Skip to content

Dataset · MBZUAI

First multitask benchmark for the Arabic language

MBZUAI/ArabicMMLU

From 10,000 to 100,000 test questions from Arabic school exams for evaluating language models.

What's inside

The first Arabic language understanding multitask benchmark (ArabicMMLU), prepared by a consortium of universities led by MBZUAI. The dataset includes 10,000 to 100,000 questions in Arabic in text and tabular formats, compiled from school exams across various disciplines.

What you can do with it

  • Comprehensive testing of knowledge and logic of language models in Arabic
  • Comparing multilingual LLMs on complex academic tasks
Tasks
Question answering
Languages
Arabic
Size
10K–100K rows
Downloads this month
3.7K
Likes
39

How to load it

from datasets import load_dataset

ds = load_dataset("MBZUAI/ArabicMMLU")