
[Daily] Self-Correct Reasoning / Verification of LLM
·
NLP
오늘은 interview 준비를 하면서 평소 관심있었던 self-correction 논문을 읽었다!1. Small Language Models Need Strong Verifiers to Self-Correct ReasoningTLDR;Small LLM 으로 self-correction data를 모으고 self-refienment ability를 가질 수 있도록 fine-tuneSelf-refining model을 만드는게 목표MotivationSelf-correction: Self-verify + Self-refineSelf-verify: LLM이 initial solution을 judgeSelf-refine: Incorrect 하면 solution을 reviseSelf-refine: Critique..