Created by Meta AI researchers, Data2vec 2.0 can train self-supervised models for vision, speech and text up to 16x faster than the most popular existing algorithm for images — achieving the same accuracy. Read more & access the code

→ View original post on X — @aiatmeta,