ci: disable ccache and toolkit cache for self-hosted OpenVINO runners#20594
ci: disable ccache and toolkit cache for self-hosted OpenVINO runners#20594
Conversation
|
Well, that was fruitful... |
Co-authored-by: ggerganov <1991296+ggerganov@users.noreply.github.com>
|
Ah, just a massive delay? |
|
Yup, very slow. copilot ngmi But I wonder if the suggestion to remove the toolkit cache from the |
Yep, will do once CI finishes. |
|
Even though the workflow is self-hosted, it seems the cache still works: https://github.com/ggml-org/llama.cpp/actions/runs/23111057155/job/67128454815 So maybe we want to keep the toolkit cache active. Not 100% sure. |
It works if it was created by that os previously, otherwise it will create a new one, might as well disable it. |
|
Yes, lets disable. It's better to put a bit more load on that runner compared to taking a bit from the GH cache. |
The cache is named with |
GitHub Actions cache causes issues on self-hosted runners due to cross-OS detection — runners alternately create their own cache entries, leading to conflicts. Both the ccache and OpenVINO toolkit cache need to be fully disabled for self-hosted runners.
build.yml: Addedif: matrix.variant != 'gpu'to theccacheandUse OpenVINO Toolkit Cachesteps. Thegpuvariant runs onself-hostedwhilecpuruns onubuntu-24.04. The existingSetup OpenVINO Toolkitcondition (cache-hit != 'true') naturally evaluates to true when the cache step is skipped, so the toolkit is always downloaded fresh on self-hosted.build-self-hosted.yml: Removed theUse OpenVINO Toolkit Cachestep entirely fromggml-ci-intel-openvino-gpu-low-perf(always self-hosted) and dropped the cache-hit condition onSetup OpenVINO Toolkitso it runs unconditionally.💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.