Using pyserini to evaluate on NQ or TriviaQA requires an additional index of Wikipedia which seems to be redundant to me.
It seems to be easier to save the doc indices that contain the answer for a question in NQ or TriviaQA so that we could compute the retrieval accuracy.
Using pyserini to evaluate on NQ or TriviaQA requires an additional index of Wikipedia which seems to be redundant to me.
It seems to be easier to save the doc indices that contain the answer for a question in NQ or TriviaQA so that we could compute the retrieval accuracy.