Overview
OpenEvolve is an open-source framework for evolutionary code discovery. It leverages MAP-Elites, island-based parallelism and LLM-assisted mutation strategies to automatically evolve and optimize algorithms, GPU kernels and code implementations. The project includes reproducible pipelines, benchmarks and visualization tools.
Key features
- Automated search combining LLMs and evolutionary algorithms for code generation and optimization.
- Multi-objective optimization and island-based parallelism to maintain diversity and avoid premature convergence.
- Rich examples and visualization tools for benchmarking and inspecting evolution results.
Use cases
- Discovering high-performance implementations (GPU kernels, algorithmic improvements).
- Research and engineering for automated algorithm exploration and benchmarking.
- Using LLMs as mutation or proposal mechanisms to expand the search space.
Technical notes
- Primarily Python-based, with examples in Rust and Metal; plugin evaluators and configurable pipelines.
- Provides reproducible configs, evaluators and visualization for integration into experiments.
- Apache-2.0 licensed, active community and growing examples/tutorials.