MALT: Improving Reasoning with Multi-Agent LLM Training

TL;DR

MALT introduces a multi-agent training pipeline for LLMs that enhances reasoning by dividing tasks into generation, verification, and refinement, leading to significant performance improvements on reasoning benchmarks.

Contribution

The paper presents a novel multi-agent post-training strategy that automatically generates training data and improves reasoning capabilities of LLMs without human supervision.

Findings

MALT achieves up to 15.66% improvement on MATH dataset.

MALT outperforms baseline LLMs on GSM8K and CSQA.

Multi-agent training enhances reasoning accuracy significantly.

Abstract

Large Language Models (LLMs) often produce answers with a single chain-of-thought, which restricts their ability to explore reasoning paths or self-correct flawed outputs in complex tasks. In this paper, we introduce MALT (Multi-Agent LLM Training), a novel post-training strategy that divides the reasoning process into generation, verification, and refinement steps using a sequential pipeline of heterogeneous agents. During data generation, each agent is repeatedly sampled to form a multi-agent search tree, where final outputs are graded against ground-truth data. We then apply value iteration to propagate reward signals back to each role-conditioned model, automatically producing multi-agent post-training data without human or teacher-model supervision. Our off-policy approach allows each agent to specialize by learning from correct and incorrect trajectories, ultimately improving the end-to-end reasoning chain. On MATH, GSM8K, and CSQA, MALT surpasses the same baseline LLM with a relative improvement of 15.66%, 7.42%, and 9.40% respectively, making it an important advance towards multi-agent cooperative training.