[CD] Skip 12.9 build on Windows (#165665)

Per title

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165665
Approved by: https://github.com/Camyll, https://github.com/malfet
This commit is contained in:
Huy Do
2025-10-16 19:11:27 +00:00
committed by PyTorch MergeBot
parent a303d6dda9
commit 6dedd34c31
4 changed files with 10 additions and 2168 deletions

View File

@ -241,7 +241,11 @@ def generate_libtorch_matrix(
arches += CUDA_ARCHES
arches += ROCM_ARCHES
elif os == "windows":
arches += CUDA_ARCHES
# TODO (huydhn): Only build CUDA 12.9 for Linux. This logic is to be cleaned up
# in 2.10
windows_cuda_arches = CUDA_ARCHES.copy()
windows_cuda_arches.remove("12.9")
arches += windows_cuda_arches
if libtorch_variants is None:
libtorch_variants = [
"shared-with-deps",
@ -305,7 +309,11 @@ def generate_wheels_matrix(
if os == "linux":
arches += CUDA_ARCHES + ROCM_ARCHES + XPU_ARCHES
elif os == "windows":
arches += CUDA_ARCHES + XPU_ARCHES
# TODO (huydhn): Only build CUDA 12.9 for Linux. This logic is to be cleaned up
# in 2.10
windows_cuda_arches = CUDA_ARCHES.copy()
windows_cuda_arches.remove("12.9")
arches += windows_cuda_arches + XPU_ARCHES
elif os == "linux-aarch64":
# Separate new if as the CPU type is different and
# uses different build/test scripts

View File

@ -788,256 +788,6 @@ jobs:
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
libtorch-cuda12_9-shared-with-deps-debug-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 360
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: libtorch
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu129
GPU_ARCH_VERSION: "12.9"
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
LIBTORCH_CONFIG: debug
LIBTORCH_VARIANT: shared-with-deps
# This is a dummy value for libtorch to work correctly with our batch scripts
# without this value pip does not get installed for some reason
DESIRED_PYTHON: "3.10"
steps:
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: Display EC2 information
shell: bash
run: |
set -euo pipefail
function get_ec2_metadata() {
# Pulled from instance metadata endpoint for EC2
# see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
category=$1
curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
}
echo "ami-id: $(get_ec2_metadata ami-id)"
echo "instance-id: $(get_ec2_metadata instance-id)"
echo "instance-type: $(get_ec2_metadata instance-type)"
echo "system info $(uname -a)"
- name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
uses: pytorch/test-infra/.github/actions/setup-ssh@main
continue-on-error: true
with:
github-secret: ${{ secrets.GITHUB_TOKEN }}
- name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
shell: bash
run: |
git config --global core.longpaths true
git config --global core.symlinks true
# https://git-scm.com/docs/git-fsmonitor--daemon. The daemon could lock
# the directory on Windows and prevent GHA from checking out as reported
# in https://github.com/actions/checkout/issues/1018
git config --global core.fsmonitor false
# Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
- name: Enable long paths on Windows
shell: powershell
run: |
Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
# Since it's just a defensive command, the workflow should continue even the command fails. This step can be
# removed once Windows Defender is removed from the AMI
- name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
continue-on-error: true
shell: powershell
run: |
Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
# Let's both exclude the path and disable Windows Defender completely just to be sure
# that it doesn't interfere
Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
- name: Checkout PyTorch
uses: actions/checkout@v4
with:
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
submodules: recursive
path: pytorch
show-progress: false
- name: Clean PyTorch checkout
run: |
# Remove any artifacts from the previous checkouts
git clean -fxd
working-directory: pytorch
- name: Populate binary env
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
- name: Build PyTorch binary
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: libtorch-cuda12_9-shared-with-deps-debug
retention-days: 14
if-no-files-found: error
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- name: Wait until all sessions have drained
shell: powershell
working-directory: pytorch
if: always()
timeout-minutes: 120
run: |
.github\scripts\wait_for_ssh_to_drain.ps1
- name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
shell: powershell
working-directory: pytorch
if: always()
run: |
.github\scripts\kill_active_ssh_sessions.ps1
libtorch-cuda12_9-shared-with-deps-debug-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- libtorch-cuda12_9-shared-with-deps-debug-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 360
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: libtorch
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu129
GPU_ARCH_VERSION: "12.9"
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
LIBTORCH_CONFIG: debug
LIBTORCH_VARIANT: shared-with-deps
# This is a dummy value for libtorch to work correctly with our batch scripts
# without this value pip does not get installed for some reason
DESIRED_PYTHON: "3.10"
steps:
- name: Display EC2 information
shell: bash
run: |
set -euo pipefail
function get_ec2_metadata() {
# Pulled from instance metadata endpoint for EC2
# see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
category=$1
curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
}
echo "ami-id: $(get_ec2_metadata ami-id)"
echo "instance-id: $(get_ec2_metadata instance-id)"
echo "instance-type: $(get_ec2_metadata instance-type)"
echo "system info $(uname -a)"
- name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
uses: pytorch/test-infra/.github/actions/setup-ssh@main
continue-on-error: true
with:
github-secret: ${{ secrets.GITHUB_TOKEN }}
- name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
shell: bash
run: |
git config --global core.longpaths true
git config --global core.symlinks true
# https://git-scm.com/docs/git-fsmonitor--daemon. The daemon could lock
# the directory on Windows and prevent GHA from checking out as reported
# in https://github.com/actions/checkout/issues/1018
git config --global core.fsmonitor false
# Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
- name: Enable long paths on Windows
shell: powershell
run: |
Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
# Since it's just a defensive command, the workflow should continue even the command fails. This step can be
# removed once Windows Defender is removed from the AMI
- name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
continue-on-error: true
shell: powershell
run: |
Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
# Let's both exclude the path and disable Windows Defender completely just to be sure
# that it doesn't interfere
Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
- name: Checkout PyTorch
uses: actions/checkout@v4
with:
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
submodules: recursive
path: pytorch
show-progress: false
- name: Clean PyTorch checkout
run: |
# Remove any artifacts from the previous checkouts
git clean -fxd
working-directory: pytorch
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- uses: actions/download-artifact@v4.1.7
name: Download Build Artifacts
with:
name: libtorch-cuda12_9-shared-with-deps-debug
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- name: Populate binary env
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
- name: Test PyTorch binary
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_test.sh"
- name: Wait until all sessions have drained
shell: powershell
working-directory: pytorch
if: always()
timeout-minutes: 120
run: |
.github\scripts\wait_for_ssh_to_drain.ps1
- name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
shell: powershell
working-directory: pytorch
if: always()
run: |
.github\scripts\kill_active_ssh_sessions.ps1
libtorch-cuda12_9-shared-with-deps-debug-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: libtorch-cuda12_9-shared-with-deps-debug-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: libtorch
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu129
GPU_ARCH_VERSION: "12.9"
GPU_ARCH_TYPE: cuda
LIBTORCH_CONFIG: debug
LIBTORCH_VARIANT: shared-with-deps
# This is a dummy value for libtorch to work correctly with our batch scripts
# without this value pip does not get installed for some reason
DESIRED_PYTHON: "3.10"
build_name: libtorch-cuda12_9-shared-with-deps-debug
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
libtorch-cuda13_0-shared-with-deps-debug-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type

View File

@ -788,256 +788,6 @@ jobs:
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
libtorch-cuda12_9-shared-with-deps-release-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 360
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: libtorch
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu129
GPU_ARCH_VERSION: "12.9"
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
LIBTORCH_CONFIG: release
LIBTORCH_VARIANT: shared-with-deps
# This is a dummy value for libtorch to work correctly with our batch scripts
# without this value pip does not get installed for some reason
DESIRED_PYTHON: "3.10"
steps:
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: Display EC2 information
shell: bash
run: |
set -euo pipefail
function get_ec2_metadata() {
# Pulled from instance metadata endpoint for EC2
# see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
category=$1
curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
}
echo "ami-id: $(get_ec2_metadata ami-id)"
echo "instance-id: $(get_ec2_metadata instance-id)"
echo "instance-type: $(get_ec2_metadata instance-type)"
echo "system info $(uname -a)"
- name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
uses: pytorch/test-infra/.github/actions/setup-ssh@main
continue-on-error: true
with:
github-secret: ${{ secrets.GITHUB_TOKEN }}
- name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
shell: bash
run: |
git config --global core.longpaths true
git config --global core.symlinks true
# https://git-scm.com/docs/git-fsmonitor--daemon. The daemon could lock
# the directory on Windows and prevent GHA from checking out as reported
# in https://github.com/actions/checkout/issues/1018
git config --global core.fsmonitor false
# Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
- name: Enable long paths on Windows
shell: powershell
run: |
Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
# Since it's just a defensive command, the workflow should continue even the command fails. This step can be
# removed once Windows Defender is removed from the AMI
- name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
continue-on-error: true
shell: powershell
run: |
Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
# Let's both exclude the path and disable Windows Defender completely just to be sure
# that it doesn't interfere
Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
- name: Checkout PyTorch
uses: actions/checkout@v4
with:
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
submodules: recursive
path: pytorch
show-progress: false
- name: Clean PyTorch checkout
run: |
# Remove any artifacts from the previous checkouts
git clean -fxd
working-directory: pytorch
- name: Populate binary env
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
- name: Build PyTorch binary
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: libtorch-cuda12_9-shared-with-deps-release
retention-days: 14
if-no-files-found: error
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- name: Wait until all sessions have drained
shell: powershell
working-directory: pytorch
if: always()
timeout-minutes: 120
run: |
.github\scripts\wait_for_ssh_to_drain.ps1
- name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
shell: powershell
working-directory: pytorch
if: always()
run: |
.github\scripts\kill_active_ssh_sessions.ps1
libtorch-cuda12_9-shared-with-deps-release-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- libtorch-cuda12_9-shared-with-deps-release-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 360
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: libtorch
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu129
GPU_ARCH_VERSION: "12.9"
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
LIBTORCH_CONFIG: release
LIBTORCH_VARIANT: shared-with-deps
# This is a dummy value for libtorch to work correctly with our batch scripts
# without this value pip does not get installed for some reason
DESIRED_PYTHON: "3.10"
steps:
- name: Display EC2 information
shell: bash
run: |
set -euo pipefail
function get_ec2_metadata() {
# Pulled from instance metadata endpoint for EC2
# see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
category=$1
curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
}
echo "ami-id: $(get_ec2_metadata ami-id)"
echo "instance-id: $(get_ec2_metadata instance-id)"
echo "instance-type: $(get_ec2_metadata instance-type)"
echo "system info $(uname -a)"
- name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
uses: pytorch/test-infra/.github/actions/setup-ssh@main
continue-on-error: true
with:
github-secret: ${{ secrets.GITHUB_TOKEN }}
- name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
shell: bash
run: |
git config --global core.longpaths true
git config --global core.symlinks true
# https://git-scm.com/docs/git-fsmonitor--daemon. The daemon could lock
# the directory on Windows and prevent GHA from checking out as reported
# in https://github.com/actions/checkout/issues/1018
git config --global core.fsmonitor false
# Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
- name: Enable long paths on Windows
shell: powershell
run: |
Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
# Since it's just a defensive command, the workflow should continue even the command fails. This step can be
# removed once Windows Defender is removed from the AMI
- name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
continue-on-error: true
shell: powershell
run: |
Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
# Let's both exclude the path and disable Windows Defender completely just to be sure
# that it doesn't interfere
Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
- name: Checkout PyTorch
uses: actions/checkout@v4
with:
ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
submodules: recursive
path: pytorch
show-progress: false
- name: Clean PyTorch checkout
run: |
# Remove any artifacts from the previous checkouts
git clean -fxd
working-directory: pytorch
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- uses: actions/download-artifact@v4.1.7
name: Download Build Artifacts
with:
name: libtorch-cuda12_9-shared-with-deps-release
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- name: Populate binary env
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
- name: Test PyTorch binary
shell: bash
run: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_test.sh"
- name: Wait until all sessions have drained
shell: powershell
working-directory: pytorch
if: always()
timeout-minutes: 120
run: |
.github\scripts\wait_for_ssh_to_drain.ps1
- name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
shell: powershell
working-directory: pytorch
if: always()
run: |
.github\scripts\kill_active_ssh_sessions.ps1
libtorch-cuda12_9-shared-with-deps-release-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: libtorch-cuda12_9-shared-with-deps-release-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: libtorch
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu129
GPU_ARCH_VERSION: "12.9"
GPU_ARCH_TYPE: cuda
LIBTORCH_CONFIG: release
LIBTORCH_VARIANT: shared-with-deps
# This is a dummy value for libtorch to work correctly with our batch scripts
# without this value pip does not get installed for some reason
DESIRED_PYTHON: "3.10"
build_name: libtorch-cuda12_9-shared-with-deps-release
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
libtorch-cuda13_0-shared-with-deps-release-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type

File diff suppressed because it is too large Load Diff