Modelpublic

volkan-aslan/whisper-tiny-tr-optimized-v2

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes4downloads

Model Card

Whisper Tiny TR Optimized V2- Volkan ASLAN

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

—Loss: 0.7111
—Wer: 38.2594

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: 7e-05
—trainbatchsize: 128
—evalbatchsize: 128
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 600
—training_steps: 8000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3522	1.3774	500	0.5757	44.1437
0.2275	2.7548	1000	0.5423	42.0455
0.0836	4.1322	1500	0.5472	41.2382
0.0609	5.5096	2000	0.5756	40.3372
0.0432	6.8871	2500	0.6074	40.9146
0.018	8.2645	3000	0.6295	39.7496
0.0156	9.6419	3500	0.6646	40.7170
0.0086	11.0193	4000	0.6617	40.3321
0.0052	12.3967	4500	0.6812	39.7224
0.0024	13.7741	5000	0.6888	39.1586
0.001	15.1515	5500	0.6927	38.5881
0.0009	16.5289	6000	0.6959	38.5557
0.0008	17.9063	6500	0.7012	38.2134
0.0005	19.2837	7000	0.7066	38.3054
0.0005	20.6612	7500	0.7097	38.3871
0.0005	22.0386	8000	0.7111	38.2594

Framework versions

—Transformers 4.51.3
—Pytorch 2.5.1+cu124
—Datasets 3.5.0
—Tokenizers 0.21.0