Modelpublic

davidilag/wav2vec2-xls-r-300m-cpt-1000h_faroese-best-faroese-100h-30-epochs_run3_2025-08-26

sourceHugging Faceupdated 1y agoView on Hugging Face

0likes6downloads

Model Card

wav2vec2-xls-r-300m-pretrained-1000hfaroese-best-faroese-100h-30-epochsrun32025-08-26

This model was fine-tuned with 100h Faroese (Ravnursson) dataset. It achieves the following results on the test set:

—Loss: 0.0964
—Wer: 7.75
—Cer: 2.18

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 0.0001
—trainbatchsize: 16
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: cosine
—lrschedulerwarmup_steps: 5000
—num_epochs: 30
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3343	0.4877	1000	3.2618	100.0	99.3017
0.8421	0.9754	2000	0.5013	42.3228	11.7106
0.4042	1.4628	3000	0.2241	29.6162	7.4728
0.3608	1.9505	4000	0.1943	27.8495	6.8834
0.3014	2.4379	5000	0.1740	26.4396	6.4857
0.2691	2.9256	6000	0.1538	25.4571	6.1662
0.203	3.4131	7000	0.1446	24.5143	5.8813
0.2211	3.9008	8000	0.1359	24.2147	5.7409
0.1772	4.3882	9000	0.1354	23.4348	5.6091
0.1904	4.8759	10000	0.1304	22.8576	5.4418
0.1564	5.3633	11000	0.1436	23.4348	5.5563
0.1604	5.8510	12000	0.1261	22.7519	5.3827
0.1332	6.3385	13000	0.1231	22.5052	5.2422
0.1518	6.8261	14000	0.1174	22.2320	5.1341
0.1262	7.3136	15000	0.1247	22.3862	5.1941
0.1277	7.8013	16000	0.1230	22.0470	5.1144
0.1359	8.2887	17000	0.1081	21.8619	4.9787
0.1217	8.7764	18000	0.1179	21.7650	5.0039
0.1026	9.2638	19000	0.1103	21.3817	4.8619
0.1069	9.7515	20000	0.1113	21.3597	4.8422
0.097	10.2390	21000	0.1144	21.2319	4.7988
0.1014	10.7267	22000	0.1127	21.3244	4.8682
0.0817	11.2141	23000	0.1073	20.8706	4.6836
0.0872	11.7018	24000	0.1062	20.9940	4.7160
0.0829	12.1892	25000	0.1014	20.6459	4.6268
0.0745	12.6769	26000	0.1135	20.7957	4.6710
0.0783	13.1644	27000	0.1125	20.5666	4.5952
0.0737	13.6520	28000	0.1058	20.2494	4.5290
0.0712	14.1395	29000	0.1051	20.2185	4.5266
0.0797	14.6272	30000	0.1048	20.3155	4.5479
0.0702	15.1146	31000	0.1009	20.2802	4.4611
0.0689	15.6023	32000	0.0954	19.9806	4.3601
0.0729	16.0897	33000	0.1010	19.9982	4.4201
0.0635	16.5774	34000	0.1045	19.8572	4.4130
0.053	17.0649	35000	0.1017	19.6854	4.3167
0.0474	17.5525	36000	0.1000	19.6766	4.3333
0.0565	18.0400	37000	0.1020	19.7075	4.3025
0.0492	18.5277	38000	0.1006	19.5312	4.2946
0.0503	19.0151	39000	0.1011	19.3902	4.2575
0.0464	19.5028	40000	0.0999	19.4475	4.2378
0.0433	19.9905	41000	0.1020	19.5400	4.2560
0.0416	20.4779	42000	0.1039	19.3638	4.2260
0.0352	20.9656	43000	0.1017	19.2008	4.1581
0.0384	21.4531	44000	0.0972	19.2316	4.1660
0.0559	21.9407	45000	0.0953	19.0201	4.1013
0.0427	22.4282	46000	0.0989	18.9937	4.1084
0.0483	22.9159	47000	0.0960	19.0113	4.0989
0.049	23.4033	48000	0.0929	18.7514	4.0461
0.0354	23.8910	49000	0.0985	18.8615	4.0524
0.0381	24.3784	50000	0.0967	18.8131	4.0445
0.038	24.8661	51000	0.0960	18.9056	4.0634
0.0358	25.3536	52000	0.0974	18.7558	4.0413
0.0348	25.8413	53000	0.0956	18.7205	4.0303
0.0383	26.3287	54000	0.0968	18.6412	3.9956
0.032	26.8164	55000	0.0955	18.6456	3.9987
0.0391	27.3038	56000	0.0955	18.7161	3.9916
0.0409	27.7915	57000	0.0961	18.6324	3.9964
0.0368	28.2790	58000	0.0967	18.6368	3.9814
0.0308	28.7666	59000	0.0968	18.6721	3.9861
0.0406	29.2541	60000	0.0965	18.6412	3.9869
0.0376	29.7418	61000	0.0964	18.6368	3.9869

Framework versions

—Transformers 4.55.2
—Pytorch 2.8.0+cu126
—Datasets 4.0.0
—Tokenizers 0.21.4