whisper-large-basque

This model is a fine-tuned version of openai/whisper-large on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2281
• Wer: 11.8969

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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1763	0.42	500	0.2915	17.9244
0.1415	0.84	1000	0.2468	15.0738
0.1182	1.26	1500	0.2483	14.3144
0.1234	1.68	2000	0.2520	14.1380
0.0961	2.1	2500	0.2397	13.6557
0.0952	2.52	3000	0.2360	12.7999
0.0912	2.94	3500	0.2264	12.4284
0.0717	3.36	4000	0.2295	12.0590
0.0698	3.78	4500	0.2260	12.2581
0.0582	4.2	5000	0.2281	11.8969

Framework versions

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