Beyond Semantic Similarity: Reducing Unnecessary API Calls via Behavior-Aligned Retriever

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Yixin, Xiong, Ying, Wu, Shangyu, Cui, Yufei, Liu, Xue, Guan, Nan, Xue, Chun Jason
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915459371106304
author Chen, Yixin
Xiong, Ying
Wu, Shangyu
Cui, Yufei
Liu, Xue
Guan, Nan
Xue, Chun Jason
author_facet Chen, Yixin
Xiong, Ying
Wu, Shangyu
Cui, Yufei
Liu, Xue
Guan, Nan
Xue, Chun Jason
contents Tool-augmented large language models (LLMs) leverage external functions to extend their capabilities, but inaccurate function calls can lead to inefficiencies and increased costs.Existing methods address this challenge by fine-tuning LLMs or using demonstration-based prompting, yet they often suffer from high training overhead and fail to account for inconsistent demonstration samples, which misguide the model's invocation behavior. In this paper, we trained a behavior-aligned retriever (BAR), which provides behaviorally consistent demonstrations to help LLMs make more accurate tool-using decisions. To train the BAR, we construct a corpus including different function-calling behaviors, i.e., calling or non-calling.We use the contrastive learning framework to train the BAR with customized positive/negative pairs and a dual-negative contrastive loss, ensuring robust retrieval of behaviorally consistent examples.Experiments demonstrate that our approach significantly reduces erroneous function calls while maintaining high task performance, offering a cost-effective and efficient solution for tool-augmented LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2508_14323
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Semantic Similarity: Reducing Unnecessary API Calls via Behavior-Aligned Retriever
Chen, Yixin
Xiong, Ying
Wu, Shangyu
Cui, Yufei
Liu, Xue
Guan, Nan
Xue, Chun Jason
Computation and Language
Tool-augmented large language models (LLMs) leverage external functions to extend their capabilities, but inaccurate function calls can lead to inefficiencies and increased costs.Existing methods address this challenge by fine-tuning LLMs or using demonstration-based prompting, yet they often suffer from high training overhead and fail to account for inconsistent demonstration samples, which misguide the model's invocation behavior. In this paper, we trained a behavior-aligned retriever (BAR), which provides behaviorally consistent demonstrations to help LLMs make more accurate tool-using decisions. To train the BAR, we construct a corpus including different function-calling behaviors, i.e., calling or non-calling.We use the contrastive learning framework to train the BAR with customized positive/negative pairs and a dual-negative contrastive loss, ensuring robust retrieval of behaviorally consistent examples.Experiments demonstrate that our approach significantly reduces erroneous function calls while maintaining high task performance, offering a cost-effective and efficient solution for tool-augmented LLMs.
title Beyond Semantic Similarity: Reducing Unnecessary API Calls via Behavior-Aligned Retriever
topic Computation and Language
url https://arxiv.org/abs/2508.14323