Compiler-Assisted Speculative Sampling for Accelerated LLM Inference on Heterogeneous Edge Devices

TL;DR

This paper presents a compiler-assisted approach for speculative sampling that accelerates large language model inference on heterogeneous edge devices by optimizing partitioning strategies, achieving significant speedups.

Contribution

It introduces an analytical cost model that guides heterogeneous hardware partitioning for speculative decoding, addressing integration and resource exploitation challenges at the edge.

Findings

Achieves up to 1.68× speedup on edge devices for translation tasks.

Validates the cost model with real hardware, closely matching predictions.

Demonstrates effective integration of speculative sampling into compiler workflows.

Abstract

LLM deployment on resource-constrained edge devices faces severe latency constraints, particularly in real-time applications where delayed responses can compromise safety or usability. Among many approaches to mitigate the inefficiencies of sequential token-by-token generation, Speculative Decoding (SD) has emerged as a promising technique. However, SD at the edge is hindered by two major challenges: (1) integrating SD into a compiler-based workflow without sacrificing performance or programmability, and (2) exploiting the heterogeneous compute resources of modern SoCs through carefully designed partitioning strategies. This work addresses these challenges by using an analytical cost model that explores heterogeneous hardware configurations and guides coarse-grained partitioning of LLM subgraphs, particularly with edge-typical short input sequence lengths. The cost model predicts when speculative sampling and heterogeneous execution are jointly beneficial and is validated on an edge device featuring a hexacore Cortex-A CPU and a Mali GPU, revealing up to 1.68$\times$ speedup for translation tasks, closely matching analytic expectations.