MIRACL (Multilingual Information Retrieval Across a Continuum of Languages) is a multilingual retrieval dataset that focuses on search across 18 different languages. The hard negative version has been created by pooling the 250 top documents per query from BM25, e5-multilingual-large and e5-mistral-instruct.

Task category t2t

Domains Encyclopaedic, Written

Reference http://miracl.ai/… See the full description on the dataset page: https://huggingface.co/datasets/mteb/MIRACLRetrievalHardNegatives.