Skip to content

ci: fix SYCL build #12546

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

Merged
merged 1 commit into from
Mar 24, 2025
Merged

ci: fix SYCL build #12546

merged 1 commit into from
Mar 24, 2025

Conversation

qnixsynapse
Copy link
Collaborator

It seems debug build flags set by cmake when -DCMAKE_BUILD_TYPE=Debug is set breaks the SYCL build.

Different errors on different distros. Disable it in SYCL build and use only release build test.

@qnixsynapse qnixsynapse requested a review from ggerganov as a code owner March 24, 2025 12:51
@github-actions github-actions bot added the devops improvements to build systems and github actions label Mar 24, 2025
@ggerganov ggerganov merged commit 48d7021 into ggml-org:master Mar 24, 2025
2 checks passed
@qnixsynapse qnixsynapse deleted the fix/ci branch March 24, 2025 12:58
@ggerganov
Copy link
Member

We should update the ggml and whisper.cpp scripts in a similar way - they will also use the new SYCL node to run ggml-ci.

@qnixsynapse
Copy link
Collaborator Author

Yeah.. Will do once I confirm the test passes here.

@ggerganov
Copy link
Member

Hm, even though the tests are green, there seems to be an issue with the text generation of some quants:

https://github.com./ggml-org/ci/blob/results/llama.cpp/48/d7021c61ceda6fcf1a7294d2115b8e1a53ae95/ggml-6-x86-sycl/pythia_1_4b-tg-q4_0.log#L159

I believe the meaning of life isEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

@qnixsynapse
Copy link
Collaborator Author

I think I have seen this before..

Is this because of it is detecting two GPUs I wonder...
Possible that multi GPU is broken. sigh

@Alcpz
Copy link
Collaborator

Alcpz commented Mar 24, 2025

I may have missed some, but the errors all seem to contain q4_0 quants, which is the type that was reworked recently. Maybe support for multi-gpu wasn't considered for the patch?

@qnixsynapse
Copy link
Collaborator Author

@Alcpz I am currently testing it. For far I can confirm that it wasn't multi gpu.

@NeoZhangJianyu
Copy link
Collaborator

@ggerganov

Hm, even though the tests are green, there seems to be an issue with the text generation of some quants:

https://github.com./ggml-org/ci/blob/results/llama.cpp/48/d7021c61ceda6fcf1a7294d2115b8e1a53ae95/ggml-6-x86-sycl/pythia_1_4b-tg-q4_0.log#L159

I believe the meaning of life isEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE

There are iGPU and dGPU in the PC.
Both are used as default.

Please use the dGPU by:
export ONEAPI_DEVICE_SELECTOR="level_zero:0"

@NeoZhangJianyu
Copy link
Collaborator

q4_0 doesn't support iGPU, but support dGPU.
In this case, the iGPU+dGPU lead to unknown issue.
In pure dGPUs, it should work well.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
devops improvements to build systems and github actions
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants