Saved in:
Bibliographic Details
Main Authors: Huang, Zixiao, Zeng, Wen, Fu, Tianyu, Liu, Tengxuan, Sun, Yizhou, Hong, Ke, Yang, Xinhao, Liu, Chengchun, Li, Yan, Zhang, Quanlu, Dai, Guohao, Zhu, Zhenhua, Wang, Yu
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.20048
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908674014838784
author Huang, Zixiao
Zeng, Wen
Fu, Tianyu
Liu, Tengxuan
Sun, Yizhou
Hong, Ke
Yang, Xinhao
Liu, Chengchun
Li, Yan
Zhang, Quanlu
Dai, Guohao
Zhu, Zhenhua
Wang, Yu
author_facet Huang, Zixiao
Zeng, Wen
Fu, Tianyu
Liu, Tengxuan
Sun, Yizhou
Hong, Ke
Yang, Xinhao
Liu, Chengchun
Li, Yan
Zhang, Quanlu
Dai, Guohao
Zhu, Zhenhua
Wang, Yu
contents LLM-based search agents achieve strong performance but suffer from severe latency, as each step requires serialized LLM reasoning followed by action of tool execution. We revisit this bottleneck through the lens of speculation. While traditional predict-verify speculation paradigm can break serial execution, its benefit remains limited, as it retains the full original workload and adds extra inference overhead. We observe that early agent steps often involve simple evidence-gathering, where correct actions can often be predicted without full reasoning. Building on these observations, we present SPAgent, an algorithm-system co-design framework that expands the role of speculation in search agents to reduce latency. Algorithmically, SPAgent introduces a two-phase adaptive speculation mechanism that selectively omits verification when safe. System-wise, a two-level scheduler regulates speculative requests based on engine load to ensure speculation remains beneficial. We implement SPAgent in real-world systems. Across extensive experimental settings, SPAgent achieves up to $1.65\times$ end-to-end speedup while maintaining same or even achieving higher accuracy, enabling practical deployment of multi-step search agents.
format Preprint
id arxiv_https___arxiv_org_abs_2511_20048
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reducing Latency of LLM Search Agent via Speculation-based Algorithm-System Co-Design
Huang, Zixiao
Zeng, Wen
Fu, Tianyu
Liu, Tengxuan
Sun, Yizhou
Hong, Ke
Yang, Xinhao
Liu, Chengchun
Li, Yan
Zhang, Quanlu
Dai, Guohao
Zhu, Zhenhua
Wang, Yu
Artificial Intelligence
Machine Learning
Performance
LLM-based search agents achieve strong performance but suffer from severe latency, as each step requires serialized LLM reasoning followed by action of tool execution. We revisit this bottleneck through the lens of speculation. While traditional predict-verify speculation paradigm can break serial execution, its benefit remains limited, as it retains the full original workload and adds extra inference overhead. We observe that early agent steps often involve simple evidence-gathering, where correct actions can often be predicted without full reasoning. Building on these observations, we present SPAgent, an algorithm-system co-design framework that expands the role of speculation in search agents to reduce latency. Algorithmically, SPAgent introduces a two-phase adaptive speculation mechanism that selectively omits verification when safe. System-wise, a two-level scheduler regulates speculative requests based on engine load to ensure speculation remains beneficial. We implement SPAgent in real-world systems. Across extensive experimental settings, SPAgent achieves up to $1.65\times$ end-to-end speedup while maintaining same or even achieving higher accuracy, enabling practical deployment of multi-step search agents.
title Reducing Latency of LLM Search Agent via Speculation-based Algorithm-System Co-Design
topic Artificial Intelligence
Machine Learning
Performance
url https://arxiv.org/abs/2511.20048