Fine-tuning Large Language Models for Inertial Confinement Fusion – a collaboration with NVIDIA
ORAL
Abstract
Inertial Confinement Fusion (ICF) has applications to stockpile stewardship and fusion energy. Predictive capabilities, however, are limited as ICF is a highly coupled, multi-physics, and energy-limited system. Success in this field has largely been a combination of simulations that guide experimental design and systematic semi-empirical tuning of experiments. More recently machine learning techniques have contributed to the interpretation and design of experiments. Domain specific Large Language Models offer the promise of contextually retrieving information from literature, combining them with data, interacting with subject matter experts using natural language processing, and potentially proposing hypotheses to explain measurements. In this talk, the process of developing the pre-trained and fine-tuned NVIDIA AI assisted agent for ICF is described. The goal is to develop a trusted aide that can interact collaboratively with ICF subject matter experts. The challenges associated with this project are discussed. Use cases that can potentially advance the field, using data from the OMEGA laser at the University of Rochester and the National Ignition Facility are described.
LA-UR-25-26028
LA-UR-25-26028
*This work was supported by the U.S. Department of Energy through the Los Alamos National Laboratory. Los Alamos National Laboratory is operated by Triad National Security, LLC, for the National Nuclear Security Administration of U.S. Department of Energy (Contract No. 89233218CNA000001).
–
Presenters
-
Radha Bahukutumbi
- Los Alamos National Laboratory
- University of Rochester