Modelpublic

nsomazr/whisper-small-sukuma

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face

0likes5downloads

Model Card

whisper-small-sukuma

This model is a fine-tuned version of openai/whisper-small on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.2000
—Wer: 13.49
—Cer: 3.67

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: 1e-05
—trainbatchsize: 16
—evalbatchsize: 8
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—training_steps: 4000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.2403	1.2920	500	0.2914	29.29	7.41
0.1005	2.5840	1000	0.1915	17.35	4.47
0.0557	3.8760	1500	0.1722	15.47	4.46
0.0106	5.1680	2000	0.1860	13.78	3.63
0.0052	6.4599	2500	0.1907	13.63	3.65
0.0018	7.7519	3000	0.1954	13.5	3.58
0.0019	9.0439	3500	0.1977	13.52	3.64
0.0011	10.3359	4000	0.2000	13.49	3.67

Framework versions

—Transformers 5.2.0
—Pytorch 2.9.0+cu126
—Datasets 4.0.0
—Tokenizers 0.22.2