Modelpublic

4m1g0/wav2vec2-large-xls-r-53m-gl-jupyter7

sourceHugging Faceapache-2.0updated 4y agoView on Hugging Face

0likes70downloads

Model Card

wav2vec2-large-xls-r-53m-gl-jupyter7

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.1000
—Wer: 0.0639

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: 60
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.8697	3.36	400	0.2631	0.2756
0.1569	6.72	800	0.1243	0.1300
0.0663	10.08	1200	0.1124	0.1153
0.0468	13.44	1600	0.1118	0.1037
0.0356	16.8	2000	0.1102	0.0978
0.0306	20.17	2400	0.1095	0.0935
0.0244	23.53	2800	0.1072	0.0844
0.0228	26.89	3200	0.1014	0.0874
0.0192	30.25	3600	0.1084	0.0831
0.0174	33.61	4000	0.1048	0.0772
0.0142	36.97	4400	0.1063	0.0764
0.0131	40.33	4800	0.1046	0.0770
0.0116	43.69	5200	0.0999	0.0716
0.0095	47.06	5600	0.1044	0.0729
0.0077	50.42	6000	0.1024	0.0670
0.0071	53.78	6400	0.0968	0.0631
0.0064	57.14	6800	0.1000	0.0639

Framework versions

—Transformers 4.11.3
—Pytorch 1.9.1+cu111
—Datasets 1.18.3
—Tokenizers 0.10.3