Hugging Face's logo

Datasets:
hakari-bench
/
NanoRARb

Modalities:
Text
Formats:
parquet
Languages:
English
Size:
100K - 1M
Tags:
information-retrieval
retrieval
nano
bm25
dense-retrieval
reranking
Libraries:
Datasets
pandas
Polars
Dataset card Data Studio Files
xet
 Community
Dataset Viewer
The dataset viewer is not available for this dataset.
Unexpected token '<', "<html> <h"... is not valid JSON

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

NanoRARb

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

NanoRARb contains 17 Nano retrieval splits derived from RAR-b. 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/NanoRARb"
split = "NanoARCChallenge"

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
NanoARCChallenge 200 9350 200 126.7 108.5 171.2 30.9 30.0 43.0
NanoAlphaNLI 200 10000 200 103.8 102.5 119.0 43.8 42.0 51.0
NanoHellaSwag 200 10000 200 114.7 115.5 148.2 62.2 55.0 77.0
NanoPIQA 200 10000 200 37.9 37.0 46.0 98.0 65.0 124.0
NanoQuail 200 10000 200 1813.8 1823.5 1861.0 25.0 22.0 33.0
NanoRARbCode 200 10000 200 470.1 430.5 542.0 256.0 244.0 333.0
NanoRARbMath 200 10000 200 201.3 166.5 256.8 481.3 386.5 603.0
NanoSIQA 200 10000 200 126.9 124.0 137.2 21.5 19.0 27.0
NanoSpartQA 200 1592 384 654.9 621.0 727.5 49.8 55.0 62.0
NanoTempReasonL1 200 10000 200 49.9 51.0 52.0 9.0 9.0 9.0
NanoTempReasonL2Context 200 10000 200 28755.2 6562.0 75421.0 19.9 17.0 23.0
NanoTempReasonL2Fact 200 10000 200 1744.4 672.0 4822.0 19.9 17.0 23.0
NanoTempReasonL2Pure 200 10000 200 53.0 53.0 61.0 19.9 17.0 23.0
NanoTempReasonL3Context 200 10000 200 31804.1 7032.0 75374.0 19.9 17.0 23.0
NanoTempReasonL3Fact 200 10000 200 1981.1 836.0 4775.0 19.9 17.0 23.0
NanoTempReasonL3Pure 200 10000 200 65.1 62.0 76.0 19.9 17.0 23.0
NanoWinoGrande 200 5095 200 112.0 110.0 122.2 7.7 7.0 9.0

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 - 15.36 24.69 23.10 46.01 65.99 64.45 - 1170
NanoARCChallenge english_porter_stop 3.86 11.13 6.42 22.50 36.00 35.50 100-101 129
NanoAlphaNLI english_porter_stop 32.88 58.98 47.77 67.50 91.50 89.50 100-101 21
NanoHellaSwag english_porter_stop 13.93 12.53 15.51 52.50 52.50 59.50 100-101 81
NanoPIQA english_porter_stop 24.43 40.17 37.41 59.50 68.00 67.50 100-101 65
NanoQuail english_porter_stop 5.22 11.74 9.82 28.50 46.00 46.50 100-101 107
NanoRARbCode regex 13.18 11.73 17.73 44.50 43.50 57.50 100-101 85
NanoRARbMath english_porter_stop 61.47 78.18 73.50 94.50 94.00 97.50 100-101 5
NanoSIQA english_porter_stop 2.39 6.18 4.05 18.50 38.50 33.50 100-101 133
NanoSpartQA english_porter_stop 18.88 26.34 34.19 60.67 54.83 62.17 100-101 37
NanoTempReasonL1 english_porter_stop 1.25 4.88 1.29 3.50 64.50 37.00 100-101 126
NanoTempReasonL2Context english_porter_stop 11.14 21.71 20.49 46.00 79.50 77.00 100-101 46
NanoTempReasonL2Fact english_porter_stop 6.15 30.05 25.13 72.00 89.50 92.50 100-101 15
NanoTempReasonL2Pure english_porter_stop 0.00 4.83 0.33 0.50 48.50 33.00 100-101 134
NanoTempReasonL3Context english_porter_stop 9.45 19.26 16.68 36.00 76.00 73.00 100-101 54
NanoTempReasonL3Fact english_porter_stop 5.47 25.49 19.81 66.00 87.00 92.50 100-101 15
NanoTempReasonL3Pure english_porter_stop 0.74 7.07 2.38 9.50 54.00 41.50 100-101 117
NanoWinoGrande english_porter_stop 50.67 49.46 60.20 100.00 98.00 100.00 100 0

Hybrid Safeguard Summary

  • Safeguard positives: 1170
  • Rows limited by corpus size: 0
  • Metadata file: reranking_hybrid_metadata.json

Source Links

License

NanoRARb is a derived dataset. Users must comply with the licenses, terms, and attribution requirements of the upstream datasets and benchmarks.
Downloads last month
855
Total file size:
87.9 MB