Modelpublic

tarunanand/wav2vec2-common_voice-tr-demo

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes5downloads

Model Card

wav2vec2-common_voice-tr-demo

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

—Loss: 0.3760
—Wer: 0.3521

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: 0.0003
—trainbatchsize: 16
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 32
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 15.0
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.92	100	3.5999	1.0
No log	1.83	200	2.9942	0.9999
No log	2.75	300	0.9031	0.7883
No log	3.67	400	0.5930	0.6226
3.1501	4.59	500	0.4967	0.5234
3.1501	5.5	600	0.4888	0.5053
3.1501	6.42	700	0.4393	0.4745
3.1501	7.34	800	0.4362	0.4370
3.1501	8.26	900	0.4384	0.4224
0.2259	9.17	1000	0.4169	0.4009
0.2259	10.09	1100	0.3965	0.3887
0.2259	11.01	1200	0.4072	0.3840
0.2259	11.93	1300	0.3937	0.3703
0.2259	12.84	1400	0.3901	0.3655
0.1024	13.76	1500	0.3835	0.3559
0.1024	14.68	1600	0.3780	0.3534

Framework versions

—Transformers 4.28.0.dev0
—Pytorch 2.0.0+cu117
—Datasets 2.11.0
—Tokenizers 0.13.3