shamilmohammedi/Azhar_Model_v0.2_Final
Azhar_Model_v0.2_Final is a fine-tuned Qwen2.5-7B model developed by Shamil Al-Mohammedi, specifically optimized for Islamic Jurisprudence (Fiqh). Trained on 20,000 high-quality juristic records from the Shamela Library corpus, it demonstrates significant reduction in juristic hallucinations and adopts classical 'Shamela' phrasing. This model excels at providing evidence-based Fiqh rulings with high consistency, making it suitable for academic and research assistance in navigating classical Islamic texts.
Loading preview...
Azhar_Model_v0.2_Final: Islamic Jurisprudence LLM
This model, developed by MSc. Shamil Al-Mohammedi, is a fine-tuned version of Qwen2.5-7B specifically optimized for Islamic Jurisprudence (Fiqh). It leverages 20,000 high-quality juristic records from the Shamela Library corpus, processing approximately 24.5 million tokens over 3,000 optimization steps.
Key Capabilities & Performance
- Domain Adaptation: Achieved an 85.21% improvement in Cross-Entropy Loss and a final perplexity of 9.68, indicating successful adaptation to the Fiqh domain.
- Reduced Hallucinations: Qualitatively demonstrated a significant reduction in juristic hallucinations compared to the base model.
- Stylistic Authenticity: Successfully adopted classical 'Shamela' phrasing and scholarly terminology, providing high consistency in evidence-based Fiqh rulings.
- Hybrid Approach: The 'Azhar Hybrid' approach (Fine-Tuning + RAG) was identified as optimal for factual accuracy and stylistic authenticity.
Intended Use Cases
- Academic Research: Designed to assist scholars in navigating classical Islamic texts.
- Supplementary Tool: Intended for research purposes, to be used alongside traditional scholarly verification.
Limitations
- The model is intended as a supplementary tool and requires traditional scholarly verification for critical applications.