SLM Ensembles: Smarter, Smaller AI for Code Debugging

SLM Ensembles: Smarter, Smaller AI for Code Debugging

Harnessing small language models to efficiently locate software bugs

COSMosFL demonstrates how ensembles of smaller, open-source language models can effectively compete with large closed-source models for software fault localization.

  • Combines multiple small language models to achieve performance comparable to larger models
  • Reduces deployment costs, energy consumption, and security concerns
  • Enables local deployment without compromising effectiveness
  • Demonstrates practical viability for software engineering tools without reliance on massive language models

This research matters because it provides an accessible, cost-effective approach for building AI-powered debugging tools that respects privacy concerns while maintaining high performance standards.

COSMosFL: Ensemble of Small Language Models for Fault Localisation

125 | 323