TurboEvolve: Towards Fast and Robust LLM-Driven Program Evolution

2026-04-12

AI Summary

This work addresses the challenges of high computational cost, large variance, and performance instability under limited evaluation budgets in large language model (LLM)-driven program evolution. To overcome these issues, the authors propose TurboEvolve, a multi-island evolutionary framework that innovatively integrates LLM-based multi-candidate verbal sampling, an adaptive online scheduling mechanism, and a clustering-based seed pool injection strategy. This approach enhances elite preservation while maintaining population diversity. TurboEvolve significantly improves sample efficiency and evolutionary robustness, outperforming state-of-the-art methods on multiple program optimization benchmarks with fewer evaluations and establishing new best-known solutions for several tasks.

Technology Category

• Search and Optimization: Evolutionary Computation
• Machine Learning: Evolutionary Learning
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

LLM-driven program evolution can discover high-quality programs, but its cost and run-to-run variance hinder reliable progress. We propose TurboEvolve, a multi-island evolutionary framework that improves sample efficiency and robustness under fixed evaluation budgets. Inspired by the multiple-offspring strategy in evolutionary algorithms, TurboEvolve introduces verbalized Sampling, prompting the LLM to emit K diverse candidates with explicit self-assigned sampling weights, and an online scheduler that adapts K to expand exploration under stagnation and reduce overhead during steady progress. To exploit existing solution pools, we further propose "seed-pool injection," which clusters seeds and assigns them across islands with controlled perturbations and elitist preservation to balance diversity and refinement. Across multiple program-optimization benchmarks, TurboEvolve consistently achieves stronger performance at lower budgets and improves best-known solutions on several tasks.

Problem

• LLM-driven program evolution
• sample efficiency
• run-to-run variance
• evaluation budget
• program optimization

Innovation

• LLM-driven program evolution
• multi-island evolutionary framework
• verbalized sampling
• online scheduler
• seed-pool injection