Code for ALBEF: a new vision-language pre-training method - View it on GitHub