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.

Home Page:https://selfrag.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Bug report & Self-RAG never predicts `[Retrieval]` when set `threshold=None`

ZhangzihanGit opened this issue · comments

Hi @AkariAsai,

I noticed that you fixed the bug in this issue. However, there is no pred in the context, do you mean pred_text?

In addition, when I set threshold = None and let Self-RAG do the prediction of [Retrieval], it seems that the pred_text never contains the token, and therefore it does not retrieve at all on the PopQA dataset, i.e. the retrieval frequency = 0.

Is there something wrong?

Thank you!