datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
results
HAKARI-Bench Results
This dataset stores raw benchmark result artifacts generated by HAKARI-Bench.
Raw results: per-task JSON (.xz) result files measured by HAKARI-bench.
Leaderboard: https://huggingface.co/spaces/hakari-bench/leaderboard
GitHub repository: https://github.com/hakari-bench/hakari-bench
Contributing official model results: follow the new model evaluation workflow to evaluate a model and submit results for HAKARI-Bench review:… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/results.leaderboard_database
HAKARI-Bench Leaderboard Database
This dataset hosts the DuckDB database used by the
HAKARI-Bench leaderboard.
It is derived from the raw benchmark result artifacts in
hakari-bench/results
and is packaged for leaderboard, viewer, notebook, and SQL use.
The database is produced by the HAKARI-Bench implementation in
hakari-bench/hakari-bench.
Because the benchmark code, schema, and build workflow evolve over time, this
dataset card intentionally points to the canonical… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/leaderboard_database.NanoMTEB-Scandinavian
NanoMTEB-Scandinavian
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMTEB-Scandinavian is a compact retrieval benchmark for Scandinavian-language MTEB-style task families. It includes Danish, Norwegian, and Swedish retrieval tasks spanning fact verification, question answering, news, encyclopedic content, FAQ retrieval, and social-media retrieval.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoMTEB-Scandinavian"
split… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMTEB-Scandinavian.NanoMTEB-Dutch
NanoMTEB-Dutch
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMTEB-Dutch is a compact Dutch retrieval benchmark containing the MTEB-NL retrieval-family splits. It combines Dutch BEIR-style tasks, legal and public-domain QA, news, tender, web FAQ, Wikipedia, and cross-lingual Belebele retrieval splits.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoMTEB-Dutch"
split = "argu_ana_nl"
queries = load_dataset(dataset_id… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMTEB-Dutch.NanoBRIGHT
NanoBRIGHT
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBRIGHT contains 20 Nano retrieval splits derived from BRIGHT(v1.1). 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/NanoBRIGHT"
split = "NanoBrightAops"
queries = load_dataset(dataset_id… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBRIGHT.NanoMMTEB-v2
NanoMMTEB-v2
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMMTEB-v2 is a compact multilingual and multi-domain retrieval benchmark assembled from MMTEB/MTEB retrieval tasks. It includes legal, QA, long-context, dialogue, social, code/community, and reasoning-oriented retrieval splits in the Nano query-corpus-qrels format.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoMMTEB-v2"
split = "ailastatutes"
queries =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMMTEB-v2.NanoMIRACL
NanoMIRACL
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMIRACL is a compact multilingual benchmark derived from MIRACL. Each language split evaluates monolingual retrieval of Wikipedia passages for natural-language questions.
This rebuild uses hotchpotch/miracl-hf-unified dev queries at source revision 21ad00eb467639e927b5badb7c49f4947c6c24ca. For each sampled query, it preserves all source positive passages and expands the split-local corpus to 10,000… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMIRACL.NanoDAPFAM
NanoDAPFAM
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoDAPFAM contains 18 Nano retrieval splits derived from DAPFAM. 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/NanoDAPFAM"
split = "NanoDAPFAMAllTitlAbsClmToFullText"
queries =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoDAPFAM.NanoBEIR-en
NanoBEIR-en
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-en is a compact English benchmark derived from BEIR retrieval datasets. It keeps the query-corpus-qrels retrieval format while using small task splits for fast, repeatable evaluation.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-en"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-en.NanoMuPLeR
NanoMuPLeR
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMuPLeR is a compact multilingual parallel legal retrieval benchmark derived from mteb/MuPLeR-retrieval. It uses MuPLeR retrieval tasks built from European Union DGT-Acquis parallel legal text, covering 14 European languages with human-translated parallel passages and synthetic queries.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoMuPLeR"
split = "el"
queries… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMuPLeR.NanoBEIR-th
NanoBEIR-th
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-th is the Thai language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-th"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-th.NanoMTEB-Polish
NanoMTEB-Polish
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMTEB-Polish is a compact Polish retrieval benchmark assembled from Polish MTEB-family retrieval tasks. It includes Polish CQADupStack domains, FiQA, Natural Questions, PUGG information retrieval, and Quora-style duplicate-question retrieval.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoMTEB-Polish"
split = "cqadupstack_android"
queries =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMTEB-Polish.bird-species-10k
Bird-10K SigLIP Training Dataset
Taxonomy-aware image-text dataset for fine-tuning SigLIP on 10,753 bird species.
Dataset Structure
taxonomy.json # eBird v2025 taxonomy (species/genus/family/order hierarchy)
hard_negatives.json # Same-genus & same-family negative sampling index
siglip_train.jsonl # 692,236 training image-text pairs
siglip_val.jsonl # 32,125 validation pairs
images_224_*.tar.gz # 224x224 bird images organized by species… See the full description on the dataset page: https://huggingface.co/datasets/Hakureirm/bird-species-10k.NanoBEIR-no
NanoBEIR-no
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-no is the Norwegian language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-no"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-no.NanoMedical
NanoMedical
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMedical contains 10 Nano retrieval splits derived from MTEB(Medical, v1). 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/NanoMedical"
split = "NanoCMedQAv2reranking"
queries =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMedical.NanoBEIR-es
NanoBEIR-es
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-es is the Spanish language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-es"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-es.NanoBEIR-de
NanoBEIR-de
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-de is the German language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-de"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-de.NanoFaMTEB-v2
NanoFaMTEB-v2
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoFaMTEB-v2 is a compact Persian retrieval benchmark assembled from FaMTEB and related MTEB-family sources. It includes Persian BEIR-style tasks, MIRACL and NeuCLIR retrieval, Persian web and FAQ retrieval, and synthetic Persian QA/RAG splits in the Nano query-corpus-qrels format.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoFaMTEB-v2"
split = "argu_ana_fa"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoFaMTEB-v2.NanoLaw
NanoLaw
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoLaw contains 8 Nano retrieval splits derived from MTEB(Law, v1). 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/NanoLaw"
split = "NanoAILACasedocs"
queries = load_dataset(dataset_id, "queries"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoLaw.NanoR2MED
NanoR2MED
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoR2MED contains 8 Nano retrieval splits derived from R2MED. 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/NanoR2MED"
split = "NanoR2MEDBioinformatics"
queries = load_dataset(dataset_id, "queries"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoR2MED.NanoBEIR-sr
NanoBEIR-sr
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-sr is the Serbian language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-sr"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-sr.NanoBEIR-it
NanoBEIR-it
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-it is the Italian language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-it"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-it.NanoBEIR-sv
NanoBEIR-sv
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-sv is the Swedish language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-sv"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-sv.NanoVNMTEB
NanoVNMTEB
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoVNMTEB is a compact Vietnamese retrieval benchmark assembled from VN-MTEB full retrieval tasks. It includes Vietnamese BEIR-style tasks, MIRACL, MS MARCO, Natural Questions, Quora, SciFact, SCIDOCS, Touché, TREC-COVID, and related Vietnamese retrieval sources.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoVNMTEB"
split = "argu_ana_vn"
queries =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoVNMTEB.NanoJMTEB-v2
NanoJMTEB-v2
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoJMTEB-v2 is a Nano-style Japanese retrieval dataset recreated from MTEB/JMTEB sources. The ja-cwir, jaqket, miracl-ja, and mr-tidy-japanese splits are rebuilt from sbintuitions/JMTEB-lite retrieval query/corpus tables with 200 queries and 10,000 corpus documents per split.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoJMTEB-v2"
split = "ja_cwir"
queries =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoJMTEB-v2.NanoBEIR-pt
NanoBEIR-pt
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-pt is the Portuguese language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-pt"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-pt.NanoMLDR
NanoMLDR
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoMLDR contains multilingual retrieval splits derived from Shitao/MLDR revision refs/convert/parquet. Malformed or non-question query strings are filtered before sampling.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoMLDR"
split = "ar"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus", split=split)
qrels =… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoMLDR.HakushoBench
HakushoBench: A Japanese Chart and Table VQA Benchmark from Governmental White Papers
|
🤗 Dataset
|
📄 Paper
|
🧑💻 Code
|
Overview
HakushoBench is a challenging Japanese chart and table VQA benchmark built from 33 governmental white papers.
HakushoBench contains 2,053 images spanning over 10 image types, with manually annotated QA pairs, designed to assess deep and holistic understanding of charts and tables, rather than local… See the full description on the dataset page: https://huggingface.co/datasets/llm-jp/HakushoBench.hakidame_loraNanoBEIR-vi
NanoBEIR-vi
This dataset is a Nano-style retrieval dataset for HAKARI-bench.
NanoBEIR-vi is the Vietnamese language-specific component of MNanoBEIR. It groups compact BEIR-derived retrieval tasks for efficient evaluation of document ranking in that language.
Usage
from datasets import load_dataset
dataset_id = "hakari-bench/NanoBEIR-vi"
split = "NanoArguAna"
queries = load_dataset(dataset_id, "queries", split=split)
corpus = load_dataset(dataset_id, "corpus"… See the full description on the dataset page: https://huggingface.co/datasets/hakari-bench/NanoBEIR-vi.