Data-Driven Dynamic Algorithm Dispatch with Large Language Models

Researchers from MIT and DARPA introduced an LLM-driven approach that generates dynamic algorithmic dispatch heuristics for high-performance linear algebra, using LLaMA 3 and a curated performance database to synthesize selection heuristics. In a case study on LU factorization, the model replicated expert-designed strategies, demonstrating the potential of LLMs for algorithmic discovery and adaptive software development.

Data-Driven Dynamic Algorithm Dispatch with Large Language Models By Rushil Shah, Emmanuel Lujan, Rabab Alomairy, Alan EdelmanSource: arXiv cs.AI https://arxiv.org/list/cs.AI/recent arXiv:2608.21584v1 Announce Type: new Abstract: We introduce a large language model /glossary/large-language-model LLM /glossary/llm -driven approach for generating dynamic algorithmic dispatch heuristics in high-performance linear algebra. By combining prompt engineering /glossary/prompt-engineering with LLaMA /glossary/llama 3 and a curated performance database, the model learns to synthesize selection heuristics that exploit structural patterns to identify fast algorithmic choices. A case study on LU factorization demonstrates the model's ability to replicate expert-designed strategies. This work, developed as part of the DARPA-MIT SmartSolve project, highlights the promise of LLMs for algorithmic discovery and the development of more adaptive, fast linear algebra software.Get AI news in your inbox Daily digest of what matters in AI.