Linear transformers have their own downsides, and there have been ongoing efforts in the community on improving their performance while maintaining training efficiency. Recently, CSAIL researchers have proposed a novel parallel algorithm for efficiently training an expressive

→ View original post on X — @mit_csail