[vLLM] Move vLLM test schedules into per-arch wrapper workflows#7560
Draft
quinnlp wants to merge 7 commits into
Draft
[vLLM] Move vLLM test schedules into per-arch wrapper workflows#7560quinnlp wants to merge 7 commits into
quinnlp wants to merge 7 commits into
Conversation
quinnlp
force-pushed
the
quinnlp/vllm-ci-schedule-consistency
branch
4 times, most recently
from
July 23, 2026 15:49
5049833 to
207ea0b
Compare
quinnlp
changed the title
Split the PVC and BMG test schedules out of the vllm-tests.yml dispatch workflow into thin per-arch wrapper workflows, keeping the shared job body in the vllm-tests-reusable.yml reusable (unchanged). All three entry points are now symmetric thin callers of the reusable: - vllm-tests.yml - workflow_dispatch only (no schedule) - vllm-tests-pvc.yml - PVC schedule, runner_label max1550 (new) - vllm-tests-bmg.yml - BMG schedule, runner_label b580, skip_list xe2 This mirrors the intent of the *-pvc / *-bmg benchmark wrappers and keeps the dispatch input surface separate from the reusable's call contract, so future trigger surfaces (e.g. pull_request) are added as new wrappers without touching the body. The dispatch wrapper's input->reusable mapping is simplified to pass values through directly; this also fixes use_pyenv_python, which previously used `|| true` and could never be disabled from the dispatch form. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
quinnlp
force-pushed
the
quinnlp/vllm-ci-schedule-consistency
branch
from
July 23, 2026 16:14
b3ff477 to
4d612c2
Compare
quinnlp
changed the title
quinnlp
changed the title
The constant -ci concurrency group meant a PR run labeled run-vllm-tests and the nightly scheduled run shared one group and cancelled each other. Key the group by PR number (falling back to a shared ci lane for scheduled runs) so each PR gets its own lane. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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.
This patch splits the PVC scheduled vllm-test runs out of the
vllm-tests.ymldispatch workflow intovllm-tests-pvc.ymlworkflow. The BMG schedule is already implemented this way.The main motivation for this change is to consistently follow a cleaner structure that will simplify adding PR + label triggered runs.
This dispatch interface is unchanged.
This patch also moves the
cancel-in-progresslogic for BMG tovllm-tests-bmg.yml.This patch also removes some dead code in the
vllm-tests-reusuablecall in the dispatch path and fixes a bug that always passedtrueforuse_pyenv_python.vLLM tests: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/30060545420
vLLM tests BMG: https://github.com/intel/intel-xpu-backend-for-triton/actions/runs/30060556629