CaraServe: CPU-Assisted and Rank-Aware LoRA Serving for Generative LLM Inference

TL;DR

CaraServe is a system that enhances LoRA adapter serving for large language models by using CPU-assisted loading and rank-aware scheduling, significantly reducing latency and improving service level objectives.

Contribution

It introduces a novel CPU-assisted loading mechanism and a rank-aware scheduling algorithm for efficient LoRA adapter serving in LLM inference.

Findings

Speeds up request serving latency by up to 1.4 times

Achieves up to 99% service level objective attainment

Outperforms state-of-the-art LoRA serving systems

Abstract

Pre-trained large language models (LLMs) often need specialization for domain-specific tasks. Low-Rank Adaptation (LoRA) is a popular approach that adapts a base model to multiple tasks by adding lightweight trainable adapters. In this paper, we present CaraServe, a system that efficiently serves many LoRA adapters derived from a common base model. CaraServe maintains the base model on GPUs and dynamically loads activated LoRA adapters from main memory. As GPU loading results in a cold-start that substantially delays token generation, CaraServe employs a CPU-assisted approach. It early starts the activated adapters on CPUs for prefilling as they are being loaded onto GPUs; after loading completes, it then switches to the GPUs for generative LoRA inference. CaraServe develops a highly optimized synchronization mechanism to efficiently coordinate LoRA computation on the CPU and GPU. Moreover, CaraServe employs a rank-aware scheduling algorithm to optimally schedule heterogeneous LoRA requests for maximum service-level objective (SLO) attainment. We have implemented CaraServe and evaluated it against state-of-the-art LoRA serving systems. Our results demonstrate that CaraServe can speed up the average request serving latency by up to 1.4$\times$ and achieve an SLO attainment of up to 99%.