igarciahuidobro/whisper-tiny-300-audios-v3

Whisper Pre Tuned 300 Audios - Nacho v3.0

This model is a fine-tuned version of rasel35/whisper-base-es-medical-terms on the 300 audios 1.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3444
• Wer: 16.1793

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use 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_steps: 50
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.8732	1.0	18	1.1907	60.6238
0.676	2.0	36	0.4489	21.0526
0.2633	3.0	54	0.4061	17.9337
0.132	4.0	72	0.3804	17.9337
0.0802	5.0	90	0.3507	41.7154
0.0498	6.0	108	0.3660	18.5185
0.036	7.0	126	0.3614	17.3489
0.0213	8.0	144	0.3329	15.9844
0.0152	9.0	162	0.3453	15.7895
0.0042	9.4507	170	0.3444	16.1793

Framework versions

• Transformers 4.48.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.21.0