[ENCCS Webinar] Machine Learning Interatomic Potentials on HPC: From Installation to Benchmarking
Online
A joint webinar by ENCCS and the Sweden AI Factory.
About this webinar
Machine-learning interatomic potentials (MLIPs) can approach the accuracy of density functional theory (DFT) at a much lower cost, and many pretrained "foundation" models are now available. Using them is still painful. Each framework needs its own versions of PyTorch or TensorFlow, CUDA and accelerated kernels; they conflict when installed together; and a wrong setup can silently fall back to the CPU or to the wrong model.
oh-my-mlip is an open-source hub that brings the major MLIP frameworks (MACE, UMA, SevenNet, NequIP, ORB, GRACE and more) into one registry. It installs each framework in its own pinned conda environment and counts a model as installed only after it computes energies and forces on the GPU. It is designed to be driven by LLM coding agents such as Claude Code and Codex, but every step also runs from the command line.
In this first part, we will show how to install, verify and run MLIPs on GPU clusters, including a demo in which an LLM agent sets up a model from scratch. We will then benchmark many MLIPs on adsorption energies for heterogeneous catalysis with CatBench. Hands-on materials will be provided after the webinar so that participants can practice on their own.
Part 2, "Machine Learning Interatomic Potentials on HPC: From Fine-tuning to Distillation", follows on 21 January 2027: see Part 2.
Who is the webinar for
This webinar is for:
- Computational chemists and materials scientists who want to use MLIPs alongside or instead of DFT
- Researchers who have struggled to install several MLIP frameworks side by side
- HPC support staff and research software engineers who install and maintain MLIP software for users
- Anyone curious about how LLM coding agents can handle the setup of scientific software
Basic familiarity with Linux, Python and conda is helpful. No machine-learning background is required.
Key takeaways
After this webinar, participants will know:
- Why installing MLIPs is hard (conflicting framework and CUDA versions, compiled kernels, gated weights) and how isolated, pinned environments avoid these problems
- How to install a model and confirm that it really computes energies and forces on the GPU
- How to get a ready-to-use ASE calculator for any supported model in their own scripts
- How to let an LLM agent do the setup through the Claude Code plugin or AGENTS.md, and how to check its work
- How to benchmark many MLIPs against DFT adsorption energies with CatBench, including their own VASP data
- Practical tips for running MLIPs on shared GPU clusters with Slurm
- The current scope: Linux x86-64 systems with NVIDIA GPUs (Arm-based GH200 nodes and AMD GPUs are not supported yet)
Speaker and moderators
Jinuk Moon (Seoul National University)
Jinuk Moon is a PhD student in the Department of Materials Science and Engineering at Seoul National University, working in Prof. Jeong Woo Han’s group. His research combines density functional theory and machine-learning interatomic potentials (MLIPs) for computational catalysis. He develops CatBench, a benchmark framework for evaluating MLIPs on heterogeneous catalysis, and oh-my-mlip, an open-source hub for installing, running, fine-tuning and distilling MLIPs with LLM agents. More about his work on his website.
Moderators: Yonglei Wang and Wei Li (ENCCS)
More events & contact
Check out more upcoming events from ENCCS and our European network at https://enccs.se/events/, as well as available ENCCS lesson materials, suitable also for self-learning.
For questions regarding this webinar or general questions about ENCCS training events, please contact training@enccs.se.
ENCCS and Sweden AI Factory