Skip to content

Commit 87c0a30

Browse files
yaoyu-33yaoyu-33
authored
Merge branch 'main' into yuya/unified-perf-recipes
2 parents 9ec6a55 + 7583593 commit 87c0a30

422 files changed

Lines changed: 42070 additions & 2947 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎.github/workflows/cicd-main.yml‎

Lines changed: 57 additions & 35 deletions
Original file line numberDiff line numberDiff line change
@@ -89,12 +89,26 @@ jobs:
8989
expect_l1: ${{ steps.configure.outputs.expect_l1 }}
9090
expect_l2: ${{ steps.configure.outputs.expect_l2 }}
9191
perf_scripts_only: ${{ steps.configure.outputs.perf_scripts_only }}
92+
merge_sha: ${{ steps.merge-sha.outputs.merge_sha }}
9293
steps:
9394
- name: Get PR info
9495
id: get-pr-info
9596
if: startsWith(github.ref, 'refs/heads/pull-request/')
9697
uses: nv-gha-runners/get-pr-info@main
9798

99+
- name: Resolve merge commit sha
100+
id: merge-sha
101+
shell: bash -e -u -o pipefail {0}
102+
env:
103+
IS_PR: ${{ startsWith(github.ref, 'refs/heads/pull-request/') }}
104+
run: |
105+
if [[ "$IS_PR" == "true" ]]; then
106+
SHA=${{ fromJSON(steps.get-pr-info.outputs.pr-info || '{}').merge_commit_sha }}
107+
else
108+
SHA=${GITHUB_SHA}
109+
fi
110+
echo "merge_sha=${SHA}" | tee -a "$GITHUB_OUTPUT"
111+
98112
- name: Configure
99113
id: configure
100114
shell: bash -x -e -u -o pipefail {0}
@@ -220,7 +234,7 @@ jobs:
220234
lint-check:
221235
name: Lint check
222236
runs-on: ubuntu-latest
223-
needs: [pre-flight]
237+
needs: [pre-flight, configure]
224238
if: |
225239
needs.pre-flight.outputs.is_deployment_workflow == 'false'
226240
|| github.event_name == 'workflow_dispatch'
@@ -229,6 +243,7 @@ jobs:
229243
uses: actions/checkout@v6
230244
with:
231245
submodules: "recursive"
246+
ref: ${{ needs.configure.outputs.merge_sha }}
232247

233248
- name: Update MCore submodule (if triggered from MCore)
234249
if: ${{ github.event.inputs.mcore_ref != '' }}
@@ -325,24 +340,11 @@ jobs:
325340
if: startsWith(github.ref, 'refs/heads/pull-request/')
326341
uses: nv-gha-runners/get-pr-info@main
327342

328-
- name: Get merge commit sha
329-
shell: bash -x -e -u -o pipefail {0}
330-
id: sha
331-
env:
332-
IS_PR: ${{ startsWith(github.ref, 'refs/heads/pull-request/') }}
333-
run: |
334-
if [[ "$IS_PR" == "true" ]]; then
335-
SHA=${{ fromJSON(steps.get-pr-info.outputs.pr-info || '{}').merge_commit_sha }}
336-
else
337-
SHA=${GITHUB_SHA}
338-
fi
339-
echo "main=${SHA}" | tee -a "$GITHUB_OUTPUT"
340-
341343
- name: Checkout
342344
uses: actions/checkout@v6
343345
with:
344346
submodules: recursive
345-
ref: ${{ steps.sha.outputs.main }}
347+
ref: ${{ needs.configure.outputs.merge_sha }}
346348

347349
- name: Update MCore submodule (if triggered from MCore)
348350
if: ${{ github.event.inputs.mcore_ref != '' }}
@@ -464,7 +466,7 @@ jobs:
464466
no-cache: false
465467
tags: |
466468
${{ matrix.registry }}/megatron-bridge:${{ steps.cache_keys.outputs.key }}
467-
${{ matrix.registry }}/megatron-bridge:${{ github.sha }}
469+
${{ matrix.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
468470
secrets: |
469471
GH_TOKEN=${{ secrets.PAT }}
470472
@@ -484,11 +486,13 @@ jobs:
484486
steps:
485487
- name: Checkout
486488
uses: actions/checkout@v6
489+
with:
490+
ref: ${{ needs.configure.outputs.merge_sha }}
487491

488492
- name: Run venv import check
489493
shell: bash -e -u -o pipefail {0}
490494
env:
491-
IMAGE: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
495+
IMAGE: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
492496
run: |
493497
docker run --rm \
494498
-v "${{ github.workspace }}/docker/common:/opt/import-check:ro" \
@@ -507,7 +511,7 @@ jobs:
507511
)
508512
&& !cancelled()
509513
&& (github.event.inputs.test_suite == '' || github.event.inputs.test_suite == 'all' || github.event.inputs.test_suite == 'unit-only' || contains('L0 L1 L2', github.event.inputs.test_suite))
510-
needs: [pre-flight, cicd-wait-in-queue, cicd-container-build]
514+
needs: [pre-flight, configure, cicd-wait-in-queue, cicd-container-build]
511515
runs-on: ${{ needs.pre-flight.outputs.runner_prefix }}
512516
name: Launch_Unit_Tests_Core
513517
env:
@@ -518,6 +522,7 @@ jobs:
518522
uses: actions/checkout@v6
519523
with:
520524
submodules: recursive
525+
ref: ${{ needs.configure.outputs.merge_sha }}
521526

522527
- name: main
523528
uses: ./.github/actions/test-template
@@ -526,7 +531,7 @@ jobs:
526531
timeout: 18
527532
is_unit_test: "true"
528533
PAT: ${{ secrets.PAT }}
529-
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
534+
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
530535
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
531536
runner: ${{ needs.pre-flight.outputs.runner_prefix }}
532537

@@ -540,7 +545,7 @@ jobs:
540545
)
541546
&& !cancelled()
542547
&& (github.event.inputs.test_suite == '' || github.event.inputs.test_suite == 'all' || github.event.inputs.test_suite == 'unit-only' || contains('L0 L1 L2', github.event.inputs.test_suite))
543-
needs: [pre-flight, cicd-wait-in-queue, cicd-container-build]
548+
needs: [pre-flight, configure, cicd-wait-in-queue, cicd-container-build]
544549
runs-on: ${{ needs.pre-flight.outputs.runner_prefix }}
545550
name: Launch_Unit_Tests_Diffusion
546551
env:
@@ -551,6 +556,7 @@ jobs:
551556
uses: actions/checkout@v6
552557
with:
553558
submodules: recursive
559+
ref: ${{ needs.configure.outputs.merge_sha }}
554560

555561
- name: main
556562
uses: ./.github/actions/test-template
@@ -559,12 +565,12 @@ jobs:
559565
timeout: 18
560566
is_unit_test: "true"
561567
PAT: ${{ secrets.PAT }}
562-
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
568+
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
563569
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
564570
runner: ${{ needs.pre-flight.outputs.runner_prefix }}
565571

566572
generate-test-matrix:
567-
needs: [pre-flight, cicd-container-build]
573+
needs: [pre-flight, configure, cicd-container-build]
568574
runs-on: ubuntu-latest
569575
outputs:
570576
matrix_l0: ${{ steps.scan.outputs.matrix_l0 }}
@@ -581,6 +587,8 @@ jobs:
581587
&& !cancelled()
582588
steps:
583589
- uses: actions/checkout@v6
590+
with:
591+
ref: ${{ needs.configure.outputs.merge_sha }}
584592
- id: scan
585593
shell: bash
586594
env:
@@ -629,7 +637,7 @@ jobs:
629637
fail-fast: false
630638
max-parallel: 16
631639
matrix: ${{ fromJSON(needs.generate-test-matrix.outputs.matrix_l0) }}
632-
needs: [pre-flight, generate-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
640+
needs: [pre-flight, configure, generate-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
633641
runs-on: ${{ matrix.runner }}
634642
if: |
635643
(
@@ -650,6 +658,7 @@ jobs:
650658
uses: actions/checkout@v6
651659
with:
652660
submodules: recursive
661+
ref: ${{ needs.configure.outputs.merge_sha }}
653662

654663
- name: main
655664
uses: ./.github/actions/test-template
@@ -658,7 +667,7 @@ jobs:
658667
timeout: ${{ fromJSON(matrix.timeout || '30') }}
659668
is_unit_test: "false"
660669
PAT: ${{ secrets.PAT }}
661-
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
670+
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
662671
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
663672
runner: ${{ matrix.runner }}
664673

@@ -690,6 +699,7 @@ jobs:
690699
uses: actions/checkout@v6
691700
with:
692701
submodules: recursive
702+
ref: ${{ needs.configure.outputs.merge_sha }}
693703

694704
- name: main
695705
uses: ./.github/actions/test-template
@@ -698,7 +708,7 @@ jobs:
698708
timeout: ${{ fromJSON(matrix.timeout || '30') }}
699709
is_unit_test: "false"
700710
PAT: ${{ secrets.PAT }}
701-
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
711+
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
702712
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
703713
runner: ${{ matrix.runner }}
704714

@@ -730,6 +740,7 @@ jobs:
730740
uses: actions/checkout@v6
731741
with:
732742
submodules: recursive
743+
ref: ${{ needs.configure.outputs.merge_sha }}
733744

734745
- name: main
735746
uses: ./.github/actions/test-template
@@ -738,7 +749,7 @@ jobs:
738749
timeout: ${{ fromJSON(matrix.timeout || '30') }}
739750
is_unit_test: "false"
740751
PAT: ${{ secrets.PAT }}
741-
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
752+
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
742753
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
743754
runner: ${{ matrix.runner }}
744755

@@ -748,7 +759,7 @@ jobs:
748759
fail-fast: false
749760
max-parallel: 16
750761
matrix: ${{ fromJSON(needs.generate-test-matrix.outputs.matrix_flaky) }}
751-
needs: [pre-flight, generate-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
762+
needs: [pre-flight, configure, generate-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
752763
runs-on: ${{ matrix.runner }}
753764
name: ${{ matrix.script }}
754765
env:
@@ -760,6 +771,7 @@ jobs:
760771
uses: actions/checkout@v6
761772
with:
762773
submodules: recursive
774+
ref: ${{ needs.configure.outputs.merge_sha }}
763775

764776
- name: main
765777
uses: ./.github/actions/test-template
@@ -769,12 +781,12 @@ jobs:
769781
timeout: ${{ fromJSON(matrix.timeout || '30') }}
770782
is_unit_test: "false"
771783
PAT: ${{ secrets.PAT }}
772-
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ github.sha }}
784+
container-image: ${{ needs.pre-flight.outputs.registry }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
773785
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
774786
runner: ${{ matrix.runner }}
775787

776788
generate-gb200-test-matrix:
777-
needs: [pre-flight, cicd-container-build]
789+
needs: [pre-flight, configure, cicd-container-build]
778790
runs-on: ubuntu-latest
779791
outputs:
780792
matrix_gb200_l0: ${{ steps.scan.outputs.matrix_gb200_l0 }}
@@ -792,6 +804,8 @@ jobs:
792804
&& needs.pre-flight.outputs.is_member == 'true'
793805
steps:
794806
- uses: actions/checkout@v6
807+
with:
808+
ref: ${{ needs.configure.outputs.merge_sha }}
795809
- id: scan
796810
shell: bash
797811
run: |
@@ -838,7 +852,7 @@ jobs:
838852
fail-fast: false
839853
max-parallel: 16
840854
matrix: ${{ fromJSON(needs.generate-gb200-test-matrix.outputs.matrix_gb200_l0) }}
841-
needs: [pre-flight, generate-gb200-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
855+
needs: [pre-flight, configure, generate-gb200-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
842856
runs-on: ${{ matrix.runner }}
843857
if: |
844858
(
@@ -861,6 +875,7 @@ jobs:
861875
uses: actions/checkout@v6
862876
with:
863877
submodules: recursive
878+
ref: ${{ needs.configure.outputs.merge_sha }}
864879

865880
- name: main
866881
uses: ./.github/actions/test-template
@@ -874,7 +889,7 @@ jobs:
874889
azure-tenant-id: ${{ secrets.AZURE_TENANT_ID }}
875890
azure-subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}
876891
PAT: ${{ secrets.PAT }}
877-
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ github.sha }}
892+
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
878893
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
879894
runner: ${{ matrix.runner }}
880895

@@ -908,6 +923,7 @@ jobs:
908923
uses: actions/checkout@v6
909924
with:
910925
submodules: recursive
926+
ref: ${{ needs.configure.outputs.merge_sha }}
911927

912928
- name: main
913929
uses: ./.github/actions/test-template
@@ -921,7 +937,7 @@ jobs:
921937
azure-tenant-id: ${{ secrets.AZURE_TENANT_ID }}
922938
azure-subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}
923939
PAT: ${{ secrets.PAT }}
924-
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ github.sha }}
940+
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
925941
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
926942
runner: ${{ matrix.runner }}
927943

@@ -955,6 +971,7 @@ jobs:
955971
uses: actions/checkout@v6
956972
with:
957973
submodules: recursive
974+
ref: ${{ needs.configure.outputs.merge_sha }}
958975

959976
- name: main
960977
uses: ./.github/actions/test-template
@@ -968,7 +985,7 @@ jobs:
968985
azure-tenant-id: ${{ secrets.AZURE_TENANT_ID }}
969986
azure-subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}
970987
PAT: ${{ secrets.PAT }}
971-
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ github.sha }}
988+
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
972989
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
973990
runner: ${{ matrix.runner }}
974991

@@ -978,7 +995,7 @@ jobs:
978995
fail-fast: false
979996
max-parallel: 16
980997
matrix: ${{ fromJSON(needs.generate-gb200-test-matrix.outputs.matrix_gb200_flaky) }}
981-
needs: [pre-flight, generate-gb200-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
998+
needs: [pre-flight, configure, generate-gb200-test-matrix, cicd-unit-tests-core, cicd-unit-tests-diffusion]
982999
runs-on: ${{ matrix.runner }}
9831000
name: gb200_${{ matrix.script }}
9841001
environment: ${{ contains(needs.pre-flight.outputs.registry, 'azure') && 'nemo-ci' || '' }}
@@ -991,6 +1008,7 @@ jobs:
9911008
uses: actions/checkout@v6
9921009
with:
9931010
submodules: recursive
1011+
ref: ${{ needs.configure.outputs.merge_sha }}
9941012

9951013
- name: main
9961014
uses: ./.github/actions/test-template
@@ -1004,7 +1022,7 @@ jobs:
10041022
azure-tenant-id: ${{ secrets.AZURE_TENANT_ID }}
10051023
azure-subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}
10061024
PAT: ${{ secrets.PAT }}
1007-
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ github.sha }}
1025+
container-image: ${{ env.container-registry-gb200 }}/megatron-bridge:${{ needs.configure.outputs.merge_sha }}
10081026
test-data-path: ${{ needs.pre-flight.outputs.test_data_path }}
10091027
runner: ${{ matrix.runner }}
10101028

@@ -1027,6 +1045,8 @@ jobs:
10271045
steps:
10281046
- name: Checkout
10291047
uses: actions/checkout@v6
1048+
with:
1049+
ref: ${{ needs.configure.outputs.merge_sha }}
10301050

10311051
- name: Get workflow result
10321052
id: result
@@ -1126,6 +1146,8 @@ jobs:
11261146

11271147
- name: Checkout
11281148
uses: actions/checkout@v6
1149+
with:
1150+
ref: ${{ needs.configure.outputs.merge_sha }}
11291151

11301152
- name: Download coverage reports of current branch
11311153
uses: actions/download-artifact@v7

0 commit comments

Comments
 (0)