zuazo/whisper-large-eu-cv21.0

Whisper Large Basque

This model is a fine-tuned version of openai/whisper-large on the common_voice_21_0_eu dataset. It achieves the following results on the evaluation set:

• Loss: 0.2974
• Wer: 9.7643

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: 3.75e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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_steps: 500
• training_steps: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0114	11.1112	5000	0.2477	10.4770
0.0052	22.2225	10000	0.2739	9.9325
0.0043	33.3337	15000	0.2855	9.8441
0.0034	44.4449	20000	0.2974	9.7643
0.0031	55.5562	25000	0.3101	10.1424
0.0034	66.6674	30000	0.3182	10.3530
0.0026	77.7786	35000	0.3195	10.3609
0.0032	88.8899	40000	0.3279	10.4146
0.0022	100.0	45000	0.3253	9.9915

Framework versions

• Transformers 4.52.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1