[AMD] Stop rocm720 pr auto-runs (#29768)
Co-authored-by: bingxche <bingxche@users.noreply.github.com>
This commit is contained in:
co-authored by
bingxche
parent
f2756f53f3
commit
c70cc96905
@@ -16,14 +16,14 @@ on:
|
|||||||
# - "sgl-kernel/**"
|
# - "sgl-kernel/**"
|
||||||
# - ".github/workflows/pr-test-amd-rocm720.yml"
|
# - ".github/workflows/pr-test-amd-rocm720.yml"
|
||||||
# - "docker/rocm.Dockerfile"
|
# - "docker/rocm.Dockerfile"
|
||||||
pull_request:
|
# pull_request:
|
||||||
paths:
|
# paths:
|
||||||
- "python/**"
|
# - "python/**"
|
||||||
- "scripts/ci/**"
|
# - "scripts/ci/**"
|
||||||
- "test/**"
|
# - "test/**"
|
||||||
- "sgl-kernel/**"
|
# - "sgl-kernel/**"
|
||||||
- ".github/workflows/pr-test-amd-rocm720.yml"
|
# - ".github/workflows/pr-test-amd-rocm720.yml"
|
||||||
- "docker/rocm.Dockerfile"
|
# - "docker/rocm.Dockerfile"
|
||||||
workflow_dispatch:
|
workflow_dispatch:
|
||||||
inputs:
|
inputs:
|
||||||
target_stage_select:
|
target_stage_select:
|
||||||
@@ -115,21 +115,17 @@ env:
|
|||||||
DOCKERHUB_AMD_TOKEN: ${{ secrets.DOCKERHUB_AMD_TOKEN }}
|
DOCKERHUB_AMD_TOKEN: ${{ secrets.DOCKERHUB_AMD_TOKEN }}
|
||||||
|
|
||||||
concurrency:
|
concurrency:
|
||||||
# When called via workflow_call with run_all_tests=true, use a unique group per run to
|
# Scheduled and manual full runs use unique groups so they never cancel each other.
|
||||||
# avoid collisions with direct schedule/workflow_dispatch triggers. We use run_all_tests
|
# PR cancellation is kept only for pull_request events if this trigger is re-enabled.
|
||||||
# (not github.event_name) to detect this, because github.event_name inherits from the caller.
|
group: pr-test-amd-rocm720-${{ github.event_name == 'schedule' && format('schedule-{0}', github.run_id) || (inputs.run_all_tests || github.event_name == 'workflow_dispatch') && format('full-{0}', github.run_id) || inputs.pr_head_sha || inputs.ref || github.ref }}
|
||||||
# Manual dispatch runs also get unique groups so they never cancel each other.
|
cancel-in-progress: ${{ github.event_name == 'pull_request' }}
|
||||||
group: pr-test-amd-rocm720-${{ (inputs.run_all_tests || github.event_name == 'workflow_dispatch') && format('full-{0}', github.run_id) || inputs.pr_head_sha || inputs.ref || github.ref }}
|
|
||||||
cancel-in-progress: ${{ !inputs.run_all_tests && github.event_name != 'workflow_call' && github.event_name != 'workflow_dispatch' }}
|
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
call-gate:
|
call-gate:
|
||||||
# Runs on PRs (pr-gate.yml enforces the run-ci label / blocks drafts), the daily
|
# Runs on the daily '30 17 * * *' cron, workflow_dispatch and workflow_call.
|
||||||
# '30 17 * * *' cron, workflow_dispatch and workflow_call. It is skipped only on the
|
# It is skipped only on the 6h ('0 */6 * * *') cron, which cascades a skip
|
||||||
# 6h ('0 */6 * * *') cron, which cascades a skip to check-changes and every existing
|
# to check-changes and every existing test job so that cron runs the two
|
||||||
# test job, so that cron runs the two dsv4 jobs only. On PRs the existing jobs are
|
# dsv4 jobs only.
|
||||||
# additionally kept off via the change-detection guard in check-changes, so only the
|
|
||||||
# dsv4 jobs run there -- and only when this gate passes (they `needs` it).
|
|
||||||
if: github.event.schedule != '0 */6 * * *'
|
if: github.event.schedule != '0 */6 * * *'
|
||||||
uses: ./.github/workflows/pr-gate.yml
|
uses: ./.github/workflows/pr-gate.yml
|
||||||
secrets: inherit
|
secrets: inherit
|
||||||
@@ -183,9 +179,7 @@ jobs:
|
|||||||
- name: Detect file changes
|
- name: Detect file changes
|
||||||
id: filter
|
id: filter
|
||||||
uses: dorny/paths-filter@v3
|
uses: dorny/paths-filter@v3
|
||||||
# On pull_request only the dsv4 jobs run; skip change detection on PRs so the
|
if: steps.run-mode.outputs.run_all_tests != 'true'
|
||||||
# per-stage outputs stay 'false' and the existing test jobs are gated off.
|
|
||||||
if: steps.run-mode.outputs.run_all_tests != 'true' && github.event_name != 'pull_request'
|
|
||||||
with:
|
with:
|
||||||
filters: |
|
filters: |
|
||||||
main_package:
|
main_package:
|
||||||
@@ -1141,12 +1135,10 @@ jobs:
|
|||||||
-w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-large-8-gpu-mi35x-disaggregation-amd --timeout-per-file 1800 ${{ needs.check-changes.outputs.continue_on_error == 'true' && '--continue-on-error' || '' }}
|
-w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-large-8-gpu-mi35x-disaggregation-amd --timeout-per-file 1800 ${{ needs.check-changes.outputs.continue_on_error == 'true' && '--continue-on-error' || '' }}
|
||||||
|
|
||||||
# =============================================== DeepSeek-V4 (MI35x, 8-GPU) ====================================================
|
# =============================================== DeepSeek-V4 (MI35x, 8-GPU) ====================================================
|
||||||
# On `pull_request` and the 6h ('0 */6 * * *') cron these are the ONLY jobs that run;
|
# On the 6h ('0 */6 * * *') cron these are the only jobs that run; they also
|
||||||
# they also run on the daily cron alongside the full suite. On PRs they run only when
|
# run on the daily cron alongside the full suite. Scheduled runs pass
|
||||||
# call-gate succeeds, i.e. the PR carries the run-ci label (and is not a draft), so
|
# --continue-on-error. They stay selectable via workflow_dispatch / run on
|
||||||
# they are gated just like every other PR test job. On PR they hard-fail (a failure
|
# workflow_call full runs.
|
||||||
# blocks merge); on any scheduled run they pass --continue-on-error. They stay
|
|
||||||
# selectable via workflow_dispatch / run on workflow_call full runs.
|
|
||||||
dsv4-flash-fp4-fp8-amd-rocm720:
|
dsv4-flash-fp4-fp8-amd-rocm720:
|
||||||
needs: [call-gate]
|
needs: [call-gate]
|
||||||
if: |
|
if: |
|
||||||
|
|||||||
Reference in New Issue
Block a user