Modelpublic

Hanhpt23/whisper-small-chinesemed-free_E0-8

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes6downloads

Model Card

openai/whisper-small

This model is a fine-tuned version of openai/whisper-small on the pphuc25/ChiMed dataset. It achieves the following results on the evaluation set:

—Loss: 1.1189
—Wer: 80.7466
—Cer: 23.1952

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: 8
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 100
—num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.6947	1.0	161	0.8771	224.7544	75.6462
0.3688	2.0	322	0.8461	107.2692	32.0187
0.2168	3.0	483	0.9040	84.8723	26.0472
0.1117	4.0	644	0.9532	90.5697	28.1640
0.0928	5.0	805	0.9663	89.7839	28.2977
0.0672	6.0	966	1.0584	87.2299	31.1275
0.0534	7.0	1127	1.0810	86.4440	28.5651
0.0443	8.0	1288	1.0709	83.8900	29.1889
0.0415	9.0	1449	1.0984	85.0688	26.4929
0.0198	10.0	1610	1.1180	89.9804	27.3841
0.0182	11.0	1771	1.0824	86.4440	27.1613
0.0116	12.0	1932	1.1320	85.6582	26.1809
0.0107	13.0	2093	1.1042	82.5147	24.6658
0.0075	14.0	2254	1.1034	80.9430	24.5989
0.0021	15.0	2415	1.0967	78.9784	23.3734
0.0003	16.0	2576	1.1061	81.1395	23.4848
0.0002	17.0	2737	1.1144	81.1395	23.5294
0.0002	18.0	2898	1.1173	81.1395	23.3957
0.0004	19.0	3059	1.1184	80.7466	23.1506
0.0003	20.0	3220	1.1189	80.7466	23.1952

Framework versions

—Transformers 4.41.1
—Pytorch 2.3.0
—Datasets 2.19.1
—Tokenizers 0.19.1