer": 0.03920394468948573,
+      "eval_loss": 0.2582501471042633,
+      "eval_runtime": 1389.4184,
+      "eval_samples_per_second": 5.407,
+      "eval_steps_per_second": 0.677,
+      "step": 65300
+    }
+  ],
+  "max_steps": 183120,
+  "num_train_epochs": 20,
+  "total_flos": 1.5636429250868596e+21,
+  "trial_name": null,
+  "trial_params": null
+}