Skip to content

Dataset · hotpotqa

Multi-hop Wikipedia questions

hotpotqa/hotpot_qa

113k Wikipedia question-answer pairs with supporting facts for multi-document answering.

What's inside

The dataset contains 113k English question-answer pairs built from Wikipedia articles. Questions require finding and matching information from multiple documents; supporting sentences are provided for reasoning. Data is presented in text format with supporting fact markup.

What you can do with it

  • Train and test a QA model for multi-document answers
  • Evaluate the retrieval and reasoning pipeline
  • Fine-tune an LLM to answer questions grounded in retrieved facts
Tasks
Question answering
Languages
English
Size
100K–1M rows
Downloads this month
85.8K
Likes
365

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("hotpotqa/hotpot_qa")