Data-Driven Dynamic Algorithm Dispatch with Large Language Models

1. 1Combine prompt engineering with LLaMA 3 to generate dynamic algorithmic dispatch heuristics
2. 2Curate a performance database to train the model on structural patterns
3. 3Apply the trained model to identify fast algorithmic choices for LU factorization
4. 4Evaluate the model's performance using a case study
5. 5Refine the model by incorporating expert-designed strategies

Researchers and developers in high-performance linear algebra can benefit from this approach to optimize algorithmic choices, while data scientists and machine learning engineers can apply similar techniques to other domains

💡 Large language models can learn to synthesize selection heuristics that exploit structural patterns for fast algorithmic choices

🤖 Use LLMs for dynamic algorithm dispatch in linear algebra! 📊