transformers

8ce50c5d - Make it possible to predict_with_generate for vision and speech models