search

AkariAsai / self-rag

This includes the original implementation of SELF-RAG: Learning to Retrieve, Generate and Critique through self-reflection by Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi.
https://selfrag.github.io/
MIT License
1.76k stars 162 forks source link

Explanation needed for [Continue to Use Evidence] #66

Closed zhuzihan728 closed 5 months ago

zhuzihan728 commented 5 months ago

Hi, I noticed in your code there is another special token [Continue to Use Evidence] that also falls into the retrieval token group. As it seems not mentioned in your paper, could you explain when and how it is supposed to be generated and used by a LM? Thx in advance.

zhuzihan728 commented 5 months ago

I found that is indeed mentioned in your paper! Sorry for any confusion this issue might have caused.

We additionally have continue to use evidence, which indicates that a model can continue to use the evidence retrieved previously. For instance, a passage may contain rich factual information, and thus SELF-RAG generates multiple segments based on the passage.