Azhar_Model_v0.3: Specialized Islamic Jurisprudence LLM

Azhar_Model_v0.3, developed by Dr. Shamil Al-Mohammedi, is a 7.6 billion parameter large language model with a 32768 token context length, uniquely optimized for Islamic Jurisprudence (Fiqh). This model employs a hybrid architecture, combining fine-tuning with Retrieval Augmented Generation (RAG), specifically leveraging the extensive Shamela Library for its training data.

Key Capabilities & Performance:

• Specialized Fiqh Understanding: Fine-tuned on the Shamela Library, enabling deep contextual understanding and generation of content in Islamic jurisprudence.
• Hybrid Architecture: Integrates Fine-Tuning with RAG, which has been shown to achieve superior contextual alignment and reduced perplexity.
• Strong Benchmarking: Achieves a BERTScore of 75.01%, indicating precise semantic accuracy and alignment with original Fiqh sources. It also boasts a low perplexity of 3.47, reflecting high linguistic fluency.
• Scholarly Tone: The fine-tuning process successfully imbues the model with the scholarly tone characteristic of the Shamela library, as evidenced by reduced perplexity.

Ideal Use Cases:

• Islamic Jurisprudence Research: Generating or analyzing texts related to Fiqh.
• Academic Applications: Supporting studies and research within Islamic law and theology.
• Content Generation: Creating linguistically fluent and contextually accurate content in the domain of Islamic jurisprudence.