Summarization with a Custom Fine-Tuned LLaMA Model via Instruction-Based Training

  • Unique Paper ID: 174041
  • Volume: 11
  • Issue: 10
  • PageNo: 2400-2405
  • Abstract:
  • In the contemporary digital landscape, legal professionals are increasingly confronted with voluminous, complex documents that require efficient summarization techniques capable of preserving critical information and context. This paper introduces a novel summarization framework that leverages instruction-based fine-tuning applied to both Tiny LLaMA 1.1B and Gemma 2-2B models, specifically tailored for processing intricate legal texts. The proposed framework incorporates an instruction-driven training paradigm, which enhances the models’ ability to comprehend and condense legal documents while maintaining domain-specific accuracy, contextual relevance, and terminological precision. A carefully curated legal corpus, encompassing diverse categories such as contracts, regulatory filings, case law, and legislative documents, serves as the foundation for the fine-tuning process. To further enhance scalability and computational efficiency, Databricks infrastructure is employed, enabling streamlined data processing, model training, and performance evaluation. By aligning the models’ training objectives with legal practitioners' real-world summarization needs, the framework achieves superior performance in generating coherent, concise summaries that retain essential legal arguments, obligations, and references. Comprehensive evaluations demonstrate that the fine-tuned models consistently outperform conventional summarization techniques in terms of factual accuracy, semantic coherence, and relevance to legal inquiries. This research not only highlights the transformative potential of instruction-fine-tuned language models in the legal domain but also establishes a scalable blueprint for integrating advanced natural language processing techniques into legal workflows. Ultimately, the framework contributes to more efficient legal research, enhanced document review processes, and improved decision-making capabilities for legal practitioners, thereby fostering more accessible, technology-driven legal systems.

Cite This Article

  • ISSN: 2349-6002
  • Volume: 11
  • Issue: 10
  • PageNo: 2400-2405

Summarization with a Custom Fine-Tuned LLaMA Model via Instruction-Based Training

Related Articles