Modelpublic

davidilag/wav2vec2-xls-r-300m-pre_trained-converted-faroese-100h-30-epochs_2025-07-10

sourceHugging Faceupdated 1y agoView on Hugging Face

0likes11downloads

Model Card

wav2vec2-xls-r-300m-pretrained-converted-faroese-100h-30-epochs2025-07-10

This model was trained from scratch on the Ravnursson 100h dataset. It achieves the following results on the Test set:

—Loss: 0.0986
—Wer: 8.21
—Cer: 2.28

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.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=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
3.3152	0.4877	1000	3.2279	100.0	99.4935
0.9943	0.9754	2000	0.6195	53.4035	15.3503
0.5408	1.4628	3000	0.2845	36.1149	9.5203
0.4766	1.9505	4000	0.2440	33.9825	8.6137
0.3776	2.4379	5000	0.1999	30.5283	7.6093
0.3464	2.9256	6000	0.1867	28.9289	7.1225
0.2626	3.4131	7000	0.1706	27.6248	6.7627
0.2949	3.9008	8000	0.1612	26.5454	6.4439
0.2243	4.3882	9000	0.1570	25.9858	6.3421
0.2466	4.8759	10000	0.1505	25.3866	6.0762
0.1965	5.3633	11000	0.1538	25.3470	6.0786
0.2083	5.8510	12000	0.1383	24.6068	5.8821
0.1897	6.3385	13000	0.1318	24.1398	5.7077
0.1919	6.8261	14000	0.1301	24.2587	5.6943
0.1547	7.3136	15000	0.1318	23.4965	5.5634
0.1603	7.8013	16000	0.1234	22.7607	5.2880
0.1525	8.2887	17000	0.1323	23.1572	5.3811
0.1476	8.7764	18000	0.1262	22.8621	5.3077
0.1356	9.2638	19000	0.1211	22.6814	5.2856
0.1319	9.7515	20000	0.1246	22.3289	5.1878
0.1285	10.2390	21000	0.1200	22.2100	5.1089
0.1194	10.7267	22000	0.1179	22.0822	5.0892
0.1044	11.2141	23000	0.1192	21.9985	5.0734
0.1034	11.7018	24000	0.1210	21.9544	5.0560
0.1072	12.1892	25000	0.1158	21.6989	4.9629
0.0933	12.6769	26000	0.1140	21.9016	5.0395
0.0973	13.1644	27000	0.1154	21.3068	4.8556
0.0974	13.6520	28000	0.1114	21.5447	4.8840
0.0834	14.1395	29000	0.1084	21.3332	4.8193
0.0987	14.6272	30000	0.1062	21.2099	4.7736
0.0834	15.1146	31000	0.1088	20.9323	4.7270
0.0785	15.6023	32000	0.1059	21.0336	4.7183
0.0857	16.0897	33000	0.1021	20.7913	4.6449
0.0815	16.5774	34000	0.1043	20.4917	4.5889
0.0656	17.0649	35000	0.1143	20.7957	4.6292
0.0578	17.5525	36000	0.1070	20.5622	4.5747
0.0687	18.0400	37000	0.1054	20.5005	4.5408
0.0592	18.5277	38000	0.1100	20.2185	4.4927
0.065	19.0151	39000	0.1052	20.2317	4.4990
0.059	19.5028	40000	0.1032	20.3066	4.4832
0.0508	19.9905	41000	0.1080	20.2846	4.4856
0.0553	20.4779	42000	0.1074	20.3287	4.4808
0.0469	20.9656	43000	0.0979	19.8749	4.3285
0.0447	21.4531	44000	0.1020	20.0070	4.3885
0.0695	21.9407	45000	0.1017	19.9321	4.3491
0.0576	22.4282	46000	0.1026	19.8837	4.3585
0.0642	22.9159	47000	0.1034	19.8132	4.3388
0.0597	23.4033	48000	0.1019	19.7427	4.2899
0.0473	23.8910	49000	0.1054	19.7471	4.3285
0.0544	24.3784	50000	0.1001	19.6370	4.2859
0.0479	24.8661	51000	0.1018	19.5973	4.2575
0.0467	25.3536	52000	0.0991	19.6017	4.2662
0.0443	25.8413	53000	0.0983	19.5136	4.2441
0.0483	26.3287	54000	0.0997	19.5444	4.2410
0.0436	26.8164	55000	0.1007	19.5532	4.2425
0.0459	27.3038	56000	0.0992	19.5268	4.2418
0.0508	27.7915	57000	0.0987	19.5004	4.2299
0.0526	28.2790	58000	0.0991	19.4828	4.2291
0.0417	28.7666	59000	0.0989	19.4960	4.2268
0.0517	29.2541	60000	0.0985	19.4960	4.2276
0.057	29.7418	61000	0.0986	19.4916	4.2276

Framework versions

—Transformers 4.53.1
—Pytorch 2.6.0+cu124
—Datasets 2.14.4
—Tokenizers 0.21.2