learning_rate: 0.0001
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 5000
num_epochs: 30
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3132	0.4877	1000	3.2545	100.0	98.0148
0.7509	0.9754	2000	0.4356	40.8644	11.0904
0.3943	1.4628	3000	0.2208	30.6208	7.6653
0.3328	1.9505	4000	0.1868	27.9376	6.8747
0.2716	2.4379	5000	0.1718	26.9243	6.5275
0.2452	2.9256	6000	0.1580	26.2810	6.2743
0.1875	3.4131	7000	0.1456	25.0430	5.9318
0.1862	3.9008	8000	0.1351	24.1177	5.6296
0.154	4.3882	9000	0.1322	23.5053	5.5547
0.1691	4.8759	10000	0.1361	23.5406	5.5657
0.1328	5.3633	11000	0.1265	22.6682	5.2533
0.1426	5.8510	12000	0.1199	22.8576	5.2556
0.1176	6.3385	13000	0.1146	22.2849	5.1373
0.1361	6.8261	14000	0.1164	22.2540	5.0899
0.1068	7.3136	15000	0.1174	22.0249	5.0355
0.1163	7.8013	16000	0.1137	21.6681	4.9345
0.1069	8.2887	17000	0.1147	21.5139	4.8958
0.1051	8.7764	18000	0.1150	21.5535	4.8722
0.0938	9.2638	19000	0.1084	21.1526	4.7799
0.1029	9.7515	20000	0.1126	21.3068	4.8169
0.0815	10.2390	21000	0.1055	21.0909	4.7254
0.0819	10.7267	22000	0.1134	21.2627	4.8059
0.0723	11.2141	23000	0.1076	20.9719	4.6765
0.0728	11.7018	24000	0.1064	20.7957	4.6189
0.0761	12.1892	25000	0.1044	20.4432	4.5424
0.0668	12.6769	26000	0.1111	20.3199	4.5432
0.0693	13.1644	27000	0.1076	20.4917	4.5258
0.06	13.6520	28000	0.1074	20.1657	4.4572
0.0625	14.1395	29000	0.1092	20.3860	4.4942
0.0642	14.6272	30000	0.1090	20.2714	4.4643
0.0587	15.1146	31000	0.1030	20.1657	4.4406
0.0574	15.6023	32000	0.1073	20.0908	4.4493
0.0623	16.0897	33000	0.1053	20.0423	4.4161
0.0545	16.5774	34000	0.1068	19.8617	4.3767
0.0474	17.0649	35000	0.1043	19.7383	4.3104
0.0471	17.5525	36000	0.1022	19.6237	4.3065
0.0492	18.0400	37000	0.1007	19.4123	4.2260
0.0402	18.5277	38000	0.1084	19.5665	4.2725
0.0465	19.0151	39000	0.0996	19.3594	4.2062
0.0446	19.5028	40000	0.1052	19.3462	4.1952
0.0377	19.9905	41000	0.1015	19.2669	4.1913
0.0389	20.4779	42000	0.1088	19.3374	4.2086
0.0318	20.9656	43000	0.1062	19.1435	4.1542
0.032	21.4531	44000	0.1026	19.1567	4.1431
0.0506	21.9407	45000	0.1036	18.9981	4.0926
0.0383	22.4282	46000	0.1026	19.1038	4.1053
0.0448	22.9159	47000	0.1046	19.0289	4.1060
0.0405	23.4033	48000	0.1017	18.9100	4.0879
0.0341	23.8910	49000	0.1014	18.9232	4.0595
0.0354	24.3784	50000	0.1018	18.8791	4.0461
0.0344	24.8661	51000	0.1025	18.9673	4.0477
0.0315	25.3536	52000	0.1020	18.9364	4.0516
0.0312	25.8413	53000	0.1008	18.8880	4.0358
0.0352	26.3287	54000	0.1009	18.8924	4.0279
0.0303	26.8164	55000	0.1008	18.8659	4.0153
0.0325	27.3038	56000	0.1020	18.8351	4.0074
0.0353	27.7915	57000	0.1017	18.8703	4.0185
0.0401	28.2790	58000	0.1017	18.8615	4.0137
0.0294	28.7666	59000	0.1010	18.8615	4.0169
0.0401	29.2541	60000	0.1010	18.8439	4.0137
0.0352	29.7418	61000	0.1009	18.8351	4.0114

Framework versions

Transformers 4.56.1
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.22.0

davidilag/wav2vec2-xls-r-300m-pt-500h-FO-500h-DK-cp-best-ft-faroese-100h-30-epochs_run9_2025-09-11

wav2vec2-xls-r-300m-pt-500h-FO-500h-DK-cp-best-ft-faroese-100h-30-epochs_run9_2025-09-11

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-pt-500h-FO-500h-DK-cp-best-ft-faroese-100h-30-epochs_run9_2025-09-11