wav2vec2-base-960h_041426

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the UIT-ViMD dataset. It achieves the following results on the evaluation set:

• Loss: 3.3332
• 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: 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
13.3195	0.9787	23	11.5720	0.9997
8.8443	1.9362	46	4.6994	0.9997
4.4565	2.8936	69	3.6900	0.9997
3.5367	3.8511	92	3.3994	0.9997
3.3029	4.8085	115	3.3583	0.9997
3.2993	5.7660	138	3.3439	0.9997
3.2675	6.7234	161	3.3632	0.9997
3.258	7.6809	184	3.3283	0.9997
3.2519	8.6383	207	3.3472	0.9997
3.2528	9.5957	230	3.3368	0.9997
3.2501	10.5532	253	3.3426	0.9998
3.247	11.5106	276	3.3382	0.9997
3.2484	12.4681	299	3.3332	0.9997

Framework versions

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