license: apache-2.0
task_categories:
- text-generation
language:
- zh
tags:
- medical
size_categories:
- 1M<n<10M
MedicalQA-1.4M
MedicalQA is an integrated large-scale, high-quality Chinese SFT dataset, designed for medical knowledge injection into LLMs by SFT or RAG.
Each sample is reviewed by the free LLM (ERNIE-Speed) using our proposed Quality Evaluation Algorithm.
MedArk-KI involving in two types of medical knowledge: Traditional Chinese Medicine (TCM) and Western Medicine(WM). It consists of 4 subsets as shown in the tabel:
| Name | Volume | Author | Reviewer | Type | Source | Source Format |
|---|---|---|---|---|---|---|
| DX | 5,3554 | Doctor | Doctor | WM | Web | HTML |
| HT | 135,8093 | ChatGPT | ERNIE-Speed | WM | Books, Encyclopedia, Web | PDF, TXT, HTML |
| TCM | 3,8294 | ERNIE 4.0 | ERNIE-Speed | TCM | Books, KG | PDF, Latex, JSON |
| MB | 1,4543 | ERNIE-Speed | ERNIE-Speed | WM | Books | Latex |
| Totol | 146,4484 |
1.DX
A total of 5,3554 high-quality Q&A pairs were collected from the DingXiang website, 丁香医生 (https://dxy.com/diseases), and underwent rigorous data cleaning. Each answer was authored by a doctor and reviewed by another doctor, ensuring that all information is accurate and reliable. These Q&A pairs encompass 8 key areas of medical knowledge across 3412 diseases in 31 departments: disease introduction, symptoms, causes, diagnosis, treatment, lifestyle, prevention, and consultation guidance.
2.HT
https://huggingface.co/datasets/FreedomIntelligence/HuatuoGPT2-Pretraining-Instruction
Medical_Web_Corpus_cn Medical_Encyclopedia_cn Medical_Books_cn
3.TCM
