lightonai/nanobeir-multilingual

This multilingual collection is derived from the original English NanoBEIR datasets, which are smaller versions of BEIR datasets.

The compact size of these datasets makes them ideal for conducting quick and efficient evaluations during training. To facilitate broader research in cross-lingual information retrieval, our dataset has been machine-translated from the original English into eight additional languages: Arabic (ar), German (de), Spanish (es), French (fr), Italian (it), Norwegian (no), Portuguese (pt), and Swedish (sv).

The original dataset is available at zeta-alpha-ai.

from datasets import load_dataset

languages = [
    "ar",
    "de",
    "en",
    "es",
    "fr",
    "it",
    "no",
    "pt",
    "sv",
]

datasets = [
    "NanoArguAna",
    "NanoClimateFEVER",
    "NanoDBPedia",
    "NanoFEVER",
    "NanoFiQA2018",
    "NanoHotpotQA",
    "NanoMSMARCO",
    "NanoNFCorpus",
    "NanoNQ",
    "NanoQuoraRetrieval",
    "NanoSCIDOCS",
    "NanoSciFact",
    "NanoTouche2020",
]

language = "fr"

corpus = load_dataset(
    "lightonai/nanobeir-multilingual",
    f"NanoQuoraRetrieval_{language}",
    split="corpus"
)

queries = load_dataset(
    "lightonai/nanobeir-multilingual",
    f"NanoQuoraRetrieval_{language}",
    split="queries"
)

qrels = load_dataset(
    "lightonai/nanobeir-multilingual",
    "NanoQuoraRetrieval",
    split="qrels"
)

@misc{nanobeir-multilingual,
  author = {Sourty, Raphaël},
  title = {NanoBeir-Multilingual: Multilingual version of NanoBeir for quick evaluation.},
  year = {2025},
  url = {https://huggingface.co/datasets/lightonai/nanobeir-multilingual}
}