Exploration of Large Language Model (LLM) Tools

About this project

This project looks to explore using LLMs as an automated tool to port our community models to other languages and to run on heterogeneous architectures. This project will use a rigorous testing process to assess the evaluated LLM approaches for correctness, performance, cost, and ease of use.

This work will be performed as part of the infrastructure project to investigate the autogeneration of bridge code, enabling code written in different languages to communicate correctly within a single code base.

Why this work is important

Porting millions of lines of code to a different language is a large investment. Automating this process could make it easier to switch to different languages as compiler support, hardware architecture, and community needs evolve.

How does this fit within the CSF

1. Fits under Better Practices
2. Increases code portability across different architectures
3. Increase interoperability of code written in different languages
4. Integrates the use of AI/ML

NSF NCAR Labs involved in this project

CISL
CGD

External partners

Columbia University
University of Wisconsin, Milwaukee