Skip to content

Dataset · lavita

Medical question-answer pairs for chatbot training

lavita/medical-qa-datasets

A collection of English medical question-answer pairs for SFT and verifying the truthfulness of answers.

What's inside

An English text dataset with a volume of 1 to 10 million examples, which is a concatenation of the medical-meadow-* and chatdoctor_healthcaremagic sets. In the chatdoctor part, the term Chat Doctor is replaced with chatbot, and medical_meadow_cord19 is thinned out to 50,000 examples. It also includes the truthful-qa-* benchmark for evaluating generation truthfulness (used in the Llama 2 paper), including 55 and 16 health questions.

What you can do with it

  • SFT of a medical chatbot on question-answer pairs
  • Evaluating the truthfulness of LLM answers on truthful-qa questions
  • LoRA fine-tuning for the medical domain
Tasks
Question answering
Languages
English
Size
1M–10M rows
Downloads this month
4.6K
Likes
68

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("lavita/medical-qa-datasets")