Early testing suggest it's better than the "raw" 120b instruct.
#2
by
6346y9uey
- opened
Solid work, this one's a frontrunner!
It may still be slightly behind c4ai command + in reasoning, but it wins in a few other key areas; less likely to get into repetitive statements (so far at least), a (debatably) better tone for roleplaying; follows instruct-commands well. Models like this give hope that the L3 finetunes will show strong progress.