116 - Update CUDA in Dockerfile and bump requirement versions

[MaxJa4]

Description

Update CUDA in Dockerfile and bump requirement versions. See issue below for reasons why.

Fixes #116

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)

Does this PR introduce a breaking change?

No

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• My changes generate no new warnings
• New and existing unit tests pass locally with my changes (might be obsolete with CI later on)

[MaxJa4]

@okrusch You can use this branch to check whether #102 is fixed with this. Be aware that this now installs Torch 2.1.1 instead of 1.13.1

[samuelkuehnel]

I had an error when cloning the repo and running the command b5 install.

[MaxJa4]

I had an error when cloning the repo and running the command b5 install.

Would be great to see the actual source of the error which seems to be cut off at the top :)

[samuelkuehnel]

Sorry for the bad screenshot, I will add a better one tomorrow.

[samuelkuehnel]

This one contains the whole error message:

[MaxJa4]

@samuelkuehnel Thanks. I changed the installation method (which is also faster and simpler) for CUDA. Maybe that helps. Also removed libgit as it was a workaround for DVC and it seems to be fine without it now.

[samuelkuehnel]

The error seems to be on the PC as the git action that builds the image completes without an error. I still get an error when running b5 install.

@JuliusMiller also tried to run b5 install after cloning your branch and he sees the same error.

[MaxJa4]

The error seems to be on the PC as the git action that builds the image completes without an error. I still get an error when running b5 install.

@JuliusMiller also tried to run b5 install after cloning your branch and he sees the same error.

Thanks for testing it. Hmm maybe the lab pcs need an apt upgrade as they may use an old kernel.
I'll try to find a workaround.
It also works for me on my machine (Ubuntu).

[MaxJa4]

Found a possible workaround for the error. Can you or @JuliusMiller try again? Thanks!

[samuelkuehnel]

Error sadly still exists 😢

[MaxJa4]

Did some more research, as there is no helpful error message anymore.
Turns out, that the lab pc graphics cards are too "old" for CUDA 12.X, so I downgraded to 11.8 (which still works with PyTorch 2).
Hope it works now. I won't be able to work on PAF tomorrow, so no hurry from my side :) Thanks!

[samuelkuehnel]

I tried the b5 installcommand again and it still fails 😢

I am in the laboratory today until 4:30 pm if you want me to test other solutions.

[samuelkuehnel]

Update works. Image is built and container starts as expected