2T tokens
Arabic Pre-training Corpus
A 2-trillion-token proprietary Arabic corpus, the largest clean dataset built for Arabic LLM pre-training.
Kawn Lab
Datasets
Arabic corpora built and released by Kawn Lab for pre-training, evaluation and multimodal work.
Every corpus, released or not
35B+ tokensOpen Source
Open Arabic Corpus
35B+ curated Arabic tokens released publicly, growing regularly, for researchers to build on.
Open Source
Wasm - Structured Arabic Multimodal Corpus
An open pipeline for building structured Arabic multimodal corpora from Common Crawl.
Open Source
MSDD - Misraj Structured Data Dump
A large-scale Arabic multimodal dataset built with the Wasm pipeline from Common Crawl.
4,758,338 rowsOpen Source
MUDD - Misraj Unstructured Data Dump
A large-scale Arabic plain-text dataset of 4.76M rows, translated from SlimPajama-627B for pre-training.
Open Source
Sadeed Tashkeela
A large, high-quality Arabic diacritized corpus for training and evaluating diacritization models.
100M captionsOpen Source
Arabic Image Captioning (100M)
A large-scale dataset of 100 million Arabic image captions, generated with the Mutarjim translation model.
Open Source
KITAB PDF-to-Markdown (Reviewed)
A reviewed and corrected version of the KITAB-Bench PDF-to-Markdown subset for Arabic OCR evaluation.
Get started
Let's talk about what you're trying to build.
Tell us the problem. We'll tell you honestly whether AI is the right answer.