Outwards-facing setup and documentation for running WG

[florianscheidl]

Describe the task. It can be a feature, documentation, etc.

We want to prepare an easy (minimal) setup for anyone to run WG code and documentation for this.
This is particularly relevant for external collaborations (e.g. Nvidia for code optimization) but also the wider reach of the project.

Specifically, this should support:

• hardware and installation requirements
• commands for common workflows, such as training, inference
• direct execution and slurm systems

Hedgedoc URL, if you are keeping notes, plots, logs in hedgedoc.

No response

Area

• datasets, data readers, data preparation and transfer
• model
• science
• infrastructure and engineering
• evaluation, export and visualization
• documentation
• performance

[tjhunter]

I think the harder part will be to specify access to toy data.

[florianscheidl]

Notes from meeting with eng-team:

Running outside of HPCs:

• Datasets: https://www.ecmwf.int/en/about/media-centre/aifs-blog/2025/introducing-anemoi-training-ready-version-era5
• FlashAttn
• Platform-env module: Where to find the configs and the data paths?
• Private config: Could be any file really, as long as it has the paths
• Priority: 1. NV: could be a separate branch for now before merging, 2. Theme 3

[tjhunter]

I started to wiret down instructions here:
https://gitlab.jsc.fz-juelich.de/hedgedoc/s/5ykm9w_kV

Eventually we should move them to the wiki when they are more stable.

[clessig]

I started to wiret down instructions here: https://gitlab.jsc.fz-juelich.de/hedgedoc/s/5ykm9w_kV

Eventually we should move them to the wiki when they are more stable.

The wiki is private. The goal is to have it in a public space when it's stable (hedged would be one option, but others are possible).