NanoMTEB-Misc

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

NanoMTEB-Misc groups compact retrieval tasks that belong to the broader NanoMTEB family but do not fit a single official language benchmark family. It includes multilingual, cross-lingual, translated, and separate benchmark-family sources such as NeuCLIR, RuSciBench, EuroPIRQ, WMT, and other miscellaneous retrieval tasks.

Usage

from datasets import load_dataset

dataset_id = "hakari-bench/NanoMTEB-Misc"
split = "2022_fa"

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
2022_fa	45	8882	1131	83.1	83.0	96.0	2818.8	1951.0	3576.0
2022_ru	44	8722	1664	85.6	81.0	106.2	2448.9	1534.0	2792.5
2022_zh	47	10000	1643	24.0	23.0	28.0	1107.6	794.0	1344.0
cite_ru	200	10000	1000	1399.1	1392.5	1847.0	926.9	718.0	1278.2
cocite_ru	200	10000	1000	961.8	743.5	1297.5	908.9	702.0	1222.0
en	100	9422	100	140.4	139.5	161.5	550.1	552.0	588.0
fi	100	9422	100	146.5	140.0	165.2	594.5	593.0	632.0
pt	100	9517	100	149.8	146.0	173.2	583.8	586.0	623.0
wmt19_de_fr	200	7364	200	159.1	146.0	193.8	147.5	131.0	181.0
wmt19_fr_de	200	7365	200	149.0	137.5	189.5	154.2	140.0	188.0
wmt21_de_fr	200	4465	200	170.1	154.0	216.5	177.3	162.0	228.0
wmt21_fr_de	200	4465	200	175.0	158.0	226.2	174.5	161.0	224.0

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	-	49.39	74.67	63.62	72.12	86.97	87.85	-	33
2022_fa	regex	26.00	49.15	41.38	58.96	77.99	77.85	100-101	1
2022_ru	stemmer@russian	34.90	58.07	60.11	51.09	72.65	66.83	100-101	1
2022_zh	wordseg@zh	29.31	51.01	40.72	39.87	70.73	61.47	100-101	2
cite_ru	stemmer@russian	55.66	61.82	61.34	78.40	82.60	84.00	100-101	3
cocite_ru	stemmer@russian	39.20	42.49	43.46	59.60	66.20	68.10	100-101	18
en	english_porter_stop	94.14	92.55	94.38	100.00	99.00	100.00	100	0
fi	stemmer@finnish	90.92	85.42	88.13	99.00	93.00	100.00	100	0
pt	regex	91.86	86.23	89.01	100.00	96.00	100.00	100	0
wmt19_de_fr	stemmer@french	22.04	91.51	54.47	61.50	96.50	98.00	100-101	4
wmt19_fr_de	stemmer@german	30.78	95.74	60.54	67.00	98.50	99.50	100-101	1
wmt21_de_fr	stemmer@french	31.27	92.49	59.88	69.50	97.00	99.50	100-101	1
wmt21_fr_de	stemmer@german	46.58	89.54	69.99	80.50	93.50	99.00	100-101	2

Hybrid Safeguard Summary

• Safeguard positives: 33
• Rows limited by corpus size: 0
• Metadata file: reranking_hybrid_metadata.json

Source Links

• mteb/NeuCLIR2022RetrievalHardNegatives
• mlsa-iai-msu-lab/ru_sci_bench_cite_retrieval
• mlsa-iai-msu-lab/ru_sci_bench_cocite_retrieval
• eherra/EuroPIRQ-retrieval
• Andrianos/clsd_wmt19_21

License

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