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
3.451	1.0	672	3.0870	1.0
2.7	2.0	1344	2.6840	1.0
2.3201	3.0	2016	2.3206	0.9491
2.191	4.0	2688	2.0507	0.9347
1.7364	5.0	3360	1.2831	0.6323
1.372	6.0	4032	0.8211	0.2372
1.193	7.0	4704	0.6767	0.1889
1.1232	8.0	5376	0.5950	0.1746
1.0251	9.0	6048	0.5333	0.1697
1.0245	10.0	6720	0.4998	0.1665
0.9766	11.0	7392	0.4926	0.1651
0.9366	12.0	8064	0.4714	0.1602
0.9169	13.0	8736	0.4420	0.1557
0.8847	14.0	9408	0.4339	0.1585
0.8566	15.0	10080	0.4267	0.1565
0.8517	16.0	10752	0.4202	0.1528
0.8177	17.0	11424	0.4153	0.1529
0.8142	18.0	12096	0.4039	0.1532
0.817	19.0	12768	0.3959	0.1524
0.7499	20.0	13440	0.3948	0.1526
0.7748	21.0	14112	0.3902	0.1506
0.792	22.0	14784	0.3869	0.1515
0.7698	23.0	15456	0.3844	0.1535
0.7498	24.0	16128	0.3791	0.1498
0.7694	25.0	16800	0.3768	0.1508
0.7523	26.0	17472	0.3784	0.1481
0.7342	27.0	18144	0.3776	0.1469
0.7581	28.0	18816	0.3779	0.1487
0.7135	29.0	19488	0.3773	0.1490
0.7198	30.0	20160	0.3769	0.1492

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

xls-r-300m-engphon-base