Saved in:
Bibliographic Details
Main Authors: Lv, Bo, Sun, Jingbo
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.25558
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917531073118208
author Lv, Bo
Sun, Jingbo
author_facet Lv, Bo
Sun, Jingbo
contents Optimizing the trade-off among predictive performance and computational cost is a central focus in the deployment of Large Language Models (LLMs). Current routing methods primarily rely on direct mapping from queries to models based on surface-level features, making them susceptible to the memorization trap and leading to poor generalizability on out-of-distribution (OOD) data. In this paper, we propose DecoR, a novel routing framework that recasts the routing task as a matching process of sifting similar queries from historical logs, effectively mitigating the memorization trap. To enhance matching accuracy, we introduce a query capability deconstruction method that decouples linguistic surface forms from task-intrinsic requirements, directing matching toward capability dimensions to ground decisions in essential task attributes. Furthermore, we develop CodaSet, a comprehensive benchmark for assessing routing generalization, where experimental results demonstrate that DecoR maintains superior accuracy while substantially lowering inference costs across both in-distribution and OOD settings. All the codes and data are available at https://github.com/lvbotenbest/DecoR.
format Preprint
id arxiv_https___arxiv_org_abs_2605_25558
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Beyond Query Memorization: Large Language Model Routing with Query Decomposition and Historical Matching
Lv, Bo
Sun, Jingbo
Artificial Intelligence
Optimizing the trade-off among predictive performance and computational cost is a central focus in the deployment of Large Language Models (LLMs). Current routing methods primarily rely on direct mapping from queries to models based on surface-level features, making them susceptible to the memorization trap and leading to poor generalizability on out-of-distribution (OOD) data. In this paper, we propose DecoR, a novel routing framework that recasts the routing task as a matching process of sifting similar queries from historical logs, effectively mitigating the memorization trap. To enhance matching accuracy, we introduce a query capability deconstruction method that decouples linguistic surface forms from task-intrinsic requirements, directing matching toward capability dimensions to ground decisions in essential task attributes. Furthermore, we develop CodaSet, a comprehensive benchmark for assessing routing generalization, where experimental results demonstrate that DecoR maintains superior accuracy while substantially lowering inference costs across both in-distribution and OOD settings. All the codes and data are available at https://github.com/lvbotenbest/DecoR.
title Beyond Query Memorization: Large Language Model Routing with Query Decomposition and Historical Matching
topic Artificial Intelligence
url https://arxiv.org/abs/2605.25558