StructTuning introduces a structure-aware approach to efficiently transform foundation LLMs into domain specialists, inspired by human learning processes.

• Reduces training corpus requirements by 95% while maintaining full performance
• Employs a two-stage strategy: Structure-aware Continual Pre-Training (SCPT) and Structure-aware Supervised Fine-tuning
• Organizes domain knowledge in a structured manner that mirrors human education principles
• Demonstrates particular effectiveness in medical applications, allowing faster development of specialized healthcare AI with less data

This breakthrough significantly lowers the resource barriers for creating domain-specific AI systems in medicine, potentially accelerating the development of specialized clinical decision support tools and medical knowledge systems.

Structure-aware Domain Knowledge Injection for Large Language Models