datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
BioMed-R1-Eval
Disentangling Reasoning and Knowledge in Medical Large Language Models
This is the evaluation dataset accompanying our paper, comprising 11 publicly available biomedical benchmarks. We disentangle each benchmark question into either medical reasoning or medical knowledge categories.
Additionally, we provide a set of adversarial reasoning traces designed to evaluate the robustness of medical reasoning models.
For more details, please refer to our GitHub.
If you find this work useful… See the full description on the dataset page: https://huggingface.co/datasets/zou-lab/BioMed-R1-Eval.chinese-biomedicine-and-genomics-open-intelligence
🔬 Chinese Biomedicine, Cell Therapy & Genomics Open Intelligence Dataset
Curated open intelligence dataset providing English briefs, clinical trial benchmarks, verified abstracts, and DOIs of frontier Chinese research in Cellular Therapeutics, Gene Editing, ADCs, and NMPA Clinical Approvals.
[!IMPORTANT]
Data Completeness & Research Authenticity Notice:
Included in this Hugging Face Open Dataset: English structured abstracts, core quantitative takeaways, author… See the full description on the dataset page: https://huggingface.co/datasets/simpleG2023/chinese-biomedicine-and-genomics-open-intelligence.bio-overrefusal-v0.1
Bio Over-Refusal Dataset v0.1.0
Dataset Summary
The Bio Over-Refusal Dataset is a domain-expert-authored and tier-annotated benchmark of 201 legitimate biology research queries stratified by sensitivity tier. It is designed to measure the false-positive refusal rate (FPR) of large language models — specifically, the rate at which models refuse or hedge on questions that credentialed biology researchers would consider appropriate to answer.
The dataset does not… See the full description on the dataset page: https://huggingface.co/datasets/jang1563/bio-overrefusal-v0.1.OpenSciReasoning-Biology-20K
OpenSciReasoning-Biology-20K
Three-domain release derived from nvidia/OpenScienceReasoning-2 for
domain-specific reasoner training and cross-domain transfer experiments.
Each row preserves the stable source_row_id and has exactly one mutually
exclusive domain value: BIOLOGY. Domain acceptance was checked from the
question and choices with two independent question-only verifiers; answer and
source-ID gates were also replayed.
The audit records list any remaining source-output… See the full description on the dataset page: https://huggingface.co/datasets/TerryJCZhang/OpenSciReasoning-Biology-20K.rag-mini-bioasq-with-metadataThis dataset is an extension of the rag-mini-bioasq dataset.
Its difference resides in the text-corpus part of the aforementioned set where the metadata was added for each passage.
Metadata contains six separate categories, each in a dedicated column:
Year of the publication (publish_year)
Type of the publication (publish_type)
Country of the publication - often correlated with the homeland of the authors (country)
Number of pages (no_pages)
Authors (authors)
Keywords (keywords)
bio-faiss-longevity-v1
bio-faiss-longevity-v1
A FAISS index + metadata for scientific retrieval
Contents
index.faiss: FAISS index (cosine w/ inner product).
meta.jsonl: one JSON per chunk; fields include chunk_id, paper_id, title, section, subsection, paragraph_index, keywords, boost.
index.info.json: (optional) dimensions, index type, faiss version.
Build provenance
Chunking: hierarchical (section→paragraph→~480-token chunks, ~15% overlap)
Embedder:… See the full description on the dataset page: https://huggingface.co/datasets/bio-protocol/bio-faiss-longevity-v1.neophyte-faiss-index-v1
neophyte-faiss-index-v1
A FAISS index + metadata for scientific retrieval
Contents
index.faiss: FAISS index (cosine w/ inner product).
meta.jsonl: one JSON per chunk; fields include chunk_id, paper_id, title, section, subsection, paragraph_index, keywords, boost.
index.info.json: (optional) dimensions, index type, faiss version.
Build provenance
Chunking: hierarchical (section→paragraph→~480-token chunks, ~15% overlap)
Embedder:… See the full description on the dataset page: https://huggingface.co/datasets/bio-protocol/neophyte-faiss-index-v1.bioasq-rag-13b-resplit
BioASQ RAG 13B (Resplit)
Reshuffled version of DinoStackAI/bioasq-rag-13b for Retrieval-Augmented Generation (RAG).
All original train, dev and test queries were merged, shuffled with seed 42, and reassigned using:
0.2 of all queries → test
0.2 of the remaining queries → dev
the rest → train
The shared PubMed corpus is unchanged from the source dataset.
Structure
Subset
Splits
Description
corpus
train (default)
PubMed abstracts shared across all query… See the full description on the dataset page: https://huggingface.co/datasets/DinoStackAI/bioasq-rag-13b-resplit.biosum-cuh
BioSum-CUH
A Biography Summarization Benchmark with Token-Level Correctness, Uncertainty, and Hallucination Annotations
Paper: UT-ACA: Uncertainty-Triggered Adaptive Context Allocation for Long-Context Inference | Code: github.com/Tommy307/UT-ACA
BioSum-CUH is a benchmark for studying factual generation over biography
contexts. It combines biography-based question answering and structured
summarization with token-aligned model predictions, final-layer attention
activations… See the full description on the dataset page: https://huggingface.co/datasets/Saria307/biosum-cuh.BiochemForge
BiochemForge
BiochemForge is a provenance-first biology, chemistry, and biochemistry post-training mixture for
mechanistic explanation, quantitative derivation, experimental inference, and consistency between
reasoning and final answers.
Dataset summary
Slice
Records
Purpose
SFT train
99,773
Supervised post-training
SFT validation
2,052
Model selection and early stopping
SFT test
1,093
Internal held-out evaluation
Solver-verified records
27,657… See the full description on the dataset page: https://huggingface.co/datasets/0xKitkat/BiochemForge.bio-faiss-d1ckgpt-v1
bio-faiss-d1ckgpt-v1
A FAISS index + metadata for scientific retrieval
Contents
index.faiss: FAISS index (cosine w/ inner product).
meta.jsonl: one JSON per chunk; fields include chunk_id, paper_id, title, section, subsection, paragraph_index, keywords, boost.
Build provenance
Chunking: hierarchical (section→paragraph→~380-token chunks, ~15% overlap)
Embedder: bio-protocol/scientific-retriever (mean-pooled, L2-normalized)
Similarity: cosine via inner… See the full description on the dataset page: https://huggingface.co/datasets/bio-protocol/bio-faiss-d1ckgpt-v1.biomni-eval1
Biomni-Eval1 data
This directory is the local input for the native Biomni-Eval1 harness in
mega_eval/benchmarks/biomni_eval1.py.
Default mirror: ssswwwxxx/biomni-eval1
Official source: biomni/Eval1
Split: official val (433 instances across 10 tasks)
License: Apache-2.0
Local file: validation.parquet (ignored by Git)
Reproducibility: manifest.json pins the resolved Hub commit and file hash
Official evaluator reference: Biomni commit 400c1f366b96a35ca253e13c9b06c5076af41d65… See the full description on the dataset page: https://huggingface.co/datasets/ssswwwxxx/biomni-eval1.bioasq_factoid
BioASQ Factoid Test Set
Processed BioASQ factoid test set with golden answers for evaluation.
Purpose
This dataset is a convenience collection of BioASQ factoid questions with added golden answer synonyms for exact match evaluation. The original source data is from the BioASQ Challenge. We created this processed version to facilitate evaluation in our PaperSearchQA work.
IMPORTANT: This is not the original BioASQ dataset. We have simply reformatted the BioASQ factoid test… See the full description on the dataset page: https://huggingface.co/datasets/jmhb/bioasq_factoid.predator-biomedical
PREDATOR Biomedical Dataset
50K+ curated biomedical abstracts from PubMed/EuropePMC
Curated biomedical research data extracted from PubMed, EuropePMC, and clinical trial databases. Each entry includes DOI/PMID, title, source, domain classification, commercial value score, and quality assessment.
Fields
Column
Description
id
DOI or PMID identifier
title
Article title or abstract summary
source
Data source (EuropePMC, PubMed, ClinicalTrials.gov)… See the full description on the dataset page: https://huggingface.co/datasets/iservice/predator-biomedical.bio-faiss-microbiome-v1
bio-faiss-microbiome-v1
A FAISS index + metadata for scientific retrieval
Contents
index.faiss: FAISS index (cosine w/ inner product).
meta.jsonl: one JSON per chunk; fields include chunk_id, paper_id, title, section, subsection, paragraph_index, keywords, boost.
Build provenance
Chunking: hierarchical (section→paragraph→~380-token chunks, ~15% overlap)
Embedder: bio-protocol/scientific-retriever (mean-pooled, L2-normalized)
Similarity: cosine via inner… See the full description on the dataset page: https://huggingface.co/datasets/bio-protocol/bio-faiss-microbiome-v1.BioASQ_11B_DutchBio ASQ challenge 11b
This can be used to finetune a decoder model for Q/A interaction,
alternatively it can be used to create (question, positive, negative)
triplets to train a sentence encoder using SBERT.
Reference:
@inbook{Nentidis_2023,
title={Overview of BioASQ 2023: The Eleventh BioASQ Challenge on Large-Scale Biomedical Semantic Indexing and Question Answering},
ISBN={9783031424489},
ISSN={1611-3349},
url={http://dx.doi.org/10.1007/978-3-031-42448-9_19}… See the full description on the dataset page: https://huggingface.co/datasets/UMCU/BioASQ_11B_Dutch.MedQA-CS-ExamBenchmarking LLMs Clinical Skills for Patient-Centered Diagnostics and Documentation
Project github: https://github.com/bio-nlp/MedQA-CS
MedQA-CS-Student dataset: https://huggingface.co/datasets/bio-nlp-umass/MedQA-CS-Student
islamic-biographies
islamlab — Islamic Biographical Notices
219,364 biographical notices taken out of the ṭabaqāt, tarājim and
chronicle literature and given one row each: who the notice is about, which
work it stands in, and what that work says about him.
The Muslim scholarly tradition kept biographical records for a thousand years,
mostly so that a chain of transmission could be checked. Read at scale that
record is a prosopography — who taught whom, who lived where, who was trusted
and by whom.… See the full description on the dataset page: https://huggingface.co/datasets/islamlab/islamic-biographies.NCERT_Biology_11threwrite-questions-nonsensical-biology
nonsensical_biology.csv - Question Rewriting Dataset
This dataset contains question rewriting outputs from the file nonsensical_biology.csv.
Dataset Structure
The dataset contains the following columns:
custom_id: Unique identifier for each question
style: Rewriting style applied (e.g., "gibberish")
index: Numerical index
original: Original question text
rewritten: Rewritten version of the question
options: Multiple choice options (list format)
correct: Index of the… See the full description on the dataset page: https://huggingface.co/datasets/NLie2/rewrite-questions-nonsensical-biology.NCERT_Biology_12thbiomedical-forecasting-lightningrod
Biomedical Forecasting Dataset
A dataset of 1444 binary forecasting questions about biomedical and public health outcomes. Each question is a forward-looking prediction (Yes/No) about a real event, grounded in news and labeled with the actual outcome.
What is in this dataset?
Questions: FDA drug approvals, clinical trial results (Phase 2/3), WHO and CDC declarations, vaccine development, disease outbreaks, gene therapy, and public health policy.
Grounded in real news:… See the full description on the dataset page: https://huggingface.co/datasets/Ainoafv/biomedical-forecasting-lightningrod.mini-bioasq-with-metadataThis dataset is an extension of the rag-mini-bioasq dataset.
Its difference resides in the text-corpus part of the aforementioned set where the metadata was added for each passage.
Metadata contains six separate categories, each in a dedicated column:
Year of the publication (publish_year)
Type of the publication (publish_type)
Country of the publication - often correlated with the homeland of the authors (country)
Number of pages (no_pages)
Authors (authors)
Keywords (keywords)
biosciences-evaluation-metrics
Biosciences RAG Evaluation Metrics
Dataset Description
This dataset contains detailed RAGAS evaluation results with per-question metric scores for 4 retrieval strategies tested on the biosciences RAG system. Each record includes the full evaluation context (question, contexts, response) plus 4 RAGAS metric scores.
Dataset Summary
Total Examples: 48 records (12 questions x 4 retrievers)
Retrievers Evaluated: Naive, BM25, Ensemble, Cohere Rerank
Metrics Per… See the full description on the dataset page: https://huggingface.co/datasets/open-biosciences/biosciences-evaluation-metrics.cleand_moremilk_ToT-Biology元データ: https://huggingface.co/datasets/moremilk/ToT-Biology
データ件数: 5,752
平均トークン数: 675
最大トークン数: 1,105
合計トークン数: 3,881,334
ファイル形式: JSONL
ファイル分割数: 1
合計ファイルサイズ: 19.3 MB
加工内容：
長文フィルタリング: トークナイズ処理の負荷を軽減するため、事前に文字列が極端に長い行を除外します。
question 列: 6,000文字を超える行を除外。
metadata 列: 80,000文字を超える行を除外。
metadata フィールドの展開:
metadata 列に含まれるJSON形式のデータから reasoning と difficulty の値を抽出します。
reasoning は thought という新しい列に格納します。
difficulty は difficulty という新しい列に格納します。
処理後、元の metadata 列は削除されます。
繰り返し表現の除去:
thought… See the full description on the dataset page: https://huggingface.co/datasets/LLMTeamAkiyama/cleand_moremilk_ToT-Biology.biosciences-cq-validations
Open Biosciences — Competency Question Validations
Execution results from running structured competency questions (CQs) against live biomedical APIs using the Fuzzy-to-Fact protocol.
What's in this dataset?
Each row represents one validated competency question with:
Column
Description
cq_id
Competency question identifier (e.g., cq14)
question
The research question text
category
Domain category (e.g., "Synthetic Lethality Validation")
overall_status… See the full description on the dataset page: https://huggingface.co/datasets/open-biosciences/biosciences-cq-validations.NCERT_Biology_11th