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

Start an Asahi version #369

Merged
merged 1 commit into from
Nov 6, 2024
Merged

Start an Asahi version #369

merged 1 commit into from
Nov 6, 2024

Conversation

ericcurtin
Copy link
Collaborator

Asahi has a forked version of mesa while it upstream.

@rhatdan
Copy link
Member

rhatdan commented Nov 5, 2024

Does this mean on Asahi we can't use the standard RamaLama image?

@ericcurtin
Copy link
Collaborator Author

ericcurtin commented Nov 5, 2024

Fedora Asahi Remix is 99% percent Fedora... The two significant packages it forks are the kernel and mesa (mesa being where the userspace GPU driver is).

This Vulkan driver "Honeykrisp" is weeks old, so the patches have not made their way to mainline Fedora mesa yet:

https://www.phoronix.com/news/Mesa-HoneyKrisp-October

anyway, with the other images being UBI based, even besides this mesa fork, we would have had to make a special container image for Asahi because Asahi is only enabled for Fedora, not UBI.

So yeah, we need a specific container image for Asahi.

I don't expect to make this exception often, but the Asahi community is so vibrant and kinda is a flagbearer for ARM Linux efforts and ARM GPU drivers. I think it's worth it.

We could use the standard image and do CPU inferencing on Asahi, but whats the point in doing CPU inferencing.

@ericcurtin
Copy link
Collaborator Author

ericcurtin commented Nov 5, 2024

@Conan-Kudo expressed an interest in running llama.cpp outside of containers which is fine too, as long as one is using the forked mesa and building/using llama.cpp built with -DGGML_VULKAN=1 it should work

@ericcurtin ericcurtin force-pushed the fedora-asahi branch 2 times, most recently from f6fa180 to c5e4d33 Compare November 6, 2024 00:12
@rhatdan
Copy link
Member

rhatdan commented Nov 6, 2024

We should be able to build the asahi image from the RamaLama Containerfile.

podman build --from fedora:41 -t quay.io/ramalama/asahi:latest containers-images/ramalama

That way we don't need to update multiple containerfiles.

@ericcurtin
Copy link
Collaborator Author

ericcurtin commented Nov 6, 2024

We should be able to build the asahi image from the RamaLama Containerfile.

podman build --from fedora:41 -t quay.io/ramalama/asahi:latest containers-images/ramalama

That way we don't need to update multiple containerfiles.

More like:

podman build --from fedora:41 -t quay.io/ramalama/asahi:latest containers-images/vulkan

We can do, there's gonna be quite a few "if asahi" type statements though... Even our UBI image uses it's own mesa fork, which is different to this one.

Asahi has a forked version of mesa while it upstream.

Signed-off-by: Eric Curtin <[email protected]>
@rhatdan
Copy link
Member

rhatdan commented Nov 6, 2024

No, I was thinking they were the same Containerfiles, other then the From, I will merge and we can revisit later.

@rhatdan rhatdan merged commit 8d8af5e into main Nov 6, 2024
12 checks passed
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.

2 participants