Use gfx11-ci Docker container for wheel build and kernel tests#873
Merged
mgehre-amd merged 4 commits intogfx11from Apr 14, 2026
Merged
Use gfx11-ci Docker container for wheel build and kernel tests#873mgehre-amd merged 4 commits intogfx11from
mgehre-amd merged 4 commits intogfx11from
Conversation
Switch build-wheel and test-kernels jobs to use the pre-built ghcr.io/rocm/vllm/gfx11-ci container image which has ROCm SDK, PyTorch, sccache, and uv pre-baked. This eliminates ~15 minutes of setup overhead per CI run by removing: - free-disk-space step (ROCm/PyTorch already in image layers) - setup-python, pip cache, sccache-action steps - ROCm SDK pip install and env var configuration - uv/system-deps installation in the test job The test-kernels job uses --device /dev/kfd --device /dev/dri for GPU passthrough on the self-hosted Strix Halo runner. Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
The GHA cache service URL (ACTIONS_CACHE_URL) is not forwarded into container jobs, causing sccache to fail on startup. Remove the flag so sccache operates as a local-only cache within the build. Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
Container jobs don't automatically inherit ACTIONS_CACHE_URL, ACTIONS_RESULTS_URL, and ACTIONS_RUNTIME_TOKEN from the runner. Use actions/github-script to export them so sccache can use the GitHub Actions cache backend for cross-run compilation caching. Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
mgehre-amd
force-pushed
the
matthias.gfx11-ci-use-docker
branch
from
460f873 to
a65bdff
Compare
sccache v0.8.1 uses the legacy GitHub Actions cache v1 API which was sunset April 2025. v0.14.0 supports the v2 API (ACTIONS_RESULTS_URL). Upgrade in both the Dockerfile and as a runtime override in the workflow until the next image rebuild picks up the Dockerfile change. Signed-off-by: Matthias Gehre <matthias.gehre@amd.com>
mgehre-amd
force-pushed
the
matthias.gfx11-ci-use-docker
branch
from
a65bdff to
d499123
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
build-wheelandtest-kernelsjobs to use the pre-builtghcr.io/rocm/vllm/gfx11-ci:latestcontainer image (from Add Docker CI image for gfx11 wheel builds #872)--device /dev/kfd --device /dev/drifor GPU passthrough on the self-hosted Strix Halo runnerVerified locally
gfx11-ci:localcontainertest_hip_w4a16.py(106/106 passed) with GPU passthroughTest plan
build-wheeljob succeeds with the containertest-kernelsjob detects the GPU and passes kernel testsupload-wheeljob (unchanged) still works on push to gfx11