metadata
language:
- vi
size_categories:
- 10M<n<100M
This is a filtered subset of the triples.train.ids.small.tsv file from the unicamp-dl/mmarco dataset.
Selection
Filtered using a custom-trained crossencoder.
Each query-positive pair in the original triples.train.ids.small.tsv was scored, and only pairs with a score greater than 0.5 were retained.