AWTO: A latency-optimized task offloading scheme for LLM-driven agentic workflows on heterogeneous edge

Key Points

• Improved latency in task offloading enhances performance across heterogeneous edge systems, supporting efficient processing.
• Key performance metric shows optimization reduces task completion time significantly across various workflows.
• Observational analysis of workflows demonstrates effective task offloading mechanisms for LLM-driven applications.
• The findings suggest potential for greater efficiency in edge computing, calling for broader implementation and testing.