datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
MedQA-USMLE-4-optionsOriginal dataset introduced by Jin et al. in What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Exams
Citation information:
@article{jin2020disease,
title={What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Exams},
author={Jin, Di and Pan, Eileen and Oufattole, Nassim and Weng, Wei-Hung and Fang, Hanyi and Szolovits, Peter},
journal={arXiv preprint arXiv:2009.13081},
year={2020}
}
MedQA-USMLE-4-options-hfOriginal dataset introduced by Jin et al. in What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Exams
Citation information:
@article{jin2020disease,
title={What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Exams},
author={Jin, Di and Pan, Eileen and Oufattole, Nassim and Weng, Wei-Hung and Fang, Hanyi and Szolovits, Peter},
journal={arXiv preprint arXiv:2009.13081},
year={2020}
}
MedQA-USMLE-4-options-hfmedqa_usmle
Dataset Card for "medqa_usmle"
More Information needed
medqa-usmleusmle_step_1
Dataset Card for "usmle_self_eval_step1"
More Information needed
usmle_step_2
Dataset Card for "usmle_self_eval_step2"
More Information needed
usmle_textbooksTextbooks:
Epidemiology https://vtechworks.lib.vt.edu/items/3f46b25c-c8e7-46bc-a7ef-64e0db8556be (ISBN 13: 9781957213651)
Neuroscience for Pre-Clinical Students https://pressbooks.lib.vt.edu/neuroscience/ (ISBN 978-1-949373-80-6)
Microbiology, Pharmacology, and Immunology for Pre-Clinical Students https://pressbooks.lib.vt.edu/micropharmimmuno/ (ISBN 978-1-962841-04-7)
usmle_step_3
Dataset Card for "usmle_self_eval_step3"
More Information needed
USMLE-Test-TrainMedQA-USMLE
MedQA-USMLE
HuggingFace upload of the MedQA-USMLE dataset with deduping. If used, please cite the original authors using the citation below.
A small number of exact-duplicate questions were identified within train and us_qbank. The question text was identical, but the options were formatted slightly differently or had a different distractor. The main difference was the listed correct letter, so the incorrect duplicates were removed. Each split was then reindexed to keep indices… See the full description on the dataset page: https://huggingface.co/datasets/mkieffer/MedQA-USMLE.gbaker_medqa_usmle_4_options_hf_generic_to_brandusmle-crackers-question-bank
USMLE Crackers Question Bank
198,379 medical multiple-choice questions, every one assigned a topic and
a chapter from a closed taxonomy of 20 topics and
228 chapters.
This is a re-annotation of two existing open datasets, not new questions. What
it adds is complete, consistent categorization:
Upstream MedQA has no topic labels at all.
Upstream MedMCQA has 21 coarse subjects, one of which is literally
Unknown, and a topic_name field that is null on 53% of rows and spread
over 2… See the full description on the dataset page: https://huggingface.co/datasets/kernelvectortech/usmle-crackers-question-bank.usmle_self_assessment
Dataset Card for "usmle_self_assesment"
More Information needed
gbaker_medqa_usmle_4_options_hf_originalusmle-step1-form31
USMLE Step 1 — Form 31, Form 30 & NBME Form 27
Structured multiple-choice questions extracted from USMLE Step 1 and NBME practice forms.
Files
data/questions.jsonl — 183 USMLE Step 1 Form 31 questions (35 SOTA-cropped images)
questions_nbme27.jsonl — 198 NBME Form 27 questions (41 SOTA-cropped images)
questions_form30.jsonl — 200 NBME Form 30 questions (45 SOTA-cropped images)
Structure
Each record contains:
id — unique identifier (e.g. form31_page-0… See the full description on the dataset page: https://huggingface.co/datasets/agentN0/usmle-step1-form31.gbaker_medqa_usmle_4_options_hf_brand_to_genericmedqa-usmle-4-options
MedQA USMLE Four Options (Parquet)
This repository provides the English USMLE four-option subset of
bigbio/med_qa as one
Parquet file: medqa_usmle_4_options.parquet.
It contains 12,723 multiple-choice questions. The original data partitions are
preserved in the split column:
Split
Rows
train
10,178
dev
1,272
test
1,273
Schema
question: exam question text
options: mapping of answer labels A–D to option text
answer_idx: correct answer label… See the full description on the dataset page: https://huggingface.co/datasets/ssswwwxxx/medqa-usmle-4-options.MedQA-USMLE-4-MultiTurnRobust
MedQA Multi-Turn Robustness Benchmark
Paper: Shallow Robustness, Deep Vulnerabilities: Multi-Turn Evaluation of Medical LLMsCode: https://github.com/bmanczak/medqa_deep_robustnessVenue: NeurIPS 2025 Workshop - The Second Workshop on GenAI for Health: Potential, Trust, and Policy Compliance
1,050 USMLE questions with adversarial follow-up contexts that test whether medical LLMs maintain correct answers across conversation turns.
Why This Dataset
Medical LLMs achieve… See the full description on the dataset page: https://huggingface.co/datasets/dynamoai-ml/MedQA-USMLE-4-MultiTurnRobust.MedQA-USMLE-4-options
Mirrored by Aurigene AI
Discovery stage: Evidence and literature
US Medical Licensing Exam style questions in four-option multiple choice form.
Rows: 11,451 (phrases_no_exclude_test.jsonl 1,273, phrases_no_exclude_train.jsonl 10,178)
Pairs with Aurigene-AI/BioMistral-7B from our model catalogue.
Upstream: GBaker/MedQA-USMLE-4-options - all credit to the original authors and to the researchers who produced the underlying data; the dataset card and licence below are theirs.… See the full description on the dataset page: https://huggingface.co/datasets/Aurigene-AI/MedQA-USMLE-4-options.MedQA-USMLE-4-options-hf-MPNet-IR
Dataset Card for "MedQA-USMLE-4-options-hf-MPNet-IR"
More Information needed
fineweb-edu-usmle
FineWeb-Edu USMLE
This is a paragraph-level subset of LeoZotos/fineweb-edu-topics ranked by
usmle_similarity. The 2.5B configuration is the highest-ranked
core. The 5B configuration contains that same core plus the extension; the
shared core files are stored only once.
Token budgets use allenai/OLMo-2-0425-1B at revision
stage1-step1907359-tokens4001B and include one EOS document boundary per
paragraph. The paragraph crossing each target is retained, so the actual token
count is… See the full description on the dataset page: https://huggingface.co/datasets/LeoZotos/fineweb-edu-usmle.USMLEMedQA-USMLE-4-options-clean
MedQA-USMLE-4-options-clean Dataset
Overview
MedQA-USMLE-4-options-clean is an enhanced medical question-answering benchmark that builds upon the MedQA-USMLE dataset. Physicians analyzed the 1373 questions in the original dataset and moved 52 questions that were either malformed or incomplete to another split incomplete.
Key Features
Relabeled malformed/incorrect questions
Dataset Details
Size: 1373
Language: English
Data Source… See the full description on the dataset page: https://huggingface.co/datasets/maximegmd/MedQA-USMLE-4-options-clean.MedQA-USMLE-combined-synonym-firstusmle-step1-qbank-v3GBaker-MedQA-USMLE-4-options-KoreanMedQA-USMLE-Benchmark
💻 Dataset Usage
Run the following command to load the testing set (1,273 examples):
from datasets import load_dataset
dataset = load_dataset("shuyuej/MedQA-USMLE-Benchmark", split="test")
print(dataset)
usmle-qbankMedQA-USMLE-4-options-hf