Scaling vision transformers to 22 billion parameters

Dehghani, Mostafa, Djolonga, Josip, Mustafa, Basil · arXiv · 2023