Repository navigation
[pipeline-generator] VLLM_CI_PLATFORM: build only one platform's jobs (amd or cuda) - #842
Merged
AndreasKaratzas merged 5 commits intoNov 4, 2026
Merged
Conversation
… (amd or cuda) TORCH_NIGHTLY now implies VLLM_CI_PLATFORM=cuda instead of dropping only the AMD lane. Signed-off-by: Micah Williamson <micah.williamson@amd.com> Co-authored-by: Cursor <cursoragent@cursor.com>
Signed-off-by: Micah Williamson <micah.williamson@amd.com> Co-authored-by: Cursor <cursoragent@cursor.com>
Drops platforms.py and its CUDA device list; the check is now AMD vs. not AMD, next to is_amd_device. TORCH_NIGHTLY keeps exactly the jobs it kept before. Signed-off-by: Micah Williamson <micah.williamson@amd.com> Co-authored-by: Cursor <cursoragent@cursor.com>
Member
|
Wrong button sorry !! |
Co-authored-by: Cursor <cursoragent@cursor.com> # Conflicts: # buildkite/pipeline_generator/buildkite_step.py # buildkite/pipeline_generator/pipeline_generator.py # buildkite/tests/pipeline_generator/test_step.py
…TFORM Signed-off-by: Micah Williamson <micah.williamson@amd.com> Co-authored-by: Cursor <cursoragent@cursor.com>
micah-wil
changed the title
khluu
approved these changes
Nov 4, 2026
khluu
left a comment
khluu
left a comment
Member
There was a problem hiding this comment.
Kevin's review agent: approve for CI changes only
Contributor
Author
|
Test build showing only AMD jobs being selected with Test build showing no impact to a regular build: https://buildkite.com/vllm/ci/builds/100040/list |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Adds
VLLM_CI_PLATFORMto the pipeline generator. Set it toamdorcudaand the build contains only that platform's jobs; every other step is left out of the pipeline (not deferred into "Run more jobs").Each step belongs to the platform of its
device:amd:amd_cpu(the AMD build steps), the MI250/MI300/MI325/MI355 queues, andamd-mirrors. A CUDA step with an AMD mirror keeps the mirror and drops the CUDA job.cuda: the NVIDIA GPUs, thecpu/cpu-small/cpu-mediumqueues, and steps with no device (image builds, default GPU jobs).Within the platform the usual selection rules still apply (add
RUN_ALL=1to run everything).VLLM_CI_STEP_KEYSnaming a step outside the platform fails generation. AMD-only builds skip the test selector, since it never picks AMD steps.Torch nightly
TORCH_NIGHTLY=1now impliesVLLM_CI_PLATFORM=cuda, replacing the AMD-only exclusion (include_amd, thetorch_nightlyargument ofselect_steps_and_dependencies). Behavior change: the torch nightly also drops the Intel CPU/HPU, Arm CPU and Ascend tests, which pin their own torch and gave no torch-nightly signal.TORCH_NIGHTLY=1withVLLM_CI_PLATFORM=amdfails generation.Why a platform value and not an
AMD_ONLYflag: the torch nightly needs the opposite filter (CUDA only), and #777 adds a third (AMD on gfx942/gfx950). One platform value replaces those special cases instead of adding another.Testing
pytest buildkite/tests: new tests for the device classification, step conversion per platform, key validation, env parsing and the selector skip.main's real.buildkite/ci_config.yamlwithRUN_ALL=1in each mode: 368 jobs with no platform, 181 withamd(all on AMD queues, plus the "Run more jobs" step), 177 withcuda, and the same 177 withTORCH_NIGHTLY=1. No mode left adepends_onpointing at a dropped step.Notes
device. The torch nightly already builds them today.AITER_NIGHTLY=1implyVLLM_CI_PLATFORM=amd.