arch-wiki-docs / README.md
hannah-eee's picture
Update README.md
7631eeb verified
---
license: cc-by-sa-3.0
language:
- en
- de
- fr
- es
- it
- ja
- ru
- zh
size_categories:
- 100K-1M
---
# Arch Wiki - JSONL Dataset
This dataset provides the complete Arch Linux Wiki in a clean, machine‑readable JSON Lines format (.jsonl).
It is derived from the official arch-wiki-docs package, which contains static HTML exports of all language versions.
Each line is one wiki page, with extracted plain text, headings, language code, and a relative URL path – ready for RAG (Retrieval-Augmented Generation), embedding, full‑text search, or offline browsing.
## Dataset Structure
One JSON object per line. Example:
{
"lang": "en",
"title": "Arch Linux",
"url_path": "en/Arch_Linux.html",
"content": "Arch Linux is an independently developed, x86-64 general-purpose Linux distribution that strives to provide the latest stable versions of most software...",
"headings": ["Arch Linux", "History", "Installation", "Package management"]
}
| Field | Type | Description |
|------------|--------------|-------------|
| lang | string | Two‑letter language code (e.g., en, fr, de). Chinese variants are zh-hans/zh-hant. |
| title | string | Page title as displayed in the wiki (without “ - ArchWiki” suffix). |
| url_path | string | Path relative to the root of the offline dump (e.g., en/Arch_Linux.html). |
| content | string | Clean plain‑text version of the article. All HTML tags, edit links, tables of contents, and navigation elements have been removed. |
| headings | list[string] | All h1, h2, h3 headings found on the page, in order of appearance. |
## Statistics
- Total pages: 5,713 (may vary slightly with updates)
- Languages: All languages served by the official Arch Wiki (determined by the arch-wiki-docs package)
- Average content length: ~2,500 characters per page
- File size (compressed): ~50‑80 MB (.7z or .gz)
- File size (uncompressed JSONL): ~200‑300 MB
## Usage
Load with Hugging Face datasets:
from datasets import load_dataset
dataset = load_dataset("your-username/arch-wiki-jsonl", split="train")
# Examine a page
print(dataset[0]["title"])
print(dataset[0]["content"][:500])
Load from raw JSONL (pure Python):
import json
pages = []
with open("arch_wiki.jsonl", "r", encoding="utf-8") as f:
for line in f:
pages.append(json.loads(line))
# Filter by language
en_pages = [p for p in pages if p["lang"] == "en"]
Build a simple RAG pipeline (example with LangChain):
from langchain.document_loaders import JSONLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
loader = JSONLoader(
file_path="arch_wiki.jsonl",
jq_schema=".content",
text_content=False
)
documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
chunks = text_splitter.split_documents(documents)
# → now embed and index
## Licenses and Attribution
The content of the Arch Wiki is licensed under the GNU Free Documentation License 1.3 and Creative Commons Attribution-ShareAlike 3.0 Unported (CC BY-SA 3.0).
This dataset is a derivative work. Full attribution to the Arch Linux community and all wiki contributors is required. When using this dataset, please include:
> Based on the Arch Linux Wiki (https://wiki.archlinux.org/), available under CC BY‑SA 3.0 and GFDL.
The conversion script is provided under the Apache 2.0 License.
## Generation Process
* Download from the arch-wiki-docs package (`sudo pacman -S arch-wiki-docs`)
* Upload to my cloud storage (I downloaded it in a VM, I use Debian on my system)
* Download it and run this python script to convert it to .JSON: https://github.com/dapper-11/docs-to-json
* Upload to huggingface.co
## Changelog
- 2026-05-27: Initial release based on arch-wiki-docs version as of May 27 2026.
## Notes
- This dataset is static. For the very latest wiki content, regenerate from an updated arch-wiki-docs package or pull from the live wiki.
- If you need the original HTML (e.g., for full styling), install arch-wiki-docs directly on an Arch Linux system.
- The headings field is useful for document structure‑aware chunking (e.g., split by heading boundaries).