Modelpublic

Leonel-Maia/mHuBERT-147-Hausa

sourceHugging Facecc-by-nc-sa-4.0updated 1y agoView on Hugging Face

0likes8downloads

Model Card

mHuBERT-147-Hausa

This model is a fine-tuned version of utter-project/mHuBERT-147 on the LAFRICAMOBILE/HAUSA - DEFAULT dataset. It achieves the following results on the evaluation set:

—Loss: 0.2528
—Wer: 0.3308
—Cer: 0.0817

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: 8
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 4
—totaltrainbatch_size: 32
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 1000
—num_epochs: 60.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.4357	1.6595	500	6.1556	1.0	1.0
2.7478	3.3215	1000	2.7490	1.0	1.0
2.0378	4.9809	1500	1.7083	1.0	0.4821
0.6111	6.6429	2000	0.5210	0.5636	0.1398
0.4629	8.3049	2500	0.3899	0.4782	0.1190
0.3658	9.9644	3000	0.3376	0.4359	0.1111
0.3326	11.6263	3500	0.3015	0.4042	0.1014
0.301	13.2883	4000	0.2863	0.3849	0.0970
0.2782	14.9478	4500	0.2747	0.3638	0.0919
0.2705	16.6098	5000	0.2692	0.3640	0.0911
0.2538	18.2717	5500	0.2670	0.3504	0.0867
0.2611	19.9312	6000	0.2537	0.3476	0.0850
0.2315	21.5932	6500	0.2668	0.3515	0.0863
0.232	23.2552	7000	0.2659	0.3350	0.0829
0.2245	24.9147	7500	0.2526	0.3303	0.0816
0.2264	26.5766	8000	0.2555	0.3325	0.0820
0.2144	28.2386	8500	0.2547	0.3236	0.0794
0.2153	29.8981	9000	0.2539	0.3243	0.0801

Framework versions

—Transformers 4.50.3
—Pytorch 2.7.0+cu126
—Datasets 3.5.0
—Tokenizers 0.21.1