Modelpublic

KasuleTrevor/cdli-whisper-lg-sulaimank-full-a40-5e-5-constant-warmup-10k

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes10downloads

Model Card

cdli-whisper-lg-sulaimank-full-a40-5e-5-constant-warmup-10k

This model is a fine-tuned version of sulaimank/whisper-small-luganda-400hrs on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 3.3498
—Wer: 0.7083
—Cer: 0.4358

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: 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: constantwithwarmup
—lrschedulerwarmup_steps: 500
—training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.192	1.3550	500	1.4079	0.6726	0.3833
0.9627	2.7100	1000	1.1817	0.6500	0.3635
0.7192	4.0650	1500	1.1718	0.6331	0.3472
0.5879	5.4201	2000	1.1887	0.6328	0.3507
0.4554	6.7751	2500	1.2667	0.6279	0.3477
0.3708	8.1301	3000	1.4589	0.6472	0.3651
0.2921	9.4851	3500	1.6562	0.6522	0.3721
0.2028	10.8401	4000	1.8895	0.6608	0.3814
0.187	12.1951	4500	2.3483	0.6829	0.4060
0.1028	13.5501	5000	2.6295	0.6899	0.4139
0.0633	14.9051	5500	2.8355	0.6932	0.4169
0.0468	16.2602	6000	3.0647	0.6993	0.4306
0.0212	17.6152	6500	3.1580	0.7010	0.4287
0.0218	18.9702	7000	3.2333	0.7066	0.4325
0.0107	20.3252	7500	3.2798	0.7038	0.4359
0.0097	21.6802	8000	3.3077	0.7025	0.4334
0.0109	23.0352	8500	3.3265	0.7082	0.4420
0.005	24.3902	9000	3.3377	0.7039	0.4334
0.0045	25.7453	9500	3.3447	0.7033	0.4328
0.0042	27.1003	10000	3.3498	0.7083	0.4358

Framework versions

—Transformers 4.57.6
—Pytorch 2.7.1+cu118
—Datasets 3.6.0
—Tokenizers 0.22.2