Positive users see OPSD as a gold standard quicker fix that works well especially for thinking models, while negative users worry about implementation difficulties and expect it to fail against PPO.
Based on 8 visible X reactions from 27 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@vivek_2332 c2.5 is the gold standard use case for it, agreed it is a quicker fix for things that rlvr can fix eventually, but more surgical, and more elegant than sft-based methods
@willccbb thats why we have people like you to make it work opsd would be good if solved with actual better outcomes
@TheGoldenAnvil @PrimeIntellect the algorithm we’ve bet the most on is policy gradient seems to be working out
@willccbb I am currently emotionally disturbed just trying to get my implementation to work 😔
Positive users see OPSD as a gold standard quicker fix that works well especially for thinking models, while negative users worry about implementation difficulties and expect it to fail against PPO.
Based on 8 visible X reactions from 27 accounts; directional sample.
Ask a question below.
Published answers will appear here.