Wav2Vec_VinData

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8569
• Wer: 0.3143
• Cer: 0.1316
• Syer: 0.3143

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 1000
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Syer
27.7293	0.6591	500	3.4983	1.0	1.0	1.0
13.7701	1.3177	1000	1.4881	0.6608	0.2736	0.6608
11.0362	1.9768	1500	1.1284	0.4733	0.1992	0.4733
9.4754	2.6354	2000	0.9490	0.3819	0.1600	0.3819
8.2131	3.2940	2500	0.9020	0.3377	0.1412	0.3377
7.2151	3.9530	3000	0.8558	0.3147	0.1316	0.3147
7.2151	4.0	3036	0.8569	0.3143	0.1316	0.3143

Framework versions

• Transformers 5.16.1
• Pytorch 2.11.0+cu128
• Datasets 4.0.0
• Tokenizers 0.23.1