A Flexible Framework for AI-Driven Scientific and Algorithmic Discovery
LLM-driven evolutionary search is a powerful approach for discovering algorithms and designs, but existing frameworks are difficult to extend and compare.
We introduce SkyDiscover, a flexible and modular framework for AI-driven scientific and algorithmic discovery. We use it to build two adaptive evolutionary algorithms about 2.5K LoC and set SOTA results 200+ tasks:
- 🚀 Concrete discoveries with real impact: 41% lower cross-cloud transfer cost, 14% better GPU load balance for MoE serving, and 29% lower KV-cache pressure via GPU model placement.
- 🏆 Best open-source performance: improving median scores by ~34% on 172 Frontier-CS programming problems over OpenEvolve, GEPA, and ShinkaEvolve.
- 💪 Matching or exceeding AlphaEvolve and human SOTA: on 6/8 math and 6/6 systems optimization tasks.
Contributors
Shu Liu, Mert Cemri, Shubham Agarwal, Alexander Krentsel, Ashwin Naren, Qiuyang Mang, Zhifei Li, Akshat Gupta, Monishwaran Maheswaran, Audrey Cheng, Melissa Pan, Ethan Boneh, Kannan Ramchandran, Koushik Sen, Alexandros G. Dimakis, Matei Zaharia, Ion Stoica
Publications
CoRR – AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization