ToolACE-DEV: Self-Improving Tool Learning via Decomposition and EVolution

TL;DR

ToolACE-DEV introduces a self-evolving framework that decomposes tool learning tasks, enabling lightweight models to self-improve and reducing dependence on costly advanced models, with validated effectiveness across various model scales.

Contribution

It presents a novel self-improving paradigm for tool learning that decomposes objectives and enhances lightweight models without heavy reliance on advanced LLMs.

Findings

Effective across different model scales and architectures

Reduces reliance on advanced models for tool learning

Improves basic tool-making and tool-using abilities

Abstract

The tool-using capability of large language models (LLMs) enables them to access up-to-date external information and handle complex tasks. Current approaches to enhancing this capability primarily rely on distilling advanced models by data synthesis. However, this method incurs significant costs associated with advanced model usage and often results in data compatibility issues, led by the high discrepancy in the knowledge scope between the advanced model and the target model. To address these challenges, we propose ToolACE-DEV, a self-improving framework for tool learning. First, we decompose the tool-learning objective into sub-tasks that enhance basic tool-making and tool-using abilities. Then, we introduce a self-evolving paradigm that allows lightweight models to self-improve, reducing reliance on advanced LLMs. Extensive experiments validate the effectiveness of our approach across models of varying scales and architectures.