This release packages the Natural Questions (NQ) slice of the BEIR (Benchmarking IR) benchmark as a flat, table-oriented dataset for passage retrieval evaluation and tooling. The underlying questions come from Google’s Natural Questions corpus: real user queries paired with evidence in Wikipedia. BEIR reformulates NQ as ad-hoc retrieval: each row is a query (question text), and systems must rank a large corpus of… See the full description on the dataset page: https://huggingface.co/datasets/orgrctera/beir_nq.
