This dataset is a Nano-style retrieval dataset for HAKARI-bench. NanoIFIR contains 7 Nano retrieval splits derived from IFIR. Each split keeps up to 200 eligible queries and up to 10000 corpus documents, with exact duplicate query and document text removed where the generator records that policy.

Usage

from datasets import load_dataset

dataset_id = "hakari-bench/NanoIFIR" split = "NanoIFIRAila"

queries = load_dataset(dataset_id, "queries", split=split)… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoIFIR.