learning_rate: 3e-05
train_batch_size: 4
eval_batch_size: 8
seed: 42
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 2000
num_epochs: 30
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer
3.311	1.0	672	2.9834	1.0
2.6148	2.0	1344	2.6014	0.9870
2.231	3.0	2016	2.2240	0.9415
2.0945	4.0	2688	1.9552	0.9338
1.6302	5.0	3360	1.0949	0.5048
1.2185	6.0	4032	0.6218	0.1922
1.0618	7.0	4704	0.4965	0.1658
0.999	8.0	5376	0.4223	0.1550
0.897	9.0	6048	0.3839	0.1529
0.9083	10.0	6720	0.3515	0.1484
0.854	11.0	7392	0.3328	0.1439
0.8205	12.0	8064	0.3197	0.1440
0.8207	13.0	8736	0.3111	0.1412
0.7778	14.0	9408	0.3009	0.1409
0.754	15.0	10080	0.2925	0.1382
0.7466	16.0	10752	0.2893	0.1332
0.7152	17.0	11424	0.2842	0.1314
0.7196	18.0	12096	0.2798	0.1311
0.7231	19.0	12768	0.2715	0.1306
0.6543	20.0	13440	0.2706	0.1330
0.6655	21.0	14112	0.2716	0.1328
0.6783	22.0	14784	0.2693	0.1301
0.6834	23.0	15456	0.2627	0.1282
0.6551	24.0	16128	0.2650	0.1321
0.6753	25.0	16800	0.2615	0.1306
0.666	26.0	17472	0.2621	0.1308
0.638	27.0	18144	0.2611	0.1279
0.6629	28.0	18816	0.2616	0.1282
0.6131	29.0	19488	0.2612	0.1301
0.6348	30.0	20160	0.2624	0.1305

duck-hug-567/xls-r-300m-engphon-collapsed

xls-r-300m-engphon-collapsed