data2vec-audio-base-960h_060210

This model is a fine-tuned version of facebook/data2vec-audio-base-960h on the UIT-ViMD dataset. It achieves the following results on the evaluation set:

• Loss: nan
• Wer: 0.9997

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.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_ratio: 0.05
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
13.7284	0.0639	30	13.3308	1.0
11.8389	0.1278	60	9.1456	0.9997
8.4786	0.1917	90	5.2784	0.9997
5.5272	0.2556	120	4.1963	0.9998
4.3002	0.3195	150	3.8152	0.9998
4.3512	0.3834	180	3.6461	0.9997
6.6518	0.4473	210	nan	0.9997
0.0	0.5112	240	nan	0.9997
0.0	0.5751	270	nan	0.9997
0.0	0.6390	300	nan	0.9997
0.0	0.7029	330	nan	0.9997

Framework versions

• Transformers 4.48.3
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0