HeRo: Adaptive Orchestration of Agentic RAG on Heterogeneous Mobile SoC

TL;DR

HeRo is a framework that optimizes the deployment of agentic retrieval-augmented generation on mobile SoCs, significantly reducing latency by intelligently scheduling heterogeneous models and workflows.

Contribution

HeRo introduces a profiling-based performance modeling and a lightweight online scheduler for efficient, low-latency agentic RAG on mobile heterogeneous SoCs.

Findings

Up to 10.94x latency reduction compared to existing methods.

Effective handling of heterogeneous models and dynamic workflows.

Practical on-device agentic RAG enabled.

Abstract

With the increasing computational capability of mobile devices, deploying agentic retrieval-augmented generation (RAG) locally on heterogeneous System-on-Chips (SoCs) has become a promising way to enhance LLM-based applications. However, agentic RAG induces multi-stage workflows with heterogeneous models and dynamic execution flow, while mobile SoCs exhibit strong accelerator affinity, shape sensitivity, and shared-memory bandwidth contention, making naive scheduling ineffective. We present HeRo, a heterogeneous-aware framework for low-latency agentic RAG on mobile SoCs. HeRo builds profiling-based performance models for each sub-stage and model-PU configuration, capturing latency, workload shape, and contention-induced slowdown, and leverages them in a lightweight online scheduler that combines shape-aware sub-stage partitioning, criticality-based accelerator mapping, and bandwidth-aware concurrency control. Experiments on commercial mobile devices show that HeRo reduces end-to-end latency by up to $10.94\times$ over existing deployment strategies, enabling practical on-device agentic RAG.