@pedropverani@0xCVYH pior q nem esta tao quantizado bro, o modelo continua com todos os parametros por isso a velocidade ser s/t e nao t/s kkkkkk, se fosse quantizado pra cabaer em 8GB + cpu seria bem mais rapido so q tbm bem mais burro
@daniel_mac8 Bro i dont think that luna was trained to act as subagent, cause to use it as subagent we gotta change codex files cause by default it's not even allowed
@thsottiaux i love that GPT team is always giving us opportunity to keep working, things that i don't see in any other product of the market and beyond all of that community care we also have the best model rn