how to botain the original dev Subset in a tsv file?

[xiaoyu]

Hi, I am doing the "Experiments on MS MARCO Passage Retrieval - Dev Subset - with GPU"，I want to get the original dev Subset in a tsv file(containing 105 queries)，just like 《 Passage Re-ranking with BERT》provide us with "top1000.dev.tsv" and others requirements. In 《 Passage Re-ranking with BERT》，I can use convert scripts convert tsv file to tfrecord format，but it is too big, I just want to convert 105 queries，not almost 6800 queries, but how to get that Dev Subset?

[Ronak]

The data prep section has details on how to get this subset, you are downloading it. It should be in the msmarco_ans_small folder after extracting. I'll attempt to add some clarity, the dev subset we use is nothing official, it was just a randomly curated subset of MS MARCO Passage that most users can quickly run these systems on instead of running on all the 6xxx queries.

[xiaoyu]

The data prep section has details on how to get this subset, you are downloading it. It should be in the msmarco_ans_small folder after extracting. I'll attempt to add some clarity, the dev subset we use is nothing official, it was just a randomly curated subset of MS MARCO Passage that most users can quickly run these systems on instead of running on all the 6xxx queries.

OK, thanks for your reply!