logoalt Hacker News

porridgeraisinyesterday at 8:33 PM0 repliesview on HN

The training here is RL training, the rollouts there are not different from inference and have access to the same tools as regular inference.