Modelpublic

Gwenn-LR/whisper-tiny

sourceHugging Faceapache-2.0updated 2y agoView on Hugging Face

0likes5downloads

Model Card

whisper-tiny

This model is a fine-tuned version of openai/whisper-tiny on the PolyAI/minds14 dataset. It achieves the following results on the evaluation set:

—Loss: 0.5900
—Wer Ortho: 0.3103
—Wer: 0.3103

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 adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: constantwithwarmup
—lrschedulerwarmup_steps: 500
—training_steps: 500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.7824	1.7857	50	1.0732	0.4565	0.4565
0.3528	3.5714	100	0.4932	0.3745	0.3745
0.1313	5.3571	150	0.5215	0.3430	0.3430
0.035	7.1429	200	0.5468	0.3387	0.3387
0.0103	8.9286	250	0.5900	0.3103	0.3103
0.0085	10.7143	300	0.6345	0.3307	0.3307
0.009	12.5	350	0.6771	0.3418	0.3418
0.0137	14.2857	400	0.6456	0.3374	0.3374
0.0138	16.0714	450	0.6171	0.3294	0.3294
0.0151	17.8571	500	0.7379	0.4312	0.4312

Framework versions

—Transformers 4.49.0.dev0
—Pytorch 2.5.1+cu124
—Datasets 3.2.0
—Tokenizers 0.21.0