@turchin yes. but i don't think that will result in the same model. the policy that sonnet 3.6 learned from RL is optimized for its own architecture/initialization
same thread: 1955802026241351771 1955806704400261623 1955807323001381254 1955811638667567324 1955856679595368449 1955952763747360905 1956115436086681987 1956115702143967495 1956475633825177911 1956476139259810116 1956521268683104275
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.