Skip to content

Dataset · citeright

Scientific paper corpus with chunks and FAISS indices for RAG

citeright/corpus

From 10k to 100k examples from English scientific articles with ready-made chunks and FAISS indices for verifying citation accuracy.

What's inside

A research dataset of 10 to 100 thousand entries created as part of the Faithful RAG project (EPFL, spring 2026) to study citation accuracy in scientific literature. Inside are JSON files with pre-cut semantic chunks of two sizes (large ~2,000 characters and small ~300 characters), as well as ready-made FAISS vector indices.

What you can do with it

  • Evaluating and fine-tuning RAG systems for searching scientific publications without citation hallucinations
  • Comparing search quality across different text splitting granularities (300 vs 2,000 characters)
Languages
English
Size
10K–100K rows
Downloads this month
39.1K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("citeright/corpus")