CoEvol: Constructing Better Responses for Instruction Finetuning through Multi-Agent Cooperation

TL;DR

CoEvol is a multi-agent cooperation framework that iteratively improves instruction responses in large language models by leveraging debate, advice, editing, and judging, leading to enhanced instruction-following performance.

Contribution

It introduces a novel multi-agent debate and editing framework, CoEvol, that leverages LLMs to enhance data quality for instruction fine-tuning, surpassing existing methods.

Findings

Models with CoEvol outperform baselines on MT-Bench.

CoEvol improves instruction-following capabilities.

Two-stage debate strategy ensures diverse and reliable edits.

Abstract

In recent years, instruction fine-tuning (IFT) on large language models (LLMs) has garnered considerable attention to enhance model performance on unseen tasks. Attempts have been made on automatic construction and effective selection for IFT data. However, we posit that previous methods have not fully harnessed the potential of LLMs for enhancing data quality. The responses within IFT data could be further enhanced by leveraging the capabilities of LLMs themselves. In this paper, we propose CoEvol, an LLM-based multi-agent cooperation framework for the improvement of responses to instructions. To effectively refine the responses, we develop an iterative framework following a debate-advise-edit-judge paradigm. A two-stage multi-agent debate strategy is further devised to ensure the diversity and reliability of editing suggestions within the framework. Empirically, models equipped with CoEvol outperform competitive baselines evaluated by MT-Bench and AlpacaEval, demonstrating its effectiveness in enhancing instruction-following capabilities for LLMs.