Analysis and Sampling of Molecular Simulations by adversarial Autoencoders
Use nbstrioput to commit so that diffing/merging is not nightmare.
Find a reasonable (4+ CPU cores, 32+ GB RAM, GPU welcome) Linux machine, preferably with recent stable OS release (Ubuntu 24.04, Debian 13, ...)
Not strictly required but things will be slooooooow otherwise ...
Things work with Nvidia/cuda now, will play with AMD/rocm and Apple/Metal one day ...
So make sure decent Nvidia GPU is installed and follow installation instructions
for your Linux flavor.
If drivers find it, nvidia-smi tells you.
We run Gromacs from our Docker image, hence Docker is needed.
- Install according to the instructions for your OS: https://docs.docker.com/engine/install/
- Install Nvidia container toolkit: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html
Check with docker run --rm --gpus all ubuntu nvidia-smi, it should say the same as plain nvidia-smi.
Troubleshoot by asking your favourite AI tool, they are fairly good in it.
Though Conda is my least prefered package manager, we want to be compatible with Binder environment etc.
Install Miniforge: https://github.com/conda-forge/miniforge (or some other distribution which will provide mamba command
and access to conda-forge channel.
Run make install_local from this directory. It takes few minutes to download all the required packages.
Run make run_local -- besides activating the environment it prepends PATH to have the right gmx command, and maybe other specific settings.
Jupyterlab starts, the command yiedls URL to go to. In this setup, we don't complicate things with SSL, the server runs on localhost only; if it is a remote machine, one has to use ssh port forwarding etc.
TODO