PleIAs RAG-Resources
PleIAs RAG-Resources is a curated collection rather than one dataset. It pulls together material useful for building and evaluating Retrieval-Augmented Generation systems into a single place on the Hugging Face Hub: question-and- answer sets, reference corpora, and related resources sitting side by side, so you can put together an evaluation set without chasing each piece down separately.
PleIAs is a French research lab known for its work on openly licensed data and models (it built the large Common Corpus pretraining dataset), so the collection leans towards permissively licensed, reusable material and often reaches beyond English into multilingual content. That slant makes it a good match when keeping the whole pipeline open and commercially safe is part of your goal.
Treat it as a discovery tool in practice. Browse the collection on Hugging Face, open the items that match your domain, and pull the ones that fit with the datasets library. From there you pair the question-answer sets with a retrieval corpus to measure the two things a RAG evaluation has to test: whether the system retrieves the right context, and whether it grounds its answer in what it found.
Because it is a collection, plan around that shape. The contents can shift over time, the resources vary in size and format, and there is no single headline score to quote, so you will standardise formats yourself before running an evaluation. Most importantly, the licence differs from item to item: check the terms on each dataset you pull rather than assuming the whole collection shares one, especially when the output feeds a commercial product.
For a fixed, citable benchmark, something like RAG-Mini-Wikipedia or Open RAGBench is cleaner. PleIAs RAG-Resources is more valuable as a curated starting point when you want to assemble an evaluation set tailored to your own domain and openly licensed from end to end.
Related sources
CRAG (Comprehensive RAG Benchmark)
Meta's factual question answering benchmark of 4,409 QA pairs across five domains, paired with mock web-search and knowledge-graph retrieval APIs. It is distinct from "Corrective RAG (CRAG)", which is a retrieval method, not a dataset.
FinanceBench
A benchmark for evaluating LLMs and retrieval systems on questions about public financial filings. The full set holds 10,231 question, answer and evidence triplets; a 150 example sample is released openly.
FlashRAG Benchmark Datasets
A research toolkit that bundles standardised, pre-processed benchmark datasets for RAG evaluation across many scenarios. Every dataset comes in a single unified JSONL format, so you can swap benchmarks in and out without rewriting your data loading each time.
FRAMES (Fact, Fetch, and Reason)
A Google evaluation set of 824 challenging questions that tests factuality, multi-hop retrieval and reasoning together, each paired with a gold answer and the Wikipedia articles needed to answer it. Built to measure end-to-end RAG.