Modelpublic

cheesethuong/wav2vec2-vietnamese-asr-v1

sourceHugging Facecc-by-nc-4.0updated 8mo agoView on Hugging Face

0likes22downloads

Model Card

wav2vec2-vietnamese-asr-v1

This model is a fine-tuned version of nguyenvulebinh/wav2vec2-base-vietnamese-250h on the None dataset. It achieves the following results on the evaluation set:

—Loss: 3.5180
—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: 0.0003
—trainbatchsize: 32
—evalbatchsize: 16
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: linear
—lrschedulerwarmup_steps: 1000
—num_epochs: 30
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4272	1.3369	500	3.4439	1.0
3.501	2.6738	1000	3.5412	1.0
3.5189	4.0107	1500	3.5536	1.0
3.5445	5.3476	2000	3.5352	1.0
3.4916	6.6845	2500	3.5678	1.0
3.5013	8.0214	3000	3.5549	1.0
3.4873	9.3583	3500	3.5226	1.0
3.4893	10.6952	4000	3.5350	1.0
3.4986	12.0321	4500	3.5194	1.0
3.4826	13.3690	5000	3.5252	1.0
3.4875	14.7059	5500	3.5238	1.0
3.4928	16.0428	6000	3.5334	1.0
3.4837	17.3797	6500	3.5296	1.0
3.4758	18.7166	7000	3.5277	1.0
3.4817	20.0535	7500	3.5083	1.0
3.4796	21.3904	8000	3.5263	1.0
3.4904	22.7273	8500	3.5196	1.0
3.5063	24.0642	9000	3.5140	1.0
3.4833	25.4011	9500	3.5206	1.0
3.4763	26.7380	10000	3.5206	1.0
3.4652	28.0749	10500	3.5262	1.0
3.4807	29.4118	11000	3.5180	1.0

Framework versions

—Transformers 4.57.1
—Pytorch 2.8.0+cu126
—Datasets 4.4.2
—Tokenizers 0.22.1