Multi-Agent Evolve: LLM Self-Improve through Co-evolution

TL;DR

Multi-Agent Evolve (MAE) is a scalable framework where LLMs self-improve their reasoning skills through co-evolving agents that generate, solve, and evaluate tasks without heavy reliance on human data.

Contribution

The paper introduces MAE, a novel multi-agent co-evolution framework enabling LLMs to self-improve across diverse reasoning tasks without extensive human supervision.

Findings

Achieves 4.54% average improvement on multiple benchmarks.

Demonstrates scalability and data efficiency in LLM self-improvement.

Operates effectively across mathematics, reasoning, and general knowledge tasks.

Abstract

Reinforcement Learning (RL) has demonstrated significant potential in enhancing the reasoning capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets and verifiable rewards, which limit their scalability and generality. Recent Self-Play RL methods, inspired by the success of the paradigm in games and Go, aim to enhance LLM reasoning capabilities without human-annotated data. However, their methods primarily depend on a grounded environment for feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose Multi-Agent Evolve (MAE), a framework that enables LLMs to self-evolve in solving diverse tasks, including mathematics, reasoning, and general knowledge Q&A. The core design of MAE is based on a triplet of interacting agents (Proposer, Solver, Judge) that are instantiated from a single LLM, and applies reinforcement learning to optimize their behaviors. The Proposer generates questions, the Solver attempts solutions, and the Judge evaluates both while co-evolving. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.54% on multiple benchmarks. These results highlight MAE as a scalable, data-efficient method for enhancing the general reasoning abilities of LLMs with minimal reliance on human-curated supervision.