Large Language Models for Compiler Optimization

This article explores the application of Large Language Models to code optimization, specifically for reducing instruction counts in LLVM assembly. The model achieves a 3.0% improvement over the compiler and demonstrates strong code reasoning abilities.