Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Docker: Remove explicit CUDA 11.8 Reference #4343

Merged
merged 3 commits into from
Oct 21, 2023
Merged

Conversation

whiteadam
Copy link
Contributor

Since @bdashore3 added CUDA 12.1 images, the explicit CUDA 11.8 ref for Torch set by @sammcj is no longer needed.

I tested this build locally with RTX 3060 and found no issues (other than a DEPRECATION NOTICE! from NVIDIA on 12.1).

Checklist:

@whiteadam whiteadam changed the title Remove explicit CUDA 11.8 Reference Docker: Remove explicit CUDA 11.8 Reference Oct 21, 2023
@oobabooga
Copy link
Owner

Can you replace cu118 with cu121 instead of removing it? For future-proofness when a new pytorch version gets released.

@whiteadam
Copy link
Contributor Author

It is done and tested.

I am curious about the impact on this line, though.

@oobabooga
Copy link
Owner

This one can be changed as well, as this file exists under bitsandbytes. Thanks for pointing out and for the fixes.

@oobabooga oobabooga merged commit 5a5bc13 into oobabooga:main Oct 21, 2023
@nalf3in
Copy link

nalf3in commented Oct 27, 2023

This unfortunately breaks compatibility with debian stable as the latest driver only supports up to cuda 12.0. Users trying to run this without a server gpu (which supports forward compatibility) receive a weird error message. Could we keep a second dockerfile with 11.8 with instructions in the readme for that use case ? I would be happy to make a pr

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants