whisper-large-basque

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

• Loss: 0.2113
• Wer: 6.4467

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: 1e-05
• train_batch_size: 256
• eval_batch_size: 32
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 512
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0913	0.66	500	0.1614	9.1333
0.0529	1.32	1000	0.1456	7.5385
0.0495	1.98	1500	0.1379	7.3054
0.0324	2.64	2000	0.1365	6.8024
0.0202	3.3	2500	0.1440	6.7656
0.0202	3.96	3000	0.1453	6.8576
0.0123	4.62	3500	0.1569	6.6552
0.0064	5.28	4000	0.1666	6.7718
0.0071	5.94	4500	0.1667	6.6429
0.0037	6.61	5000	0.1761	6.4957
0.0019	7.27	5500	0.1780	6.5019
0.002	7.93	6000	0.1830	6.4835
0.0013	8.59	6500	0.1901	6.6429
0.0006	9.25	7000	0.1926	6.4344
0.0007	9.91	7500	0.1940	6.4957
0.0003	10.57	8000	0.2000	6.5693
0.0002	11.23	8500	0.2037	6.4957
0.0002	11.89	9000	0.2080	6.4712
0.0002	12.55	9500	0.2103	6.4712
0.0002	13.21	10000	0.2113	6.4467

Framework versions

• Transformers 4.38.0
• Pytorch 2.1.1+cu121
• Datasets 2.8.0
• Tokenizers 0.15.2