medical-reasoning-mix-100k

# Open Source Medical Reasoning Dataset Mixture This open source dataset is a custom mixture of medical reasoning datasets from Hugging Face, designed for SFT-style chat Open Source training. > **Safety Notice:** This dataset is for benign open source medical and scientific reasoning only. It must not be used for biological or chemical weapon development, pathogen enhancement, toxin production, hazardous synthesis, or any activity that enables harm. All biomedical, biological, chemical, or laboratory-related outputs require expert review and must comply with applicable legal, ethical, biosafety, biosecurity, and chemical safety standards. > > ## Dataset Description The `medical_reasoning_mix_100k` dataset contains **100000** records, specifically curated by blending various public medical reasoning datasets. The primary goal is to provide a comprehensive and diverse dataset for training language models in medical question-answering and reasoning tasks. ## Dataset Composition The dataset is composed of the following sources with their respective approximate weights and target row counts: | Dataset | Weight | Target Rows | | :----------------------------------------------- | :----- | :---------- | | `lingshu-medical-mllm/ReasonMed` | 40% | 40,000 | | `UCSC-VLAA/MedReason` | 20% | 20,000 | | `FreedomIntelligence/medical-o1-reasoning-SFT` | 10% | 10,000 | | `openlifescienceai/medmcqa` | 10% | 10,000 | | `GBaker/MedQA-USMLE-4-options` | 5% | 5,000 | | `qiaojin/PubMedQA` (config: `pqa_artificial`) | 5% | 5,000 | | `zou-lab/MedCaseReasoning` | 5% | 5,000 | | `jmhb/PaperSearchQA` | 5% | 5,000 | *Note: The exact row counts might slightly vary due to rounding and filtering of empty examples.* ## Data Structure Each record in the dataset is a dictionary with the following keys: - `source` (string): The original Hugging Face dataset ID from which the record was derived. - `prompt` (string): The normalized user prompt for the medical reasoning task. - `completion` (string): The normalized assistant's answer or completion to the prompt. - `messages` (list of dictionaries): A list of chat messages formatted for SFT training, including: - `{"role": "system", "content": "..."}`: Defines the assistant's role. - `{"role": "user", "content": "..."}`: The prompt given to the assistant. - `{"role": "assistant", "content": "..."}`: The assistant's generated response. ### Example Record ```json { "source": "UCSC-VLAA/MedReason", "prompt": " Do male and female runners demonstrate different sagittal plane mechanics as a function of static hamstring flexibility?\n\nAnswer Choices:\nAnswer Choices:\nA. Yes\nB. No", "completion": " ### Finding Reasoning Paths:\n1. Hamstring flexibility may influence sagittal plane mechanics by altering muscle tension (active or passive) and joint moments during running.\n...\nFinal answer: A. Yes", "messages": [ { "role": "system",", "content": "You are a medical reasoning assistant. Solve the task carefully using medically accurate reasoning. For exam-style questions, end with the final answer clearly." }, { "role": "user", "content": " Do male and female runners demonstrate different sagittal plane mechanics as a function of static hamstring flexibility?\n\nAnswer Choices:\nAnswer Choices:\nA. Yes\nB. No" }, { "role": "assistant", "content": " ### Finding Reasoning Paths:\n1. Hamstring flexibility may influence sagittal plane mechanics by altering muscle tension (active or passive) and joint moments during running.\n...\nFinal answer: A. Yes" } ] } ``` ## How to Load To load this dataset locally, you can use the `datasets` library: ```python from datasets import load_from_disk dataset = load_from_disk("EpistemeAI/medical-reasoning-mix-100k") print(dataset) print(dataset[0]) # Access the first record ``` ## Usage This dataset is ideal for fine-tuning large language models (LLMs) for medical reasoning tasks, question answering, and generating detailed medical explanations. The `messages` field is structured to be directly compatible with `trl.SFTTrainer` and similar SFT libraries. ## Reference ```bibtex @misc{sun2025reasonmed370kmultiagentgenerated, title={ReasonMed: A 370K Multi-Agent Generated Dataset for Advancing Medical Reasoning}, author={Yu Sun and Xingyu Qian and Weiwen Xu and Hao Zhang and Chenghao Xiao and Long Li and Yu Rong and Wenbing Huang and Qifeng Bai and Tingyang Xu}, year={2025}, eprint={2506.09513}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2506.09513} } @misc{wu2025medreasonelicitingfactualmedical, title={MedReason: Eliciting Factual Medical Reasoning Steps in LLMs via Knowledge Graphs}, author={Juncheng Wu and Wenlong Deng and Xingxuan Li and Sheng Liu and Taomian Mi and Yifan Peng and Ziyang Xu and Yi Liu and Hyunjin Cho and Chang-In Choi and Yihan Cao and Hui Ren and Xiang Li and Xiaoxiao Li and Yuyin Zhou}, year={2025}, eprint={2504.00993}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2504.00993} } @misc{chen2024huatuogpto1medicalcomplexreasoning, title={HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMs}, author={Junying Chen and Zhenyang Cai and Ke Ji and Xidong Wang and Wanlong Liu and Rongsheng Wang and Jianye Hou and Benyou Wang}, year={2024}, eprint={2412.18925}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2412.18925} } @inproceedings{pmlr-v174-pal22a, title={MedMCQA: A Large-scale Multi-Subject Multi-Choice Dataset for Medical domain Question Answering}, author={Pal, Ankit and Umapathi, Logesh Kumar and Sankarasubbu, Malaikannan}, booktitle={Proceedings of the Conference on Health, Inference, and Learning}, pages={248--260}, year={2022} } @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} } @misc{jin2019pubmedqa, title={PubMedQA: A Dataset for Biomedical Research Question Answering}, author={Qiao Jin and Bhuwan Dhingra and Zhengping Liu and William W. Cohen and Xinghua Lu}, year={2019}, eprint={1909.06146}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/1909.06146} } @misc{wu2025medcasereasoning, title={MedCaseReasoning: Evaluating and learning diagnostic reasoning from clinical case reports}, author={Kevin Wu and Eric Wu and Rahul Thapa and Kevin Wei and Angela Zhang and Arvind Suresh and Jacqueline J. Tao and Min Woo Sun and Alejandro Lozano and James Zou}, year={2025}, eprint={2505.11733}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2505.11733} } @misc{burgess2026papersearchqalearningsearchreason, title={PaperSearchQA: Learning to Search and Reason over Scientific Papers with RLVR}, author={James Burgess and Jan N. Hansen and Duo Peng and Yuhui Zhang and Alejandro Lozano and Min Woo Sun and Emma Lundberg and Serena Yeung-Levy}, year={2026}, eprint={2601.18207}, archivePrefix={arXiv}, primaryClass={cs.LG}, url={https://arxiv.org/abs/2601.18207} } ```