Skip to content

Dataset · yixuantt

MultiHop-RAG: Document RAG QA benchmark

yixuantt/MultiHopRAG

2556 queries with evidence across 2 to 4 documents to test RAG retrieval and reasoning.

What's inside

An English QA dataset for evaluating RAG pipelines. It features 2556 queries, where evidence for each is distributed across 2 to 4 documents, and queries also use document metadata. The metadata-based size ranges from 1 to 10k.

What you can do with it

  • Evaluating retrievers and rerankers for RAG
  • Testing multi-hop QA across multiple documents
  • Benchmarking RAG systems with document metadata
Tasks
Question answering, Embeddings
Languages
English
Size
1K–10K rows
Downloads this month
6.5K
Likes
77

How to load it

from datasets import load_dataset

ds = load_dataset("yixuantt/MultiHopRAG")