Skip to content

Dataset · HuggingFaceFW

Wikipedia in 325 languages: cleaned finewiki

HuggingFaceFW/finewiki

Updated Wikipedia from August 2025 HTML dumps: 325 languages, clean articles and metadata.

What's inside

Finewiki is an updated and better-extracted version of the Wikipedia dataset first released in 2023. The authors parsed August 2025 Wikipedia HTML dumps and covered 325 languages. The set renders templates, removes redirects, disambiguation pages, and other non-core articles, while adding metadata: page ID, title, modification date, wikidata ID, version, and markdown text version. Format is tabular data and text.

What you can do with it

  • Pre-training or fine-tuning multilingual LLMs
  • Building search and RAG over clean Wikipedia articles
  • Training models on texts with metadata and markdown
Tasks
Text generation
Size
10M–100M rows
Downloads this month
21.9K
Likes
346

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("HuggingFaceFW/finewiki")