Datasets:

unicamp-dl

/

mmarco

like 71

Follow

unicamp-dl 50

Hey,

are the triplets filtered, sorted or usable for "hard negatives" training?

For example https://github.com/beir-cellar/beir/blob/main/examples/retrieval/training/train_msmarco_v3.py defines hard negatives as: "Negative passage are hard negative examples, that were mined using different dense embedding methods and lexical search methods. Each positive and negative passage comes with a score from a Cross-Encoder. This allows denoising, i.e. removing false negative passages that are actually relevant for the query."

Thank you