NanoMTEB-German

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

NanoMTEB-German is a compact German retrieval benchmark containing MTEB(deu, v1) retrieval-family splits. It covers German legal and public-service retrieval, German DPR and GermanQuAD-style QA retrieval, and XMarket product/search retrieval.

Usage

from datasets import load_dataset

dataset_id = "hakari-bench/NanoMTEB-German"
split = "ger_da_lir"

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
ger_da_lir	200	10000	235	879.5	703.5	1228.0	18071.5	13329.0	22707.8
german_dpr	200	2876	200	63.7	62.0	77.0	1290.3	1146.0	1642.8
german_qu_ad	200	474	200	54.9	52.5	66.0	1941.0	1451.0	2493.2
gov_service	200	105	200	63.9	59.5	77.0	1248.5	1259.0	1335.0
xmarket_de	182	10000	4124	14.6	12.0	19.0	457.0	101.0	197.0

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	-	55.22	60.50	58.36	79.65	76.90	82.74	-	77
ger_da_lir	stemmer@german	53.60	29.20	44.61	83.38	62.12	84.12	100-101	29
german_dpr	stemmer@german	46.47	78.37	61.20	98.00	95.50	100.00	100	0
german_qu_ad	stemmer@german	94.58	93.21	94.27	100.00	96.00	100.00	100	0
gov_service	stemmer@german	61.32	79.03	69.59	99.50	99.50	100.00	100	0
xmarket_de	stemmer@german	20.12	22.68	22.10	17.38	31.37	29.56	100-101	48

Hybrid Safeguard Summary

• Safeguard positives: 77
• Rows limited by corpus size: 0
• Metadata file: reranking_hybrid_metadata.json

Source Links

• mteb/GerDaLIR
• mteb/GermanDPR
• mteb/germanquad-retrieval
• it-at-m/LHM-Dienstleistungen-QA
• mteb/XMarket

License

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