Skip to content

Dataset · miracl

Multilingual Retrieval Text Corpus

miracl/miracl-corpus

A text corpus in 16 languages from Wikipedia for training and evaluating information retrieval systems.

What's inside

The dataset is a collection of text documents for the text retrieval task in 16 languages, including English, Spanish, Arabic, Persian, Bengali, and Finnish. The materials are prepared based on Wikipedia articles and are part of the MIRACL multilingual benchmark. The total size of the corpus ranges from 10 to 100 million records.

What you can do with it

  • Training multilingual embedding models and retrievers for RAG
  • Benchmarking search quality across different languages
  • Evaluating cross-lingual information retrieval
Tasks
Text retrieval
Languages
Arabic, Bangla, English, Spanish, Persian, Finnish, French, Hindi and 4 more
Size
10M–100M rows
Downloads this month
54.8K
Likes
58

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("miracl/miracl-corpus")