mrqa/MRQA-Shared-Task-2019

dataset changed

yana-xuyan opened this issue · 2 comments

Hi! I downloaded the dev sets twice, the first time is 7.6 and the second time is 7.16. But I found that the dev sets of HotpotQA are different in two downloads. May I ask, except the dev set of HotpotQA, is there any other modification on the datasets provided with the link?

The MD5 hashes are different because they are gzipped, but the contents should be the same. HotpotQA should also be largely the same set of questions and answers besides for a few examples, though the qid's were regenerated between versions.

If you look at the README datasets section, it contains links to notes/issues explaining the changes.

Thank you for your reply!