MLIP School 2026 — Dataset craft for machine-learned interatomic potentials using ACEpotentials.jl

You will build a silicon dataset from scratch, fit ACE potentials, and learn why a low RMSE can still be a bad potential.

The opening talk is a five-minute read of the slides from the room, published flat so it keeps working after the session.

Two ways to work

In your browser

Work through the day-one exercises in order: E1, E2, E3, then C. E1x is an optional extension — take it after E1 if you are ahead. It resumes nothing and spends none of your labels, so skipping it costs you nothing.

Keep one exercise open at a time. Each of these pages carries its own Python runtime and uses about 1.5 GB of memory, so close the tab for the exercise you have finished before opening the next. Two open at once is fine on most laptops; three is not.

Day two — bring your own system

A separate one-hour session the following day. It resumes nothing from day one: you bring a system of your own and run the same loop on it.

On your own machine

Run:

uvx mograder student https://mlipschool.uk/mograder.toml

Optional: run the whole stack locally. If you have Docker, you can run both services — the oracle that labels your structures and the workbench that fits them — on your own machine. Useful if the hosted services are busy or unreachable, and it is how you keep the whole workflow after the school. About 1.3 GB to download.

curl -O https://mlipschool.uk/local.yml
docker compose -f local.yml up -d

Then set the notebook's Oracle field to http://localhost:8001 and ACE workbench to http://localhost:8002. Nothing local checks keys, so any key string works, and the label budget is your own rather than the hosted allowance. This works with the notebook you run yourself, not with one opened from this site.

Licence: The oracle image contains the MACE-MH-1 model, licensed under the Academic Software Licence (ASL): academic and non-commercial use only. https://github.com/gabor1/ASL

You will need your API key. You were given one on a printed slip at registration. Paste it into the notebook's API key field, and use the same key throughout — E2, E3 and C resume the potential you fit in E1, and models belong to the key that fitted them.

Credits. Authored by James Kermode, with feedback and support from Facundo Costa, Roman Shantsila and Svenja Janke.

AI declaration. Support from AI coding agents setting up the infrastructure, but I designed and (with help from Roman, Facundo and Svenja) fact checked everything and take responsibility for scientific decisions as well as for the accuracy of what’s there.