diff --git "a/trainer_state.json" "b/trainer_state.json"
new file mode 100644--- /dev/null
+++ "b/trainer_state.json"
@@ -0,0 +1,5905 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 99.99898063200816,
+  "global_step": 73500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.14,
+      "learning_rate": 5.940000000000001e-06,
+      "loss": 12.018,
+      "step": 100
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 1.1940000000000001e-05,
+      "loss": 5.1174,
+      "step": 200
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 1.794e-05,
+      "loss": 3.6999,
+      "step": 300
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.394e-05,
+      "loss": 3.2944,
+      "step": 400
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 2.994e-05,
+      "loss": 3.1215,
+      "step": 500
+    },
+    {
+      "epoch": 0.68,
+      "eval_cer": 1.0,
+      "eval_loss": 3.1187968254089355,
+      "eval_runtime": 183.7383,
+      "eval_samples_per_second": 31.485,
+      "eval_steps_per_second": 3.94,
+      "eval_wer": 1.0,
+      "step": 500
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 2.995931506849315e-05,
+      "loss": 3.0845,
+      "step": 600
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.9918219178082192e-05,
+      "loss": 3.0501,
+      "step": 700
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 2.9877123287671232e-05,
+      "loss": 3.025,
+      "step": 800
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 2.9836027397260276e-05,
+      "loss": 2.9543,
+      "step": 900
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 2.9794931506849316e-05,
+      "loss": 2.8562,
+      "step": 1000
+    },
+    {
+      "epoch": 1.36,
+      "eval_cer": 0.9688746571075726,
+      "eval_loss": 2.572427749633789,
+      "eval_runtime": 181.698,
+      "eval_samples_per_second": 31.839,
+      "eval_steps_per_second": 3.985,
+      "eval_wer": 1.000195798718917,
+      "step": 1000
+    },
+    {
+      "epoch": 1.5,
+      "learning_rate": 2.975383561643836e-05,
+      "loss": 2.4048,
+      "step": 1100
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 2.9712739726027397e-05,
+      "loss": 1.8048,
+      "step": 1200
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 2.9671643835616437e-05,
+      "loss": 1.4934,
+      "step": 1300
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 2.963054794520548e-05,
+      "loss": 1.3479,
+      "step": 1400
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 2.958945205479452e-05,
+      "loss": 1.2709,
+      "step": 1500
+    },
+    {
+      "epoch": 2.04,
+      "eval_cer": 0.14714593283133517,
+      "eval_loss": 0.627841591835022,
+      "eval_runtime": 182.0469,
+      "eval_samples_per_second": 31.778,
+      "eval_steps_per_second": 3.977,
+      "eval_wer": 0.6477860759139604,
+      "step": 1500
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 2.954835616438356e-05,
+      "loss": 1.2006,
+      "step": 1600
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 2.9507260273972602e-05,
+      "loss": 1.1673,
+      "step": 1700
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 2.9466164383561646e-05,
+      "loss": 1.1086,
+      "step": 1800
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 2.9425068493150686e-05,
+      "loss": 1.1001,
+      "step": 1900
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 2.938397260273973e-05,
+      "loss": 1.0817,
+      "step": 2000
+    },
+    {
+      "epoch": 2.72,
+      "eval_cer": 0.13037744578383478,
+      "eval_loss": 0.49887609481811523,
+      "eval_runtime": 182.7109,
+      "eval_samples_per_second": 31.662,
+      "eval_steps_per_second": 3.963,
+      "eval_wer": 0.5931022908450113,
+      "step": 2000
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 2.9342876712328767e-05,
+      "loss": 1.0541,
+      "step": 2100
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 2.9301780821917807e-05,
+      "loss": 1.0207,
+      "step": 2200
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 2.926068493150685e-05,
+      "loss": 1.0096,
+      "step": 2300
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 2.921958904109589e-05,
+      "loss": 1.0009,
+      "step": 2400
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 2.917849315068493e-05,
+      "loss": 0.9801,
+      "step": 2500
+    },
+    {
+      "epoch": 3.4,
+      "eval_cer": 0.12254146584918926,
+      "eval_loss": 0.4582485854625702,
+      "eval_runtime": 183.1019,
+      "eval_samples_per_second": 31.594,
+      "eval_steps_per_second": 3.954,
+      "eval_wer": 0.5667254062823418,
+      "step": 2500
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 2.913739726027397e-05,
+      "loss": 0.9825,
+      "step": 2600
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 2.9096301369863015e-05,
+      "loss": 0.9554,
+      "step": 2700
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 2.9055205479452056e-05,
+      "loss": 0.949,
+      "step": 2800
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 2.90141095890411e-05,
+      "loss": 0.9483,
+      "step": 2900
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 2.8973013698630136e-05,
+      "loss": 0.951,
+      "step": 3000
+    },
+    {
+      "epoch": 4.08,
+      "eval_cer": 0.11486213265428794,
+      "eval_loss": 0.423878937959671,
+      "eval_runtime": 182.3878,
+      "eval_samples_per_second": 31.718,
+      "eval_steps_per_second": 3.97,
+      "eval_wer": 0.5381387933204665,
+      "step": 3000
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 2.893191780821918e-05,
+      "loss": 0.9187,
+      "step": 3100
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 2.889082191780822e-05,
+      "loss": 0.9168,
+      "step": 3200
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 2.884972602739726e-05,
+      "loss": 0.9142,
+      "step": 3300
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 2.88086301369863e-05,
+      "loss": 0.9004,
+      "step": 3400
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 2.876753424657534e-05,
+      "loss": 0.8834,
+      "step": 3500
+    },
+    {
+      "epoch": 4.76,
+      "eval_cer": 0.10923377886580475,
+      "eval_loss": 0.401634156703949,
+      "eval_runtime": 181.0407,
+      "eval_samples_per_second": 31.954,
+      "eval_steps_per_second": 3.999,
+      "eval_wer": 0.5158177393639338,
+      "step": 3500
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 2.8726438356164385e-05,
+      "loss": 0.8878,
+      "step": 3600
+    },
+    {
+      "epoch": 5.03,
+      "learning_rate": 2.8685342465753425e-05,
+      "loss": 0.894,
+      "step": 3700
+    },
+    {
+      "epoch": 5.17,
+      "learning_rate": 2.864424657534247e-05,
+      "loss": 0.869,
+      "step": 3800
+    },
+    {
+      "epoch": 5.31,
+      "learning_rate": 2.8603150684931506e-05,
+      "loss": 0.8725,
+      "step": 3900
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 2.856205479452055e-05,
+      "loss": 0.857,
+      "step": 4000
+    },
+    {
+      "epoch": 5.44,
+      "eval_cer": 0.10468009457091544,
+      "eval_loss": 0.37850794196128845,
+      "eval_runtime": 181.627,
+      "eval_samples_per_second": 31.851,
+      "eval_steps_per_second": 3.986,
+      "eval_wer": 0.49920281950155243,
+      "step": 4000
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 2.852095890410959e-05,
+      "loss": 0.8585,
+      "step": 4100
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 2.847986301369863e-05,
+      "loss": 0.852,
+      "step": 4200
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 2.843876712328767e-05,
+      "loss": 0.8475,
+      "step": 4300
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 2.839808219178082e-05,
+      "loss": 0.8449,
+      "step": 4400
+    },
+    {
+      "epoch": 6.12,
+      "learning_rate": 2.8356986301369865e-05,
+      "loss": 0.8307,
+      "step": 4500
+    },
+    {
+      "epoch": 6.12,
+      "eval_cer": 0.10036320185936037,
+      "eval_loss": 0.37201541662216187,
+      "eval_runtime": 181.3973,
+      "eval_samples_per_second": 31.891,
+      "eval_steps_per_second": 3.991,
+      "eval_wer": 0.481105423624514,
+      "step": 4500
+    },
+    {
+      "epoch": 6.26,
+      "learning_rate": 2.8315890410958906e-05,
+      "loss": 0.8348,
+      "step": 4600
+    },
+    {
+      "epoch": 6.39,
+      "learning_rate": 2.8274794520547946e-05,
+      "loss": 0.824,
+      "step": 4700
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 2.8233698630136986e-05,
+      "loss": 0.8131,
+      "step": 4800
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 2.8193013698630137e-05,
+      "loss": 0.8166,
+      "step": 4900
+    },
+    {
+      "epoch": 6.8,
+      "learning_rate": 2.815191780821918e-05,
+      "loss": 0.805,
+      "step": 5000
+    },
+    {
+      "epoch": 6.8,
+      "eval_cer": 0.09372210868369381,
+      "eval_loss": 0.34496384859085083,
+      "eval_runtime": 181.4357,
+      "eval_samples_per_second": 31.885,
+      "eval_steps_per_second": 3.99,
+      "eval_wer": 0.45366563173058094,
+      "step": 5000
+    },
+    {
+      "epoch": 6.94,
+      "learning_rate": 2.8110821917808218e-05,
+      "loss": 0.8018,
+      "step": 5100
+    },
+    {
+      "epoch": 7.07,
+      "learning_rate": 2.806972602739726e-05,
+      "loss": 0.8018,
+      "step": 5200
+    },
+    {
+      "epoch": 7.21,
+      "learning_rate": 2.8028630136986302e-05,
+      "loss": 0.8098,
+      "step": 5300
+    },
+    {
+      "epoch": 7.35,
+      "learning_rate": 2.7987534246575345e-05,
+      "loss": 0.7855,
+      "step": 5400
+    },
+    {
+      "epoch": 7.48,
+      "learning_rate": 2.7946438356164386e-05,
+      "loss": 0.7828,
+      "step": 5500
+    },
+    {
+      "epoch": 7.48,
+      "eval_cer": 0.09120118905804309,
+      "eval_loss": 0.34213975071907043,
+      "eval_runtime": 182.0382,
+      "eval_samples_per_second": 31.779,
+      "eval_steps_per_second": 3.977,
+      "eval_wer": 0.4460015104472602,
+      "step": 5500
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 2.7905342465753423e-05,
+      "loss": 0.7768,
+      "step": 5600
+    },
+    {
+      "epoch": 7.75,
+      "learning_rate": 2.7864246575342466e-05,
+      "loss": 0.7881,
+      "step": 5700
+    },
+    {
+      "epoch": 7.89,
+      "learning_rate": 2.7823150684931507e-05,
+      "loss": 0.7831,
+      "step": 5800
+    },
+    {
+      "epoch": 8.03,
+      "learning_rate": 2.778205479452055e-05,
+      "loss": 0.7915,
+      "step": 5900
+    },
+    {
+      "epoch": 8.16,
+      "learning_rate": 2.7740958904109587e-05,
+      "loss": 0.7789,
+      "step": 6000
+    },
+    {
+      "epoch": 8.16,
+      "eval_cer": 0.08897170522726527,
+      "eval_loss": 0.32954761385917664,
+      "eval_runtime": 181.2086,
+      "eval_samples_per_second": 31.925,
+      "eval_steps_per_second": 3.995,
+      "eval_wer": 0.43366619115549215,
+      "step": 6000
+    },
+    {
+      "epoch": 8.3,
+      "learning_rate": 2.769986301369863e-05,
+      "loss": 0.7599,
+      "step": 6100
+    },
+    {
+      "epoch": 8.43,
+      "learning_rate": 2.765876712328767e-05,
+      "loss": 0.7774,
+      "step": 6200
+    },
+    {
+      "epoch": 8.57,
+      "learning_rate": 2.7617671232876715e-05,
+      "loss": 0.7655,
+      "step": 6300
+    },
+    {
+      "epoch": 8.71,
+      "learning_rate": 2.7576575342465756e-05,
+      "loss": 0.7513,
+      "step": 6400
+    },
+    {
+      "epoch": 8.84,
+      "learning_rate": 2.7535479452054792e-05,
+      "loss": 0.755,
+      "step": 6500
+    },
+    {
+      "epoch": 8.84,
+      "eval_cer": 0.08615206391186982,
+      "eval_loss": 0.32569244503974915,
+      "eval_runtime": 181.8099,
+      "eval_samples_per_second": 31.819,
+      "eval_steps_per_second": 3.982,
+      "eval_wer": 0.4221979804760706,
+      "step": 6500
+    },
+    {
+      "epoch": 8.98,
+      "learning_rate": 2.7494383561643836e-05,
+      "loss": 0.7566,
+      "step": 6600
+    },
+    {
+      "epoch": 9.12,
+      "learning_rate": 2.7453287671232877e-05,
+      "loss": 0.7571,
+      "step": 6700
+    },
+    {
+      "epoch": 9.25,
+      "learning_rate": 2.741219178082192e-05,
+      "loss": 0.7515,
+      "step": 6800
+    },
+    {
+      "epoch": 9.39,
+      "learning_rate": 2.7371095890410957e-05,
+      "loss": 0.7427,
+      "step": 6900
+    },
+    {
+      "epoch": 9.52,
+      "learning_rate": 2.733e-05,
+      "loss": 0.7464,
+      "step": 7000
+    },
+    {
+      "epoch": 9.52,
+      "eval_cer": 0.0846548125157102,
+      "eval_loss": 0.3269043266773224,
+      "eval_runtime": 180.9905,
+      "eval_samples_per_second": 31.963,
+      "eval_steps_per_second": 4.0,
+      "eval_wer": 0.41548488154177504,
+      "step": 7000
+    },
+    {
+      "epoch": 9.66,
+      "learning_rate": 2.728890410958904e-05,
+      "loss": 0.7414,
+      "step": 7100
+    },
+    {
+      "epoch": 9.8,
+      "learning_rate": 2.7247808219178085e-05,
+      "loss": 0.7316,
+      "step": 7200
+    },
+    {
+      "epoch": 9.93,
+      "learning_rate": 2.7206712328767125e-05,
+      "loss": 0.737,
+      "step": 7300
+    },
+    {
+      "epoch": 10.07,
+      "learning_rate": 2.7165616438356166e-05,
+      "loss": 0.7284,
+      "step": 7400
+    },
+    {
+      "epoch": 10.2,
+      "learning_rate": 2.7124520547945206e-05,
+      "loss": 0.7293,
+      "step": 7500
+    },
+    {
+      "epoch": 10.2,
+      "eval_cer": 0.08234882678877826,
+      "eval_loss": 0.3121315836906433,
+      "eval_runtime": 181.2877,
+      "eval_samples_per_second": 31.911,
+      "eval_steps_per_second": 3.994,
+      "eval_wer": 0.4025341948476966,
+      "step": 7500
+    },
+    {
+      "epoch": 10.34,
+      "learning_rate": 2.7083424657534246e-05,
+      "loss": 0.7247,
+      "step": 7600
+    },
+    {
+      "epoch": 10.48,
+      "learning_rate": 2.704232876712329e-05,
+      "loss": 0.7238,
+      "step": 7700
+    },
+    {
+      "epoch": 10.61,
+      "learning_rate": 2.7001643835616437e-05,
+      "loss": 0.7184,
+      "step": 7800
+    },
+    {
+      "epoch": 10.75,
+      "learning_rate": 2.696054794520548e-05,
+      "loss": 0.7322,
+      "step": 7900
+    },
+    {
+      "epoch": 10.88,
+      "learning_rate": 2.691945205479452e-05,
+      "loss": 0.7283,
+      "step": 8000
+    },
+    {
+      "epoch": 10.88,
+      "eval_cer": 0.07888074082979056,
+      "eval_loss": 0.29910361766815186,
+      "eval_runtime": 182.8781,
+      "eval_samples_per_second": 31.633,
+      "eval_steps_per_second": 3.959,
+      "eval_wer": 0.3940589074431484,
+      "step": 8000
+    },
+    {
+      "epoch": 11.02,
+      "learning_rate": 2.6878356164383562e-05,
+      "loss": 0.7144,
+      "step": 8100
+    },
+    {
+      "epoch": 11.16,
+      "learning_rate": 2.6837260273972602e-05,
+      "loss": 0.7199,
+      "step": 8200
+    },
+    {
+      "epoch": 11.29,
+      "learning_rate": 2.6796575342465753e-05,
+      "loss": 0.7194,
+      "step": 8300
+    },
+    {
+      "epoch": 11.43,
+      "learning_rate": 2.6755479452054797e-05,
+      "loss": 0.7171,
+      "step": 8400
+    },
+    {
+      "epoch": 11.56,
+      "learning_rate": 2.6714383561643837e-05,
+      "loss": 0.7145,
+      "step": 8500
+    },
+    {
+      "epoch": 11.56,
+      "eval_cer": 0.07858201914004583,
+      "eval_loss": 0.2960781455039978,
+      "eval_runtime": 181.1035,
+      "eval_samples_per_second": 31.943,
+      "eval_steps_per_second": 3.998,
+      "eval_wer": 0.3867863835976616,
+      "step": 8500
+    },
+    {
+      "epoch": 11.7,
+      "learning_rate": 2.6673287671232877e-05,
+      "loss": 0.7073,
+      "step": 8600
+    },
+    {
+      "epoch": 11.84,
+      "learning_rate": 2.6632191780821918e-05,
+      "loss": 0.7124,
+      "step": 8700
+    },
+    {
+      "epoch": 11.97,
+      "learning_rate": 2.6591095890410958e-05,
+      "loss": 0.7041,
+      "step": 8800
+    },
+    {
+      "epoch": 12.11,
+      "learning_rate": 2.655e-05,
+      "loss": 0.7048,
+      "step": 8900
+    },
+    {
+      "epoch": 12.24,
+      "learning_rate": 2.6508904109589042e-05,
+      "loss": 0.6963,
+      "step": 9000
+    },
+    {
+      "epoch": 12.24,
+      "eval_cer": 0.07668768647337187,
+      "eval_loss": 0.2972111105918884,
+      "eval_runtime": 181.6459,
+      "eval_samples_per_second": 31.848,
+      "eval_steps_per_second": 3.986,
+      "eval_wer": 0.3783670386842326,
+      "step": 9000
+    },
+    {
+      "epoch": 12.38,
+      "learning_rate": 2.6467808219178082e-05,
+      "loss": 0.6894,
+      "step": 9100
+    },
+    {
+      "epoch": 12.52,
+      "learning_rate": 2.6426712328767123e-05,
+      "loss": 0.6965,
+      "step": 9200
+    },
+    {
+      "epoch": 12.65,
+      "learning_rate": 2.6385616438356166e-05,
+      "loss": 0.688,
+      "step": 9300
+    },
+    {
+      "epoch": 12.79,
+      "learning_rate": 2.6344520547945207e-05,
+      "loss": 0.6864,
+      "step": 9400
+    },
+    {
+      "epoch": 12.92,
+      "learning_rate": 2.6303424657534247e-05,
+      "loss": 0.6981,
+      "step": 9500
+    },
+    {
+      "epoch": 12.92,
+      "eval_cer": 0.07572230540285534,
+      "eval_loss": 0.2879623770713806,
+      "eval_runtime": 181.933,
+      "eval_samples_per_second": 31.797,
+      "eval_steps_per_second": 3.979,
+      "eval_wer": 0.3750384604626444,
+      "step": 9500
+    },
+    {
+      "epoch": 13.06,
+      "learning_rate": 2.6262328767123287e-05,
+      "loss": 0.6897,
+      "step": 9600
+    },
+    {
+      "epoch": 13.2,
+      "learning_rate": 2.622123287671233e-05,
+      "loss": 0.682,
+      "step": 9700
+    },
+    {
+      "epoch": 13.33,
+      "learning_rate": 2.618013698630137e-05,
+      "loss": 0.6844,
+      "step": 9800
+    },
+    {
+      "epoch": 13.47,
+      "learning_rate": 2.613904109589041e-05,
+      "loss": 0.6784,
+      "step": 9900
+    },
+    {
+      "epoch": 13.6,
+      "learning_rate": 2.6097945205479452e-05,
+      "loss": 0.6888,
+      "step": 10000
+    },
+    {
+      "epoch": 13.6,
+      "eval_cer": 0.07447277443233771,
+      "eval_loss": 0.28652679920196533,
+      "eval_runtime": 180.8574,
+      "eval_samples_per_second": 31.987,
+      "eval_steps_per_second": 4.003,
+      "eval_wer": 0.3702833487175184,
+      "step": 10000
+    },
+    {
+      "epoch": 13.74,
+      "learning_rate": 2.6056849315068492e-05,
+      "loss": 0.6789,
+      "step": 10100
+    },
+    {
+      "epoch": 13.88,
+      "learning_rate": 2.6015753424657536e-05,
+      "loss": 0.6783,
+      "step": 10200
+    },
+    {
+      "epoch": 14.01,
+      "learning_rate": 2.5975068493150687e-05,
+      "loss": 0.6824,
+      "step": 10300
+    },
+    {
+      "epoch": 14.15,
+      "learning_rate": 2.5933972602739724e-05,
+      "loss": 0.6678,
+      "step": 10400
+    },
+    {
+      "epoch": 14.29,
+      "learning_rate": 2.5892876712328767e-05,
+      "loss": 0.6733,
+      "step": 10500
+    },
+    {
+      "epoch": 14.29,
+      "eval_cer": 0.07442177316823495,
+      "eval_loss": 0.28869137167930603,
+      "eval_runtime": 181.5666,
+      "eval_samples_per_second": 31.862,
+      "eval_steps_per_second": 3.988,
+      "eval_wer": 0.3663394030936198,
+      "step": 10500
+    },
+    {
+      "epoch": 14.42,
+      "learning_rate": 2.5851780821917808e-05,
+      "loss": 0.6655,
+      "step": 10600
+    },
+    {
+      "epoch": 14.56,
+      "learning_rate": 2.581068493150685e-05,
+      "loss": 0.6647,
+      "step": 10700
+    },
+    {
+      "epoch": 14.69,
+      "learning_rate": 2.576958904109589e-05,
+      "loss": 0.6739,
+      "step": 10800
+    },
+    {
+      "epoch": 14.83,
+      "learning_rate": 2.5728493150684932e-05,
+      "loss": 0.6784,
+      "step": 10900
+    },
+    {
+      "epoch": 14.97,
+      "learning_rate": 2.5687397260273973e-05,
+      "loss": 0.6701,
+      "step": 11000
+    },
+    {
+      "epoch": 14.97,
+      "eval_cer": 0.07352196515156482,
+      "eval_loss": 0.285683274269104,
+      "eval_runtime": 183.1662,
+      "eval_samples_per_second": 31.583,
+      "eval_steps_per_second": 3.953,
+      "eval_wer": 0.36236748622416154,
+      "step": 11000
+    },
+    {
+      "epoch": 15.1,
+      "learning_rate": 2.5646301369863016e-05,
+      "loss": 0.6692,
+      "step": 11100
+    },
+    {
+      "epoch": 15.24,
+      "learning_rate": 2.5605205479452057e-05,
+      "loss": 0.6654,
+      "step": 11200
+    },
+    {
+      "epoch": 15.37,
+      "learning_rate": 2.5564109589041097e-05,
+      "loss": 0.6574,
+      "step": 11300
+    },
+    {
+      "epoch": 15.51,
+      "learning_rate": 2.5523013698630137e-05,
+      "loss": 0.6599,
+      "step": 11400
+    },
+    {
+      "epoch": 15.65,
+      "learning_rate": 2.5481917808219178e-05,
+      "loss": 0.6634,
+      "step": 11500
+    },
+    {
+      "epoch": 15.65,
+      "eval_cer": 0.07227607712848311,
+      "eval_loss": 0.28218305110931396,
+      "eval_runtime": 183.0801,
+      "eval_samples_per_second": 31.598,
+      "eval_steps_per_second": 3.955,
+      "eval_wer": 0.3581438281446673,
+      "step": 11500
+    },
+    {
+      "epoch": 15.78,
+      "learning_rate": 2.544082191780822e-05,
+      "loss": 0.6693,
+      "step": 11600
+    },
+    {
+      "epoch": 15.92,
+      "learning_rate": 2.5399726027397258e-05,
+      "loss": 0.6695,
+      "step": 11700
+    },
+    {
+      "epoch": 16.05,
+      "learning_rate": 2.5358630136986302e-05,
+      "loss": 0.6716,
+      "step": 11800
+    },
+    {
+      "epoch": 16.19,
+      "learning_rate": 2.5317534246575342e-05,
+      "loss": 0.6518,
+      "step": 11900
+    },
+    {
+      "epoch": 16.33,
+      "learning_rate": 2.5276438356164386e-05,
+      "loss": 0.6484,
+      "step": 12000
+    },
+    {
+      "epoch": 16.33,
+      "eval_cer": 0.07062217899257932,
+      "eval_loss": 0.27775853872299194,
+      "eval_runtime": 180.6027,
+      "eval_samples_per_second": 32.032,
+      "eval_steps_per_second": 4.009,
+      "eval_wer": 0.3503398506335487,
+      "step": 12000
+    },
+    {
+      "epoch": 16.46,
+      "learning_rate": 2.5235342465753426e-05,
+      "loss": 0.6481,
+      "step": 12100
+    },
+    {
+      "epoch": 16.6,
+      "learning_rate": 2.5194246575342467e-05,
+      "loss": 0.6438,
+      "step": 12200
+    },
+    {
+      "epoch": 16.73,
+      "learning_rate": 2.5153150684931507e-05,
+      "loss": 0.6508,
+      "step": 12300
+    },
+    {
+      "epoch": 16.87,
+      "learning_rate": 2.5112054794520547e-05,
+      "loss": 0.6458,
+      "step": 12400
+    },
+    {
+      "epoch": 17.01,
+      "learning_rate": 2.507095890410959e-05,
+      "loss": 0.6626,
+      "step": 12500
+    },
+    {
+      "epoch": 17.01,
+      "eval_cer": 0.06972237097590919,
+      "eval_loss": 0.26972296833992004,
+      "eval_runtime": 181.5946,
+      "eval_samples_per_second": 31.857,
+      "eval_steps_per_second": 3.987,
+      "eval_wer": 0.34768258230539006,
+      "step": 12500
+    },
+    {
+      "epoch": 17.14,
+      "learning_rate": 2.5029863013698628e-05,
+      "loss": 0.6454,
+      "step": 12600
+    },
+    {
+      "epoch": 17.28,
+      "learning_rate": 2.498876712328767e-05,
+      "loss": 0.6475,
+      "step": 12700
+    },
+    {
+      "epoch": 17.41,
+      "learning_rate": 2.4947671232876712e-05,
+      "loss": 0.6402,
+      "step": 12800
+    },
+    {
+      "epoch": 17.55,
+      "learning_rate": 2.4906986301369866e-05,
+      "loss": 0.6339,
+      "step": 12900
+    },
+    {
+      "epoch": 17.69,
+      "learning_rate": 2.4865890410958903e-05,
+      "loss": 0.6341,
+      "step": 13000
+    },
+    {
+      "epoch": 17.69,
+      "eval_cer": 0.07083711289129808,
+      "eval_loss": 0.28038930892944336,
+      "eval_runtime": 180.3906,
+      "eval_samples_per_second": 32.069,
+      "eval_steps_per_second": 4.014,
+      "eval_wer": 0.3510671030180974,
+      "step": 13000
+    },
+    {
+      "epoch": 17.82,
+      "learning_rate": 2.4824794520547943e-05,
+      "loss": 0.649,
+      "step": 13100
+    },
+    {
+      "epoch": 17.96,
+      "learning_rate": 2.4784109589041098e-05,
+      "loss": 0.6398,
+      "step": 13200
+    },
+    {
+      "epoch": 18.1,
+      "learning_rate": 2.4743013698630138e-05,
+      "loss": 0.6472,
+      "step": 13300
+    },
+    {
+      "epoch": 18.23,
+      "learning_rate": 2.4701917808219178e-05,
+      "loss": 0.6236,
+      "step": 13400
+    },
+    {
+      "epoch": 18.37,
+      "learning_rate": 2.466082191780822e-05,
+      "loss": 0.6402,
+      "step": 13500
+    },
+    {
+      "epoch": 18.37,
+      "eval_cer": 0.06811947410410815,
+      "eval_loss": 0.26648715138435364,
+      "eval_runtime": 181.5264,
+      "eval_samples_per_second": 31.869,
+      "eval_steps_per_second": 3.988,
+      "eval_wer": 0.341836591983441,
+      "step": 13500
+    },
+    {
+      "epoch": 18.5,
+      "learning_rate": 2.4619726027397262e-05,
+      "loss": 0.6265,
+      "step": 13600
+    },
+    {
+      "epoch": 18.64,
+      "learning_rate": 2.4578630136986303e-05,
+      "loss": 0.6281,
+      "step": 13700
+    },
+    {
+      "epoch": 18.77,
+      "learning_rate": 2.4537534246575343e-05,
+      "loss": 0.6352,
+      "step": 13800
+    },
+    {
+      "epoch": 18.91,
+      "learning_rate": 2.4496438356164383e-05,
+      "loss": 0.6361,
+      "step": 13900
+    },
+    {
+      "epoch": 19.05,
+      "learning_rate": 2.4455342465753424e-05,
+      "loss": 0.6343,
+      "step": 14000
+    },
+    {
+      "epoch": 19.05,
+      "eval_cer": 0.06866955916693078,
+      "eval_loss": 0.27482911944389343,
+      "eval_runtime": 181.3067,
+      "eval_samples_per_second": 31.907,
+      "eval_steps_per_second": 3.993,
+      "eval_wer": 0.3425079018768706,
+      "step": 14000
+    },
+    {
+      "epoch": 19.18,
+      "learning_rate": 2.4414246575342467e-05,
+      "loss": 0.6393,
+      "step": 14100
+    },
+    {
+      "epoch": 19.32,
+      "learning_rate": 2.4373150684931508e-05,
+      "loss": 0.6341,
+      "step": 14200
+    },
+    {
+      "epoch": 19.46,
+      "learning_rate": 2.4332054794520548e-05,
+      "loss": 0.6233,
+      "step": 14300
+    },
+    {
+      "epoch": 19.59,
+      "learning_rate": 2.4290958904109588e-05,
+      "loss": 0.6284,
+      "step": 14400
+    },
+    {
+      "epoch": 19.73,
+      "learning_rate": 2.4249863013698632e-05,
+      "loss": 0.6383,
+      "step": 14500
+    },
+    {
+      "epoch": 19.73,
+      "eval_cer": 0.06774425051820927,
+      "eval_loss": 0.2695872485637665,
+      "eval_runtime": 183.3905,
+      "eval_samples_per_second": 31.545,
+      "eval_steps_per_second": 3.948,
+      "eval_wer": 0.33831221504293585,
+      "step": 14500
+    },
+    {
+      "epoch": 19.86,
+      "learning_rate": 2.4208767123287672e-05,
+      "loss": 0.6282,
+      "step": 14600
+    },
+    {
+      "epoch": 20.0,
+      "learning_rate": 2.4167671232876713e-05,
+      "loss": 0.6209,
+      "step": 14700
+    },
+    {
+      "epoch": 20.14,
+      "learning_rate": 2.4126575342465753e-05,
+      "loss": 0.6267,
+      "step": 14800
+    },
+    {
+      "epoch": 20.27,
+      "learning_rate": 2.4085479452054793e-05,
+      "loss": 0.6343,
+      "step": 14900
+    },
+    {
+      "epoch": 20.41,
+      "learning_rate": 2.4044383561643837e-05,
+      "loss": 0.6178,
+      "step": 15000
+    },
+    {
+      "epoch": 20.41,
+      "eval_cer": 0.06895370906693188,
+      "eval_loss": 0.27426353096961975,
+      "eval_runtime": 182.9058,
+      "eval_samples_per_second": 31.628,
+      "eval_steps_per_second": 3.958,
+      "eval_wer": 0.3416967357556432,
+      "step": 15000
+    },
+    {
+      "epoch": 20.54,
+      "learning_rate": 2.4003287671232877e-05,
+      "loss": 0.6249,
+      "step": 15100
+    },
+    {
+      "epoch": 20.68,
+      "learning_rate": 2.3962191780821918e-05,
+      "loss": 0.6355,
+      "step": 15200
+    },
+    {
+      "epoch": 20.82,
+      "learning_rate": 2.3921095890410958e-05,
+      "loss": 0.6267,
+      "step": 15300
+    },
+    {
+      "epoch": 20.95,
+      "learning_rate": 2.388041095890411e-05,
+      "loss": 0.6174,
+      "step": 15400
+    },
+    {
+      "epoch": 21.09,
+      "learning_rate": 2.3839315068493153e-05,
+      "loss": 0.6097,
+      "step": 15500
+    },
+    {
+      "epoch": 21.09,
+      "eval_cer": 0.06705937640025791,
+      "eval_loss": 0.26633238792419434,
+      "eval_runtime": 179.4572,
+      "eval_samples_per_second": 32.236,
+      "eval_steps_per_second": 4.034,
+      "eval_wer": 0.33523537803138376,
+      "step": 15500
+    },
+    {
+      "epoch": 21.22,
+      "learning_rate": 2.379821917808219e-05,
+      "loss": 0.6175,
+      "step": 15600
+    },
+    {
+      "epoch": 21.36,
+      "learning_rate": 2.3757123287671233e-05,
+      "loss": 0.6128,
+      "step": 15700
+    },
+    {
+      "epoch": 21.5,
+      "learning_rate": 2.3716027397260274e-05,
+      "loss": 0.6184,
+      "step": 15800
+    },
+    {
+      "epoch": 21.63,
+      "learning_rate": 2.3674931506849317e-05,
+      "loss": 0.6189,
+      "step": 15900
+    },
+    {
+      "epoch": 21.77,
+      "learning_rate": 2.3633835616438358e-05,
+      "loss": 0.6245,
+      "step": 16000
+    },
+    {
+      "epoch": 21.77,
+      "eval_cer": 0.06646193302076844,
+      "eval_loss": 0.2684832811355591,
+      "eval_runtime": 177.8727,
+      "eval_samples_per_second": 32.523,
+      "eval_steps_per_second": 4.07,
+      "eval_wer": 0.33182288607311683,
+      "step": 16000
+    },
+    {
+      "epoch": 21.9,
+      "learning_rate": 2.3592739726027398e-05,
+      "loss": 0.6116,
+      "step": 16100
+    },
+    {
+      "epoch": 22.04,
+      "learning_rate": 2.3551643835616438e-05,
+      "loss": 0.6161,
+      "step": 16200
+    },
+    {
+      "epoch": 22.18,
+      "learning_rate": 2.3510547945205482e-05,
+      "loss": 0.6066,
+      "step": 16300
+    },
+    {
+      "epoch": 22.31,
+      "learning_rate": 2.3469452054794522e-05,
+      "loss": 0.6137,
+      "step": 16400
+    },
+    {
+      "epoch": 22.45,
+      "learning_rate": 2.342835616438356e-05,
+      "loss": 0.6137,
+      "step": 16500
+    },
+    {
+      "epoch": 22.45,
+      "eval_cer": 0.06547105131820052,
+      "eval_loss": 0.26997432112693787,
+      "eval_runtime": 177.9171,
+      "eval_samples_per_second": 32.515,
+      "eval_steps_per_second": 4.069,
+      "eval_wer": 0.3262006657156443,
+      "step": 16500
+    },
+    {
+      "epoch": 22.58,
+      "learning_rate": 2.3387260273972603e-05,
+      "loss": 0.605,
+      "step": 16600
+    },
+    {
+      "epoch": 22.72,
+      "learning_rate": 2.3346164383561643e-05,
+      "loss": 0.6315,
+      "step": 16700
+    },
+    {
+      "epoch": 22.86,
+      "learning_rate": 2.3305068493150687e-05,
+      "loss": 0.6155,
+      "step": 16800
+    },
+    {
+      "epoch": 22.99,
+      "learning_rate": 2.3263972602739727e-05,
+      "loss": 0.602,
+      "step": 16900
+    },
+    {
+      "epoch": 23.13,
+      "learning_rate": 2.3222876712328768e-05,
+      "loss": 0.6018,
+      "step": 17000
+    },
+    {
+      "epoch": 23.13,
+      "eval_cer": 0.06519418731307126,
+      "eval_loss": 0.2656744122505188,
+      "eval_runtime": 177.3164,
+      "eval_samples_per_second": 32.625,
+      "eval_steps_per_second": 4.083,
+      "eval_wer": 0.3224525188106626,
+      "step": 17000
+    },
+    {
+      "epoch": 23.27,
+      "learning_rate": 2.3181780821917808e-05,
+      "loss": 0.6008,
+      "step": 17100
+    },
+    {
+      "epoch": 23.4,
+      "learning_rate": 2.314068493150685e-05,
+      "loss": 0.5924,
+      "step": 17200
+    },
+    {
+      "epoch": 23.54,
+      "learning_rate": 2.3099589041095892e-05,
+      "loss": 0.6073,
+      "step": 17300
+    },
+    {
+      "epoch": 23.67,
+      "learning_rate": 2.305849315068493e-05,
+      "loss": 0.6079,
+      "step": 17400
+    },
+    {
+      "epoch": 23.81,
+      "learning_rate": 2.3017397260273973e-05,
+      "loss": 0.6063,
+      "step": 17500
+    },
+    {
+      "epoch": 23.81,
+      "eval_cer": 0.0663235010182038,
+      "eval_loss": 0.2664479613304138,
+      "eval_runtime": 178.312,
+      "eval_samples_per_second": 32.443,
+      "eval_steps_per_second": 4.06,
+      "eval_wer": 0.3276271992391821,
+      "step": 17500
+    },
+    {
+      "epoch": 23.94,
+      "learning_rate": 2.2976301369863013e-05,
+      "loss": 0.6066,
+      "step": 17600
+    },
+    {
+      "epoch": 24.08,
+      "learning_rate": 2.2935205479452057e-05,
+      "loss": 0.61,
+      "step": 17700
+    },
+    {
+      "epoch": 24.22,
+      "learning_rate": 2.2894109589041097e-05,
+      "loss": 0.601,
+      "step": 17800
+    },
+    {
+      "epoch": 24.35,
+      "learning_rate": 2.2853013698630137e-05,
+      "loss": 0.5963,
+      "step": 17900
+    },
+    {
+      "epoch": 24.49,
+      "learning_rate": 2.2811917808219178e-05,
+      "loss": 0.5917,
+      "step": 18000
+    },
+    {
+      "epoch": 24.49,
+      "eval_cer": 0.06583898900922759,
+      "eval_loss": 0.27249768376350403,
+      "eval_runtime": 177.5086,
+      "eval_samples_per_second": 32.59,
+      "eval_steps_per_second": 4.079,
+      "eval_wer": 0.3263684931890017,
+      "step": 18000
+    },
+    {
+      "epoch": 24.63,
+      "learning_rate": 2.277082191780822e-05,
+      "loss": 0.6039,
+      "step": 18100
+    },
+    {
+      "epoch": 24.76,
+      "learning_rate": 2.2729726027397262e-05,
+      "loss": 0.6048,
+      "step": 18200
+    },
+    {
+      "epoch": 24.9,
+      "learning_rate": 2.26886301369863e-05,
+      "loss": 0.5955,
+      "step": 18300
+    },
+    {
+      "epoch": 25.03,
+      "learning_rate": 2.2647534246575342e-05,
+      "loss": 0.6069,
+      "step": 18400
+    },
+    {
+      "epoch": 25.17,
+      "learning_rate": 2.2606438356164383e-05,
+      "loss": 0.5984,
+      "step": 18500
+    },
+    {
+      "epoch": 25.17,
+      "eval_cer": 0.0642834504540934,
+      "eval_loss": 0.259331077337265,
+      "eval_runtime": 175.455,
+      "eval_samples_per_second": 32.971,
+      "eval_steps_per_second": 4.126,
+      "eval_wer": 0.3196553942547062,
+      "step": 18500
+    },
+    {
+      "epoch": 25.31,
+      "learning_rate": 2.2565342465753426e-05,
+      "loss": 0.5968,
+      "step": 18600
+    },
+    {
+      "epoch": 25.44,
+      "learning_rate": 2.2524246575342467e-05,
+      "loss": 0.5936,
+      "step": 18700
+    },
+    {
+      "epoch": 25.58,
+      "learning_rate": 2.2483150684931507e-05,
+      "loss": 0.588,
+      "step": 18800
+    },
+    {
+      "epoch": 25.71,
+      "learning_rate": 2.2442054794520547e-05,
+      "loss": 0.5983,
+      "step": 18900
+    },
+    {
+      "epoch": 25.85,
+      "learning_rate": 2.2401369863013698e-05,
+      "loss": 0.5949,
+      "step": 19000
+    },
+    {
+      "epoch": 25.85,
+      "eval_cer": 0.06349657380793652,
+      "eval_loss": 0.2580772638320923,
+      "eval_runtime": 174.8765,
+      "eval_samples_per_second": 33.08,
+      "eval_steps_per_second": 4.14,
+      "eval_wer": 0.3161030460686414,
+      "step": 19000
+    },
+    {
+      "epoch": 25.99,
+      "learning_rate": 2.2360273972602742e-05,
+      "loss": 0.6001,
+      "step": 19100
+    },
+    {
+      "epoch": 26.12,
+      "learning_rate": 2.231917808219178e-05,
+      "loss": 0.5993,
+      "step": 19200
+    },
+    {
+      "epoch": 26.26,
+      "learning_rate": 2.2278082191780823e-05,
+      "loss": 0.5972,
+      "step": 19300
+    },
+    {
+      "epoch": 26.39,
+      "learning_rate": 2.2236986301369863e-05,
+      "loss": 0.5977,
+      "step": 19400
+    },
+    {
+      "epoch": 26.53,
+      "learning_rate": 2.2195890410958907e-05,
+      "loss": 0.5863,
+      "step": 19500
+    },
+    {
+      "epoch": 26.53,
+      "eval_cer": 0.06387908328870723,
+      "eval_loss": 0.254349023103714,
+      "eval_runtime": 174.0753,
+      "eval_samples_per_second": 33.233,
+      "eval_steps_per_second": 4.159,
+      "eval_wer": 0.3196274230091466,
+      "step": 19500
+    },
+    {
+      "epoch": 26.67,
+      "learning_rate": 2.2154794520547944e-05,
+      "loss": 0.5983,
+      "step": 19600
+    },
+    {
+      "epoch": 26.8,
+      "learning_rate": 2.2113698630136987e-05,
+      "loss": 0.5795,
+      "step": 19700
+    },
+    {
+      "epoch": 26.94,
+      "learning_rate": 2.2072602739726028e-05,
+      "loss": 0.5825,
+      "step": 19800
+    },
+    {
+      "epoch": 27.07,
+      "learning_rate": 2.203150684931507e-05,
+      "loss": 0.592,
+      "step": 19900
+    },
+    {
+      "epoch": 27.21,
+      "learning_rate": 2.199041095890411e-05,
+      "loss": 0.5858,
+      "step": 20000
+    },
+    {
+      "epoch": 27.21,
+      "eval_cer": 0.06275341253101059,
+      "eval_loss": 0.26198887825012207,
+      "eval_runtime": 173.0756,
+      "eval_samples_per_second": 33.425,
+      "eval_steps_per_second": 4.183,
+      "eval_wer": 0.3135856339682806,
+      "step": 20000
+    },
+    {
+      "epoch": 27.35,
+      "learning_rate": 2.194931506849315e-05,
+      "loss": 0.5922,
+      "step": 20100
+    },
+    {
+      "epoch": 27.48,
+      "learning_rate": 2.1908219178082192e-05,
+      "loss": 0.5786,
+      "step": 20200
+    },
+    {
+      "epoch": 27.62,
+      "learning_rate": 2.1867123287671233e-05,
+      "loss": 0.5937,
+      "step": 20300
+    },
+    {
+      "epoch": 27.75,
+      "learning_rate": 2.1826027397260276e-05,
+      "loss": 0.5872,
+      "step": 20400
+    },
+    {
+      "epoch": 27.89,
+      "learning_rate": 2.1784931506849313e-05,
+      "loss": 0.5902,
+      "step": 20500
+    },
+    {
+      "epoch": 27.89,
+      "eval_cer": 0.0626914824246001,
+      "eval_loss": 0.25490838289260864,
+      "eval_runtime": 171.8594,
+      "eval_samples_per_second": 33.661,
+      "eval_steps_per_second": 4.213,
+      "eval_wer": 0.315683477385248,
+      "step": 20500
+    },
+    {
+      "epoch": 28.03,
+      "learning_rate": 2.1743835616438357e-05,
+      "loss": 0.5874,
+      "step": 20600
+    },
+    {
+      "epoch": 28.16,
+      "learning_rate": 2.1702739726027397e-05,
+      "loss": 0.584,
+      "step": 20700
+    },
+    {
+      "epoch": 28.3,
+      "learning_rate": 2.166164383561644e-05,
+      "loss": 0.5837,
+      "step": 20800
+    },
+    {
+      "epoch": 28.43,
+      "learning_rate": 2.162054794520548e-05,
+      "loss": 0.5719,
+      "step": 20900
+    },
+    {
+      "epoch": 28.57,
+      "learning_rate": 2.157945205479452e-05,
+      "loss": 0.5794,
+      "step": 21000
+    },
+    {
+      "epoch": 28.57,
+      "eval_cer": 0.06237090305023989,
+      "eval_loss": 0.2543303966522217,
+      "eval_runtime": 172.7958,
+      "eval_samples_per_second": 33.479,
+      "eval_steps_per_second": 4.19,
+      "eval_wer": 0.3136415764593997,
+      "step": 21000
+    },
+    {
+      "epoch": 28.71,
+      "learning_rate": 2.1538767123287673e-05,
+      "loss": 0.5835,
+      "step": 21100
+    },
+    {
+      "epoch": 28.84,
+      "learning_rate": 2.1497671232876713e-05,
+      "loss": 0.5854,
+      "step": 21200
+    },
+    {
+      "epoch": 28.98,
+      "learning_rate": 2.1456575342465757e-05,
+      "loss": 0.5894,
+      "step": 21300
+    },
+    {
+      "epoch": 29.12,
+      "learning_rate": 2.1415479452054793e-05,
+      "loss": 0.5807,
+      "step": 21400
+    },
+    {
+      "epoch": 29.25,
+      "learning_rate": 2.1374383561643837e-05,
+      "loss": 0.5744,
+      "step": 21500
+    },
+    {
+      "epoch": 29.25,
+      "eval_cer": 0.06195560704254598,
+      "eval_loss": 0.2541608214378357,
+      "eval_runtime": 172.9419,
+      "eval_samples_per_second": 33.451,
+      "eval_steps_per_second": 4.186,
+      "eval_wer": 0.30905429218763114,
+      "step": 21500
+    },
+    {
+      "epoch": 29.39,
+      "learning_rate": 2.1333287671232878e-05,
+      "loss": 0.5647,
+      "step": 21600
+    },
+    {
+      "epoch": 29.52,
+      "learning_rate": 2.1292191780821918e-05,
+      "loss": 0.5861,
+      "step": 21700
+    },
+    {
+      "epoch": 29.66,
+      "learning_rate": 2.1251095890410958e-05,
+      "loss": 0.5678,
+      "step": 21800
+    },
+    {
+      "epoch": 29.8,
+      "learning_rate": 2.121e-05,
+      "loss": 0.5672,
+      "step": 21900
+    },
+    {
+      "epoch": 29.93,
+      "learning_rate": 2.1168904109589042e-05,
+      "loss": 0.5899,
+      "step": 22000
+    },
+    {
+      "epoch": 29.93,
+      "eval_cer": 0.06244740494639403,
+      "eval_loss": 0.25403666496276855,
+      "eval_runtime": 172.4559,
+      "eval_samples_per_second": 33.545,
+      "eval_steps_per_second": 4.198,
+      "eval_wer": 0.3121591004447428,
+      "step": 22000
+    },
+    {
+      "epoch": 30.07,
+      "learning_rate": 2.1127808219178083e-05,
+      "loss": 0.5678,
+      "step": 22100
+    },
+    {
+      "epoch": 30.2,
+      "learning_rate": 2.1086712328767126e-05,
+      "loss": 0.5722,
+      "step": 22200
+    },
+    {
+      "epoch": 30.34,
+      "learning_rate": 2.1045616438356163e-05,
+      "loss": 0.574,
+      "step": 22300
+    },
+    {
+      "epoch": 30.48,
+      "learning_rate": 2.1004520547945207e-05,
+      "loss": 0.5609,
+      "step": 22400
+    },
+    {
+      "epoch": 30.61,
+      "learning_rate": 2.0963424657534247e-05,
+      "loss": 0.5597,
+      "step": 22500
+    },
+    {
+      "epoch": 30.61,
+      "eval_cer": 0.06086636575920846,
+      "eval_loss": 0.24997848272323608,
+      "eval_runtime": 173.1294,
+      "eval_samples_per_second": 33.414,
+      "eval_steps_per_second": 4.182,
+      "eval_wer": 0.3056697714749238,
+      "step": 22500
+    },
+    {
+      "epoch": 30.75,
+      "learning_rate": 2.0922328767123288e-05,
+      "loss": 0.5693,
+      "step": 22600
+    },
+    {
+      "epoch": 30.88,
+      "learning_rate": 2.0881232876712328e-05,
+      "loss": 0.5781,
+      "step": 22700
+    },
+    {
+      "epoch": 31.02,
+      "learning_rate": 2.0840136986301368e-05,
+      "loss": 0.5738,
+      "step": 22800
+    },
+    {
+      "epoch": 31.16,
+      "learning_rate": 2.0799041095890412e-05,
+      "loss": 0.5634,
+      "step": 22900
+    },
+    {
+      "epoch": 31.29,
+      "learning_rate": 2.0757945205479452e-05,
+      "loss": 0.5595,
+      "step": 23000
+    },
+    {
+      "epoch": 31.29,
+      "eval_cer": 0.06162409882587804,
+      "eval_loss": 0.2539171874523163,
+      "eval_runtime": 172.2419,
+      "eval_samples_per_second": 33.586,
+      "eval_steps_per_second": 4.203,
+      "eval_wer": 0.30871863724091636,
+      "step": 23000
+    },
+    {
+      "epoch": 31.43,
+      "learning_rate": 2.0716849315068496e-05,
+      "loss": 0.5787,
+      "step": 23100
+    },
+    {
+      "epoch": 31.56,
+      "learning_rate": 2.0675753424657533e-05,
+      "loss": 0.5554,
+      "step": 23200
+    },
+    {
+      "epoch": 31.7,
+      "learning_rate": 2.0634657534246577e-05,
+      "loss": 0.5579,
+      "step": 23300
+    },
+    {
+      "epoch": 31.84,
+      "learning_rate": 2.0593972602739727e-05,
+      "loss": 0.5793,
+      "step": 23400
+    },
+    {
+      "epoch": 31.97,
+      "learning_rate": 2.0552876712328768e-05,
+      "loss": 0.5664,
+      "step": 23500
+    },
+    {
+      "epoch": 31.97,
+      "eval_cer": 0.06100844070920901,
+      "eval_loss": 0.2503504753112793,
+      "eval_runtime": 173.2441,
+      "eval_samples_per_second": 33.392,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.30695644877066375,
+      "step": 23500
+    },
+    {
+      "epoch": 32.11,
+      "learning_rate": 2.0511780821917808e-05,
+      "loss": 0.5636,
+      "step": 23600
+    },
+    {
+      "epoch": 32.24,
+      "learning_rate": 2.047068493150685e-05,
+      "loss": 0.5659,
+      "step": 23700
+    },
+    {
+      "epoch": 32.38,
+      "learning_rate": 2.0429589041095892e-05,
+      "loss": 0.5583,
+      "step": 23800
+    },
+    {
+      "epoch": 32.52,
+      "learning_rate": 2.0388493150684932e-05,
+      "loss": 0.5721,
+      "step": 23900
+    },
+    {
+      "epoch": 32.65,
+      "learning_rate": 2.0347397260273973e-05,
+      "loss": 0.5608,
+      "step": 24000
+    },
+    {
+      "epoch": 32.65,
+      "eval_cer": 0.06109587144767088,
+      "eval_loss": 0.2535480260848999,
+      "eval_runtime": 172.114,
+      "eval_samples_per_second": 33.611,
+      "eval_steps_per_second": 4.207,
+      "eval_wer": 0.30664876506950856,
+      "step": 24000
+    },
+    {
+      "epoch": 32.79,
+      "learning_rate": 2.0306301369863013e-05,
+      "loss": 0.5585,
+      "step": 24100
+    },
+    {
+      "epoch": 32.92,
+      "learning_rate": 2.0265205479452057e-05,
+      "loss": 0.5556,
+      "step": 24200
+    },
+    {
+      "epoch": 33.06,
+      "learning_rate": 2.0224109589041097e-05,
+      "loss": 0.564,
+      "step": 24300
+    },
+    {
+      "epoch": 33.2,
+      "learning_rate": 2.0183013698630138e-05,
+      "loss": 0.5563,
+      "step": 24400
+    },
+    {
+      "epoch": 33.33,
+      "learning_rate": 2.0141917808219178e-05,
+      "loss": 0.5557,
+      "step": 24500
+    },
+    {
+      "epoch": 33.33,
+      "eval_cer": 0.060757077336131116,
+      "eval_loss": 0.2538328170776367,
+      "eval_runtime": 174.344,
+      "eval_samples_per_second": 33.182,
+      "eval_steps_per_second": 4.153,
+      "eval_wer": 0.3047187491258986,
+      "step": 24500
+    },
+    {
+      "epoch": 33.47,
+      "learning_rate": 2.0100821917808218e-05,
+      "loss": 0.5499,
+      "step": 24600
+    },
+    {
+      "epoch": 33.6,
+      "learning_rate": 2.0059726027397262e-05,
+      "loss": 0.553,
+      "step": 24700
+    },
+    {
+      "epoch": 33.74,
+      "learning_rate": 2.0018630136986302e-05,
+      "loss": 0.5634,
+      "step": 24800
+    },
+    {
+      "epoch": 33.88,
+      "learning_rate": 1.9977534246575343e-05,
+      "loss": 0.5565,
+      "step": 24900
+    },
+    {
+      "epoch": 34.01,
+      "learning_rate": 1.9936438356164383e-05,
+      "loss": 0.5741,
+      "step": 25000
+    },
+    {
+      "epoch": 34.01,
+      "eval_cer": 0.05957676236689581,
+      "eval_loss": 0.24800847470760345,
+      "eval_runtime": 172.8392,
+      "eval_samples_per_second": 33.47,
+      "eval_steps_per_second": 4.189,
+      "eval_wer": 0.3008866884842382,
+      "step": 25000
+    },
+    {
+      "epoch": 34.15,
+      "learning_rate": 1.9895342465753427e-05,
+      "loss": 0.5386,
+      "step": 25100
+    },
+    {
+      "epoch": 34.29,
+      "learning_rate": 1.9854657534246577e-05,
+      "loss": 0.5617,
+      "step": 25200
+    },
+    {
+      "epoch": 34.42,
+      "learning_rate": 1.9813561643835614e-05,
+      "loss": 0.5542,
+      "step": 25300
+    },
+    {
+      "epoch": 34.56,
+      "learning_rate": 1.9772465753424658e-05,
+      "loss": 0.5577,
+      "step": 25400
+    },
+    {
+      "epoch": 34.69,
+      "learning_rate": 1.97313698630137e-05,
+      "loss": 0.5614,
+      "step": 25500
+    },
+    {
+      "epoch": 34.69,
+      "eval_cer": 0.06007948911305159,
+      "eval_loss": 0.2516026198863983,
+      "eval_runtime": 173.6005,
+      "eval_samples_per_second": 33.324,
+      "eval_steps_per_second": 4.17,
+      "eval_wer": 0.30329221560236075,
+      "step": 25500
+    },
+    {
+      "epoch": 34.83,
+      "learning_rate": 1.9690273972602742e-05,
+      "loss": 0.5467,
+      "step": 25600
+    },
+    {
+      "epoch": 34.97,
+      "learning_rate": 1.9649178082191782e-05,
+      "loss": 0.555,
+      "step": 25700
+    },
+    {
+      "epoch": 35.1,
+      "learning_rate": 1.9608082191780823e-05,
+      "loss": 0.5634,
+      "step": 25800
+    },
+    {
+      "epoch": 35.24,
+      "learning_rate": 1.9566986301369863e-05,
+      "loss": 0.5521,
+      "step": 25900
+    },
+    {
+      "epoch": 35.37,
+      "learning_rate": 1.9525890410958903e-05,
+      "loss": 0.5436,
+      "step": 26000
+    },
+    {
+      "epoch": 35.37,
+      "eval_cer": 0.060064917323307945,
+      "eval_loss": 0.2540014684200287,
+      "eval_runtime": 172.7496,
+      "eval_samples_per_second": 33.488,
+      "eval_steps_per_second": 4.191,
+      "eval_wer": 0.3003552348186065,
+      "step": 26000
+    },
+    {
+      "epoch": 35.51,
+      "learning_rate": 1.9484794520547947e-05,
+      "loss": 0.5344,
+      "step": 26100
+    },
+    {
+      "epoch": 35.65,
+      "learning_rate": 1.9443698630136984e-05,
+      "loss": 0.5437,
+      "step": 26200
+    },
+    {
+      "epoch": 35.78,
+      "learning_rate": 1.9402602739726028e-05,
+      "loss": 0.5534,
+      "step": 26300
+    },
+    {
+      "epoch": 35.92,
+      "learning_rate": 1.9361506849315068e-05,
+      "loss": 0.5542,
+      "step": 26400
+    },
+    {
+      "epoch": 36.05,
+      "learning_rate": 1.9320410958904112e-05,
+      "loss": 0.555,
+      "step": 26500
+    },
+    {
+      "epoch": 36.05,
+      "eval_cer": 0.059533046997664874,
+      "eval_loss": 0.248551145195961,
+      "eval_runtime": 173.6282,
+      "eval_samples_per_second": 33.318,
+      "eval_steps_per_second": 4.17,
+      "eval_wer": 0.29926435624178344,
+      "step": 26500
+    },
+    {
+      "epoch": 36.19,
+      "learning_rate": 1.9279315068493152e-05,
+      "loss": 0.5462,
+      "step": 26600
+    },
+    {
+      "epoch": 36.33,
+      "learning_rate": 1.9238219178082192e-05,
+      "loss": 0.5573,
+      "step": 26700
+    },
+    {
+      "epoch": 36.46,
+      "learning_rate": 1.9197123287671233e-05,
+      "loss": 0.549,
+      "step": 26800
+    },
+    {
+      "epoch": 36.6,
+      "learning_rate": 1.9156027397260273e-05,
+      "loss": 0.5443,
+      "step": 26900
+    },
+    {
+      "epoch": 36.73,
+      "learning_rate": 1.9114931506849317e-05,
+      "loss": 0.5474,
+      "step": 27000
+    },
+    {
+      "epoch": 36.73,
+      "eval_cer": 0.05978441037074276,
+      "eval_loss": 0.2535877823829651,
+      "eval_runtime": 172.1924,
+      "eval_samples_per_second": 33.596,
+      "eval_steps_per_second": 4.205,
+      "eval_wer": 0.3003272635730469,
+      "step": 27000
+    },
+    {
+      "epoch": 36.87,
+      "learning_rate": 1.9073835616438354e-05,
+      "loss": 0.5453,
+      "step": 27100
+    },
+    {
+      "epoch": 37.01,
+      "learning_rate": 1.9032739726027398e-05,
+      "loss": 0.5447,
+      "step": 27200
+    },
+    {
+      "epoch": 37.14,
+      "learning_rate": 1.8991643835616438e-05,
+      "loss": 0.5458,
+      "step": 27300
+    },
+    {
+      "epoch": 37.28,
+      "learning_rate": 1.895054794520548e-05,
+      "loss": 0.5516,
+      "step": 27400
+    },
+    {
+      "epoch": 37.41,
+      "learning_rate": 1.8909452054794522e-05,
+      "loss": 0.5352,
+      "step": 27500
+    },
+    {
+      "epoch": 37.41,
+      "eval_cer": 0.05966055015792177,
+      "eval_loss": 0.2589070796966553,
+      "eval_runtime": 172.1763,
+      "eval_samples_per_second": 33.599,
+      "eval_steps_per_second": 4.205,
+      "eval_wer": 0.2985930463483539,
+      "step": 27500
+    },
+    {
+      "epoch": 37.55,
+      "learning_rate": 1.8868356164383562e-05,
+      "loss": 0.5552,
+      "step": 27600
+    },
+    {
+      "epoch": 37.69,
+      "learning_rate": 1.8827260273972603e-05,
+      "loss": 0.534,
+      "step": 27700
+    },
+    {
+      "epoch": 37.82,
+      "learning_rate": 1.8786164383561646e-05,
+      "loss": 0.5425,
+      "step": 27800
+    },
+    {
+      "epoch": 37.96,
+      "learning_rate": 1.8745068493150687e-05,
+      "loss": 0.5265,
+      "step": 27900
+    },
+    {
+      "epoch": 38.1,
+      "learning_rate": 1.8703972602739724e-05,
+      "loss": 0.5489,
+      "step": 28000
+    },
+    {
+      "epoch": 38.1,
+      "eval_cer": 0.05855309413740469,
+      "eval_loss": 0.24846366047859192,
+      "eval_runtime": 173.2453,
+      "eval_samples_per_second": 33.392,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.2924673435708092,
+      "step": 28000
+    },
+    {
+      "epoch": 38.23,
+      "learning_rate": 1.8662876712328767e-05,
+      "loss": 0.5398,
+      "step": 28100
+    },
+    {
+      "epoch": 38.37,
+      "learning_rate": 1.8621780821917808e-05,
+      "loss": 0.5359,
+      "step": 28200
+    },
+    {
+      "epoch": 38.5,
+      "learning_rate": 1.858068493150685e-05,
+      "loss": 0.5382,
+      "step": 28300
+    },
+    {
+      "epoch": 38.64,
+      "learning_rate": 1.853958904109589e-05,
+      "loss": 0.55,
+      "step": 28400
+    },
+    {
+      "epoch": 38.77,
+      "learning_rate": 1.8498493150684932e-05,
+      "loss": 0.5438,
+      "step": 28500
+    },
+    {
+      "epoch": 38.77,
+      "eval_cer": 0.058061296233556645,
+      "eval_loss": 0.2500339150428772,
+      "eval_runtime": 172.9708,
+      "eval_samples_per_second": 33.445,
+      "eval_steps_per_second": 4.186,
+      "eval_wer": 0.29076109759167573,
+      "step": 28500
+    },
+    {
+      "epoch": 38.91,
+      "learning_rate": 1.8457397260273972e-05,
+      "loss": 0.5322,
+      "step": 28600
+    },
+    {
+      "epoch": 39.05,
+      "learning_rate": 1.8416301369863016e-05,
+      "loss": 0.547,
+      "step": 28700
+    },
+    {
+      "epoch": 39.18,
+      "learning_rate": 1.8375205479452056e-05,
+      "loss": 0.5283,
+      "step": 28800
+    },
+    {
+      "epoch": 39.32,
+      "learning_rate": 1.8334109589041093e-05,
+      "loss": 0.5351,
+      "step": 28900
+    },
+    {
+      "epoch": 39.46,
+      "learning_rate": 1.8293013698630137e-05,
+      "loss": 0.541,
+      "step": 29000
+    },
+    {
+      "epoch": 39.46,
+      "eval_cer": 0.057660572015606384,
+      "eval_loss": 0.24505586922168732,
+      "eval_runtime": 172.8598,
+      "eval_samples_per_second": 33.466,
+      "eval_steps_per_second": 4.188,
+      "eval_wer": 0.2879360017901597,
+      "step": 29000
+    },
+    {
+      "epoch": 39.59,
+      "learning_rate": 1.8251917808219177e-05,
+      "loss": 0.529,
+      "step": 29100
+    },
+    {
+      "epoch": 39.73,
+      "learning_rate": 1.821082191780822e-05,
+      "loss": 0.5343,
+      "step": 29200
+    },
+    {
+      "epoch": 39.86,
+      "learning_rate": 1.817013698630137e-05,
+      "loss": 0.531,
+      "step": 29300
+    },
+    {
+      "epoch": 40.0,
+      "learning_rate": 1.8129041095890412e-05,
+      "loss": 0.5365,
+      "step": 29400
+    },
+    {
+      "epoch": 40.14,
+      "learning_rate": 1.8087945205479452e-05,
+      "loss": 0.5462,
+      "step": 29500
+    },
+    {
+      "epoch": 40.14,
+      "eval_cer": 0.058050367391248914,
+      "eval_loss": 0.2509716749191284,
+      "eval_runtime": 173.388,
+      "eval_samples_per_second": 33.364,
+      "eval_steps_per_second": 4.176,
+      "eval_wer": 0.29347430841095357,
+      "step": 29500
+    },
+    {
+      "epoch": 40.27,
+      "learning_rate": 1.8046849315068493e-05,
+      "loss": 0.5332,
+      "step": 29600
+    },
+    {
+      "epoch": 40.41,
+      "learning_rate": 1.8005753424657537e-05,
+      "loss": 0.534,
+      "step": 29700
+    },
+    {
+      "epoch": 40.54,
+      "learning_rate": 1.7964657534246573e-05,
+      "loss": 0.526,
+      "step": 29800
+    },
+    {
+      "epoch": 40.68,
+      "learning_rate": 1.7923561643835617e-05,
+      "loss": 0.523,
+      "step": 29900
+    },
+    {
+      "epoch": 40.82,
+      "learning_rate": 1.7882465753424657e-05,
+      "loss": 0.529,
+      "step": 30000
+    },
+    {
+      "epoch": 40.82,
+      "eval_cer": 0.05748206759124672,
+      "eval_loss": 0.24347859621047974,
+      "eval_runtime": 173.8958,
+      "eval_samples_per_second": 33.267,
+      "eval_steps_per_second": 4.163,
+      "eval_wer": 0.2878800592990406,
+      "step": 30000
+    },
+    {
+      "epoch": 40.95,
+      "learning_rate": 1.78413698630137e-05,
+      "loss": 0.5513,
+      "step": 30100
+    },
+    {
+      "epoch": 41.09,
+      "learning_rate": 1.7800273972602738e-05,
+      "loss": 0.5342,
+      "step": 30200
+    },
+    {
+      "epoch": 41.22,
+      "learning_rate": 1.7759178082191782e-05,
+      "loss": 0.5255,
+      "step": 30300
+    },
+    {
+      "epoch": 41.36,
+      "learning_rate": 1.7718082191780822e-05,
+      "loss": 0.5303,
+      "step": 30400
+    },
+    {
+      "epoch": 41.5,
+      "learning_rate": 1.7676986301369863e-05,
+      "loss": 0.5169,
+      "step": 30500
+    },
+    {
+      "epoch": 41.5,
+      "eval_cer": 0.057227061270732925,
+      "eval_loss": 0.2474348396062851,
+      "eval_runtime": 173.2571,
+      "eval_samples_per_second": 33.39,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.2860339570921093,
+      "step": 30500
+    },
+    {
+      "epoch": 41.63,
+      "learning_rate": 1.7635890410958906e-05,
+      "loss": 0.5259,
+      "step": 30600
+    },
+    {
+      "epoch": 41.77,
+      "learning_rate": 1.7594794520547943e-05,
+      "loss": 0.5313,
+      "step": 30700
+    },
+    {
+      "epoch": 41.9,
+      "learning_rate": 1.7553698630136987e-05,
+      "loss": 0.5214,
+      "step": 30800
+    },
+    {
+      "epoch": 42.04,
+      "learning_rate": 1.7512602739726027e-05,
+      "loss": 0.5322,
+      "step": 30900
+    },
+    {
+      "epoch": 42.18,
+      "learning_rate": 1.747150684931507e-05,
+      "loss": 0.5281,
+      "step": 31000
+    },
+    {
+      "epoch": 42.18,
+      "eval_cer": 0.057533068855349485,
+      "eval_loss": 0.247803196310997,
+      "eval_runtime": 173.4991,
+      "eval_samples_per_second": 33.343,
+      "eval_steps_per_second": 4.173,
+      "eval_wer": 0.28838354171911273,
+      "step": 31000
+    },
+    {
+      "epoch": 42.31,
+      "learning_rate": 1.7430410958904108e-05,
+      "loss": 0.5398,
+      "step": 31100
+    },
+    {
+      "epoch": 42.45,
+      "learning_rate": 1.738931506849315e-05,
+      "loss": 0.5217,
+      "step": 31200
+    },
+    {
+      "epoch": 42.58,
+      "learning_rate": 1.7348219178082192e-05,
+      "loss": 0.5252,
+      "step": 31300
+    },
+    {
+      "epoch": 42.72,
+      "learning_rate": 1.7307534246575343e-05,
+      "loss": 0.5171,
+      "step": 31400
+    },
+    {
+      "epoch": 42.86,
+      "learning_rate": 1.7266438356164383e-05,
+      "loss": 0.527,
+      "step": 31500
+    },
+    {
+      "epoch": 42.86,
+      "eval_cer": 0.05683362294765449,
+      "eval_loss": 0.2491973340511322,
+      "eval_runtime": 172.7352,
+      "eval_samples_per_second": 33.491,
+      "eval_steps_per_second": 4.191,
+      "eval_wer": 0.28452350983189284,
+      "step": 31500
+    },
+    {
+      "epoch": 42.99,
+      "learning_rate": 1.7225342465753423e-05,
+      "loss": 0.5211,
+      "step": 31600
+    },
+    {
+      "epoch": 43.13,
+      "learning_rate": 1.7184246575342467e-05,
+      "loss": 0.5275,
+      "step": 31700
+    },
+    {
+      "epoch": 43.27,
+      "learning_rate": 1.7143150684931507e-05,
+      "loss": 0.5276,
+      "step": 31800
+    },
+    {
+      "epoch": 43.4,
+      "learning_rate": 1.710205479452055e-05,
+      "loss": 0.5265,
+      "step": 31900
+    },
+    {
+      "epoch": 43.54,
+      "learning_rate": 1.7060958904109588e-05,
+      "loss": 0.5172,
+      "step": 32000
+    },
+    {
+      "epoch": 43.54,
+      "eval_cer": 0.057522140013041755,
+      "eval_loss": 0.24510592222213745,
+      "eval_runtime": 173.916,
+      "eval_samples_per_second": 33.263,
+      "eval_steps_per_second": 4.163,
+      "eval_wer": 0.28846745545579144,
+      "step": 32000
+    },
+    {
+      "epoch": 43.67,
+      "learning_rate": 1.7019863013698632e-05,
+      "loss": 0.5128,
+      "step": 32100
+    },
+    {
+      "epoch": 43.81,
+      "learning_rate": 1.6978767123287672e-05,
+      "loss": 0.5252,
+      "step": 32200
+    },
+    {
+      "epoch": 43.94,
+      "learning_rate": 1.6937671232876712e-05,
+      "loss": 0.5191,
+      "step": 32300
+    },
+    {
+      "epoch": 44.08,
+      "learning_rate": 1.6896575342465753e-05,
+      "loss": 0.5243,
+      "step": 32400
+    },
+    {
+      "epoch": 44.22,
+      "learning_rate": 1.6855479452054793e-05,
+      "loss": 0.5154,
+      "step": 32500
+    },
+    {
+      "epoch": 44.22,
+      "eval_cer": 0.05742013748483623,
+      "eval_loss": 0.24902237951755524,
+      "eval_runtime": 173.7736,
+      "eval_samples_per_second": 33.29,
+      "eval_steps_per_second": 4.166,
+      "eval_wer": 0.28729266314228974,
+      "step": 32500
+    },
+    {
+      "epoch": 44.35,
+      "learning_rate": 1.6814383561643837e-05,
+      "loss": 0.5104,
+      "step": 32600
+    },
+    {
+      "epoch": 44.49,
+      "learning_rate": 1.6773287671232877e-05,
+      "loss": 0.5134,
+      "step": 32700
+    },
+    {
+      "epoch": 44.63,
+      "learning_rate": 1.673219178082192e-05,
+      "loss": 0.523,
+      "step": 32800
+    },
+    {
+      "epoch": 44.76,
+      "learning_rate": 1.6691095890410958e-05,
+      "loss": 0.5105,
+      "step": 32900
+    },
+    {
+      "epoch": 44.9,
+      "learning_rate": 1.665e-05,
+      "loss": 0.5129,
+      "step": 33000
+    },
+    {
+      "epoch": 44.9,
+      "eval_cer": 0.05689191010662907,
+      "eval_loss": 0.24455302953720093,
+      "eval_runtime": 173.1245,
+      "eval_samples_per_second": 33.415,
+      "eval_steps_per_second": 4.182,
+      "eval_wer": 0.28527873346200106,
+      "step": 33000
+    },
+    {
+      "epoch": 45.03,
+      "learning_rate": 1.6608904109589042e-05,
+      "loss": 0.5144,
+      "step": 33100
+    },
+    {
+      "epoch": 45.17,
+      "learning_rate": 1.6567808219178082e-05,
+      "loss": 0.512,
+      "step": 33200
+    },
+    {
+      "epoch": 45.31,
+      "learning_rate": 1.6526712328767123e-05,
+      "loss": 0.5043,
+      "step": 33300
+    },
+    {
+      "epoch": 45.44,
+      "learning_rate": 1.6485616438356163e-05,
+      "loss": 0.5077,
+      "step": 33400
+    },
+    {
+      "epoch": 45.58,
+      "learning_rate": 1.6444520547945207e-05,
+      "loss": 0.5075,
+      "step": 33500
+    },
+    {
+      "epoch": 45.58,
+      "eval_cer": 0.056531258310473835,
+      "eval_loss": 0.24853074550628662,
+      "eval_runtime": 174.3487,
+      "eval_samples_per_second": 33.181,
+      "eval_steps_per_second": 4.153,
+      "eval_wer": 0.28281726385275935,
+      "step": 33500
+    },
+    {
+      "epoch": 45.71,
+      "learning_rate": 1.6403835616438357e-05,
+      "loss": 0.5256,
+      "step": 33600
+    },
+    {
+      "epoch": 45.85,
+      "learning_rate": 1.6362739726027398e-05,
+      "loss": 0.5199,
+      "step": 33700
+    },
+    {
+      "epoch": 45.99,
+      "learning_rate": 1.6321643835616438e-05,
+      "loss": 0.5362,
+      "step": 33800
+    },
+    {
+      "epoch": 46.12,
+      "learning_rate": 1.628054794520548e-05,
+      "loss": 0.517,
+      "step": 33900
+    },
+    {
+      "epoch": 46.26,
+      "learning_rate": 1.6239452054794522e-05,
+      "loss": 0.5077,
+      "step": 34000
+    },
+    {
+      "epoch": 46.26,
+      "eval_cer": 0.05590102840406116,
+      "eval_loss": 0.24523869156837463,
+      "eval_runtime": 173.6492,
+      "eval_samples_per_second": 33.314,
+      "eval_steps_per_second": 4.169,
+      "eval_wer": 0.28074739168135154,
+      "step": 34000
+    },
+    {
+      "epoch": 46.39,
+      "learning_rate": 1.6198356164383562e-05,
+      "loss": 0.513,
+      "step": 34100
+    },
+    {
+      "epoch": 46.53,
+      "learning_rate": 1.6157260273972603e-05,
+      "loss": 0.5192,
+      "step": 34200
+    },
+    {
+      "epoch": 46.67,
+      "learning_rate": 1.6116164383561643e-05,
+      "loss": 0.5071,
+      "step": 34300
+    },
+    {
+      "epoch": 46.8,
+      "learning_rate": 1.6075068493150687e-05,
+      "loss": 0.5207,
+      "step": 34400
+    },
+    {
+      "epoch": 46.94,
+      "learning_rate": 1.6033972602739727e-05,
+      "loss": 0.5004,
+      "step": 34500
+    },
+    {
+      "epoch": 46.94,
+      "eval_cer": 0.05724527600791248,
+      "eval_loss": 0.2500554025173187,
+      "eval_runtime": 174.0039,
+      "eval_samples_per_second": 33.246,
+      "eval_steps_per_second": 4.161,
+      "eval_wer": 0.28815977175463625,
+      "step": 34500
+    },
+    {
+      "epoch": 47.07,
+      "learning_rate": 1.5992876712328767e-05,
+      "loss": 0.5034,
+      "step": 34600
+    },
+    {
+      "epoch": 47.21,
+      "learning_rate": 1.5951780821917808e-05,
+      "loss": 0.5023,
+      "step": 34700
+    },
+    {
+      "epoch": 47.35,
+      "learning_rate": 1.5910684931506848e-05,
+      "loss": 0.5056,
+      "step": 34800
+    },
+    {
+      "epoch": 47.48,
+      "learning_rate": 1.5869589041095892e-05,
+      "loss": 0.515,
+      "step": 34900
+    },
+    {
+      "epoch": 47.62,
+      "learning_rate": 1.5828493150684932e-05,
+      "loss": 0.5319,
+      "step": 35000
+    },
+    {
+      "epoch": 47.62,
+      "eval_cer": 0.057460209906631254,
+      "eval_loss": 0.2516151964664459,
+      "eval_runtime": 174.3742,
+      "eval_samples_per_second": 33.176,
+      "eval_steps_per_second": 4.152,
+      "eval_wer": 0.2856423596542754,
+      "step": 35000
+    },
+    {
+      "epoch": 47.75,
+      "learning_rate": 1.5787397260273972e-05,
+      "loss": 0.5089,
+      "step": 35100
+    },
+    {
+      "epoch": 47.89,
+      "learning_rate": 1.5746301369863013e-05,
+      "loss": 0.5233,
+      "step": 35200
+    },
+    {
+      "epoch": 48.03,
+      "learning_rate": 1.5705205479452056e-05,
+      "loss": 0.5118,
+      "step": 35300
+    },
+    {
+      "epoch": 48.16,
+      "learning_rate": 1.5664109589041097e-05,
+      "loss": 0.5121,
+      "step": 35400
+    },
+    {
+      "epoch": 48.3,
+      "learning_rate": 1.5623013698630137e-05,
+      "loss": 0.4956,
+      "step": 35500
+    },
+    {
+      "epoch": 48.3,
+      "eval_cer": 0.056662404418166654,
+      "eval_loss": 0.2495414763689041,
+      "eval_runtime": 173.2309,
+      "eval_samples_per_second": 33.395,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.28211798271377025,
+      "step": 35500
+    },
+    {
+      "epoch": 48.43,
+      "learning_rate": 1.5581917808219177e-05,
+      "loss": 0.5238,
+      "step": 35600
+    },
+    {
+      "epoch": 48.57,
+      "learning_rate": 1.554082191780822e-05,
+      "loss": 0.4927,
+      "step": 35700
+    },
+    {
+      "epoch": 48.71,
+      "learning_rate": 1.549972602739726e-05,
+      "loss": 0.5027,
+      "step": 35800
+    },
+    {
+      "epoch": 48.84,
+      "learning_rate": 1.545904109589041e-05,
+      "loss": 0.5164,
+      "step": 35900
+    },
+    {
+      "epoch": 48.98,
+      "learning_rate": 1.5417945205479453e-05,
+      "loss": 0.5053,
+      "step": 36000
+    },
+    {
+      "epoch": 48.98,
+      "eval_cer": 0.05651668652073019,
+      "eval_loss": 0.24815048277378082,
+      "eval_runtime": 174.3807,
+      "eval_samples_per_second": 33.175,
+      "eval_steps_per_second": 4.152,
+      "eval_wer": 0.28245363766048504,
+      "step": 36000
+    },
+    {
+      "epoch": 49.12,
+      "learning_rate": 1.5376849315068493e-05,
+      "loss": 0.5038,
+      "step": 36100
+    },
+    {
+      "epoch": 49.25,
+      "learning_rate": 1.5335753424657537e-05,
+      "loss": 0.5158,
+      "step": 36200
+    },
+    {
+      "epoch": 49.39,
+      "learning_rate": 1.5294657534246577e-05,
+      "loss": 0.4995,
+      "step": 36300
+    },
+    {
+      "epoch": 49.52,
+      "learning_rate": 1.5253561643835619e-05,
+      "loss": 0.5088,
+      "step": 36400
+    },
+    {
+      "epoch": 49.66,
+      "learning_rate": 1.5212465753424658e-05,
+      "loss": 0.5014,
+      "step": 36500
+    },
+    {
+      "epoch": 49.66,
+      "eval_cer": 0.0559156001938048,
+      "eval_loss": 0.24410146474838257,
+      "eval_runtime": 172.7818,
+      "eval_samples_per_second": 33.482,
+      "eval_steps_per_second": 4.19,
+      "eval_wer": 0.28077536292691113,
+      "step": 36500
+    },
+    {
+      "epoch": 49.8,
+      "learning_rate": 1.5171369863013698e-05,
+      "loss": 0.4999,
+      "step": 36600
+    },
+    {
+      "epoch": 49.93,
+      "learning_rate": 1.513027397260274e-05,
+      "loss": 0.4958,
+      "step": 36700
+    },
+    {
+      "epoch": 50.07,
+      "learning_rate": 1.508917808219178e-05,
+      "loss": 0.5001,
+      "step": 36800
+    },
+    {
+      "epoch": 50.2,
+      "learning_rate": 1.5048082191780822e-05,
+      "loss": 0.5027,
+      "step": 36900
+    },
+    {
+      "epoch": 50.34,
+      "learning_rate": 1.5006986301369863e-05,
+      "loss": 0.4945,
+      "step": 37000
+    },
+    {
+      "epoch": 50.34,
+      "eval_cer": 0.05617424946175452,
+      "eval_loss": 0.24599410593509674,
+      "eval_runtime": 173.1481,
+      "eval_samples_per_second": 33.411,
+      "eval_steps_per_second": 4.181,
+      "eval_wer": 0.2806914491902324,
+      "step": 37000
+    },
+    {
+      "epoch": 50.48,
+      "learning_rate": 1.4966301369863013e-05,
+      "loss": 0.5061,
+      "step": 37100
+    },
+    {
+      "epoch": 50.61,
+      "learning_rate": 1.4925205479452055e-05,
+      "loss": 0.4998,
+      "step": 37200
+    },
+    {
+      "epoch": 50.75,
+      "learning_rate": 1.4884109589041096e-05,
+      "loss": 0.5158,
+      "step": 37300
+    },
+    {
+      "epoch": 50.88,
+      "learning_rate": 1.4843013698630138e-05,
+      "loss": 0.5096,
+      "step": 37400
+    },
+    {
+      "epoch": 51.02,
+      "learning_rate": 1.480191780821918e-05,
+      "loss": 0.51,
+      "step": 37500
+    },
+    {
+      "epoch": 51.02,
+      "eval_cer": 0.05470249869764629,
+      "eval_loss": 0.24342398345470428,
+      "eval_runtime": 173.6876,
+      "eval_samples_per_second": 33.307,
+      "eval_steps_per_second": 4.168,
+      "eval_wer": 0.27409023523817516,
+      "step": 37500
+    },
+    {
+      "epoch": 51.16,
+      "learning_rate": 1.4760821917808218e-05,
+      "loss": 0.5041,
+      "step": 37600
+    },
+    {
+      "epoch": 51.29,
+      "learning_rate": 1.471972602739726e-05,
+      "loss": 0.4983,
+      "step": 37700
+    },
+    {
+      "epoch": 51.43,
+      "learning_rate": 1.46786301369863e-05,
+      "loss": 0.4929,
+      "step": 37800
+    },
+    {
+      "epoch": 51.56,
+      "learning_rate": 1.4637534246575343e-05,
+      "loss": 0.492,
+      "step": 37900
+    },
+    {
+      "epoch": 51.7,
+      "learning_rate": 1.4596438356164383e-05,
+      "loss": 0.5095,
+      "step": 38000
+    },
+    {
+      "epoch": 51.7,
+      "eval_cer": 0.055824526507907014,
+      "eval_loss": 0.24342697858810425,
+      "eval_runtime": 173.4435,
+      "eval_samples_per_second": 33.354,
+      "eval_steps_per_second": 4.174,
+      "eval_wer": 0.2789572319655394,
+      "step": 38000
+    },
+    {
+      "epoch": 51.84,
+      "learning_rate": 1.4555342465753425e-05,
+      "loss": 0.4991,
+      "step": 38100
+    },
+    {
+      "epoch": 51.97,
+      "learning_rate": 1.4514246575342466e-05,
+      "loss": 0.4944,
+      "step": 38200
+    },
+    {
+      "epoch": 52.11,
+      "learning_rate": 1.4473150684931508e-05,
+      "loss": 0.4887,
+      "step": 38300
+    },
+    {
+      "epoch": 52.24,
+      "learning_rate": 1.443205479452055e-05,
+      "loss": 0.4987,
+      "step": 38400
+    },
+    {
+      "epoch": 52.38,
+      "learning_rate": 1.439095890410959e-05,
+      "loss": 0.5026,
+      "step": 38500
+    },
+    {
+      "epoch": 52.38,
+      "eval_cer": 0.055966601457907564,
+      "eval_loss": 0.2477608025074005,
+      "eval_runtime": 173.5664,
+      "eval_samples_per_second": 33.33,
+      "eval_steps_per_second": 4.171,
+      "eval_wer": 0.2787054907555033,
+      "step": 38500
+    },
+    {
+      "epoch": 52.52,
+      "learning_rate": 1.434986301369863e-05,
+      "loss": 0.4932,
+      "step": 38600
+    },
+    {
+      "epoch": 52.65,
+      "learning_rate": 1.430876712328767e-05,
+      "loss": 0.4985,
+      "step": 38700
+    },
+    {
+      "epoch": 52.79,
+      "learning_rate": 1.4267671232876713e-05,
+      "loss": 0.4933,
+      "step": 38800
+    },
+    {
+      "epoch": 52.92,
+      "learning_rate": 1.4226575342465753e-05,
+      "loss": 0.4926,
+      "step": 38900
+    },
+    {
+      "epoch": 53.06,
+      "learning_rate": 1.4185479452054795e-05,
+      "loss": 0.5081,
+      "step": 39000
+    },
+    {
+      "epoch": 53.06,
+      "eval_cer": 0.056593188416884335,
+      "eval_loss": 0.2484765350818634,
+      "eval_runtime": 173.2703,
+      "eval_samples_per_second": 33.387,
+      "eval_steps_per_second": 4.178,
+      "eval_wer": 0.28209001146821067,
+      "step": 39000
+    },
+    {
+      "epoch": 53.2,
+      "learning_rate": 1.4144383561643835e-05,
+      "loss": 0.495,
+      "step": 39100
+    },
+    {
+      "epoch": 53.33,
+      "learning_rate": 1.4103287671232877e-05,
+      "loss": 0.4925,
+      "step": 39200
+    },
+    {
+      "epoch": 53.47,
+      "learning_rate": 1.406219178082192e-05,
+      "loss": 0.4795,
+      "step": 39300
+    },
+    {
+      "epoch": 53.6,
+      "learning_rate": 1.402109589041096e-05,
+      "loss": 0.5017,
+      "step": 39400
+    },
+    {
+      "epoch": 53.74,
+      "learning_rate": 1.398041095890411e-05,
+      "loss": 0.5021,
+      "step": 39500
+    },
+    {
+      "epoch": 53.74,
+      "eval_cer": 0.05513600944251976,
+      "eval_loss": 0.24097368121147156,
+      "eval_runtime": 172.8291,
+      "eval_samples_per_second": 33.472,
+      "eval_steps_per_second": 4.189,
+      "eval_wer": 0.27520908506055775,
+      "step": 39500
+    },
+    {
+      "epoch": 53.88,
+      "learning_rate": 1.393931506849315e-05,
+      "loss": 0.4966,
+      "step": 39600
+    },
+    {
+      "epoch": 54.01,
+      "learning_rate": 1.3898219178082193e-05,
+      "loss": 0.505,
+      "step": 39700
+    },
+    {
+      "epoch": 54.15,
+      "learning_rate": 1.3857123287671233e-05,
+      "loss": 0.4864,
+      "step": 39800
+    },
+    {
+      "epoch": 54.29,
+      "learning_rate": 1.3816027397260275e-05,
+      "loss": 0.5004,
+      "step": 39900
+    },
+    {
+      "epoch": 54.42,
+      "learning_rate": 1.3774931506849315e-05,
+      "loss": 0.4945,
+      "step": 40000
+    },
+    {
+      "epoch": 54.42,
+      "eval_cer": 0.05519429660149434,
+      "eval_loss": 0.24358148872852325,
+      "eval_runtime": 173.1127,
+      "eval_samples_per_second": 33.418,
+      "eval_steps_per_second": 4.182,
+      "eval_wer": 0.2765796760929764,
+      "step": 40000
+    },
+    {
+      "epoch": 54.56,
+      "learning_rate": 1.3733835616438357e-05,
+      "loss": 0.4904,
+      "step": 40100
+    },
+    {
+      "epoch": 54.69,
+      "learning_rate": 1.3692739726027398e-05,
+      "loss": 0.496,
+      "step": 40200
+    },
+    {
+      "epoch": 54.83,
+      "learning_rate": 1.3651643835616438e-05,
+      "loss": 0.4893,
+      "step": 40300
+    },
+    {
+      "epoch": 54.97,
+      "learning_rate": 1.3610547945205478e-05,
+      "loss": 0.5016,
+      "step": 40400
+    },
+    {
+      "epoch": 55.1,
+      "learning_rate": 1.356945205479452e-05,
+      "loss": 0.4882,
+      "step": 40500
+    },
+    {
+      "epoch": 55.1,
+      "eval_cer": 0.055518518923290454,
+      "eval_loss": 0.24375249445438385,
+      "eval_runtime": 172.9154,
+      "eval_samples_per_second": 33.456,
+      "eval_steps_per_second": 4.187,
+      "eval_wer": 0.27685938854857206,
+      "step": 40500
+    },
+    {
+      "epoch": 55.24,
+      "learning_rate": 1.3528356164383563e-05,
+      "loss": 0.4938,
+      "step": 40600
+    },
+    {
+      "epoch": 55.37,
+      "learning_rate": 1.3487260273972603e-05,
+      "loss": 0.4989,
+      "step": 40700
+    },
+    {
+      "epoch": 55.51,
+      "learning_rate": 1.3446164383561645e-05,
+      "loss": 0.4957,
+      "step": 40800
+    },
+    {
+      "epoch": 55.65,
+      "learning_rate": 1.3405068493150685e-05,
+      "loss": 0.4963,
+      "step": 40900
+    },
+    {
+      "epoch": 55.78,
+      "learning_rate": 1.3363972602739727e-05,
+      "loss": 0.497,
+      "step": 41000
+    },
+    {
+      "epoch": 55.78,
+      "eval_cer": 0.05499757743995512,
+      "eval_loss": 0.24226950109004974,
+      "eval_runtime": 173.1033,
+      "eval_samples_per_second": 33.419,
+      "eval_steps_per_second": 4.182,
+      "eval_wer": 0.2757964812173086,
+      "step": 41000
+    },
+    {
+      "epoch": 55.92,
+      "learning_rate": 1.3322876712328768e-05,
+      "loss": 0.4895,
+      "step": 41100
+    },
+    {
+      "epoch": 56.05,
+      "learning_rate": 1.3281780821917808e-05,
+      "loss": 0.5024,
+      "step": 41200
+    },
+    {
+      "epoch": 56.19,
+      "learning_rate": 1.3240684931506848e-05,
+      "loss": 0.4848,
+      "step": 41300
+    },
+    {
+      "epoch": 56.33,
+      "learning_rate": 1.319958904109589e-05,
+      "loss": 0.4944,
+      "step": 41400
+    },
+    {
+      "epoch": 56.46,
+      "learning_rate": 1.3158493150684932e-05,
+      "loss": 0.4925,
+      "step": 41500
+    },
+    {
+      "epoch": 56.46,
+      "eval_cer": 0.05595931556303574,
+      "eval_loss": 0.24736058712005615,
+      "eval_runtime": 173.1065,
+      "eval_samples_per_second": 33.419,
+      "eval_steps_per_second": 4.182,
+      "eval_wer": 0.2790411457022181,
+      "step": 41500
+    },
+    {
+      "epoch": 56.6,
+      "learning_rate": 1.3117397260273973e-05,
+      "loss": 0.4815,
+      "step": 41600
+    },
+    {
+      "epoch": 56.73,
+      "learning_rate": 1.3076301369863015e-05,
+      "loss": 0.489,
+      "step": 41700
+    },
+    {
+      "epoch": 56.87,
+      "learning_rate": 1.3035205479452055e-05,
+      "loss": 0.4917,
+      "step": 41800
+    },
+    {
+      "epoch": 57.01,
+      "learning_rate": 1.2994109589041097e-05,
+      "loss": 0.4833,
+      "step": 41900
+    },
+    {
+      "epoch": 57.14,
+      "learning_rate": 1.2953013698630137e-05,
+      "loss": 0.4894,
+      "step": 42000
+    },
+    {
+      "epoch": 57.14,
+      "eval_cer": 0.055944743773292095,
+      "eval_loss": 0.24967217445373535,
+      "eval_runtime": 174.0793,
+      "eval_samples_per_second": 33.232,
+      "eval_steps_per_second": 4.159,
+      "eval_wer": 0.27971245559564767,
+      "step": 42000
+    },
+    {
+      "epoch": 57.28,
+      "learning_rate": 1.291191780821918e-05,
+      "loss": 0.4819,
+      "step": 42100
+    },
+    {
+      "epoch": 57.41,
+      "learning_rate": 1.2870821917808218e-05,
+      "loss": 0.4714,
+      "step": 42200
+    },
+    {
+      "epoch": 57.55,
+      "learning_rate": 1.282972602739726e-05,
+      "loss": 0.4868,
+      "step": 42300
+    },
+    {
+      "epoch": 57.69,
+      "learning_rate": 1.278904109589041e-05,
+      "loss": 0.5001,
+      "step": 42400
+    },
+    {
+      "epoch": 57.82,
+      "learning_rate": 1.2747945205479453e-05,
+      "loss": 0.4767,
+      "step": 42500
+    },
+    {
+      "epoch": 57.82,
+      "eval_cer": 0.055602306714316416,
+      "eval_loss": 0.25277969241142273,
+      "eval_runtime": 173.3781,
+      "eval_samples_per_second": 33.366,
+      "eval_steps_per_second": 4.176,
+      "eval_wer": 0.27999216805124333,
+      "step": 42500
+    },
+    {
+      "epoch": 57.96,
+      "learning_rate": 1.2706849315068493e-05,
+      "loss": 0.4886,
+      "step": 42600
+    },
+    {
+      "epoch": 58.1,
+      "learning_rate": 1.2665753424657535e-05,
+      "loss": 0.4922,
+      "step": 42700
+    },
+    {
+      "epoch": 58.23,
+      "learning_rate": 1.2624657534246577e-05,
+      "loss": 0.4815,
+      "step": 42800
+    },
+    {
+      "epoch": 58.37,
+      "learning_rate": 1.2583561643835616e-05,
+      "loss": 0.4785,
+      "step": 42900
+    },
+    {
+      "epoch": 58.5,
+      "learning_rate": 1.2542465753424658e-05,
+      "loss": 0.4796,
+      "step": 43000
+    },
+    {
+      "epoch": 58.5,
+      "eval_cer": 0.0549283614386728,
+      "eval_loss": 0.24632307887077332,
+      "eval_runtime": 173.4291,
+      "eval_samples_per_second": 33.357,
+      "eval_steps_per_second": 4.175,
+      "eval_wer": 0.2754608262705938,
+      "step": 43000
+    },
+    {
+      "epoch": 58.64,
+      "learning_rate": 1.2501369863013698e-05,
+      "loss": 0.4859,
+      "step": 43100
+    },
+    {
+      "epoch": 58.77,
+      "learning_rate": 1.246027397260274e-05,
+      "loss": 0.4821,
+      "step": 43200
+    },
+    {
+      "epoch": 58.91,
+      "learning_rate": 1.241917808219178e-05,
+      "loss": 0.4844,
+      "step": 43300
+    },
+    {
+      "epoch": 59.05,
+      "learning_rate": 1.2378082191780823e-05,
+      "loss": 0.4867,
+      "step": 43400
+    },
+    {
+      "epoch": 59.18,
+      "learning_rate": 1.2336986301369863e-05,
+      "loss": 0.4767,
+      "step": 43500
+    },
+    {
+      "epoch": 59.18,
+      "eval_cer": 0.054768071751492696,
+      "eval_loss": 0.2451842874288559,
+      "eval_runtime": 173.5286,
+      "eval_samples_per_second": 33.337,
+      "eval_steps_per_second": 4.172,
+      "eval_wer": 0.27526502755167687,
+      "step": 43500
+    },
+    {
+      "epoch": 59.32,
+      "learning_rate": 1.2295890410958905e-05,
+      "loss": 0.4842,
+      "step": 43600
+    },
+    {
+      "epoch": 59.46,
+      "learning_rate": 1.2254794520547947e-05,
+      "loss": 0.469,
+      "step": 43700
+    },
+    {
+      "epoch": 59.59,
+      "learning_rate": 1.2213698630136987e-05,
+      "loss": 0.4837,
+      "step": 43800
+    },
+    {
+      "epoch": 59.73,
+      "learning_rate": 1.2172602739726028e-05,
+      "loss": 0.4799,
+      "step": 43900
+    },
+    {
+      "epoch": 59.86,
+      "learning_rate": 1.2131506849315068e-05,
+      "loss": 0.4786,
+      "step": 44000
+    },
+    {
+      "epoch": 59.86,
+      "eval_cer": 0.05511415175790429,
+      "eval_loss": 0.24800312519073486,
+      "eval_runtime": 172.3947,
+      "eval_samples_per_second": 33.557,
+      "eval_steps_per_second": 4.2,
+      "eval_wer": 0.27685938854857206,
+      "step": 44000
+    },
+    {
+      "epoch": 60.0,
+      "learning_rate": 1.209041095890411e-05,
+      "loss": 0.4847,
+      "step": 44100
+    },
+    {
+      "epoch": 60.14,
+      "learning_rate": 1.204931506849315e-05,
+      "loss": 0.4767,
+      "step": 44200
+    },
+    {
+      "epoch": 60.27,
+      "learning_rate": 1.2008219178082192e-05,
+      "loss": 0.4883,
+      "step": 44300
+    },
+    {
+      "epoch": 60.41,
+      "learning_rate": 1.1967123287671233e-05,
+      "loss": 0.4762,
+      "step": 44400
+    },
+    {
+      "epoch": 60.54,
+      "learning_rate": 1.1926027397260275e-05,
+      "loss": 0.4804,
+      "step": 44500
+    },
+    {
+      "epoch": 60.54,
+      "eval_cer": 0.05562052145149598,
+      "eval_loss": 0.25136998295783997,
+      "eval_runtime": 173.2562,
+      "eval_samples_per_second": 33.39,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.27892926071997987,
+      "step": 44500
+    },
+    {
+      "epoch": 60.68,
+      "learning_rate": 1.1884931506849317e-05,
+      "loss": 0.4731,
+      "step": 44600
+    },
+    {
+      "epoch": 60.82,
+      "learning_rate": 1.1843835616438357e-05,
+      "loss": 0.4795,
+      "step": 44700
+    },
+    {
+      "epoch": 60.95,
+      "learning_rate": 1.1802739726027399e-05,
+      "loss": 0.4638,
+      "step": 44800
+    },
+    {
+      "epoch": 61.09,
+      "learning_rate": 1.1761643835616438e-05,
+      "loss": 0.4847,
+      "step": 44900
+    },
+    {
+      "epoch": 61.22,
+      "learning_rate": 1.172054794520548e-05,
+      "loss": 0.4794,
+      "step": 45000
+    },
+    {
+      "epoch": 61.22,
+      "eval_cer": 0.05385369194507893,
+      "eval_loss": 0.23912762105464935,
+      "eval_runtime": 173.9168,
+      "eval_samples_per_second": 33.263,
+      "eval_steps_per_second": 4.163,
+      "eval_wer": 0.2714889094011356,
+      "step": 45000
+    },
+    {
+      "epoch": 61.36,
+      "learning_rate": 1.167945205479452e-05,
+      "loss": 0.4781,
+      "step": 45100
+    },
+    {
+      "epoch": 61.5,
+      "learning_rate": 1.1638356164383562e-05,
+      "loss": 0.4812,
+      "step": 45200
+    },
+    {
+      "epoch": 61.63,
+      "learning_rate": 1.1597260273972602e-05,
+      "loss": 0.4695,
+      "step": 45300
+    },
+    {
+      "epoch": 61.77,
+      "learning_rate": 1.1556164383561644e-05,
+      "loss": 0.4703,
+      "step": 45400
+    },
+    {
+      "epoch": 61.9,
+      "learning_rate": 1.1515068493150686e-05,
+      "loss": 0.4789,
+      "step": 45500
+    },
+    {
+      "epoch": 61.9,
+      "eval_cer": 0.05456042374764575,
+      "eval_loss": 0.24608446657657623,
+      "eval_runtime": 173.52,
+      "eval_samples_per_second": 33.339,
+      "eval_steps_per_second": 4.172,
+      "eval_wer": 0.27249587424127997,
+      "step": 45500
+    },
+    {
+      "epoch": 62.04,
+      "learning_rate": 1.1473972602739727e-05,
+      "loss": 0.474,
+      "step": 45600
+    },
+    {
+      "epoch": 62.18,
+      "learning_rate": 1.1432876712328769e-05,
+      "loss": 0.4721,
+      "step": 45700
+    },
+    {
+      "epoch": 62.31,
+      "learning_rate": 1.1391780821917807e-05,
+      "loss": 0.4758,
+      "step": 45800
+    },
+    {
+      "epoch": 62.45,
+      "learning_rate": 1.135068493150685e-05,
+      "loss": 0.4745,
+      "step": 45900
+    },
+    {
+      "epoch": 62.58,
+      "learning_rate": 1.130958904109589e-05,
+      "loss": 0.4683,
+      "step": 46000
+    },
+    {
+      "epoch": 62.58,
+      "eval_cer": 0.054097769423284994,
+      "eval_loss": 0.2443952113389969,
+      "eval_runtime": 174.6552,
+      "eval_samples_per_second": 33.122,
+      "eval_steps_per_second": 4.145,
+      "eval_wer": 0.2707336857710274,
+      "step": 46000
+    },
+    {
+      "epoch": 62.72,
+      "learning_rate": 1.1268493150684932e-05,
+      "loss": 0.4857,
+      "step": 46100
+    },
+    {
+      "epoch": 62.86,
+      "learning_rate": 1.1227397260273972e-05,
+      "loss": 0.4765,
+      "step": 46200
+    },
+    {
+      "epoch": 62.99,
+      "learning_rate": 1.1186301369863014e-05,
+      "loss": 0.4561,
+      "step": 46300
+    },
+    {
+      "epoch": 63.13,
+      "learning_rate": 1.1145205479452056e-05,
+      "loss": 0.4716,
+      "step": 46400
+    },
+    {
+      "epoch": 63.27,
+      "learning_rate": 1.1104109589041096e-05,
+      "loss": 0.4721,
+      "step": 46500
+    },
+    {
+      "epoch": 63.27,
+      "eval_cer": 0.053850048997643014,
+      "eval_loss": 0.24683703482151031,
+      "eval_runtime": 172.5355,
+      "eval_samples_per_second": 33.529,
+      "eval_steps_per_second": 4.196,
+      "eval_wer": 0.26933512349304917,
+      "step": 46500
+    },
+    {
+      "epoch": 63.4,
+      "learning_rate": 1.1063424657534245e-05,
+      "loss": 0.4679,
+      "step": 46600
+    },
+    {
+      "epoch": 63.54,
+      "learning_rate": 1.1022328767123288e-05,
+      "loss": 0.4683,
+      "step": 46700
+    },
+    {
+      "epoch": 63.67,
+      "learning_rate": 1.098123287671233e-05,
+      "loss": 0.4863,
+      "step": 46800
+    },
+    {
+      "epoch": 63.81,
+      "learning_rate": 1.094013698630137e-05,
+      "loss": 0.4638,
+      "step": 46900
+    },
+    {
+      "epoch": 63.94,
+      "learning_rate": 1.0899041095890412e-05,
+      "loss": 0.4792,
+      "step": 47000
+    },
+    {
+      "epoch": 63.94,
+      "eval_cer": 0.05462235385405624,
+      "eval_loss": 0.2479219138622284,
+      "eval_runtime": 174.0485,
+      "eval_samples_per_second": 33.238,
+      "eval_steps_per_second": 4.16,
+      "eval_wer": 0.2737545802914604,
+      "step": 47000
+    },
+    {
+      "epoch": 64.08,
+      "learning_rate": 1.0857945205479452e-05,
+      "loss": 0.4724,
+      "step": 47100
+    },
+    {
+      "epoch": 64.22,
+      "learning_rate": 1.0816849315068494e-05,
+      "loss": 0.4807,
+      "step": 47200
+    },
+    {
+      "epoch": 64.35,
+      "learning_rate": 1.0775753424657535e-05,
+      "loss": 0.4783,
+      "step": 47300
+    },
+    {
+      "epoch": 64.49,
+      "learning_rate": 1.0734657534246577e-05,
+      "loss": 0.4676,
+      "step": 47400
+    },
+    {
+      "epoch": 64.63,
+      "learning_rate": 1.0693561643835615e-05,
+      "loss": 0.4712,
+      "step": 47500
+    },
+    {
+      "epoch": 64.63,
+      "eval_cer": 0.05465514038097944,
+      "eval_loss": 0.24659140408039093,
+      "eval_runtime": 172.9304,
+      "eval_samples_per_second": 33.453,
+      "eval_steps_per_second": 4.187,
+      "eval_wer": 0.274230091465973,
+      "step": 47500
+    },
+    {
+      "epoch": 64.76,
+      "learning_rate": 1.0652465753424657e-05,
+      "loss": 0.4616,
+      "step": 47600
+    },
+    {
+      "epoch": 64.9,
+      "learning_rate": 1.06113698630137e-05,
+      "loss": 0.4685,
+      "step": 47700
+    },
+    {
+      "epoch": 65.03,
+      "learning_rate": 1.057027397260274e-05,
+      "loss": 0.4625,
+      "step": 47800
+    },
+    {
+      "epoch": 65.17,
+      "learning_rate": 1.0529178082191782e-05,
+      "loss": 0.4645,
+      "step": 47900
+    },
+    {
+      "epoch": 65.31,
+      "learning_rate": 1.0488493150684932e-05,
+      "loss": 0.4607,
+      "step": 48000
+    },
+    {
+      "epoch": 65.31,
+      "eval_cer": 0.0539411226835408,
+      "eval_loss": 0.25033918023109436,
+      "eval_runtime": 173.9135,
+      "eval_samples_per_second": 33.264,
+      "eval_steps_per_second": 4.163,
+      "eval_wer": 0.2707336857710274,
+      "step": 48000
+    },
+    {
+      "epoch": 65.44,
+      "learning_rate": 1.0447397260273974e-05,
+      "loss": 0.4803,
+      "step": 48100
+    },
+    {
+      "epoch": 65.58,
+      "learning_rate": 1.0406301369863013e-05,
+      "loss": 0.4705,
+      "step": 48200
+    },
+    {
+      "epoch": 65.71,
+      "learning_rate": 1.0365205479452055e-05,
+      "loss": 0.459,
+      "step": 48300
+    },
+    {
+      "epoch": 65.85,
+      "learning_rate": 1.0324109589041095e-05,
+      "loss": 0.463,
+      "step": 48400
+    },
+    {
+      "epoch": 65.99,
+      "learning_rate": 1.0283013698630137e-05,
+      "loss": 0.4712,
+      "step": 48500
+    },
+    {
+      "epoch": 65.99,
+      "eval_cer": 0.05427263090020874,
+      "eval_loss": 0.24582460522651672,
+      "eval_runtime": 173.4675,
+      "eval_samples_per_second": 33.349,
+      "eval_steps_per_second": 4.174,
+      "eval_wer": 0.2718245643478504,
+      "step": 48500
+    },
+    {
+      "epoch": 66.12,
+      "learning_rate": 1.0241917808219178e-05,
+      "loss": 0.476,
+      "step": 48600
+    },
+    {
+      "epoch": 66.26,
+      "learning_rate": 1.020082191780822e-05,
+      "loss": 0.4629,
+      "step": 48700
+    },
+    {
+      "epoch": 66.39,
+      "learning_rate": 1.015972602739726e-05,
+      "loss": 0.4702,
+      "step": 48800
+    },
+    {
+      "epoch": 66.53,
+      "learning_rate": 1.0118630136986302e-05,
+      "loss": 0.464,
+      "step": 48900
+    },
+    {
+      "epoch": 66.67,
+      "learning_rate": 1.0077534246575344e-05,
+      "loss": 0.4647,
+      "step": 49000
+    },
+    {
+      "epoch": 66.67,
+      "eval_cer": 0.053835477207899364,
+      "eval_loss": 0.2473554164171219,
+      "eval_runtime": 173.249,
+      "eval_samples_per_second": 33.391,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.2693071522474896,
+      "step": 49000
+    },
+    {
+      "epoch": 66.8,
+      "learning_rate": 1.0036438356164383e-05,
+      "loss": 0.4598,
+      "step": 49100
+    },
+    {
+      "epoch": 66.94,
+      "learning_rate": 9.995342465753425e-06,
+      "loss": 0.4709,
+      "step": 49200
+    },
+    {
+      "epoch": 67.07,
+      "learning_rate": 9.954246575342465e-06,
+      "loss": 0.4641,
+      "step": 49300
+    },
+    {
+      "epoch": 67.21,
+      "learning_rate": 9.913150684931507e-06,
+      "loss": 0.4616,
+      "step": 49400
+    },
+    {
+      "epoch": 67.35,
+      "learning_rate": 9.872054794520548e-06,
+      "loss": 0.4736,
+      "step": 49500
+    },
+    {
+      "epoch": 67.35,
+      "eval_cer": 0.0540686258437977,
+      "eval_loss": 0.2513918876647949,
+      "eval_runtime": 174.1671,
+      "eval_samples_per_second": 33.215,
+      "eval_steps_per_second": 4.157,
+      "eval_wer": 0.26958686470308524,
+      "step": 49500
+    },
+    {
+      "epoch": 67.48,
+      "learning_rate": 9.83095890410959e-06,
+      "loss": 0.4622,
+      "step": 49600
+    },
+    {
+      "epoch": 67.62,
+      "learning_rate": 9.78986301369863e-06,
+      "loss": 0.4729,
+      "step": 49700
+    },
+    {
+      "epoch": 67.75,
+      "learning_rate": 9.748767123287672e-06,
+      "loss": 0.4689,
+      "step": 49800
+    },
+    {
+      "epoch": 67.89,
+      "learning_rate": 9.707671232876714e-06,
+      "loss": 0.4681,
+      "step": 49900
+    },
+    {
+      "epoch": 68.03,
+      "learning_rate": 9.666575342465754e-06,
+      "loss": 0.4718,
+      "step": 50000
+    },
+    {
+      "epoch": 68.03,
+      "eval_cer": 0.05396298036815627,
+      "eval_loss": 0.2506430745124817,
+      "eval_runtime": 174.4551,
+      "eval_samples_per_second": 33.16,
+      "eval_steps_per_second": 4.15,
+      "eval_wer": 0.26919526726525134,
+      "step": 50000
+    },
+    {
+      "epoch": 68.16,
+      "learning_rate": 9.625890410958903e-06,
+      "loss": 0.4575,
+      "step": 50100
+    },
+    {
+      "epoch": 68.3,
+      "learning_rate": 9.584794520547945e-06,
+      "loss": 0.473,
+      "step": 50200
+    },
+    {
+      "epoch": 68.43,
+      "learning_rate": 9.543698630136987e-06,
+      "loss": 0.4549,
+      "step": 50300
+    },
+    {
+      "epoch": 68.57,
+      "learning_rate": 9.502602739726028e-06,
+      "loss": 0.46,
+      "step": 50400
+    },
+    {
+      "epoch": 68.71,
+      "learning_rate": 9.46150684931507e-06,
+      "loss": 0.4695,
+      "step": 50500
+    },
+    {
+      "epoch": 68.71,
+      "eval_cer": 0.05384276310277119,
+      "eval_loss": 0.2499210387468338,
+      "eval_runtime": 173.5237,
+      "eval_samples_per_second": 33.338,
+      "eval_steps_per_second": 4.172,
+      "eval_wer": 0.26748902128611785,
+      "step": 50500
+    },
+    {
+      "epoch": 68.84,
+      "learning_rate": 9.420821917808219e-06,
+      "loss": 0.4656,
+      "step": 50600
+    },
+    {
+      "epoch": 68.98,
+      "learning_rate": 9.37972602739726e-06,
+      "loss": 0.4615,
+      "step": 50700
+    },
+    {
+      "epoch": 69.12,
+      "learning_rate": 9.338630136986301e-06,
+      "loss": 0.462,
+      "step": 50800
+    },
+    {
+      "epoch": 69.25,
+      "learning_rate": 9.297534246575343e-06,
+      "loss": 0.4493,
+      "step": 50900
+    },
+    {
+      "epoch": 69.39,
+      "learning_rate": 9.256438356164383e-06,
+      "loss": 0.4549,
+      "step": 51000
+    },
+    {
+      "epoch": 69.39,
+      "eval_cer": 0.05339468056815408,
+      "eval_loss": 0.24913272261619568,
+      "eval_runtime": 172.9495,
+      "eval_samples_per_second": 33.449,
+      "eval_steps_per_second": 4.186,
+      "eval_wer": 0.266901625129367,
+      "step": 51000
+    },
+    {
+      "epoch": 69.52,
+      "learning_rate": 9.215342465753426e-06,
+      "loss": 0.464,
+      "step": 51100
+    },
+    {
+      "epoch": 69.66,
+      "learning_rate": 9.174246575342466e-06,
+      "loss": 0.4534,
+      "step": 51200
+    },
+    {
+      "epoch": 69.8,
+      "learning_rate": 9.133150684931508e-06,
+      "loss": 0.453,
+      "step": 51300
+    },
+    {
+      "epoch": 69.93,
+      "learning_rate": 9.092054794520548e-06,
+      "loss": 0.4694,
+      "step": 51400
+    },
+    {
+      "epoch": 70.07,
+      "learning_rate": 9.050958904109589e-06,
+      "loss": 0.4605,
+      "step": 51500
+    },
+    {
+      "epoch": 70.07,
+      "eval_cer": 0.0532416767758458,
+      "eval_loss": 0.24969758093357086,
+      "eval_runtime": 173.5701,
+      "eval_samples_per_second": 33.329,
+      "eval_steps_per_second": 4.171,
+      "eval_wer": 0.26600654527146095,
+      "step": 51500
+    },
+    {
+      "epoch": 70.2,
+      "learning_rate": 9.00986301369863e-06,
+      "loss": 0.4647,
+      "step": 51600
+    },
+    {
+      "epoch": 70.34,
+      "learning_rate": 8.968767123287671e-06,
+      "loss": 0.4559,
+      "step": 51700
+    },
+    {
+      "epoch": 70.48,
+      "learning_rate": 8.927671232876713e-06,
+      "loss": 0.4512,
+      "step": 51800
+    },
+    {
+      "epoch": 70.61,
+      "learning_rate": 8.886575342465753e-06,
+      "loss": 0.4587,
+      "step": 51900
+    },
+    {
+      "epoch": 70.75,
+      "learning_rate": 8.845479452054795e-06,
+      "loss": 0.4538,
+      "step": 52000
+    },
+    {
+      "epoch": 70.75,
+      "eval_cer": 0.05356954204507783,
+      "eval_loss": 0.24719761312007904,
+      "eval_runtime": 173.5837,
+      "eval_samples_per_second": 33.327,
+      "eval_steps_per_second": 4.171,
+      "eval_wer": 0.26841207238958353,
+      "step": 52000
+    },
+    {
+      "epoch": 70.88,
+      "learning_rate": 8.804383561643836e-06,
+      "loss": 0.467,
+      "step": 52100
+    },
+    {
+      "epoch": 71.02,
+      "learning_rate": 8.763287671232878e-06,
+      "loss": 0.4569,
+      "step": 52200
+    },
+    {
+      "epoch": 71.16,
+      "learning_rate": 8.722191780821918e-06,
+      "loss": 0.4575,
+      "step": 52300
+    },
+    {
+      "epoch": 71.29,
+      "learning_rate": 8.681095890410958e-06,
+      "loss": 0.4583,
+      "step": 52400
+    },
+    {
+      "epoch": 71.43,
+      "learning_rate": 8.64e-06,
+      "loss": 0.4571,
+      "step": 52500
+    },
+    {
+      "epoch": 71.43,
+      "eval_cer": 0.052250795073277886,
+      "eval_loss": 0.24405916035175323,
+      "eval_runtime": 173.5172,
+      "eval_samples_per_second": 33.34,
+      "eval_steps_per_second": 4.172,
+      "eval_wer": 0.2628737657687897,
+      "step": 52500
+    },
+    {
+      "epoch": 71.56,
+      "learning_rate": 8.59890410958904e-06,
+      "loss": 0.458,
+      "step": 52600
+    },
+    {
+      "epoch": 71.7,
+      "learning_rate": 8.557808219178083e-06,
+      "loss": 0.4524,
+      "step": 52700
+    },
+    {
+      "epoch": 71.84,
+      "learning_rate": 8.516712328767123e-06,
+      "loss": 0.4587,
+      "step": 52800
+    },
+    {
+      "epoch": 71.97,
+      "learning_rate": 8.475616438356165e-06,
+      "loss": 0.4622,
+      "step": 52900
+    },
+    {
+      "epoch": 72.11,
+      "learning_rate": 8.434931506849316e-06,
+      "loss": 0.4608,
+      "step": 53000
+    },
+    {
+      "epoch": 72.11,
+      "eval_cer": 0.052946598033536976,
+      "eval_loss": 0.24687132239341736,
+      "eval_runtime": 173.425,
+      "eval_samples_per_second": 33.357,
+      "eval_steps_per_second": 4.175,
+      "eval_wer": 0.26522335039579314,
+      "step": 53000
+    },
+    {
+      "epoch": 72.24,
+      "learning_rate": 8.393835616438356e-06,
+      "loss": 0.4498,
+      "step": 53100
+    },
+    {
+      "epoch": 72.38,
+      "learning_rate": 8.352739726027396e-06,
+      "loss": 0.4671,
+      "step": 53200
+    },
+    {
+      "epoch": 72.52,
+      "learning_rate": 8.311643835616438e-06,
+      "loss": 0.4483,
+      "step": 53300
+    },
+    {
+      "epoch": 72.65,
+      "learning_rate": 8.270547945205479e-06,
+      "loss": 0.4451,
+      "step": 53400
+    },
+    {
+      "epoch": 72.79,
+      "learning_rate": 8.22945205479452e-06,
+      "loss": 0.4541,
+      "step": 53500
+    },
+    {
+      "epoch": 72.79,
+      "eval_cer": 0.05332910751430767,
+      "eval_loss": 0.24980194866657257,
+      "eval_runtime": 173.8551,
+      "eval_samples_per_second": 33.275,
+      "eval_steps_per_second": 4.164,
+      "eval_wer": 0.26734916505832007,
+      "step": 53500
+    },
+    {
+      "epoch": 72.92,
+      "learning_rate": 8.188356164383561e-06,
+      "loss": 0.4579,
+      "step": 53600
+    },
+    {
+      "epoch": 73.06,
+      "learning_rate": 8.147260273972603e-06,
+      "loss": 0.4578,
+      "step": 53700
+    },
+    {
+      "epoch": 73.2,
+      "learning_rate": 8.106164383561645e-06,
+      "loss": 0.4514,
+      "step": 53800
+    },
+    {
+      "epoch": 73.33,
+      "learning_rate": 8.065068493150686e-06,
+      "loss": 0.46,
+      "step": 53900
+    },
+    {
+      "epoch": 73.47,
+      "learning_rate": 8.023972602739728e-06,
+      "loss": 0.4424,
+      "step": 54000
+    },
+    {
+      "epoch": 73.47,
+      "eval_cer": 0.052986670455332,
+      "eval_loss": 0.25042009353637695,
+      "eval_runtime": 173.0787,
+      "eval_samples_per_second": 33.424,
+      "eval_steps_per_second": 4.183,
+      "eval_wer": 0.26583871779810353,
+      "step": 54000
+    },
+    {
+      "epoch": 73.6,
+      "learning_rate": 7.982876712328766e-06,
+      "loss": 0.4652,
+      "step": 54100
+    },
+    {
+      "epoch": 73.74,
+      "learning_rate": 7.941780821917808e-06,
+      "loss": 0.4482,
+      "step": 54200
+    },
+    {
+      "epoch": 73.88,
+      "learning_rate": 7.900684931506849e-06,
+      "loss": 0.4551,
+      "step": 54300
+    },
+    {
+      "epoch": 74.01,
+      "learning_rate": 7.85958904109589e-06,
+      "loss": 0.4608,
+      "step": 54400
+    },
+    {
+      "epoch": 74.15,
+      "learning_rate": 7.818493150684931e-06,
+      "loss": 0.4482,
+      "step": 54500
+    },
+    {
+      "epoch": 74.15,
+      "eval_cer": 0.05340560941046182,
+      "eval_loss": 0.2516920566558838,
+      "eval_runtime": 173.9431,
+      "eval_samples_per_second": 33.258,
+      "eval_steps_per_second": 4.162,
+      "eval_wer": 0.26838410114402395,
+      "step": 54500
+    },
+    {
+      "epoch": 74.29,
+      "learning_rate": 7.777397260273973e-06,
+      "loss": 0.4488,
+      "step": 54600
+    },
+    {
+      "epoch": 74.42,
+      "learning_rate": 7.736301369863015e-06,
+      "loss": 0.444,
+      "step": 54700
+    },
+    {
+      "epoch": 74.56,
+      "learning_rate": 7.695205479452055e-06,
+      "loss": 0.4447,
+      "step": 54800
+    },
+    {
+      "epoch": 74.69,
+      "learning_rate": 7.654109589041097e-06,
+      "loss": 0.4538,
+      "step": 54900
+    },
+    {
+      "epoch": 74.83,
+      "learning_rate": 7.613013698630138e-06,
+      "loss": 0.4554,
+      "step": 55000
+    },
+    {
+      "epoch": 74.83,
+      "eval_cer": 0.052932026243793326,
+      "eval_loss": 0.24706895649433136,
+      "eval_runtime": 173.541,
+      "eval_samples_per_second": 33.335,
+      "eval_steps_per_second": 4.172,
+      "eval_wer": 0.26558697658806746,
+      "step": 55000
+    },
+    {
+      "epoch": 74.97,
+      "learning_rate": 7.571917808219178e-06,
+      "loss": 0.4442,
+      "step": 55100
+    },
+    {
+      "epoch": 75.1,
+      "learning_rate": 7.530821917808219e-06,
+      "loss": 0.4519,
+      "step": 55200
+    },
+    {
+      "epoch": 75.24,
+      "learning_rate": 7.48972602739726e-06,
+      "loss": 0.457,
+      "step": 55300
+    },
+    {
+      "epoch": 75.37,
+      "learning_rate": 7.4486301369863014e-06,
+      "loss": 0.4548,
+      "step": 55400
+    },
+    {
+      "epoch": 75.51,
+      "learning_rate": 7.407534246575343e-06,
+      "loss": 0.444,
+      "step": 55500
+    },
+    {
+      "epoch": 75.51,
+      "eval_cer": 0.0534638965694364,
+      "eval_loss": 0.24925513565540314,
+      "eval_runtime": 174.4345,
+      "eval_samples_per_second": 33.164,
+      "eval_steps_per_second": 4.151,
+      "eval_wer": 0.26751699253167743,
+      "step": 55500
+    },
+    {
+      "epoch": 75.65,
+      "learning_rate": 7.366438356164384e-06,
+      "loss": 0.4543,
+      "step": 55600
+    },
+    {
+      "epoch": 75.78,
+      "learning_rate": 7.325342465753425e-06,
+      "loss": 0.4513,
+      "step": 55700
+    },
+    {
+      "epoch": 75.92,
+      "learning_rate": 7.284246575342465e-06,
+      "loss": 0.4556,
+      "step": 55800
+    },
+    {
+      "epoch": 76.05,
+      "learning_rate": 7.2431506849315065e-06,
+      "loss": 0.4527,
+      "step": 55900
+    },
+    {
+      "epoch": 76.19,
+      "learning_rate": 7.2020547945205485e-06,
+      "loss": 0.4464,
+      "step": 56000
+    },
+    {
+      "epoch": 76.19,
+      "eval_cer": 0.05243294244507346,
+      "eval_loss": 0.24606037139892578,
+      "eval_runtime": 173.8609,
+      "eval_samples_per_second": 33.274,
+      "eval_steps_per_second": 4.164,
+      "eval_wer": 0.2634891331711001,
+      "step": 56000
+    },
+    {
+      "epoch": 76.33,
+      "learning_rate": 7.16095890410959e-06,
+      "loss": 0.4547,
+      "step": 56100
+    },
+    {
+      "epoch": 76.46,
+      "learning_rate": 7.119863013698631e-06,
+      "loss": 0.4401,
+      "step": 56200
+    },
+    {
+      "epoch": 76.6,
+      "learning_rate": 7.078767123287671e-06,
+      "loss": 0.4518,
+      "step": 56300
+    },
+    {
+      "epoch": 76.73,
+      "learning_rate": 7.037671232876712e-06,
+      "loss": 0.4454,
+      "step": 56400
+    },
+    {
+      "epoch": 76.87,
+      "learning_rate": 6.9965753424657535e-06,
+      "loss": 0.4436,
+      "step": 56500
+    },
+    {
+      "epoch": 76.87,
+      "eval_cer": 0.052615089816869035,
+      "eval_loss": 0.2479296326637268,
+      "eval_runtime": 173.7241,
+      "eval_samples_per_second": 33.3,
+      "eval_steps_per_second": 4.168,
+      "eval_wer": 0.26410450057341056,
+      "step": 56500
+    },
+    {
+      "epoch": 77.01,
+      "learning_rate": 6.955479452054795e-06,
+      "loss": 0.4546,
+      "step": 56600
+    },
+    {
+      "epoch": 77.14,
+      "learning_rate": 6.914383561643836e-06,
+      "loss": 0.4394,
+      "step": 56700
+    },
+    {
+      "epoch": 77.28,
+      "learning_rate": 6.873287671232876e-06,
+      "loss": 0.4462,
+      "step": 56800
+    },
+    {
+      "epoch": 77.41,
+      "learning_rate": 6.832191780821917e-06,
+      "loss": 0.4541,
+      "step": 56900
+    },
+    {
+      "epoch": 77.55,
+      "learning_rate": 6.791095890410959e-06,
+      "loss": 0.4432,
+      "step": 57000
+    },
+    {
+      "epoch": 77.55,
+      "eval_cer": 0.05262237571174085,
+      "eval_loss": 0.25127509236335754,
+      "eval_runtime": 174.4639,
+      "eval_samples_per_second": 33.159,
+      "eval_steps_per_second": 4.15,
+      "eval_wer": 0.2641324718189701,
+      "step": 57000
+    },
+    {
+      "epoch": 77.69,
+      "learning_rate": 6.750000000000001e-06,
+      "loss": 0.4458,
+      "step": 57100
+    },
+    {
+      "epoch": 77.82,
+      "learning_rate": 6.708904109589042e-06,
+      "loss": 0.4444,
+      "step": 57200
+    },
+    {
+      "epoch": 77.96,
+      "learning_rate": 6.667808219178082e-06,
+      "loss": 0.456,
+      "step": 57300
+    },
+    {
+      "epoch": 78.1,
+      "learning_rate": 6.627123287671234e-06,
+      "loss": 0.4599,
+      "step": 57400
+    },
+    {
+      "epoch": 78.23,
+      "learning_rate": 6.586027397260274e-06,
+      "loss": 0.4459,
+      "step": 57500
+    },
+    {
+      "epoch": 78.23,
+      "eval_cer": 0.05214150665020054,
+      "eval_loss": 0.24595066905021667,
+      "eval_runtime": 174.0217,
+      "eval_samples_per_second": 33.243,
+      "eval_steps_per_second": 4.16,
+      "eval_wer": 0.2624541970853962,
+      "step": 57500
+    },
+    {
+      "epoch": 78.37,
+      "learning_rate": 6.544931506849315e-06,
+      "loss": 0.4558,
+      "step": 57600
+    },
+    {
+      "epoch": 78.5,
+      "learning_rate": 6.503835616438356e-06,
+      "loss": 0.4328,
+      "step": 57700
+    },
+    {
+      "epoch": 78.64,
+      "learning_rate": 6.462739726027398e-06,
+      "loss": 0.4497,
+      "step": 57800
+    },
+    {
+      "epoch": 78.77,
+      "learning_rate": 6.421643835616439e-06,
+      "loss": 0.449,
+      "step": 57900
+    },
+    {
+      "epoch": 78.91,
+      "learning_rate": 6.380547945205479e-06,
+      "loss": 0.4433,
+      "step": 58000
+    },
+    {
+      "epoch": 78.91,
+      "eval_cer": 0.05212329191302099,
+      "eval_loss": 0.24571694433689117,
+      "eval_runtime": 173.8557,
+      "eval_samples_per_second": 33.275,
+      "eval_steps_per_second": 4.164,
+      "eval_wer": 0.2621744846298006,
+      "step": 58000
+    },
+    {
+      "epoch": 79.05,
+      "learning_rate": 6.33945205479452e-06,
+      "loss": 0.4438,
+      "step": 58100
+    },
+    {
+      "epoch": 79.18,
+      "learning_rate": 6.298356164383562e-06,
+      "loss": 0.4394,
+      "step": 58200
+    },
+    {
+      "epoch": 79.32,
+      "learning_rate": 6.2572602739726035e-06,
+      "loss": 0.4483,
+      "step": 58300
+    },
+    {
+      "epoch": 79.46,
+      "learning_rate": 6.216164383561645e-06,
+      "loss": 0.447,
+      "step": 58400
+    },
+    {
+      "epoch": 79.59,
+      "learning_rate": 6.175068493150685e-06,
+      "loss": 0.4407,
+      "step": 58500
+    },
+    {
+      "epoch": 79.59,
+      "eval_cer": 0.05277537950404913,
+      "eval_loss": 0.25311264395713806,
+      "eval_runtime": 173.098,
+      "eval_samples_per_second": 33.42,
+      "eval_steps_per_second": 4.183,
+      "eval_wer": 0.2658666890436631,
+      "step": 58500
+    },
+    {
+      "epoch": 79.73,
+      "learning_rate": 6.133972602739726e-06,
+      "loss": 0.4518,
+      "step": 58600
+    },
+    {
+      "epoch": 79.86,
+      "learning_rate": 6.093287671232877e-06,
+      "loss": 0.4477,
+      "step": 58700
+    },
+    {
+      "epoch": 80.0,
+      "learning_rate": 6.052191780821918e-06,
+      "loss": 0.4356,
+      "step": 58800
+    },
+    {
+      "epoch": 80.14,
+      "learning_rate": 6.011095890410959e-06,
+      "loss": 0.4569,
+      "step": 58900
+    },
+    {
+      "epoch": 80.27,
+      "learning_rate": 5.9700000000000004e-06,
+      "loss": 0.4389,
+      "step": 59000
+    },
+    {
+      "epoch": 80.27,
+      "eval_cer": 0.05206864770148231,
+      "eval_loss": 0.24845506250858307,
+      "eval_runtime": 172.065,
+      "eval_samples_per_second": 33.621,
+      "eval_steps_per_second": 4.208,
+      "eval_wer": 0.2630975357332662,
+      "step": 59000
+    },
+    {
+      "epoch": 80.41,
+      "learning_rate": 5.928904109589041e-06,
+      "loss": 0.4429,
+      "step": 59100
+    },
+    {
+      "epoch": 80.54,
+      "learning_rate": 5.887808219178082e-06,
+      "loss": 0.4407,
+      "step": 59200
+    },
+    {
+      "epoch": 80.68,
+      "learning_rate": 5.846712328767123e-06,
+      "loss": 0.4427,
+      "step": 59300
+    },
+    {
+      "epoch": 80.82,
+      "learning_rate": 5.805616438356164e-06,
+      "loss": 0.4346,
+      "step": 59400
+    },
+    {
+      "epoch": 80.95,
+      "learning_rate": 5.764520547945206e-06,
+      "loss": 0.4384,
+      "step": 59500
+    },
+    {
+      "epoch": 80.95,
+      "eval_cer": 0.05223622328353424,
+      "eval_loss": 0.25022611021995544,
+      "eval_runtime": 174.0525,
+      "eval_samples_per_second": 33.237,
+      "eval_steps_per_second": 4.16,
+      "eval_wer": 0.26527929288691227,
+      "step": 59500
+    },
+    {
+      "epoch": 81.09,
+      "learning_rate": 5.723424657534247e-06,
+      "loss": 0.4509,
+      "step": 59600
+    },
+    {
+      "epoch": 81.22,
+      "learning_rate": 5.682328767123288e-06,
+      "loss": 0.4423,
+      "step": 59700
+    },
+    {
+      "epoch": 81.36,
+      "learning_rate": 5.641232876712329e-06,
+      "loss": 0.4375,
+      "step": 59800
+    },
+    {
+      "epoch": 81.5,
+      "learning_rate": 5.60013698630137e-06,
+      "loss": 0.4485,
+      "step": 59900
+    },
+    {
+      "epoch": 81.63,
+      "learning_rate": 5.559041095890411e-06,
+      "loss": 0.4306,
+      "step": 60000
+    },
+    {
+      "epoch": 81.63,
+      "eval_cer": 0.05276080771430549,
+      "eval_loss": 0.2479526847600937,
+      "eval_runtime": 174.2383,
+      "eval_samples_per_second": 33.202,
+      "eval_steps_per_second": 4.155,
+      "eval_wer": 0.26648205644597356,
+      "step": 60000
+    },
+    {
+      "epoch": 81.77,
+      "learning_rate": 5.517945205479452e-06,
+      "loss": 0.4342,
+      "step": 60100
+    },
+    {
+      "epoch": 81.9,
+      "learning_rate": 5.476849315068493e-06,
+      "loss": 0.4365,
+      "step": 60200
+    },
+    {
+      "epoch": 82.04,
+      "learning_rate": 5.435753424657534e-06,
+      "loss": 0.4524,
+      "step": 60300
+    },
+    {
+      "epoch": 82.18,
+      "learning_rate": 5.394657534246576e-06,
+      "loss": 0.4384,
+      "step": 60400
+    },
+    {
+      "epoch": 82.31,
+      "learning_rate": 5.353561643835617e-06,
+      "loss": 0.4505,
+      "step": 60500
+    },
+    {
+      "epoch": 82.31,
+      "eval_cer": 0.05226536686302154,
+      "eval_loss": 0.24607928097248077,
+      "eval_runtime": 173.8187,
+      "eval_samples_per_second": 33.282,
+      "eval_steps_per_second": 4.165,
+      "eval_wer": 0.2637129031355766,
+      "step": 60500
+    },
+    {
+      "epoch": 82.45,
+      "learning_rate": 5.3124657534246576e-06,
+      "loss": 0.4383,
+      "step": 60600
+    },
+    {
+      "epoch": 82.58,
+      "learning_rate": 5.271369863013699e-06,
+      "loss": 0.4502,
+      "step": 60700
+    },
+    {
+      "epoch": 82.72,
+      "learning_rate": 5.23027397260274e-06,
+      "loss": 0.4393,
+      "step": 60800
+    },
+    {
+      "epoch": 82.86,
+      "learning_rate": 5.189589041095891e-06,
+      "loss": 0.4395,
+      "step": 60900
+    },
+    {
+      "epoch": 82.99,
+      "learning_rate": 5.148493150684932e-06,
+      "loss": 0.4442,
+      "step": 61000
+    },
+    {
+      "epoch": 82.99,
+      "eval_cer": 0.052320011074560205,
+      "eval_loss": 0.2518540024757385,
+      "eval_runtime": 174.1941,
+      "eval_samples_per_second": 33.21,
+      "eval_steps_per_second": 4.156,
+      "eval_wer": 0.2641324718189701,
+      "step": 61000
+    },
+    {
+      "epoch": 83.13,
+      "learning_rate": 5.107397260273973e-06,
+      "loss": 0.4562,
+      "step": 61100
+    },
+    {
+      "epoch": 83.27,
+      "learning_rate": 5.066301369863014e-06,
+      "loss": 0.4403,
+      "step": 61200
+    },
+    {
+      "epoch": 83.4,
+      "learning_rate": 5.0252054794520545e-06,
+      "loss": 0.4402,
+      "step": 61300
+    },
+    {
+      "epoch": 83.54,
+      "learning_rate": 4.984109589041096e-06,
+      "loss": 0.4379,
+      "step": 61400
+    },
+    {
+      "epoch": 83.67,
+      "learning_rate": 4.943013698630137e-06,
+      "loss": 0.4349,
+      "step": 61500
+    },
+    {
+      "epoch": 83.67,
+      "eval_cer": 0.05220343675661104,
+      "eval_loss": 0.2508690655231476,
+      "eval_runtime": 173.9649,
+      "eval_samples_per_second": 33.254,
+      "eval_steps_per_second": 4.162,
+      "eval_wer": 0.2625381108220749,
+      "step": 61500
+    },
+    {
+      "epoch": 83.81,
+      "learning_rate": 4.901917808219178e-06,
+      "loss": 0.4392,
+      "step": 61600
+    },
+    {
+      "epoch": 83.94,
+      "learning_rate": 4.86082191780822e-06,
+      "loss": 0.4414,
+      "step": 61700
+    },
+    {
+      "epoch": 84.08,
+      "learning_rate": 4.8197260273972604e-06,
+      "loss": 0.44,
+      "step": 61800
+    },
+    {
+      "epoch": 84.22,
+      "learning_rate": 4.778630136986302e-06,
+      "loss": 0.4403,
+      "step": 61900
+    },
+    {
+      "epoch": 84.35,
+      "learning_rate": 4.737534246575343e-06,
+      "loss": 0.4398,
+      "step": 62000
+    },
+    {
+      "epoch": 84.35,
+      "eval_cer": 0.052345511706611586,
+      "eval_loss": 0.2509627640247345,
+      "eval_runtime": 173.2451,
+      "eval_samples_per_second": 33.392,
+      "eval_steps_per_second": 4.179,
+      "eval_wer": 0.2658946602892227,
+      "step": 62000
+    },
+    {
+      "epoch": 84.49,
+      "learning_rate": 4.696438356164384e-06,
+      "loss": 0.4362,
+      "step": 62100
+    },
+    {
+      "epoch": 84.63,
+      "learning_rate": 4.655342465753424e-06,
+      "loss": 0.4389,
+      "step": 62200
+    },
+    {
+      "epoch": 84.76,
+      "learning_rate": 4.6142465753424655e-06,
+      "loss": 0.4436,
+      "step": 62300
+    },
+    {
+      "epoch": 84.9,
+      "learning_rate": 4.573150684931507e-06,
+      "loss": 0.4394,
+      "step": 62400
+    },
+    {
+      "epoch": 85.03,
+      "learning_rate": 4.532054794520548e-06,
+      "loss": 0.4398,
+      "step": 62500
+    },
+    {
+      "epoch": 85.03,
+      "eval_cer": 0.05256408855276627,
+      "eval_loss": 0.2507339119911194,
+      "eval_runtime": 173.9958,
+      "eval_samples_per_second": 33.248,
+      "eval_steps_per_second": 4.161,
+      "eval_wer": 0.26483175295795924,
+      "step": 62500
+    },
+    {
+      "epoch": 85.17,
+      "learning_rate": 4.49095890410959e-06,
+      "loss": 0.4287,
+      "step": 62600
+    },
+    {
+      "epoch": 85.31,
+      "learning_rate": 4.44986301369863e-06,
+      "loss": 0.4262,
+      "step": 62700
+    },
+    {
+      "epoch": 85.44,
+      "learning_rate": 4.408767123287671e-06,
+      "loss": 0.4406,
+      "step": 62800
+    },
+    {
+      "epoch": 85.58,
+      "learning_rate": 4.3676712328767125e-06,
+      "loss": 0.4387,
+      "step": 62900
+    },
+    {
+      "epoch": 85.71,
+      "learning_rate": 4.326575342465754e-06,
+      "loss": 0.4355,
+      "step": 63000
+    },
+    {
+      "epoch": 85.71,
+      "eval_cer": 0.052320011074560205,
+      "eval_loss": 0.2500060200691223,
+      "eval_runtime": 174.9316,
+      "eval_samples_per_second": 33.07,
+      "eval_steps_per_second": 4.139,
+      "eval_wer": 0.2653072641324718,
+      "step": 63000
+    },
+    {
+      "epoch": 85.85,
+      "learning_rate": 4.285479452054795e-06,
+      "loss": 0.4333,
+      "step": 63100
+    },
+    {
+      "epoch": 85.99,
+      "learning_rate": 4.244383561643835e-06,
+      "loss": 0.4353,
+      "step": 63200
+    },
+    {
+      "epoch": 86.12,
+      "learning_rate": 4.203287671232876e-06,
+      "loss": 0.4424,
+      "step": 63300
+    },
+    {
+      "epoch": 86.26,
+      "learning_rate": 4.1621917808219175e-06,
+      "loss": 0.4424,
+      "step": 63400
+    },
+    {
+      "epoch": 86.39,
+      "learning_rate": 4.121506849315068e-06,
+      "loss": 0.4373,
+      "step": 63500
+    },
+    {
+      "epoch": 86.39,
+      "eval_cer": 0.05235644054891932,
+      "eval_loss": 0.2522575259208679,
+      "eval_runtime": 174.2545,
+      "eval_samples_per_second": 33.199,
+      "eval_steps_per_second": 4.155,
+      "eval_wer": 0.2649995804313166,
+      "step": 63500
+    },
+    {
+      "epoch": 86.53,
+      "learning_rate": 4.0804109589041095e-06,
+      "loss": 0.4331,
+      "step": 63600
+    },
+    {
+      "epoch": 86.67,
+      "learning_rate": 4.039315068493151e-06,
+      "loss": 0.4359,
+      "step": 63700
+    },
+    {
+      "epoch": 86.8,
+      "learning_rate": 3.998219178082192e-06,
+      "loss": 0.4316,
+      "step": 63800
+    },
+    {
+      "epoch": 86.94,
+      "learning_rate": 3.957123287671233e-06,
+      "loss": 0.4304,
+      "step": 63900
+    },
+    {
+      "epoch": 87.07,
+      "learning_rate": 3.916027397260274e-06,
+      "loss": 0.4391,
+      "step": 64000
+    },
+    {
+      "epoch": 87.07,
+      "eval_cer": 0.052272652757893355,
+      "eval_loss": 0.2508656978607178,
+      "eval_runtime": 174.5282,
+      "eval_samples_per_second": 33.147,
+      "eval_steps_per_second": 4.148,
+      "eval_wer": 0.26354507566221924,
+      "step": 64000
+    },
+    {
+      "epoch": 87.21,
+      "learning_rate": 3.874931506849315e-06,
+      "loss": 0.4358,
+      "step": 64100
+    },
+    {
+      "epoch": 87.35,
+      "learning_rate": 3.8338356164383566e-06,
+      "loss": 0.4265,
+      "step": 64200
+    },
+    {
+      "epoch": 87.48,
+      "learning_rate": 3.7927397260273977e-06,
+      "loss": 0.4316,
+      "step": 64300
+    },
+    {
+      "epoch": 87.62,
+      "learning_rate": 3.751643835616438e-06,
+      "loss": 0.4312,
+      "step": 64400
+    },
+    {
+      "epoch": 87.75,
+      "learning_rate": 3.7105479452054792e-06,
+      "loss": 0.4381,
+      "step": 64500
+    },
+    {
+      "epoch": 87.75,
+      "eval_cer": 0.052108720123277344,
+      "eval_loss": 0.250161349773407,
+      "eval_runtime": 173.481,
+      "eval_samples_per_second": 33.347,
+      "eval_steps_per_second": 4.173,
+      "eval_wer": 0.26354507566221924,
+      "step": 64500
+    },
+    {
+      "epoch": 87.89,
+      "learning_rate": 3.669452054794521e-06,
+      "loss": 0.4314,
+      "step": 64600
+    },
+    {
+      "epoch": 88.03,
+      "learning_rate": 3.628356164383562e-06,
+      "loss": 0.4437,
+      "step": 64700
+    },
+    {
+      "epoch": 88.16,
+      "learning_rate": 3.5872602739726028e-06,
+      "loss": 0.4316,
+      "step": 64800
+    },
+    {
+      "epoch": 88.3,
+      "learning_rate": 3.546164383561644e-06,
+      "loss": 0.4346,
+      "step": 64900
+    },
+    {
+      "epoch": 88.43,
+      "learning_rate": 3.5050684931506847e-06,
+      "loss": 0.4297,
+      "step": 65000
+    },
+    {
+      "epoch": 88.43,
+      "eval_cer": 0.052086862438661875,
+      "eval_loss": 0.25211092829704285,
+      "eval_runtime": 173.7361,
+      "eval_samples_per_second": 33.298,
+      "eval_steps_per_second": 4.167,
+      "eval_wer": 0.2631814494699449,
+      "step": 65000
+    },
+    {
+      "epoch": 88.57,
+      "learning_rate": 3.4639726027397263e-06,
+      "loss": 0.4337,
+      "step": 65100
+    },
+    {
+      "epoch": 88.71,
+      "learning_rate": 3.422876712328767e-06,
+      "loss": 0.436,
+      "step": 65200
+    },
+    {
+      "epoch": 88.84,
+      "learning_rate": 3.3817808219178082e-06,
+      "loss": 0.4368,
+      "step": 65300
+    },
+    {
+      "epoch": 88.98,
+      "learning_rate": 3.3406849315068494e-06,
+      "loss": 0.4413,
+      "step": 65400
+    },
+    {
+      "epoch": 89.12,
+      "learning_rate": 3.2995890410958906e-06,
+      "loss": 0.44,
+      "step": 65500
+    },
+    {
+      "epoch": 89.12,
+      "eval_cer": 0.05202857527968729,
+      "eval_loss": 0.2506903111934662,
+      "eval_runtime": 174.6918,
+      "eval_samples_per_second": 33.115,
+      "eval_steps_per_second": 4.144,
+      "eval_wer": 0.26237028334871754,
+      "step": 65500
+    },
+    {
+      "epoch": 89.25,
+      "learning_rate": 3.2584931506849317e-06,
+      "loss": 0.435,
+      "step": 65600
+    },
+    {
+      "epoch": 89.39,
+      "learning_rate": 3.2173972602739725e-06,
+      "loss": 0.4417,
+      "step": 65700
+    },
+    {
+      "epoch": 89.52,
+      "learning_rate": 3.1763013698630137e-06,
+      "loss": 0.4367,
+      "step": 65800
+    },
+    {
+      "epoch": 89.66,
+      "learning_rate": 3.135205479452055e-06,
+      "loss": 0.4269,
+      "step": 65900
+    },
+    {
+      "epoch": 89.8,
+      "learning_rate": 3.094109589041096e-06,
+      "loss": 0.4313,
+      "step": 66000
+    },
+    {
+      "epoch": 89.8,
+      "eval_cer": 0.05187921443481492,
+      "eval_loss": 0.24967756867408752,
+      "eval_runtime": 173.7921,
+      "eval_samples_per_second": 33.287,
+      "eval_steps_per_second": 4.166,
+      "eval_wer": 0.26234231210315795,
+      "step": 66000
+    },
+    {
+      "epoch": 89.93,
+      "learning_rate": 3.053013698630137e-06,
+      "loss": 0.4314,
+      "step": 66100
+    },
+    {
+      "epoch": 90.07,
+      "learning_rate": 3.011917808219178e-06,
+      "loss": 0.4399,
+      "step": 66200
+    },
+    {
+      "epoch": 90.2,
+      "learning_rate": 2.970821917808219e-06,
+      "loss": 0.4251,
+      "step": 66300
+    },
+    {
+      "epoch": 90.34,
+      "learning_rate": 2.9297260273972603e-06,
+      "loss": 0.4338,
+      "step": 66400
+    },
+    {
+      "epoch": 90.48,
+      "learning_rate": 2.8886301369863015e-06,
+      "loss": 0.4402,
+      "step": 66500
+    },
+    {
+      "epoch": 90.48,
+      "eval_cer": 0.051686138220711615,
+      "eval_loss": 0.24882683157920837,
+      "eval_runtime": 173.6629,
+      "eval_samples_per_second": 33.312,
+      "eval_steps_per_second": 4.169,
+      "eval_wer": 0.26083186484294146,
+      "step": 66500
+    },
+    {
+      "epoch": 90.61,
+      "learning_rate": 2.8475342465753427e-06,
+      "loss": 0.4338,
+      "step": 66600
+    },
+    {
+      "epoch": 90.75,
+      "learning_rate": 2.8064383561643834e-06,
+      "loss": 0.4274,
+      "step": 66700
+    },
+    {
+      "epoch": 90.88,
+      "learning_rate": 2.7653424657534246e-06,
+      "loss": 0.4252,
+      "step": 66800
+    },
+    {
+      "epoch": 91.02,
+      "learning_rate": 2.724246575342466e-06,
+      "loss": 0.4452,
+      "step": 66900
+    },
+    {
+      "epoch": 91.16,
+      "learning_rate": 2.683150684931507e-06,
+      "loss": 0.4324,
+      "step": 67000
+    },
+    {
+      "epoch": 91.16,
+      "eval_cer": 0.05123076979122268,
+      "eval_loss": 0.24848763644695282,
+      "eval_runtime": 173.7308,
+      "eval_samples_per_second": 33.299,
+      "eval_steps_per_second": 4.167,
+      "eval_wer": 0.2585102514614976,
+      "step": 67000
+    },
+    {
+      "epoch": 91.29,
+      "learning_rate": 2.642054794520548e-06,
+      "loss": 0.4407,
+      "step": 67100
+    },
+    {
+      "epoch": 91.43,
+      "learning_rate": 2.600958904109589e-06,
+      "loss": 0.4244,
+      "step": 67200
+    },
+    {
+      "epoch": 91.56,
+      "learning_rate": 2.55986301369863e-06,
+      "loss": 0.4277,
+      "step": 67300
+    },
+    {
+      "epoch": 91.7,
+      "learning_rate": 2.5187671232876716e-06,
+      "loss": 0.4247,
+      "step": 67400
+    },
+    {
+      "epoch": 91.84,
+      "learning_rate": 2.4776712328767124e-06,
+      "loss": 0.4317,
+      "step": 67500
+    },
+    {
+      "epoch": 91.84,
+      "eval_cer": 0.05132912937199229,
+      "eval_loss": 0.2487734705209732,
+      "eval_runtime": 174.374,
+      "eval_samples_per_second": 33.176,
+      "eval_steps_per_second": 4.152,
+      "eval_wer": 0.25867807893485495,
+      "step": 67500
+    },
+    {
+      "epoch": 91.97,
+      "learning_rate": 2.4365753424657536e-06,
+      "loss": 0.4233,
+      "step": 67600
+    },
+    {
+      "epoch": 92.11,
+      "learning_rate": 2.3954794520547943e-06,
+      "loss": 0.4274,
+      "step": 67700
+    },
+    {
+      "epoch": 92.24,
+      "learning_rate": 2.3543835616438355e-06,
+      "loss": 0.4411,
+      "step": 67800
+    },
+    {
+      "epoch": 92.38,
+      "learning_rate": 2.313287671232877e-06,
+      "loss": 0.4272,
+      "step": 67900
+    },
+    {
+      "epoch": 92.52,
+      "learning_rate": 2.272191780821918e-06,
+      "loss": 0.437,
+      "step": 68000
+    },
+    {
+      "epoch": 92.52,
+      "eval_cer": 0.051296342845069086,
+      "eval_loss": 0.24733413755893707,
+      "eval_runtime": 173.9217,
+      "eval_samples_per_second": 33.262,
+      "eval_steps_per_second": 4.163,
+      "eval_wer": 0.25901373388156973,
+      "step": 68000
+    },
+    {
+      "epoch": 92.65,
+      "learning_rate": 2.231095890410959e-06,
+      "loss": 0.4239,
+      "step": 68100
+    },
+    {
+      "epoch": 92.79,
+      "learning_rate": 2.1899999999999998e-06,
+      "loss": 0.4285,
+      "step": 68200
+    },
+    {
+      "epoch": 92.92,
+      "learning_rate": 2.1489041095890414e-06,
+      "loss": 0.4319,
+      "step": 68300
+    },
+    {
+      "epoch": 93.06,
+      "learning_rate": 2.1078082191780826e-06,
+      "loss": 0.4387,
+      "step": 68400
+    },
+    {
+      "epoch": 93.2,
+      "learning_rate": 2.0667123287671233e-06,
+      "loss": 0.4389,
+      "step": 68500
+    },
+    {
+      "epoch": 93.2,
+      "eval_cer": 0.05115426789506854,
+      "eval_loss": 0.24715159833431244,
+      "eval_runtime": 174.6504,
+      "eval_samples_per_second": 33.123,
+      "eval_steps_per_second": 4.145,
+      "eval_wer": 0.2581466252692232,
+      "step": 68500
+    },
+    {
+      "epoch": 93.33,
+      "learning_rate": 2.0256164383561645e-06,
+      "loss": 0.4306,
+      "step": 68600
+    },
+    {
+      "epoch": 93.47,
+      "learning_rate": 1.9845205479452052e-06,
+      "loss": 0.4287,
+      "step": 68700
+    },
+    {
+      "epoch": 93.6,
+      "learning_rate": 1.943424657534247e-06,
+      "loss": 0.4227,
+      "step": 68800
+    },
+    {
+      "epoch": 93.74,
+      "learning_rate": 1.9023287671232876e-06,
+      "loss": 0.4238,
+      "step": 68900
+    },
+    {
+      "epoch": 93.88,
+      "learning_rate": 1.8612328767123288e-06,
+      "loss": 0.4428,
+      "step": 69000
+    },
+    {
+      "epoch": 93.88,
+      "eval_cer": 0.0511724826322481,
+      "eval_loss": 0.2475104033946991,
+      "eval_runtime": 174.3396,
+      "eval_samples_per_second": 33.182,
+      "eval_steps_per_second": 4.153,
+      "eval_wer": 0.2586501076892954,
+      "step": 69000
+    },
+    {
+      "epoch": 94.01,
+      "learning_rate": 1.8201369863013697e-06,
+      "loss": 0.4334,
+      "step": 69100
+    },
+    {
+      "epoch": 94.15,
+      "learning_rate": 1.7790410958904111e-06,
+      "loss": 0.4347,
+      "step": 69200
+    },
+    {
+      "epoch": 94.29,
+      "learning_rate": 1.7383561643835617e-06,
+      "loss": 0.4308,
+      "step": 69300
+    },
+    {
+      "epoch": 94.42,
+      "learning_rate": 1.6972602739726026e-06,
+      "loss": 0.4249,
+      "step": 69400
+    },
+    {
+      "epoch": 94.56,
+      "learning_rate": 1.6561643835616438e-06,
+      "loss": 0.4294,
+      "step": 69500
+    },
+    {
+      "epoch": 94.56,
+      "eval_cer": 0.051310914634812736,
+      "eval_loss": 0.24888011813163757,
+      "eval_runtime": 173.9117,
+      "eval_samples_per_second": 33.264,
+      "eval_steps_per_second": 4.163,
+      "eval_wer": 0.25957315879276105,
+      "step": 69500
+    },
+    {
+      "epoch": 94.69,
+      "learning_rate": 1.615068493150685e-06,
+      "loss": 0.4224,
+      "step": 69600
+    },
+    {
+      "epoch": 94.83,
+      "learning_rate": 1.5739726027397262e-06,
+      "loss": 0.4297,
+      "step": 69700
+    },
+    {
+      "epoch": 94.97,
+      "learning_rate": 1.5328767123287671e-06,
+      "loss": 0.4211,
+      "step": 69800
+    },
+    {
+      "epoch": 95.1,
+      "learning_rate": 1.491780821917808e-06,
+      "loss": 0.4284,
+      "step": 69900
+    },
+    {
+      "epoch": 95.24,
+      "learning_rate": 1.4506849315068493e-06,
+      "loss": 0.4247,
+      "step": 70000
+    },
+    {
+      "epoch": 95.24,
+      "eval_cer": 0.05148213316430057,
+      "eval_loss": 0.24985703825950623,
+      "eval_runtime": 174.6527,
+      "eval_samples_per_second": 33.123,
+      "eval_steps_per_second": 4.145,
+      "eval_wer": 0.2597409862661184,
+      "step": 70000
+    },
+    {
+      "epoch": 95.37,
+      "learning_rate": 1.4095890410958905e-06,
+      "loss": 0.4311,
+      "step": 70100
+    },
+    {
+      "epoch": 95.51,
+      "learning_rate": 1.3684931506849316e-06,
+      "loss": 0.4307,
+      "step": 70200
+    },
+    {
+      "epoch": 95.65,
+      "learning_rate": 1.3273972602739726e-06,
+      "loss": 0.4226,
+      "step": 70300
+    },
+    {
+      "epoch": 95.78,
+      "learning_rate": 1.2863013698630138e-06,
+      "loss": 0.4286,
+      "step": 70400
+    },
+    {
+      "epoch": 95.92,
+      "learning_rate": 1.2452054794520547e-06,
+      "loss": 0.4309,
+      "step": 70500
+    },
+    {
+      "epoch": 95.92,
+      "eval_cer": 0.051358272951479586,
+      "eval_loss": 0.24932163953781128,
+      "eval_runtime": 174.1672,
+      "eval_samples_per_second": 33.215,
+      "eval_steps_per_second": 4.157,
+      "eval_wer": 0.25901373388156973,
+      "step": 70500
+    },
+    {
+      "epoch": 96.05,
+      "learning_rate": 1.204109589041096e-06,
+      "loss": 0.429,
+      "step": 70600
+    },
+    {
+      "epoch": 96.19,
+      "learning_rate": 1.163013698630137e-06,
+      "loss": 0.4335,
+      "step": 70700
+    },
+    {
+      "epoch": 96.33,
+      "learning_rate": 1.121917808219178e-06,
+      "loss": 0.4278,
+      "step": 70800
+    },
+    {
+      "epoch": 96.46,
+      "learning_rate": 1.0808219178082192e-06,
+      "loss": 0.4334,
+      "step": 70900
+    },
+    {
+      "epoch": 96.6,
+      "learning_rate": 1.0397260273972602e-06,
+      "loss": 0.4366,
+      "step": 71000
+    },
+    {
+      "epoch": 96.6,
+      "eval_cer": 0.051238055686094505,
+      "eval_loss": 0.249197855591774,
+      "eval_runtime": 174.0449,
+      "eval_samples_per_second": 33.239,
+      "eval_steps_per_second": 4.16,
+      "eval_wer": 0.25918156135492715,
+      "step": 71000
+    },
+    {
+      "epoch": 96.73,
+      "learning_rate": 9.986301369863016e-07,
+      "loss": 0.4261,
+      "step": 71100
+    },
+    {
+      "epoch": 96.87,
+      "learning_rate": 9.575342465753425e-07,
+      "loss": 0.4188,
+      "step": 71200
+    },
+    {
+      "epoch": 97.01,
+      "learning_rate": 9.164383561643836e-07,
+      "loss": 0.4171,
+      "step": 71300
+    },
+    {
+      "epoch": 97.14,
+      "learning_rate": 8.753424657534247e-07,
+      "loss": 0.4306,
+      "step": 71400
+    },
+    {
+      "epoch": 97.28,
+      "learning_rate": 8.342465753424659e-07,
+      "loss": 0.4245,
+      "step": 71500
+    },
+    {
+      "epoch": 97.28,
+      "eval_cer": 0.05127812810788953,
+      "eval_loss": 0.24934062361717224,
+      "eval_runtime": 173.1587,
+      "eval_samples_per_second": 33.409,
+      "eval_steps_per_second": 4.181,
+      "eval_wer": 0.2586501076892954,
+      "step": 71500
+    },
+    {
+      "epoch": 97.41,
+      "learning_rate": 7.931506849315069e-07,
+      "loss": 0.4311,
+      "step": 71600
+    },
+    {
+      "epoch": 97.55,
+      "learning_rate": 7.520547945205479e-07,
+      "loss": 0.4389,
+      "step": 71700
+    },
+    {
+      "epoch": 97.69,
+      "learning_rate": 7.109589041095891e-07,
+      "loss": 0.428,
+      "step": 71800
+    },
+    {
+      "epoch": 97.82,
+      "learning_rate": 6.698630136986301e-07,
+      "loss": 0.4347,
+      "step": 71900
+    },
+    {
+      "epoch": 97.96,
+      "learning_rate": 6.287671232876712e-07,
+      "loss": 0.4346,
+      "step": 72000
+    },
+    {
+      "epoch": 97.96,
+      "eval_cer": 0.0511724826322481,
+      "eval_loss": 0.24779224395751953,
+      "eval_runtime": 174.3069,
+      "eval_samples_per_second": 33.189,
+      "eval_steps_per_second": 4.154,
+      "eval_wer": 0.25834242398814017,
+      "step": 72000
+    },
+    {
+      "epoch": 98.1,
+      "learning_rate": 5.876712328767124e-07,
+      "loss": 0.4297,
+      "step": 72100
+    },
+    {
+      "epoch": 98.23,
+      "learning_rate": 5.465753424657535e-07,
+      "loss": 0.4373,
+      "step": 72200
+    },
+    {
+      "epoch": 98.37,
+      "learning_rate": 5.054794520547945e-07,
+      "loss": 0.4235,
+      "step": 72300
+    },
+    {
+      "epoch": 98.5,
+      "learning_rate": 4.6438356164383565e-07,
+      "loss": 0.441,
+      "step": 72400
+    },
+    {
+      "epoch": 98.64,
+      "learning_rate": 4.232876712328767e-07,
+      "loss": 0.4289,
+      "step": 72500
+    },
+    {
+      "epoch": 98.64,
+      "eval_cer": 0.05115426789506854,
+      "eval_loss": 0.24892671406269073,
+      "eval_runtime": 174.4572,
+      "eval_samples_per_second": 33.16,
+      "eval_steps_per_second": 4.15,
+      "eval_wer": 0.2585102514614976,
+      "step": 72500
+    },
+    {
+      "epoch": 98.77,
+      "learning_rate": 3.821917808219178e-07,
+      "loss": 0.4245,
+      "step": 72600
+    },
+    {
+      "epoch": 98.91,
+      "learning_rate": 3.410958904109589e-07,
+      "loss": 0.4271,
+      "step": 72700
+    },
+    {
+      "epoch": 99.05,
+      "learning_rate": 3.0000000000000004e-07,
+      "loss": 0.4287,
+      "step": 72800
+    },
+    {
+      "epoch": 99.18,
+      "learning_rate": 2.589041095890411e-07,
+      "loss": 0.4311,
+      "step": 72900
+    },
+    {
+      "epoch": 99.32,
+      "learning_rate": 2.178082191780822e-07,
+      "loss": 0.4246,
+      "step": 73000
+    },
+    {
+      "epoch": 99.32,
+      "eval_cer": 0.051263556318145886,
+      "eval_loss": 0.24873462319374084,
+      "eval_runtime": 174.0209,
+      "eval_samples_per_second": 33.243,
+      "eval_steps_per_second": 4.16,
+      "eval_wer": 0.2589018488993315,
+      "step": 73000
+    },
+    {
+      "epoch": 99.46,
+      "learning_rate": 1.7671232876712328e-07,
+      "loss": 0.4247,
+      "step": 73100
+    },
+    {
+      "epoch": 99.59,
+      "learning_rate": 1.3602739726027397e-07,
+      "loss": 0.4186,
+      "step": 73200
+    },
+    {
+      "epoch": 99.73,
+      "learning_rate": 9.493150684931508e-08,
+      "loss": 0.4377,
+      "step": 73300
+    },
+    {
+      "epoch": 99.86,
+      "learning_rate": 5.383561643835616e-08,
+      "loss": 0.4296,
+      "step": 73400
+    },
+    {
+      "epoch": 100.0,
+      "learning_rate": 1.2739726027397261e-08,
+      "loss": 0.4241,
+      "step": 73500
+    },
+    {
+      "epoch": 100.0,
+      "eval_cer": 0.05127812810788953,
+      "eval_loss": 0.2491369992494583,
+      "eval_runtime": 174.2027,
+      "eval_samples_per_second": 33.208,
+      "eval_steps_per_second": 4.156,
+      "eval_wer": 0.25884590640821237,
+      "step": 73500
+    },
+    {
+      "epoch": 100.0,
+      "step": 73500,
+      "total_flos": 3.141686635566631e+20,
+      "train_loss": 0.0,
+      "train_runtime": 32.0497,
+      "train_samples_per_second": 73442.278,
+      "train_steps_per_second": 2293.316
+    }
+  ],
+  "max_steps": 73500,
+  "num_train_epochs": 100,
+  "total_flos": 3.141686635566631e+20,
+  "trial_name": null,
+  "trial_params": null
+}