Datasets:

Misraj
/

mudd

MUDD is a substantial dataset comprising 4,758,338 rows of unstructured, plain Arabic text. Each entry includes Arabic text and retains its UUID from the original source. This dataset provides high-quality Arabic content specifically designed for pretraining large language models (LLMs) and advancing Arabic natural language processing (NLP) research.

🌟 Key Features

📏 Size: 4,758,338 rows of Arabic text
🗣️ Language: Arabic (translated from English)
📌 Source: Selected subset of SlimPajama-627B
🤖 Translation Model: Mutarjim
📝 Format: Plain text with UUID identifiers

🗃️ Dataset Details

🌐 Source Data

The foundation of MUDD is SlimPajama-627B, a high-quality English text dataset containing:

📈 627 billion tokens
♻️ Deduplicated content
🌍 Diverse sources including web pages, Wikipedia, GitHub, and books

Note: MUDD is derived from a carefully selected subset of SlimPajama-627B, not the complete dataset.

🔄 Translation Process

The Arabic content was generated using Mutarjim, a high-performance Arabic-English translation model built on the Kuwain-1.5B architecture. The translation involved:

🛠️ Pre-training: On extensive monolingual Arabic and English corpora
🎯 Fine-tuning: Using high-quality, human-curated parallel sentence pairs for accurate Arabic translations

📂 Dataset Structure

{
    "uuid": {
        "dtype": "string",
        "_type": "Value"
    },
    "plain_text": {
        "dtype": "string",
        "_type": "Value"
    }
}

💡 Usage

📥 Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("Misraj/mudd")

📋 Example Usage

# Access the first example
example = dataset['train'][0]
print(f"UUID: {example['uuid']}")
print(f"Arabic Text: {example['plain_text']}")

🎯 Intended Use Cases

🤗 Pretraining Arabic LLMs: Large-scale, high-quality Arabic text corpus for training new language models
🔍 Arabic NLP Research: Supporting research initiatives focused on Arabic language processing
🚦 Downstream Applications: Reliable source for projects requiring extensive and diverse Arabic text data

📊 Dataset Statistics

📏 Metric	📌 Value
Total Rows	4,758,338
Language	Arabic
Source	SlimPajama-627B (subset)
Translation Model	Mutarjim
Format	Plain text

📖 Citations

If you use this dataset, please cite:

@misc{misraj2025mudd, 
  title        = {Misraj Unstructured Data Dump (MUDD)},
  author       = {Khalil Hennara, Muhammad Hreden, Mohamed Motaism Hamed, Zeina Aldallal, Sara Chrouf, Safwan AlModhayan, Ahmed Bustati},
  year         = {2025},
  publisher    = {MisrajAI},
  howpublished = {\url{[https://huggingface.co/datasets/Misraj/mudd](https://huggingface.co/datasets/Misraj/mudd)}}
}