Modelpublic

MohammadFazli/Whisper-small-model-FullFinetuning-train-test

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes4downloads

Model Card

Whisper-small-model-FullFinetuning-train-test

This model is a fine-tuned version of openai/whisper-small on the commonvoice13_0 dataset. It achieves the following results on the evaluation set:

—Loss: 1.6703
—Wer: 96.9697

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: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 50
—training_steps: 2000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.131	0.05	100	2.2796	100.0
0.6661	0.1	200	2.1839	96.9697
0.5889	0.15	300	2.1523	107.5758
0.5005	0.2	400	1.9232	107.5758
0.421	0.25	500	1.9957	101.5152
0.4363	0.3	600	1.9424	96.9697
0.3583	0.35	700	1.9280	95.4545
0.2703	0.4	800	2.0008	93.9394
0.2836	0.45	900	1.9656	90.9091
0.2948	0.5	1000	1.7947	100.0
0.2236	0.55	1100	1.5250	100.0
0.2029	0.6	1200	1.8172	95.4545
0.1811	0.65	1300	1.8721	90.9091
0.1186	0.7	1400	1.7841	96.9697
0.1765	0.75	1500	1.7462	95.4545
0.1328	0.8	1600	1.7392	100.0
0.1428	0.85	1700	1.7447	100.0
0.1634	1.02	1800	1.6684	90.9091
0.1678	1.07	1900	1.6602	101.5152
0.1826	1.12	2000	1.6703	96.9697

Framework versions

—Transformers 4.35.2
—Pytorch 2.1.0+cu121
—Datasets 2.16.1
—Tokenizers 0.15.1