whisper-large-turbo-finetune

This model is a fine-tuned version of adriszmar/whisper-large-v3-turbo-es on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0337
• Wer: 6.6030

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 100
• num_epochs: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5524	0.4324	200	0.4671	36.8885
0.3595	0.8649	400	0.3303	27.6716
0.219	1.2962	600	0.2166	26.2860
0.1551	1.7286	800	0.1712	19.2001
0.0996	2.16	1000	0.1329	18.1923
0.0717	2.5924	1200	0.1003	14.0248
0.0563	3.0238	1400	0.0790	12.8280
0.0338	3.4562	1600	0.0630	11.9253
0.0276	3.8886	1800	0.0480	6.6240
0.0161	4.32	2000	0.0337	6.6030
0.0104	4.7524	2200	0.0267	4.8604
0.0058	5.1838	2400	0.0208	4.9339
0.0032	5.6162	2600	0.0177	3.1388

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1