naver / bergen

Benchmarking library for RAG
Other
122 stars 12 forks source link

Multilingual RAG #2

Closed nadiinchi closed 4 months ago

nadiinchi commented 5 months ago

Add support for mRAG:

sclincha commented 5 months ago

as short is not proper english according to a native speaker

Le jeu. 13 juin 2024 à 09:46, David Rau @.***> a écrit :

@.**** commented on this pull request.

In config/prompt/basic.yaml https://github.com/naver/bergen/pull/2#discussion_r1637727711:

user: f"Background:\n{docs}\n\nQuestion:\ {question}" -system_without_docs: "You are a helpful assistant. Answer the questions as short as possible." +system_without_docs: "You are a helpful assistant. Answer the questions as briefly as possible."

is there a reason to change it?

— Reply to this email directly, view it on GitHub https://github.com/naver/bergen/pull/2#discussion_r1637727711, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABEGUKC2F4W4D3LNXFHIINTZHFE6LAVCNFSM6AAAAABJE42YWCVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDCMJUHE4DQNJYGU . You are receiving this because you commented.Message ID: @.***>

davidmrau commented 5 months ago

I see. In this case, we either have to be ok with numbers varying with the ones we report in the paper, or with re-running everything 🤔

sclincha commented 5 months ago

Personally, I am ok with number slightly varyinfg, it should not change the conclusions ....

Le jeu. 13 juin 2024 à 09:50, David Rau @.***> a écrit :

I see. In this case, we either have to be ok with numbers varying with the ones we report in the paper, or with re-running everything 🤔

— Reply to this email directly, view it on GitHub https://github.com/naver/bergen/pull/2#issuecomment-2164888215, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABEGUKAQJMW4LHZQQ2HQJDDZHFFMZAVCNFSM6AAAAABJE42YWCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNRUHA4DQMRRGU . You are receiving this because you commented.Message ID: @.***>

davidmrau commented 5 months ago

Before we merge I would like to have a look at the change of the logic (if there is any)