Interactive session with a GPU

Download the files · Source on GitHub

A marimo notebook on one GPU on the P1 DTU HPC. It shows the GPU it got, times a matrix multiply on CPU and GPU, and trains a small classifier. gpu.py runs the same check as a script, and pyproject.toml and uv.lock pin the packages. It suits installing packages, test runs and debugging on p1i, the P1 queue for interactive GPU work.

With aicentre

aicentre gallery get int-gpu                           # the example, onto the cluster
aicentre marimo examples/int-gpu/notebook.py --gpu    # the notebook on a GPU, in your browser
aicentre shell examples/int-gpu --gpu                  # or a shell on the GPU node

• gallery get has the login node download the example into your project directory, so there is nothing to clone or copy.
• --gpu asks p1i for a GPU, starts marimo with the example's own packages, and opens it in your browser through the login node. Ctrl-C ends the job; --time 2h asks for longer.