Predictive-LoRA: A Proactive and Fragmentation-Aware Serverless Inference System for LLMs

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ni, Yinan, Yang, Xiao, Tang, Yuqi, Qiu, Zhimin, Wang, Chen, Yuan, Tingzhou
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914217068593152
author Ni, Yinan
Yang, Xiao
Tang, Yuqi
Qiu, Zhimin
Wang, Chen
Yuan, Tingzhou
author_facet Ni, Yinan
Yang, Xiao
Tang, Yuqi
Qiu, Zhimin
Wang, Chen
Yuan, Tingzhou
contents The serverless computing paradigm offers compelling advantages for deploying Large Language Model (LLM) inference services, including elastic scaling and pay-per-use billing. However, serving multiple fine-tuned LLMs via Low-Rank Adaptation (LoRA) in serverless environments faces critical challenges: reactive adapter loading causes significant cold start latency, and frequent adapter swapping leads to severe GPU memory fragmentation. In this paper, we present Predictive-LoRA (P-LoRA), a proactive and fragmentation-aware serverless inference system for LoRA-based LLMs. P-LoRA introduces two key innovations: (1) a lightweight LSTM-based traffic predictor that forecasts adapter demand and proactively prefetches hot adapters from host memory to GPU, reducing cold start latency by up to 68%; and (2) a page-based adapter memory management mechanism inspired by operating system virtual memory, which keeps GPU memory utilization above 87% even under heterogeneous adapter ranks. We evaluate P-LoRA using production-like workloads derived from the Azure Functions trace. Experimental results demonstrate that P-LoRA achieves 1.52x higher throughput than S-LoRA while reducing the average Time-To-First-Token (TTFT) by 35% under high concurrency scenarios.
format Preprint
id arxiv_https___arxiv_org_abs_2512_20210
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Predictive-LoRA: A Proactive and Fragmentation-Aware Serverless Inference System for LLMs
Ni, Yinan
Yang, Xiao
Tang, Yuqi
Qiu, Zhimin
Wang, Chen
Yuan, Tingzhou
Distributed, Parallel, and Cluster Computing
The serverless computing paradigm offers compelling advantages for deploying Large Language Model (LLM) inference services, including elastic scaling and pay-per-use billing. However, serving multiple fine-tuned LLMs via Low-Rank Adaptation (LoRA) in serverless environments faces critical challenges: reactive adapter loading causes significant cold start latency, and frequent adapter swapping leads to severe GPU memory fragmentation. In this paper, we present Predictive-LoRA (P-LoRA), a proactive and fragmentation-aware serverless inference system for LoRA-based LLMs. P-LoRA introduces two key innovations: (1) a lightweight LSTM-based traffic predictor that forecasts adapter demand and proactively prefetches hot adapters from host memory to GPU, reducing cold start latency by up to 68%; and (2) a page-based adapter memory management mechanism inspired by operating system virtual memory, which keeps GPU memory utilization above 87% even under heterogeneous adapter ranks. We evaluate P-LoRA using production-like workloads derived from the Azure Functions trace. Experimental results demonstrate that P-LoRA achieves 1.52x higher throughput than S-LoRA while reducing the average Time-To-First-Token (TTFT) by 35% under high concurrency scenarios.
title Predictive-LoRA: A Proactive and Fragmentation-Aware Serverless Inference System for LLMs
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2512.20210