MetaToolAgent: Towards Generalizable Tool Usage in LLMs through Meta-Learning

TL;DR

MetaToolAgent leverages meta-learning to enhance large language models' ability to generalize tool usage across diverse and unseen tools, enabling more flexible and scalable real-world applications.

Contribution

The paper introduces a new dataset with 155 tools across 7 domains and proposes MetaToolAgent, a meta-learning method that improves generalization to novel tools in LLMs.

Findings

MetaToolAgent outperforms baselines on unseen tools

The dataset simulates realistic tool integration scenarios

MetaToolAgent enhances cross-tool generalization

Abstract

Tool learning is increasingly important for large language models (LLMs) to effectively coordinate and utilize a diverse set of tools in order to solve complex real-world tasks. By selecting and integrating appropriate tools, LLMs extend their capabilities beyond pure language understanding to perform specialized functions. However, existing methods for tool selection often focus on limited tool sets and struggle to generalize to novel tools encountered in practical deployments. To address these challenges, we introduce a comprehensive dataset spanning 7 domains, containing 155 tools and 9,377 question-answer pairs, which simulates realistic integration scenarios. Additionally, we propose MetaToolAgent (MTA), a meta-learning approach designed to improve cross-tool generalization. Experimental results show that MTA significantly outperforms baseline methods on unseen tools, demonstrating its promise for building flexible and scalable systems that require dynamic tool coordination.