Modelpublic

Alvin-Nahabwe/asr-africa-wav2vec2-xls-r-300m-mdc-rutoro-asr-1.0.0

sourceHugging Faceupdated 10mo agoView on Hugging Face

0likes24downloads

Model Card

asr-africa-wav2vec2-xls-r-300m-mdc-rutoro-asr-1.0.0

This model is a fine-tuned version of Alvin-Nahabwe/asr-africa-wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 1.0559
—Wer: 0.7247
—Cer: 0.1730

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: 5e-05
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.1
—num_epochs: 50.0
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
9.9224	1.0	84	3.4844	1.0	1.0
2.8143	2.0	168	2.0005	1.0	0.8775
1.3934	3.0	252	0.8859	0.9126	0.2768
0.9363	4.0	336	0.7422	0.8299	0.2138
0.7583	5.0	420	0.6761	0.7833	0.1947
0.6212	6.0	504	0.6583	0.7580	0.1835
0.5014	7.0	588	0.6707	0.7427	0.1785
0.4073	8.0	672	0.6948	0.7327	0.1744
0.3318	9.0	756	0.7604	0.7233	0.1745
0.2796	10.0	840	0.7809	0.7338	0.1735
0.2358	11.0	924	0.8414	0.7317	0.1734
0.205	12.0	1008	0.8859	0.7311	0.1757
0.1824	13.0	1092	0.9159	0.7319	0.1730
0.1603	14.0	1176	0.9528	0.7199	0.1724
0.1565	15.0	1260	0.9799	0.7275	0.1744
0.1424	16.0	1344	1.0055	0.7272	0.1717
0.135	17.0	1428	1.0212	0.7205	0.1712
0.1326	18.0	1512	1.0245	0.7302	0.1750
0.1219	19.0	1596	1.0559	0.7247	0.1730

Framework versions

—Transformers 4.57.1
—Pytorch 2.9.0+cu128
—Datasets 4.4.0
—Tokenizers 0.22.1