Structured Agent Distillation for Large Language Models

Learn about Structured Agent Distillation (SAD), a novel approach for large language models presented by researchers from MIT, Harvard, CMU, Northeastern University, Peking University, University of Georgia, and Florida International University. This 18-minute research presentation explores how to effectively distill AI agents before quantization, offering insights into improving efficiency while maintaining performance. The talk covers the structured approach to agent distillation developed by the collaborative research team, demonstrating advanced techniques for optimizing large language models when deployed as agents.