NightPrince/Qwen3-4B-Islamic-Arabic

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:May 5, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

NightPrince/Qwen3-4B-Islamic-Arabic is a 4 billion parameter Qwen3-4B model fine-tuned by Yahya Alnwsany (NightPrince) for Islamic Arabic question-answering. Optimized using QLoRA on a dataset of 17,944 high-quality Islamic Arabic Q&A pairs, this model excels in domains such as Fiqh, Fatwa, Aqeedah, Quran Sciences, and Islamic Finance. It is provided as a fully merged FP16 model, ready for direct inference in specialized Islamic knowledge applications.

Loading preview...

Overview

NightPrince/Qwen3-4B-Islamic-Arabic is a 4 billion parameter language model based on the Qwen3-4B architecture, specifically fine-tuned for Islamic Arabic question-answering. Developed by Yahya Alnwsany (NightPrince), this model leverages QLoRA on a comprehensive dataset of 17,944 high-quality Islamic Arabic Q&A pairs. The fine-tuning process focused on domains including Fiqh, Fatwa, Aqeedah, Quran Sciences, and Islamic Finance, making it a specialized resource for these areas.

Key Capabilities

  • Specialized Islamic Knowledge: Provides accurate answers to questions across various Islamic disciplines, drawing from the Quran, Sunnah, and classical Islamic jurisprudence.
  • Arabic Language Proficiency: Optimized for Modern Standard and Classical Arabic, ensuring high-quality responses in the target language.
  • Direct Inference: The LoRA adapter has been merged into the base weights and saved in FP16, allowing for straightforward deployment without additional adapter loading.
  • Flexible Deployment: Available in multiple variants including merged FP16, LoRA adapter, INT4 quantized, MLX 4-bit, and GGUF formats for diverse inference environments (e.g., Hugging Face Transformers, vLLM, llama.cpp).

Good for

  • Islamic Q&A Systems: Ideal for building chatbots or knowledge bases that require precise answers to Islamic legal, theological, and financial questions.
  • Educational Tools: Can be integrated into platforms for learning and studying Islamic sciences.
  • Research and Development: Useful for researchers working on Arabic NLP, particularly in religious or domain-specific contexts.

Limitations

  • Domain Specificity: Performance may degrade on general Arabic tasks or non-Islamic domains.
  • Source Verification: While trained to cite sources, generated citations should be independently verified for accuracy.
  • Jurisprudential Scope: Primarily emphasizes classical scholarship, potentially underrepresenting contemporary or minority jurisprudential views.

Finetunes
5 models