README.md

metadata

language:
  - ar
license: apache-2.0
task_categories:
  - text-generation
  - question-answering
pretty_name: Fatwa Training Dataset (Standardized)
tags:
  - islamic-jurisprudence
  - fatwa
  - sharia
  - fiqh
  - arabic
  - training
  - standardized
  - sahm-benchmark

Fatwa Training Dataset (Standardized)

Dataset Description

This is the standardized conversation-format version of the Fatwa training dataset, containing Islamic jurisprudence Q&A pairs for training language models on Islamic legal rulings and religious guidance.

Dataset Summary

• Language: Arabic
• Size: ~11,700 training examples
• Domain: Islamic jurisprudence, Sharia law, Fiqh
• Format: Conversational (human-agent pairs)
• Task: Islamic fatwa generation and religious Q&A

Dataset Structure

Data Fields

• id: Unique identifier for each example
• conversations: List containing human-agent dialogue
  • Human role: Contains the question with Islamic context framing
  • Agent role: Contains the fatwa/religious ruling
• category: Category of the fatwa (e.g., murabaha, zakat, riba, etc.)
• is_referral: Whether the answer mainly refers to other sources
• question_length: Length of the question in characters
• answer_length: Length of the answer in characters

Categories

The dataset covers various Islamic jurisprudence topics including:

• Murabaha (Islamic financing)
• Zakat (obligatory charity)
• Riba (usury/interest)
• Salah (prayer)
• Hajj (pilgrimage)
• Business transactions
• Family law
• And many more categories

Example

{
  "id": "fatwa_00446",
  "conversations": [
    {
      "role": "human",
      "content": "بناءً على أحكام الشريعة الإسلامية والفقه الإسلامي، أجب على السؤال التالي بفتوى شرعية مفصلة ومدعمة بالأدلة عند الإمكان.\n\nالسؤال: تحريم الشيء هل يقتضي الفساد؟\n\nالفتوى الشرعية:"
    },
    {
      "role": "agent",
      "content": "قد يقتضي الفساد، وقد لا يقتضي الفساد، فالنهي عن البيع يقتضي فساد البيع..."
    }
  ],
  "category": "riba",
  "is_referral": "NO",
  "question_length": 45,
  "answer_length": 234
}

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("SahmBenchmark/fatwa-training_standardized")
train_data = dataset['train']

# Access an example
example = train_data[0]
conversation = example['conversations']
print(f"Category: {example['category']}")
print(f"Question: {conversation[0]['content']}")
print(f"Fatwa: {conversation[1]['content']}")

Training Considerations

When using this dataset for training:

• Ensure the model understands the religious and cultural context
• Be aware that fatwas may vary based on different schools of Islamic thought
• The dataset represents specific interpretations and should be used accordingly
• Consider the sensitivity of religious rulings in production applications

Citation

@dataset{fatwa_training_standardized_2025,
  title={Fatwa Training Dataset (Standardized)},
  author={SahmBenchmark},
  year={2025},
  publisher={Hugging Face}
}

Disclaimer

This dataset is for academic and research purposes. Religious rulings should be sought from qualified Islamic scholars for practical application.