datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
MedThinkVQA
MedThinkVQA
MedThinkVQA is an expert-annotated benchmark for multi-image diagnostic reasoning in radiology. Unlike prior medical VQA benchmarks that typically contain at most one image per case, MedThinkVQA requires models to extract evidence from each image, integrate cross-view information, and perform differential-diagnosis reasoning.
Links
GitHub: https://github.com/benluwang/MedThinkVQA
Leaderboard: https://benluwang.github.io/MedThinkVQA/
Submission Guide:… See the full description on the dataset page: https://huggingface.co/datasets/bio-nlp-umass/MedThinkVQA.pubmed_bioasq_2022
PubMed BioASQ 2022 Corpus
This dataset contains the PubMed abstracts corpus from the BioASQ 2022 challenge, comprising approximately 23 million biomedical documents with MeSH (Medical Subject Headings) annotations.
Purpose
This is a convenience collection of the PubMed corpus from the BioASQ 2022 Challenge, reformatted for easier use in retrieval and QA systems. The original source is the BioASQ challenge data. We created this processed version with multiple formats (JSON… See the full description on the dataset page: https://huggingface.co/datasets/jmhb/pubmed_bioasq_2022.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.BioKGBench-Dataset
Agent4S-BioKG
A Knowledge Graph Checking Benchmark of AI Agent for Biomedical Science.
Github
Introduction
Pursuing artificial intelligence for biomedical science, a.k.a. AI Scientist, draws increasing attention, where one common approach is to build a copilot agent driven by Large Language Models(LLMs).However, to evaluate such systems, people either rely on direct Question-Answering(QA) to the LLM itself, or in a biomedical experimental manner. How… See the full description on the dataset page: https://huggingface.co/datasets/AutoLab-Westlake/BioKGBench-Dataset.BioMatrix-SFT
BioMatrix-SFT
This is the supervised fine-tuning (SFT) / instruction-tuning corpus used to train BioMatrix, a multimodal foundation model that natively integrates 1D sequences, 3D structures, and natural language for both molecules and proteins within a single decoder-only architecture.
📄 Paper: BioMatrix: Towards a Comprehensive Biological Foundation Model Spanning the Modality Matrix of Sequences, Structures, and Language
💻 Code: https://github.com/QizhiPei/BioMatrix… See the full description on the dataset page: https://huggingface.co/datasets/QizhiPei/BioMatrix-SFT.rag-mini-bioasqSee here for an updated version without nans in text-corpus.
In this huggingface discussion you can share what you used the dataset for.
Derives from http://participants-area.bioasq.org/Tasks/11b/trainingDataset/ we generated our own subset using generate.py.
rag-mini-bioasqThis dataset is a subset of a training dataset by the BioASQ Challenge, which is available here.
It is derived from rag-datasets/rag-mini-bioasq.
Modifications include:
filling in missing passages (some of them contained "nan" instead of actual text),
changing relevant_passage_ids' type from string to sequence of ints,
deduplicating the passages (removed 40 duplicates) and fixing the relevant_passage_ids in QAP triplets to point to the corrected, deduplicated passages' ids,
splitting QAP… See the full description on the dataset page: https://huggingface.co/datasets/enelpol/rag-mini-bioasq.HLE-BioMedX
HLE-BioMedX — Multilingual HLE Biology/Medicine
A multilingual version of the Biology/Medicine subset of Humanity's Last Exam
(HLE), released as one subset per language.
Source benchmark: Humanity's Last Exam, dataset
cais/hle.
Subsets
Group
Languages
How the target-language text was produced
Source
en
Original English questions and answers.
Machine-translated and expert-verified / revised
zh, ja, ko, fr, th
Machine translation reviewed by a human… See the full description on the dataset page: https://huggingface.co/datasets/li-lab/HLE-BioMedX.hle-gold-bio-chem
Humanity's Last Exam (HLE) Bio/Chem Gold
Humanity’s Last Exam (HLE) is a challenging question-answering AI benchmark covering advanced academic fields including Math, Physics, Chemistry, Biology, Engineering, and Computer Science.
At FutureHouse, we audited the biology and chemistry subsets of HLE using a combination of expert human evaluators and our in-house research agent, and found that around 30% of the questions contain answers directly contradicted by peer-reviewed… See the full description on the dataset page: https://huggingface.co/datasets/futurehouse/hle-gold-bio-chem.OmniGene-4-bio
gemma4-bio
Code and per-example results for the paper:
Scientific Data Composition as a Capability-Shaping Mechanism for Foundation Models:
Evidence from Biological Continued Pretraining — Liang Wang (HUST).
The study has two parts on a 26B-parameter Mixture-of-Experts model (Gemma-4-26B-A4B):
Part I — training-free re-analysis of one checkpoint lineage (instruction-tuned base →
biological CPT → SFT) across four capability axes.
Part II — a controlled seven-model experiment:… See the full description on the dataset page: https://huggingface.co/datasets/dnagpt/OmniGene-4-bio.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.BioHarness_Eval
BioHarness_Eval
This is the evaluation suite used in the BioHarness paper
(arXiv:2606.19396). It has ten biomedical
question-answering datasets with seven question types.
Paper version. This card follows the revised manuscript, which is not yet on
arXiv. arXiv v1 (June 2026) reports the earlier eight-dataset, 19,302-item suite
(BioHarness 71.0 against 65.9 for the strongest non-oracle baseline); its numbers do
not match the ten-dataset framing and
the table below. The link will… See the full description on the dataset page: https://huggingface.co/datasets/Shaow/BioHarness_Eval.events_classification_biotech
Key aspects
Event extraction;
Multi-label classification;
Biotech news domain;
31 classes;
3140 total number of examples;
Motivation
Text classification is a widespread task and a foundational step in numerous information extraction pipelines. However, a notable challenge in current NLP research lies in the oversimplification of benchmarking datasets, which predominantly focus on rudimentary tasks such as topic classification or sentiment analysis.
This dataset is… See the full description on the dataset page: https://huggingface.co/datasets/knowledgator/events_classification_biotech.bioasq-12b-rag
BioASQ 12B RAG Dataset
A processed version of the BioASQ 12B dataset optimized for Retrieval-Augmented Generation (RAG) applications in biomedical question answering.
This dataset contains two distinct subsets specifically designed for RAG applications:
A text corpus of PubMed abstracts ready for indexing and retrieval, containing detailed metadata and full abstract text.
An evaluation dataset consisting of biomedical questions, each paired with an ideal answer and a list of… See the full description on the dataset page: https://huggingface.co/datasets/mattmorgis/bioasq-12b-rag.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)
thai_buddhist_studies_exam
Thai Buddhist Studies Examination (Nak Tham)
This repository contains multiple-choice questions from the Thai Buddhist Studies
(Nak Tham) examination (2020, 2022, 2023). This dataset can be used for a benchmark for evaluating Large Language Models'
understanding of Thai Buddhist concepts and teachings.
Dataset Statistics
Year
Number of Multiple Choice Questions
2020
1,350
2022
1,400
2023
1,350
Phra Udom thought on the exam: We have reviewed the Nak… See the full description on the dataset page: https://huggingface.co/datasets/biodatlab/thai_buddhist_studies_exam.Multilingual-BioASQ-6B
Mutilingual BioASQ-6B
We translate the BioASQ-6B English Question Answering dataset to generate parallel French, Italian and Spanish versions using the NLLB200 3B parameter model. For more info read the original task description: [http://bioasq.org/participate/challenges_year_6](http://bioasq.org/participate/challenges_year_6)
We translate the body, snippets, ideal_answer and exact_answer fields. We have validated the quality of the ideal_answer field, however, the… See the full description on the dataset page: https://huggingface.co/datasets/HiTZ/Multilingual-BioASQ-6B.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.BioASQ
BioASQ - All Question Types
A comprehensive collection of BioASQ challenge questions organized by question type as separate splits.
Purpose
This dataset is a convenience collection of BioASQ questions reformatted for easier use. The original source data is from the BioASQ Challenge. We created this reorganized version (with question types as splits) to facilitate evaluation in our PaperSearchQA work.
IMPORTANT: This is not the original BioASQ dataset. We have simply… See the full description on the dataset page: https://huggingface.co/datasets/jmhb/BioASQ.bioinstruct
Dataset Card for BioInstruct
GitHub repo: https://github.com/bio-nlp/BioInstruct
Dataset Summary
BioInstruct is a dataset of 25k instructions and demonstrations generated by OpenAI's GPT-4 engine in July 2023.
This instruction data can be used to conduct instruction-tuning for language models (e.g. Llama) and make the language model follow biomedical instruction better.
Improvements of Llama on 9 common BioMedical tasks are shown in the result section.
Taking… See the full description on the dataset page: https://huggingface.co/datasets/bio-nlp-umass/bioinstruct.BioMed-R1-Train
Disentangling Reasoning and Knowledge in Medical Large Language Models
This dataset is part of our paper, which focuses on disentangling knowledge and reasoning in medical large language models (LLMs). It includes both the supervised fine-tuning (SFT) dataset (m23k_clean) and the reinforcement learning (RL) dataset (biomed_r1) used to train our final models.
For more details, please visit our GitHub.
If you find this work useful, please consider citing it as follows:… See the full description on the dataset page: https://huggingface.co/datasets/zou-lab/BioMed-R1-Train.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.neet-biology-qa
NEET Biology Questions Dataset
A comprehensive collection of NEET (National Eligibility cum Entrance Test) Biology questions designed to help students prepare for medical entrance examinations.
About This Dataset
This dataset contains 793 carefully crafted multiple-choice questions covering essential Biology topics that appear in NEET exams. Each question follows the standard NEET format with four answer choices and one correct answer.
What's Inside
Questions:… See the full description on the dataset page: https://huggingface.co/datasets/sweatSmile/neet-biology-qa.BioInstructQALarge Language Models (LLMs) have demonstrated remarkable versatility
in recent years, offering potential applications across specialized
domains such as healthcare and medicine. Despite the availability of
various open-source LLMs tailored for health contexts, adapting
general-purpose LLMs to the medical domain presents significant
challenges. In this paper, we introduce BioMistral, an open-source
LLM tailored for the biomedical domain, utilizing Mistral as its
foundation model and further pre-trained on PubMed Central. We conduct
a comprehensive evaluation of BioMistral on a benchmark comprising 10
established medical question-answering (QA) tasks in English. We also
explore lightweight models obtained through quantization and model
merging approaches. Our results demonstrate BioMistral's superior
performance compared to existing open-source medical models and its
competitive edge against proprietary counterparts. Finally, to address
the limited availability of data beyond English and to assess the multilingual
generalization of medical LLMs, we automatically translated and evaluated this
benchmark into 7 other languages. This marks the first large-scale
multilingual evaluation of LLMs in the medical domain. Datasets,
multilingual evaluation benchmarks, scripts, and all the models obtained
during our experiments are freely released.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.bioasq10b-factoid
Dataset Card for "bioasq10b"
More Information needed
CRAFT-BioQA
CRAFT-BioQA
This is a synthetic dataset generated with the CRAFT framework proposed in the paper CRAFT Your Dataset: Task-Specific Synthetic Data Generation Through Corpus Retrieval and Augmentation.
The correctness of the data has not been verified in detail, but training on this data and evaluating on human-curated biology question-answering data proved highly beneficial.
4 synthetic dataset sizes (S, M, L, XL) are available, and training on them yields consistent improvement… See the full description on the dataset page: https://huggingface.co/datasets/ingoziegler/CRAFT-BioQA.biostars_qa
Dataset Summary
This dataset contains 4803 question/answer pairs extracted from the BioStars website. The site focuses on bioinformatics, computational genomics, and biological data analysis.
Dataset Structure
Data Fields
The data contains INSTRUCTION, RESPONSE, SOURCE, and METADATA fields. The format is described for LAION-AI/Open-Assistant
Dataset Creation
Curation Rationale
Questions were included if they were an accepted answer and the… See the full description on the dataset page: https://huggingface.co/datasets/cannin/biostars_qa.