search

RUC-NLPIR / FlashRAG

⚡FlashRAG: A Python Toolkit for Efficient RAG Research
https://arxiv.org/abs/2405.13576
MIT License
891 stars 69 forks source link

Question about knowledge source #10

Closed XMHZZ closed 1 month ago

XMHZZ commented 1 month ago

Hi Team,

I'm wondering if all the Wiki-based tasks use the December 2018 snapshot (~20M 100-word passages from the DPR paper). For example, HotPotQA usually only uses the first paragraph of each Wikipedia page from a 2017 version.

Thanks!

ignorejjj commented 1 month ago

Hi, our current results are all using the dpr version of wiki dump you mentioned.

We do not reimplement the original setting of the specific dataset but only compare each method under the same setting.