AvaTaR: Optimizing LLM Agents for Tool Usage via Contrastive Reasoning

TL;DR

AvaTaR is an automated framework that enhances LLM agents' ability to effectively utilize external tools through contrastive reasoning, leading to improved performance on complex multimodal and QA tasks.

Contribution

The paper introduces AvaTaR, a novel automated method that optimizes LLM agents for tool usage using contrastive reasoning, reducing reliance on heuristic prompting techniques.

Findings

AvaTaR outperforms state-of-the-art methods across seven diverse tasks.

It achieves an average 14% improvement on retrieval tasks.

It demonstrates strong generalization to unseen cases.

Abstract

Large language model (LLM) agents have demonstrated impressive capabilities in utilizing external tools and knowledge to boost accuracy and reduce hallucinations. However, developing prompting techniques that enable LLM agents to effectively use these tools and knowledge remains a heuristic and labor-intensive task. Here, we introduce AvaTaR, a novel and automated framework that optimizes an LLM agent to effectively leverage provided tools, improving performance on a given task. During optimization, we design a comparator module to iteratively deliver insightful and comprehensive prompts to the LLM agent by contrastively reasoning between positive and negative examples sampled from training data. We demonstrate AvaTaR on four complex multimodal retrieval datasets featuring textual, visual, and relational information, and three general question-answering (QA) datasets. We find AvaTaR consistently outperforms state-of-the-art approaches across all seven tasks, exhibiting strong generalization ability when applied to novel cases and achieving an average relative improvement of 14% on the Hit@1 metric for the retrieval datasets and 13% for the QA datasets. Code and dataset are available at https://github.com/zou-group/avatar.