Wav2Vec_VinData_Small

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: 0.8539
• Wer: 0.2793
• Cer: 0.1188
• Syer: 0.2793

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: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Syer
27.7112	0.6602	500	3.4667	1.0	1.0	1.0
14.3239	1.3195	1000	1.4244	0.6617	0.2781	0.6617
11.4665	1.9797	1500	1.0741	0.4918	0.2027	0.4918
9.8071	2.6390	2000	1.0257	0.4112	0.1701	0.4112
8.0177	3.2984	2500	0.8888	0.3708	0.1521	0.3708
8.0885	3.9586	3000	0.8754	0.3321	0.1388	0.3321
7.1819	4.6179	3500	0.8762	0.3085	0.1301	0.3085
6.0268	5.2773	4000	0.8498	0.2904	0.1224	0.2904
6.0018	5.9374	4500	0.8529	0.2791	0.1188	0.2791
6.0018	6.0	4548	0.8539	0.2793	0.1188	0.2793

Framework versions

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