Modelpublic

davidilag/wav2vec2-xls-r-300m-cpt-1000h_faroese-10_epochs-faroese-100h-30-epochs_run2_2025-08-22

sourceHugging Faceupdated 1y agoView on Hugging Face

0likes6downloads

Model Card

wav2vec2-xls-r-300m-pretrained-1000hfaroese-10epochs-faroese-100h-30-epochs2025-08-22

This model was fine-tuned on the 100h Fareose ASR Ravnursson data set. It achieves the following results on the test set:

—Loss: 0.0969
—Wer: 7.42
—Cer: 2.09

📊 Word Error Rate (WER): 7.42% 📊 Character Error Rate (CER): 2.09%

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
3.2951	0.4877	1000	3.2352	100.0	98.4583
0.9295	0.9754	2000	0.5538	47.1472	13.1064
0.4669	1.4628	3000	0.2577	33.0484	8.5932
0.4318	1.9505	4000	0.2270	30.9733	7.9415
0.3422	2.4379	5000	0.1936	28.3253	7.1335
0.3198	2.9256	6000	0.1757	27.4309	6.8408
0.2391	3.4131	7000	0.1570	25.9814	6.3413
0.2622	3.9008	8000	0.1526	25.6466	6.2490
0.2132	4.3882	9000	0.1444	25.2853	6.0320
0.226	4.8759	10000	0.1488	24.6596	5.9500
0.1902	5.3633	11000	0.1453	24.1970	5.7724
0.1975	5.8510	12000	0.1280	23.8534	5.5981
0.1644	6.3385	13000	0.1342	23.7741	5.6549
0.1705	6.8261	14000	0.1302	23.3115	5.4308
0.1504	7.3136	15000	0.1278	22.8532	5.4190
0.1475	7.8013	16000	0.1236	22.7563	5.3369
0.1433	8.2887	17000	0.1235	22.4699	5.2249
0.1506	8.7764	18000	0.1221	22.3818	5.2217
0.1234	9.2638	19000	0.1102	22.1219	5.0939
0.1252	9.7515	20000	0.1162	21.8399	4.9921
0.1182	10.2390	21000	0.1161	21.6284	4.9408
0.1167	10.7267	22000	0.1089	21.7782	4.9740
0.0923	11.2141	23000	0.1138	21.1394	4.7759
0.0971	11.7018	24000	0.1070	21.2715	4.8098
0.0986	12.1892	25000	0.1151	21.1526	4.8383
0.094	12.6769	26000	0.1145	21.0072	4.7862
0.0893	13.1644	27000	0.1093	21.1173	4.7688
0.0855	13.6520	28000	0.1052	20.5710	4.5889
0.0819	14.1395	29000	0.1070	20.9719	4.7041
0.0915	14.6272	30000	0.1064	20.5181	4.5960
0.0788	15.1146	31000	0.1009	20.6415	4.5889
0.0778	15.6023	32000	0.1043	20.3111	4.5550
0.0795	16.0897	33000	0.1067	20.1921	4.5163
0.0686	16.5774	34000	0.1070	20.1524	4.4390
0.0638	17.0649	35000	0.1037	20.2626	4.4832
0.0552	17.5525	36000	0.1019	20.1921	4.5069
0.0625	18.0400	37000	0.0991	19.8969	4.3577
0.0599	18.5277	38000	0.1057	19.8925	4.3996
0.0566	19.0151	39000	0.1002	20.0467	4.4051
0.0516	19.5028	40000	0.1052	19.9277	4.3877
0.0519	19.9905	41000	0.1011	19.6898	4.3230
0.0557	20.4779	42000	0.1021	19.5621	4.2954
0.0437	20.9656	43000	0.1024	19.5180	4.2749
0.0421	21.4531	44000	0.1010	19.5136	4.2591
0.0678	21.9407	45000	0.0974	19.5841	4.2378
0.0524	22.4282	46000	0.0989	19.4739	4.2023
0.056	22.9159	47000	0.1012	19.3814	4.2339
0.0537	23.4033	48000	0.0966	19.3418	4.1755
0.0415	23.8910	49000	0.0978	19.2492	4.1786
0.0469	24.3784	50000	0.0983	19.1832	4.1510
0.0444	24.8661	51000	0.0951	19.1259	4.1092
0.0436	25.3536	52000	0.0967	19.1259	4.1273
0.0422	25.8413	53000	0.0964	19.1127	4.1289
0.0433	26.3287	54000	0.0960	19.0642	4.1076
0.038	26.8164	55000	0.0961	19.0289	4.0832
0.0447	27.3038	56000	0.0966	19.0774	4.0974
0.0468	27.7915	57000	0.0968	19.0466	4.0950
0.0444	28.2790	58000	0.0975	19.0334	4.0950
0.0368	28.7666	59000	0.0970	19.0334	4.0918
0.0496	29.2541	60000	0.0969	19.0069	4.0832
0.0512	29.7418	61000	0.0969	19.0025	4.0855

Framework versions

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