zuazo/whisper-medium-eu-cv21.0

Whisper Medium Basque

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

• Loss: 0.3958
• Wer: 8.3789

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: 64
• eval_batch_size: 32
• seed: 42
• 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
0.0102	11.1111	5000	0.2346	10.0851
0.0045	22.2222	10000	0.2662	10.2880
0.0035	33.3333	15000	0.2865	10.0383
0.0046	44.4444	20000	0.2913	9.9889
0.0018	55.5556	25000	0.3080	9.8797
0.0016	66.6667	30000	0.3096	9.8380
0.0031	77.7778	35000	0.3158	9.9612
0.0018	88.8889	40000	0.3317	10.2646
0.001	100.0	45000	0.3321	10.1380
0.0003	111.1111	50000	0.3275	9.7904
0.0007	122.2222	55000	0.3265	10.0401
0.0	133.3333	60000	0.3307	9.5641
0.0	144.4444	65000	0.3337	9.7461
0.0	155.5556	70000	0.3444	9.6820
0.0002	166.6667	75000	0.3503	9.8346
0.0	177.7778	80000	0.3586	9.1531
0.0	188.8889	85000	0.3744	8.7881
0.0	200.0	90000	0.3871	8.5323
0.0	211.1111	95000	0.3938	8.4040
0.0	222.2222	100000	0.3958	8.3789

Framework versions

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