Workshops Archive

Adopting Emerging AI Frameworks in HPC Environments


Workshop: AI on HPC: Performance Engineering, Challenges and Opportunities

Authors: Sasmita Mohapatra (UT Dallas)

Abstract: Computational scientists use HPC in two primary ways: to run simulations and to process data, iterating until a new observation emerges. As scientific computation becomes more sophisticated and resource-heavy, researchers increasingly turn to AI/MLtools to accelerate discovery. Yet before these tools can be used, scientists often spend undue time on tasks like environment setup, dependency management, model acquisition, and determining the correct way to run each framework on shared HPC systems. This demands a system that can streamline and standardize these setup tasks, so that scientists can focus on analysis rather than infrastructure. With a reproducible, well-defined workflow for deploying and running AI tools on HPC, we can accomplish this goal. In this work, we demonstrate such a workflow using four widely used AI tools: Stable Diffusion (text-to-image generation), vLLM (high-throughput large language model inference), Ollama (lightweight local LLM serving), and AlphaFold3 (protein-structure prediction).


Back to AI on HPC: Performance Engineering, Challenges and Opportunities Archive Listing

Back to Full Workshop Archive Listing