Search papers, labs, and topics across Lattice.
The paper introduces CodeEvolve, an open-source framework that leverages large language models (LLMs) and an islands-based genetic algorithm to discover and optimize algorithmic solutions. It uses execution feedback and task-specific metrics to guide the evolutionary process, employing techniques like context-aware recombination and adaptive meta-prompting. The framework achieves superior or competitive performance compared to AlphaEvolve on established benchmarks, demonstrating that open-weight models can match or surpass closed-source models with less compute.
Open-source CodeEvolve shows that LLMs, when combined with evolutionary algorithms, can outperform proprietary systems like AlphaEvolve in algorithmic discovery, even with significantly less compute.
We introduce CodeEvolve, an open-source framework that combines large language models (LLMs) with evolutionary search to synthesize high-performing algorithmic solutions. CodeEvolve couples an islands-based genetic algorithm with modular LLM orchestration, using execution feedback and task-specific metrics to guide selection and variation. Exploration and exploitation are balanced through context-aware recombination, adaptive meta-prompting, and targeted refinement of promising solutions. We evaluate CodeEvolve on benchmarks previously used to assess Google DeepMind's AlphaEvolve, showing superior performance on several tasks and competitive results overall. Notably, open-weight models often match or exceed closed-source baselines at a fraction of the compute cost. We provide extensive ablations analyzing the contribution of each component and release our framework and experimental results at https://github.com/inter-co/science-codeevolve.
