it is actually shown that The straightforward pre-training activity of predicting which caption goes with which graphic is definitely an economical and scalable way to master SOTA graphic representations from scratch on https://k2spiceshop.com/product/liquid-k2-on-paper-online/