Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
We train a sequence Transformer to auto-regressively predict pixels,
without incorporating knowledge of the 2D input structure.