Modelpublic

mikr/w2v-bert-2.0-czech-colab-cv16

sourceHugging Facemitupdated 3y agoView on Hugging Face

2likes13downloads

Model Card

w2v-bert-2.0-czech-colab-cv16

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the commonvoice16_0 dataset. It achieves the following results on the evaluation set:

—Loss: 0.1023
—Wer: 0.0573

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: 5e-05
—trainbatchsize: 64
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_steps: 500
—num_epochs: 10
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5297	0.66	300	0.1448	0.1299
0.0886	1.32	600	0.1353	0.1051
0.0717	1.98	900	0.1157	0.0861
0.0463	2.64	1200	0.0994	0.0759
0.0404	3.3	1500	0.1054	0.0724
0.0314	3.96	1800	0.0915	0.0694
0.0227	4.63	2100	0.0926	0.0664
0.0205	5.29	2400	0.0992	0.0652
0.0161	5.95	2700	0.0932	0.0654
0.0124	6.61	3000	0.0902	0.0629
0.0097	7.27	3300	0.0970	0.0612
0.0081	7.93	3600	0.0946	0.0602
0.0054	8.59	3900	0.0962	0.0588
0.0048	9.25	4200	0.1029	0.0579
0.0034	9.91	4500	0.1023	0.0573

Framework versions

—Transformers 4.38.0.dev0
—Pytorch 2.2.0+cu121
—Datasets 2.16.2.dev0
—Tokenizers 0.15.1