8/ MINT-1T – open-sources a large-scale multimodal interleaved dataset consisting of 1 trillion tokens which has 3.4 billion images; it also includes new sources such as PDFs and ArXiv papers.

→ View original post on X — @dair_ai