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
3.3286	0.4877	1000	3.2977	100.0	99.2560
0.8993	0.9754	2000	0.5522	45.3452	12.7837
0.4078	1.4628	3000	0.2254	29.7396	7.6022
0.3571	1.9505	4000	0.1875	28.1447	7.0080
0.2912	2.4379	5000	0.1804	26.4220	6.5070
0.2742	2.9256	6000	0.1603	25.3470	6.1638
0.1995	3.4131	7000	0.1520	24.7918	5.9318
0.2086	3.9008	8000	0.1364	23.8842	5.6383
0.1569	4.3882	9000	0.1303	23.1837	5.4142
0.1741	4.8759	10000	0.1288	22.8136	5.3606
0.1446	5.3633	11000	0.1301	22.8841	5.4284
0.1545	5.8510	12000	0.1207	22.6197	5.2667
0.1257	6.3385	13000	0.1198	22.4523	5.2028
0.1375	6.8261	14000	0.1176	22.0646	5.1057
0.1109	7.3136	15000	0.1246	22.2981	5.1878
0.1216	7.8013	16000	0.1164	21.6637	4.9826
0.1199	8.2887	17000	0.1133	21.3861	4.8919
0.122	8.7764	18000	0.1148	21.3508	4.8998
0.1016	9.2638	19000	0.1136	21.3024	4.8572
0.1043	9.7515	20000	0.1215	21.1350	4.8169
0.0914	10.2390	21000	0.1097	20.8001	4.7112
0.0875	10.7267	22000	0.1163	20.9896	4.8043
0.0745	11.2141	23000	0.1243	21.0292	4.8509
0.0804	11.7018	24000	0.1136	20.5622	4.6599
0.0749	12.1892	25000	0.1132	20.4344	4.6378
0.0723	12.6769	26000	0.1140	20.6900	4.6584
0.0713	13.1644	27000	0.1129	20.3815	4.5503
0.0709	13.6520	28000	0.1057	20.2626	4.5266
0.0658	14.1395	29000	0.1089	20.2670	4.5763
0.073	14.6272	30000	0.1055	19.9366	4.4493
0.0633	15.1146	31000	0.1057	20.0335	4.4595
0.0584	15.6023	32000	0.1039	19.9586	4.4035
0.0677	16.0897	33000	0.1070	19.8837	4.3893
0.0597	16.5774	34000	0.1086	19.7427	4.3948
0.0498	17.0649	35000	0.1037	19.9938	4.3972
0.0485	17.5525	36000	0.1025	19.7295	4.3648
0.0505	18.0400	37000	0.1047	19.7030	4.3191
0.0421	18.5277	38000	0.1051	19.6634	4.3333
0.0489	19.0151	39000	0.0998	19.5356	4.2765
0.0443	19.5028	40000	0.1015	19.4695	4.2709
0.0385	19.9905	41000	0.1068	19.2536	4.2212
0.0439	20.4779	42000	0.1041	19.2801	4.1960
0.0351	20.9656	43000	0.1033	19.2360	4.1976
0.0347	21.4531	44000	0.1020	19.1920	4.1613
0.0503	21.9407	45000	0.1017	19.2228	4.1463
0.0427	22.4282	46000	0.0991	19.0950	4.1337
0.0434	22.9159	47000	0.1039	19.0157	4.1321
0.0468	23.4033	48000	0.1006	19.0069	4.1195
0.0362	23.8910	49000	0.1005	18.9629	4.0879
0.039	24.3784	50000	0.1020	18.9100	4.0705
0.0353	24.8661	51000	0.1019	18.8219	4.0508
0.0336	25.3536	52000	0.1015	18.8483	4.0587
0.0354	25.8413	53000	0.1027	18.8791	4.0548
0.0373	26.3287	54000	0.1027	18.7998	4.0335
0.0304	26.8164	55000	0.1020	18.8439	4.0421
0.0395	27.3038	56000	0.1018	18.8483	4.0477
0.0396	27.7915	57000	0.1008	18.7954	4.0303
0.0356	28.2790	58000	0.1012	18.8131	4.0311
0.0293	28.7666	59000	0.1008	18.8219	4.0295
0.0425	29.2541	60000	0.1006	18.8175	4.0248
0.0384	29.7418	61000	0.1006	18.8307	4.0279

Framework versions

Transformers 4.55.4
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.21.4

davidilag/wav2vec2-xls-r-300m-pt-1000h_faroese-cp14-faroese-100h-30-epochs_run7_2025-09-03

wav2vec2-xls-r-300m-pt-1000h_faroese-cp14-faroese-100h-30-epochs_run7_2025-09-03

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-pt-1000h_faroese-cp14-faroese-100h-30-epochs_run7_2025-09-03