wav2vec2-base-vi-160h-finetuned

This model is a fine-tuned version of khanhld/wav2vec2-base-vietnamese-160h on the nguyendv02/ViMD_Dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.4200
• Wer: 0.1873

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
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 13
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.626	1.0639	500	0.5423	0.3089
0.6393	2.1278	1000	0.4949	0.2891
0.455	3.1917	1500	0.4622	0.2541
0.4141	4.2556	2000	0.4576	0.2451
0.4672	5.3195	2500	0.4485	0.2411
0.3456	6.3834	3000	0.4402	0.2242
0.3189	7.4473	3500	0.4565	0.2227
0.3851	8.5112	4000	0.4398	0.2253
0.2758	9.5751	4500	0.4432	0.2138
0.262	10.6390	5000	0.4546	0.2127
0.226	11.7029	5500	0.4668	0.2071
0.1971	12.7689	6000	0.4765	0.1946

Framework versions

• Transformers 4.53.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2