Medical Reasoning Corpus · FreedomIntelligence
medical-o1-reasoning-SFT
A medical chain-of-thought reasoning dataset for supervised fine-tuning of language models.
What medical-o1-reasoning-SFT is
medical-o1-reasoning-SFT is a medical chain-of-thought reasoning corpus created by FreedomIntelligence. It was developed to support supervised fine-tuning of HuatuoGPT-o1, a medical large language model targeted at complex clinical reasoning tasks.
The dataset addresses the challenge of instilling detailed, step-by-step reasoning capabilities into medical language models. By capturing verifiable problem-solving steps, it enables researchers to train and evaluate models on multi-step medical problem solving and clinical question answering.
How it was made
The dataset is constructed from verifiable medical problems. GPT-4o is employed to search for reasoning paths and solutions to these problems.
Generated solutions are subsequently validated using a dedicated medical verifier to ensure the correctness of the final outputs and reasoning chains.
Key facts
- Modality
- Text
- File format
- JSON
- Languages
- English, Chinese
- Primary paper
- arXiv:2412.18925
- Target model
- HuatuoGPT-o1
What you can use it for
- Medical supervised fine-tuning
- Fine-tuning base language models on clinical chain-of-thought steps to improve their medical reasoning abilities.
- Medical question answering
- Developing and evaluating systems for multi-step reasoning over verifiable clinical and biological questions.
- Instruction tuning with mixed corpora
- Combining medical problem-solving data with general instruction datasets using provided mixed configurations.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from FreedomIntelligence/medical-o1-reasoning-SFT at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 19,704 rows of en/train
Complex_CoT
string_text- min
- 834
- median
- 1,842
- mean
- 1906.97
- max
- 5,535
Question
string_text- min
- 51
- median
- 244
- mean
- 297.83
- max
- 2,380
Response
string_text- min
- 4
- median
- 570
- mean
- 638.60
- max
- 3,999
Strengths and limits
Good at
- Generated reasoning chains are checked and confirmed via an automated medical verifier.
- Includes bilingual data coverage spanning both English and Chinese.
- Provides both a purely medical subset and a combined set containing general instructions.
Watch out for
- Reasoning solutions originate from model generation (GPT-4o) rather than human clinician annotations.
- Scope is constrained to verifiable medical questions that can be assessed by an LLM-based verifier.
Questions
Can I look inside medical-o1-reasoning-SFT without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is medical-o1-reasoning-SFT under?
Its authors declare apache-2.0. Read the terms on the original page before reusing the data.
What models was this dataset designed to train?
It was designed to train HuatuoGPT-o1, a medical language model tailored for complex medical reasoning.
What data files are included in the repository?
The repository includes medical_o1_sft.json, containing purely medical SFT data, and medical_o1_sft_mix.json, containing a mixture of medical and general instructions.
What languages are represented in the dataset?
The dataset contains data in English and Chinese.
How to cite it
Junying Chen, Zhenyang Cai, Ke Ji, Xidong Wang, Wanlong Liu, Rongsheng Wang, Jianye Hou, Benyou Wang. HuatuoGPT-o1, Towards Medical Complex Reasoning with LLMs. arXiv:2412.18925 (2024).
Written from medical-o1-reasoning-SFT's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.