Population-Evolve: a Parallel Sampling and Evolutionary Method for LLM Math Reasoning

TL;DR

Population-Evolve is a training-free, genetic algorithm-inspired method that enhances LLM reasoning by maintaining and evolving a population of solutions during inference, leading to improved accuracy and efficiency.

Contribution

It introduces a novel population-based, self-evolving inference method for LLMs, unifying existing strategies under an evolutionary framework.

Findings

Achieves higher accuracy with low variance

Demonstrates computational efficiency

Unifies test-time scaling strategies

Abstract

Test-time scaling has emerged as a promising direction for enhancing the reasoning capabilities of Large Language Models in last few years. In this work, we propose Population-Evolve, a training-free method inspired by Genetic Algorithms to optimize LLM reasoning. Our approach maintains a dynamic population of candidate solutions for each problem via parallel reasoning. By incorporating an evolve prompt, the LLM self-evolves its population in all iterations. Upon convergence, the final answer is derived via majority voting. Furthermore, we establish a unification framework that interprets existing test-time scaling strategies through the lens of genetic algorithms. Empirical results demonstrate that Population-Evolve achieves superior accuracy with low performance variance and computational efficiency. Our findings highlight the potential of evolutionary strategies to unlock the reasoning power of LLMs during inference.