Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
Specifically, we propose several pretraining tasks that cover plot deconstruction and numerical reasoning which are the key capabilities in visual language modeling.