Performance-Aligned LLMs for Generating Fast Code

TL;DR

This paper presents a reinforcement learning approach to fine-tune large language models for generating high-performance code, significantly improving execution speed on benchmark tasks.

Contribution

It introduces a novel reinforcement learning methodology to align LLM outputs with code performance, enhancing their ability to generate faster code.

Findings

Expected speedup increased from 0.9 to 1.6 for serial code.

Speedup improved from 1.9 to 4.5 for OpenMP code.

Demonstrates effectiveness on benchmark tasks.

Abstract

Optimizing scientific software is a difficult task because codebases are often large and complex, and performance can depend upon several factors including the algorithm, its implementation, and hardware among others. Causes of poor performance can originate from disparate sources and be difficult to diagnose. Recent years have seen a multitude of work that use large language models (LLMs) to assist in software development tasks. However, these tools are trained to model the distribution of code as text, and are not specifically designed to understand performance aspects of code. In this work, we introduce a reinforcement learning based methodology to align the outputs of code LLMs with performance. This allows us to build upon the current code modeling capabilities of LLMs and extend them to generate better performing code. We demonstrate that our fine-tuned model improves the expected speedup of generated code over base models for a set of benchmark tasks from 0.9 to 1.6 for serial code and 1.9 to 4.5 for OpenMP code.