LLM
![[LLM] SELF-RAG: Learning to Retrieve, Generate and Critique through Self-reflect](https://img1.daumcdn.net/thumb/R750x0/?scode=mtistory2&fname=https%3A%2F%2Fblog.kakaocdn.net%2Fdna%2FeeRatj%2FbtsELtgYam1%2FAAAAAAAAAAAAAAAAAAAAAH96_NkyK5UbNDBigAGWsW0hjkp1SFzm-z5pRqOVY3ip%2Fimg.png%3Fcredential%3DyqXZFxpELC7KVnFOS48ylbz2pIh7yKj8%26expires%3D1756652399%26allow_ip%3D%26allow_referer%3D%26signature%3DD%252FL396ckqpVI0SPYx5XglpLeiqs%253D)
[LLM] SELF-RAG: Learning to Retrieve, Generate and Critique through Self-reflect
ConceptLM을 이용해서 On-demand Retrieval을 할 수 있다.Reflection Token을 통해 generation의 quality와 attribution을 증진시킨다.ProblemExpense of runtime efficiencyRobustness to irrelevant contextLack of attributionHow to Solve Reflection Token이라는 Special Token을 사용해서 Retrieval이 필요한지 판단할 수 있게 하자.Retrieval Passage와 Generated Sentence에 대해서 Reflects를 하자.Reflection TokenRetrieve : 현재 문장을 보고 Retrieve과정이 필요한지 안한지에 대한 토큰이다.Is..