Nabla: Adaptive Multi-Domain DVFS for Energy-Efficient LLM Inference on Edge SoCs

Abstract

Nabla is a residual multi-domain DVFS framework for energy-efficient edge LLM inference. It uses an interference-free execution reference and an online sensitivity map to adapt CPU, GPU, and memory operating points across prefill and decode phases while preserving inference performance.

Qianlong Sang

Fifth-Year Computer Science
Ph.D. Student

My research focuses on operating systems, mobile and edge systems, and efficient on-device AI inference.