wav2vec2-large-xlsr-53_030909

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

• Loss: 3.3345
• 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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: 0.1
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
13.8904	0.9231	3	12.5105	0.9998
10.5144	1.6154	6	12.2960	1.0
9.314	2.3077	9	9.0572	0.9997
7.7092	3.0	12	6.5385	0.9997
7.2508	3.9231	15	5.3754	0.9997
4.2192	4.6154	18	4.6707	0.9997
3.7228	5.3077	21	4.2868	0.9997
3.4506	6.0	24	4.0406	0.9997
4.1163	6.9231	27	3.8480	0.9997
2.8937	7.6154	30	3.6930	0.9997
2.8004	8.3077	33	3.5770	0.9997
2.6908	9.0	36	3.4962	0.9997
3.5181	9.9231	39	3.4494	0.9997
2.6012	10.6154	42	3.4272	0.9997
2.5532	11.3077	45	3.3969	0.9997
2.5429	12.0	48	3.3711	0.9997
3.3727	12.9231	51	3.3724	0.9997
2.5283	13.6154	54	3.3591	0.9997
2.5149	14.3077	57	3.3542	0.9997
2.5217	15.0	60	3.3539	0.9997
3.3412	15.9231	63	3.3400	0.9997
2.5332	16.6154	66	3.3409	0.9998
2.4927	17.3077	69	3.3521	0.9997
2.5114	18.0	72	3.3541	0.9997
3.3431	18.9231	75	3.3576	0.9997
2.5085	19.6154	78	3.3484	0.9997
2.5107	20.3077	81	3.3388	0.9997
2.4972	21.0	84	3.3349	0.9998
3.3369	21.9231	87	3.3343	0.9997
2.5042	22.6154	90	3.3345	0.9997

Framework versions

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