Ansu/whisper-large-v3-basque

composite_eu+faktoria+Dialektoak_54377

This model is a fine-tuned version of openai/whisper-large-v3 on the composite_eu, faktoria, Dialektoak dataset. It achieves the following results on the evaluation set:

• Loss: 0.2101
• Wer: 12.0919

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: 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1589	0.42	500	0.2720	18.1932
0.1276	0.84	1000	0.2271	15.1271
0.0953	1.26	1500	0.2117	13.5962
0.093	1.68	2000	0.2053	13.0462
0.0644	2.1	2500	0.2015	12.1739
0.0654	2.52	3000	0.2008	12.3197
0.0636	2.94	3500	0.1982	11.7142
0.0442	3.36	4000	0.2101	12.0919
0.0439	3.78	4500	0.2072	12.1945
0.0313	4.2	5000	0.2195	12.1370

Framework versions

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