OpenEvolve: Teaching LLMs to Discover Algorithms Through Evolution
2 days ago
- #evolutionary-computing
- #LLM
- #algorithm-discovery
- OpenEvolve is an open-source evolutionary coding agent that integrates large language models (LLMs) into a quality-diversity search framework for algorithm discovery.
- It uses LLM-guided edits to produce candidate programs, evaluates them with user-defined metrics, and organizes them using MAP-Elites.
- The system supports parallel, diversified exploration through an island model with migration.
- Key features include cascade staging, artifact side-channels for feedback, and optional LLM-based feedback in scoring.
- OpenEvolve has been applied in domains like systems optimization, scientific discovery, geospatial algorithms, and GPU kernel optimization.
- Architecture includes a Prompt Sampler, LLM Ensemble, Evaluator, Program Database, and Controller.
- Innovations include island-based evolution with lazy migration, MAP-Elites for diversity preservation, cascade evaluation, and a double-selection strategy.
- Sample use cases demonstrate dramatic speedups in algorithmic discovery, circle packing, GPU kernel optimization, and LLM prompt optimization.
- OpenEvolve provides both library and command-line interfaces, supports checkpoint/resume, and is available on GitHub.