zuazo/whisper-base-eu-cv21.0

Whisper Base Basque

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

• Loss: 0.5268
• Wer: 12.7885

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: 128
• eval_batch_size: 64
• 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
0.005	22.2222	5000	0.3503	16.5149
0.0022	44.4444	10000	0.3754	15.6124
0.0015	66.6667	15000	0.3950	15.3358
0.0018	88.8889	20000	0.4129	15.5708
0.0015	111.1111	25000	0.4234	15.1078
0.0004	133.3333	30000	0.4302	14.7167
0.0012	155.5556	35000	0.4466	15.0627
0.0006	177.7778	40000	0.4500	15.2569
0.0	200.0	45000	0.4556	13.8705
0.0	222.2222	50000	0.4783	13.3815
0.0	244.4444	55000	0.5086	13.0174
0.0	266.6667	60000	0.5255	12.9151
0.0	288.8889	65000	0.5255	12.8024
0.0	311.1111	70000	0.5268	12.7885
0.0	333.3333	75000	0.5301	12.8058
0.0	355.5556	80000	0.5325	12.8483
0.0	377.7778	85000	0.5351	12.8856
0.0	400.0	90000	0.5368	12.9472
0.0	422.2222	95000	0.5378	12.9662

Framework versions

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