NanoMTEB-French

This dataset is a Nano-style retrieval dataset for HAKARI-bench.

NanoMTEB-French is a compact French retrieval benchmark containing MTEB(fra, v1) retrieval-family splits. It covers French educational QA, legal retrieval, FQuAD-style question answering, Mintaka, Syntec, and cross-lingual XPQA retrieval settings.

Usage

from datasets import load_dataset

dataset_id = "hakari-bench/NanoMTEB-French"
split = "alloprof"

queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus", split=split)
qrels = load_dataset(dataset_id, "qrels", split=split)
reranking_candidates = load_dataset(dataset_id, "reranking_hybrid", split=split)

Data Layout

This dataset uses six Hugging Face Datasets configs:

• corpus: documents with _id and text
• queries: queries with _id and text
• qrels: positive relevance labels with query-id and corpus-id
• bm25: BM25 candidate lists with query-id and corpus-ids
• harrier_oss_v1_270m: dense candidate lists from microsoft/harrier-oss-v1-270m
• reranking_hybrid: RRF candidate lists built from bm25 and harrier_oss_v1_270m

Each config has the same Nano split names.

Candidate Construction

• bm25: local BM25 top-500 with automatic language-aware tokenization. The resolved tokenizer is shown in the Candidate Quality table, for example wordseg@ja.
• harrier_oss_v1_270m: dense top-500 from microsoft/harrier-oss-v1-270m. In tables this is shown as Dense; Dense means microsoft/harrier-oss-v1-270m with the web_search_query prompt for queries and cosine similarity over normalized embeddings.
• reranking_hybrid: RRF over bm25 and harrier_oss_v1_270m using rrf_k=100, keeping the RRF top-100.

Safeguard means rank 101 is appended only when RRF top-100 contains no qrels-positive document.

Split Statistics

Length statistics are character counts computed with len(str(text)).

Nano split	Queries	Corpus	Qrels	Query chars avg	Query chars p50	Query chars p75	Doc chars avg	Doc chars p50	Doc chars p75
alloprof	200	2556	200	179.3	124.5	202.8	3504.5	2527.5	4418.0
bsard	200	10000	200	145.0	140.5	169.5	793.0	430.0	890.0
fquad	200	269	200	56.2	52.0	69.0	898.3	817.0	1070.0
mintaka_fr	200	1714	200	71.6	65.5	84.0	14.4	13.0	17.0
syntec	100	90	100	72.8	71.5	88.0	1226.3	846.0	1302.0
xpqa_eng_fra	200	1674	451	54.6	53.0	66.2	137.3	100.0	159.0
xpqa_fra_eng	200	1547	437	52.1	51.0	67.0	77.0	69.0	94.0
xpqa_fra_fra	200	1547	424	54.6	53.0	66.2	77.0	69.0	94.0

Candidate Quality

nDCG@10 and Recall@100 are computed from the included candidate rankings against the included qrels, then reported as 0-100 scores such as 52.45. Recall@100 uses only the top 100 candidates; an optional rank-101 safeguard positive is not counted in Recall@100.

Dense means microsoft/harrier-oss-v1-270m with the web_search_query prompt and cosine similarity.

Nano split	BM25 tokenizer	BM25 nDCG@10	Dense nDCG@10	Hybrid nDCG@10	BM25 Recall@100	Dense Recall@100	Hybrid Recall@100	Hybrid candidates	Safeguard positives
Mean	-	42.61	56.40	50.62	66.05	85.37	83.12	-	229
alloprof	stemmer@french	34.47	51.39	52.14	79.00	89.50	91.50	100-101	17
bsard	stemmer@french	19.43	30.23	30.48	55.00	72.50	67.50	100-101	65
fquad	stemmer@french	88.99	81.02	86.66	100.00	96.00	100.00	100	0
mintaka_fr	stemmer@french	29.95	36.76	34.00	47.50	76.50	65.50	100-101	69
syntec	stemmer@french	71.80	86.60	84.63	100.00	100.00	100.00	90	0
xpqa_eng_fra	stemmer@french	10.61	36.39	17.75	27.32	70.73	66.24	100-101	49
xpqa_fra_eng	english_porter_stop	29.18	64.79	37.24	39.37	90.91	86.27	100-101	13
xpqa_fra_fra	stemmer@french	56.44	64.00	62.08	80.21	86.80	87.93	100-101	16

Hybrid Safeguard Summary

• Safeguard positives: 229
• Rows limited by corpus size: 100
• Metadata file: reranking_hybrid_metadata.json

Source Links

• mteb/AlloprofRetrieval
• mteb/BSARDRetrieval
• manu/fquad2_test
• mteb/MintakaRetrieval
• lyon-nlp/mteb-fr-retrieval-syntec-s2p
• mteb/XPQARetrieval

License

NanoMTEB-French is a derived dataset. Users must comply with the licenses, terms, and attribution requirements of the upstream MTEB task sources and their original datasets.