Wiki-RAG

This repository hosts a prebuilt FAISS index and metadata for Retrieval-Augmented Generation (RAG) over English Wikipedia.

📍 Hugging Face Hub: royrin/wiki-rag

Quick start to download entire Wikipedia and load it into a RAG for you. This RAG code gives you a RAG that directly gives you the relevant wikipedia article. It's entirely offline, so saves on requests to Wikipedia.

Note: The RAG is generated on the first 3 paragraphs of the Wikipedia page. To then get the full page from Wikipedia, you can access a local version of Wikipedia, or make an API call for that page.

There are things like this, but somehow nothing quite like this. Other things require many HTTP requests to Wikipedia (like this https://llamahub.ai/l/readers/llama-index-readers-wikipedia?from=).

Date of download of Wikipedia : April 10, 2025, from https://dumps.wikimedia.org/other/pageviews/2024/2024-12/.

🛠️ Usage

from huggingface_hub import hf_hub_download
import faiss
import pickle

# Download index and metadata
index_path = hf_hub_download("royrin/wiki-rag", PATH_TO_INDEX)
meta_path = hf_hub_download("royrin/wiki-rag", PATH_TO_METADATA)

# Load FAISS index
index = faiss.read_index(index_path)

Do it for yourself, from Scratch

Download Wikipedia full (~22 GB, 2 hours to download over Wget, ~30 min using Aria2c) aria2c -x 16 -s 16 https://dumps.wikimedia.org/enwiki/latest/enwiki-latest-pages-articles.xml.bz2
Extract Wikipedia into machine-readable code: python3 WikiExtractor.py ../enwiki-latest-pages-articles.xml.bz2 -o extracted --json
Get list of top 100k or 1M articles, by page-views from https://dumps.wikimedia.org/other/pageviews/2024/2024-12/
load abstracts into RAG

Helpful Links:

Wikipedia downloads: https://dumps.wikimedia.org/enwiki/latest/
Wikipedia page views: https://dumps.wikimedia.org/other/pageviews/2024/2024-12/