Azhar_Model_v0.3 (Islamic Jurisprudence LLM)

📜 Project Overview

A specialized Large Language Model fine-tuned on the Shamela Library and optimized for Islamic Jurisprudence (Fiqh). This version (v0.3) utilizes a Hybrid architecture (Fine-Tuning + RAG).

📈 Final Benchmarking Results

📊 Global Evaluation Metrics (Penta-Path Results):

Metric	Score (Hybrid Model)	Description
Perplexity (↓)	3.47	Measures linguistic fluency & certainty.
Semantic Sim (↑%)	55.47%	Measures meaning alignment with original Fiqh sources.
BERTScore (↑%)	75.01%	Deep contextual understanding (Semantic Accuracy).

🧪 Academic Findings:

• Hybrid Superiority: The integration of RAG with Fine-Tuning achieved the highest BERTScore (75.01%), proving precise contextual alignment.
• Linguistic Authority: The Fine-Tuning phase successfully reduced Perplexity, adopting the scholarly tone of the 'Shamela' library.

📂 Research Assets

The repository includes:

1. Global_Evaluation_Summary_v1.1.xlsx: Full statistical breakdown of the 5 evaluation paths.
2. academic_metrics_plot.png: Visualization of Model performance across different metrics.

---

Lead Researcher: Dr. Shamil Al-Mohammedi Academic Framework: Al-Azhar University Style Benchmarking