minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[diffusion] CI: dynamic load-balanced partitioning for diffusion CI (#15528)

Browse Source 
Co-authored-by: daiweitao <dwti614707404@163.com>
Co-authored-by: SGLang CI <ci@sglang.ai>
This commit is contained in:
Prozac614
2026-04-12 13:02:43 +08:00
committed by GitHub
co-authored by daiweitao SGLang CI
parent d6c9d9116b
commit 45472d70cc
11 changed files with 1877 additions and 175 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/pr-test-multimodal-gen.yml
+108 -12
View File
@@ -50,20 +50,57 @@ env:
SKIP_STAGE_HEALTH_CHECK: ${{ inputs.skip_stage_health_check == 'true' }}
jobs:
multimodal-gen-test-1-gpu:
compute-diffusion-partitions:
if: |
(inputs.target_stage == 'multimodal-gen-test-1-gpu') ||
(inputs.target_stage == 'multimodal-gen-test-2-gpu') ||
(
!inputs.target_stage &&
((github.event_name == 'schedule' || inputs.test_parallel_dispatch == 'true') || (inputs.caller_needs_failure != 'true' && !cancelled())) &&
inputs.multimodal_gen == 'true'
)
runs-on: ubuntu-latest
outputs:
matrix-1gpu: ${{ steps.compute.outputs.matrix-1gpu }}
matrix-2gpu: ${{ steps.compute.outputs.matrix-2gpu }}
partition-count-1gpu: ${{ steps.compute.outputs['partition-count-1gpu'] }}
partition-count-2gpu: ${{ steps.compute.outputs['partition-count-2gpu'] }}
plan-1gpu: ${{ steps.compute.outputs.plan-1gpu }}
plan-2gpu: ${{ steps.compute.outputs.plan-2gpu }}
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.pr_head_sha || inputs.git_ref || github.sha }}
- name: Set up Python
uses: actions/setup-python@v5
with:
python-version: '3.10'
- name: Compute partitions
id: compute
run: |
python scripts/ci/utils/diffusion/compute_diffusion_partitions.py --min-time 1200 --target-time 1800 --max-time 2400 --max-partitions 10
multimodal-gen-test-1-gpu:
needs: compute-diffusion-partitions
if: |
always() &&
needs.compute-diffusion-partitions.result == 'success' &&
needs.compute-diffusion-partitions.outputs.matrix-1gpu != '{"include":[]}' &&
(
(inputs.target_stage == 'multimodal-gen-test-1-gpu') ||
(
!inputs.target_stage &&
((github.event_name == 'schedule' || inputs.test_parallel_dispatch == 'true') || (inputs.caller_needs_failure != 'true' && !cancelled())) &&
inputs.multimodal_gen == 'true'
)
)
runs-on: 1-gpu-h100
timeout-minutes: 240
strategy:
fail-fast: false
matrix:
part: [0, 1]
matrix: ${{ fromJson(needs.compute-diffusion-partitions.outputs.matrix-1gpu) }}
steps:
- name: Checkout code
uses: actions/checkout@v4
@@ -91,33 +128,48 @@ jobs:
env:
RUNAI_STREAMER_MEMORY_LIMIT: 0
CONTINUE_ON_ERROR_FLAG: ${{ inputs.continue_on_error == 'true' && '--continue-on-error' || '' }}
PARTITION_PLAN_JSON: ${{ needs.compute-diffusion-partitions.outputs.plan-1gpu }}
run: |
cd python
python3 sglang/multimodal_gen/test/run_suite.py \
--suite 1-gpu \
--partition-id ${{ matrix.part }} \
--total-partitions 2 \
--total-partitions ${{ needs.compute-diffusion-partitions.outputs['partition-count-1gpu'] }} \
--partition-plan-json "$PARTITION_PLAN_JSON" \
$CONTINUE_ON_ERROR_FLAG
- name: Upload execution report
if: always()
uses: actions/upload-artifact@v4
with:
name: diffusion-report-1gpu-${{ matrix.part }}
path: python/sglang/multimodal_gen/test/execution_report_*.json
retention-days: 1
- uses: ./.github/actions/upload-cuda-coredumps
if: failure()
with:
artifact-suffix: ${{ matrix.part }}
multimodal-gen-test-2-gpu:
needs: compute-diffusion-partitions
if: |
(inputs.target_stage == 'multimodal-gen-test-2-gpu') ||
always() &&
needs.compute-diffusion-partitions.result == 'success' &&
needs.compute-diffusion-partitions.outputs.matrix-2gpu != '{"include":[]}' &&
(
!inputs.target_stage &&
((github.event_name == 'schedule' || inputs.test_parallel_dispatch == 'true') || (inputs.caller_needs_failure != 'true' && !cancelled())) &&
inputs.multimodal_gen == 'true'
(inputs.target_stage == 'multimodal-gen-test-2-gpu') ||
(
!inputs.target_stage &&
((github.event_name == 'schedule' || inputs.test_parallel_dispatch == 'true') || (inputs.caller_needs_failure != 'true' && !cancelled())) &&
inputs.multimodal_gen == 'true'
)
)
runs-on: 2-gpu-h100
timeout-minutes: 240
strategy:
fail-fast: false
matrix:
part: [0, 1]
matrix: ${{ fromJson(needs.compute-diffusion-partitions.outputs.matrix-2gpu) }}
steps:
- name: Checkout code
uses: actions/checkout@v4
@@ -146,14 +198,24 @@ jobs:
env:
RUNAI_STREAMER_MEMORY_LIMIT: 0
CONTINUE_ON_ERROR_FLAG: ${{ inputs.continue_on_error == 'true' && '--continue-on-error' || '' }}
PARTITION_PLAN_JSON: ${{ needs.compute-diffusion-partitions.outputs.plan-2gpu }}
run: |
cd python
python3 sglang/multimodal_gen/test/run_suite.py \
--suite 2-gpu \
--partition-id ${{ matrix.part }} \
--total-partitions 2 \
--total-partitions ${{ needs.compute-diffusion-partitions.outputs['partition-count-2gpu'] }} \
--partition-plan-json "$PARTITION_PLAN_JSON" \
$CONTINUE_ON_ERROR_FLAG
- name: Upload execution report
if: always()
uses: actions/upload-artifact@v4
with:
name: diffusion-report-2gpu-${{ matrix.part }}
path: python/sglang/multimodal_gen/test/execution_report_*.json
retention-days: 1
- uses: ./.github/actions/upload-cuda-coredumps
if: failure()
with:
@@ -354,3 +416,37 @@ jobs:
run: |
cd python
python3 sglang/multimodal_gen/test/run_suite.py --suite unit
diffusion-coverage-check:
needs: [multimodal-gen-test-1-gpu, multimodal-gen-test-2-gpu]
if: |
always() &&
inputs.multimodal_gen == 'true' &&
(
needs.multimodal-gen-test-1-gpu.result == 'success' ||
needs.multimodal-gen-test-1-gpu.result == 'failure' ||
needs.multimodal-gen-test-2-gpu.result == 'success' ||
needs.multimodal-gen-test-2-gpu.result == 'failure'
)
runs-on: ubuntu-latest
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.pr_head_sha || inputs.git_ref || github.sha }}
- name: Set up Python
uses: actions/setup-python@v5
with:
python-version: '3.10'
- name: Download all execution reports
uses: actions/download-artifact@v4
with:
path: reports/
pattern: diffusion-report-*
merge-multiple: true
- name: Verify coverage
run: |
python scripts/ci/utils/diffusion/verify_diffusion_coverage.py --reports-dir reports/