Modelpublic

sethilakshay/wav2vec2-base-finetune-flickr8k

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes5downloads

Model Card

wav2vec2-base-finetune-flickr8k

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the None dataset. It achieves the following results on the evaluation set:

—Loss: 3.0580
—Wer: 1.0

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: 12
—evalbatchsize: 2
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 15
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.2252	0.8	500	3.0787	1.0
3.1958	1.6	1000	3.0698	1.0
3.1512	2.4	1500	3.0571	1.0
3.1411	3.2	2000	3.0578	1.0
3.1186	4.0	2500	3.0553	1.0
3.0852	4.8	3000	3.0616	1.0
3.1063	5.6	3500	3.0621	1.0
3.0678	6.4	4000	3.0550	1.0
3.0239	7.2	4500	3.0592	1.0
3.0321	8.0	5000	3.0652	1.0
3.0396	8.8	5500	3.0542	1.0
3.0723	9.6	6000	3.0552	1.0
3.0567	10.4	6500	3.0545	1.0
3.0438	11.2	7000	3.0611	1.0
3.073	12.0	7500	3.0548	1.0
3.0293	12.8	8000	3.0585	1.0
3.0444	13.6	8500	3.0571	1.0
3.0213	14.4	9000	3.0580	1.0

Framework versions

—Transformers 4.27.4
—Pytorch 2.0.0
—Datasets 2.11.0
—Tokenizers 0.13.2