learning_rate: 1e-05
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 3000
num_epochs: 30
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.0058	1.4451	1000	1.3562	99.5531	42.6542
0.5946	2.8902	2000	0.2567	42.1243	7.7199
0.2822	4.3353	3000	0.1662	33.2307	5.6799
0.2782	5.7803	4000	0.1404	30.0924	5.0783
0.2994	7.2254	5000	0.1233	28.8956	4.7534
0.2259	8.6705	6000	0.1251	28.1458	4.5812
0.1798	10.1156	7000	0.1161	27.0186	4.3967
0.1734	11.5607	8000	0.1081	26.2886	4.2150
0.2183	13.0058	9000	0.1044	26.0652	4.1661
0.1639	14.4509	10000	0.1001	25.6977	4.0875
0.1519	15.8960	11000	0.0972	25.2061	4.0101
0.1255	17.3410	12000	0.0993	25.3153	4.0461
0.1181	18.7861	13000	0.0957	24.8237	3.9891
0.1274	20.2312	14000	0.0952	24.6102	3.9003
0.1302	21.6763	15000	0.0961	24.5506	3.9261
0.1073	23.1214	16000	0.0952	24.4116	3.8732
0.0998	24.5665	17000	0.0944	24.3073	3.8386
0.1149	26.0116	18000	0.0952	24.2576	3.8338
0.115	27.4566	19000	0.0946	24.1732	3.8243
0.0911	28.9017	20000	0.0947	24.1682	3.8304

davidilag/wav2vec2-xls-r-1b-distant-from-faroese-251h-30-epochs_20250120_v2

wav2vec2-xls-r-1b-distant-from-faroese-251h-30-epochs_20250120_v2