Interesting. Is there an explanation for this? I mean the embedding size is smaller but that's probably hardly it? So more like a code optimization?

→ View original post on X — @rasbt