Models were tested using lm-eval-harness (
@aieleuther
) on tasks like sentence classification, question answering, etc. The average of the scores in the eight tasks is reported. Our models achieve scores far ahead of other Japanese models!

→ View original post on X — @stabilityai,