Closed ShiyuNee closed 8 months ago
The count of samples in wikipedia-nq is 3000+ while the count in original nq dataset is nearly 8000.
I would like to know how the data is screened.
Thanks!
I got the answer in the paper of DPR.
sorry for not replying in time.
just to have a record: The training data for wikipedia-nq is converted from the original DPR repo. The difference from original NQ is due to:
sorry for not replying in time.
just to have a record: The training data for wikipedia-nq is converted from the original DPR repo. The difference from original NQ is due to:
Thanks.
The count of samples in wikipedia-nq is 3000+ while the count in original nq dataset is nearly 8000.
I would like to know how the data is screened.
Thanks!