Files
pytorch/.github/workflows/generated-windows-binary-wheel-nightly.yml
atalman 4ece056791 Nccl update to 2.25.1 for cuda 12.4-12.8 (#146073)
Should resolve: https://github.com/pytorch/pytorch/issues/144768
We use one common nccl version for cuda builds 12.4-12.8 : ``NCCL_VERSION=v2.25.1-1``
For CUDA 11.8 we use legacy ``NCCL_VERSION=v2.21.1-1``
We use pinned version of NCCL rather then submodule.
Move nccl location from ``third_party/nccl/nccl`` to ``third_party/nccl``

Pull Request resolved: https://github.com/pytorch/pytorch/pull/146073
Approved by: https://github.com/Skylion007, https://github.com/malfet, https://github.com/kwen2501, https://github.com/fduwjj
2025-02-19 03:52:26 +00:00

7147 lines
346 KiB
YAML
Generated

# @generated DO NOT EDIT MANUALLY
# Template is at: .github/templates/windows_binary_build_workflow.yml.j2
# Generation script: .github/scripts/generate_ci_workflows.py
name: windows-binary-wheel
on:
push:
# NOTE: Meta Employees can trigger new nightlies using: https://fburl.com/trigger_pytorch_nightly_build
branches:
- nightly
tags:
# NOTE: Binary build pipelines should only get triggered on release candidate builds
# Release candidate tags look like: v1.11.0-rc1
- v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
- 'ciflow/binaries/*'
- 'ciflow/binaries_wheel/*'
workflow_dispatch:
env:
# Needed for conda builds
ALPINE_IMAGE: "308535385114.dkr.ecr.us-east-1.amazonaws.com/tool/alpine"
AWS_DEFAULT_REGION: us-east-1
BUILD_ENVIRONMENT: windows-binary-wheel
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
PR_NUMBER: ${{ github.event.pull_request.number }}
SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
SKIP_ALL_TESTS: 1
concurrency:
group: windows-binary-wheel-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true
jobs:
get-label-type:
if: github.repository_owner == 'pytorch'
name: get-label-type
uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
with:
triggering_actor: ${{ github.triggering_actor }}
issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
curr_branch: ${{ github.head_ref || github.ref_name }}
curr_ref_type: ${{ github.ref_type }}
wheel-py3_9-cpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_9-cpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_9-cpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_9-cpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_9-cpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_9-cpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_9-cpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
DESIRED_PYTHON: "3.9"
build_name: wheel-py3_9-cpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_9-cuda11_8-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_9-cuda11_8
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_9-cuda11_8-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_9-cuda11_8-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_9-cuda11_8
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_9-cuda11_8-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_9-cuda11_8-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.9"
build_name: wheel-py3_9-cuda11_8
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_9-cuda12_4-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_9-cuda12_4
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_9-cuda12_4-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_9-cuda12_4-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_9-cuda12_4
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_9-cuda12_4-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_9-cuda12_4-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.9"
build_name: wheel-py3_9-cuda12_4
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_9-cuda12_6-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_9-cuda12_6
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_9-cuda12_6-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_9-cuda12_6-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_9-cuda12_6
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_9-cuda12_6-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_9-cuda12_6-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.9"
build_name: wheel-py3_9-cuda12_6
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_9-xpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: intel-cmplr-lib-rt==2025.0.2 | intel-cmplr-lib-ur==2025.0.2 | intel-cmplr-lic-rt==2025.0.2 | intel-sycl-rt==2025.0.2 | tcmlib==1.2.0 | umf==0.9.1 | intel-pti==0.10.0
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_9-xpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_9-xpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_9-xpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.9"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_9-xpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_9-xpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_9-xpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
DESIRED_PYTHON: "3.9"
build_name: wheel-py3_9-xpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_10-cpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.10"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_10-cpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_10-cpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_10-cpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_10-cpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_10-cpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_10-cpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
DESIRED_PYTHON: "3.10"
build_name: wheel-py3_10-cpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_10-cuda11_8-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.10"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_10-cuda11_8
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_10-cuda11_8-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_10-cuda11_8-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_10-cuda11_8
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_10-cuda11_8-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_10-cuda11_8-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.10"
build_name: wheel-py3_10-cuda11_8
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_10-cuda12_4-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.10"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_10-cuda12_4
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_10-cuda12_4-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_10-cuda12_4-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_10-cuda12_4
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_10-cuda12_4-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_10-cuda12_4-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.10"
build_name: wheel-py3_10-cuda12_4
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_10-cuda12_6-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.10"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_10-cuda12_6
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_10-cuda12_6-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_10-cuda12_6-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_10-cuda12_6
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_10-cuda12_6-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_10-cuda12_6-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.10"
build_name: wheel-py3_10-cuda12_6
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_10-xpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.10"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: intel-cmplr-lib-rt==2025.0.2 | intel-cmplr-lib-ur==2025.0.2 | intel-cmplr-lic-rt==2025.0.2 | intel-sycl-rt==2025.0.2 | tcmlib==1.2.0 | umf==0.9.1 | intel-pti==0.10.0
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_10-xpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_10-xpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_10-xpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_10-xpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_10-xpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_10-xpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
DESIRED_PYTHON: "3.10"
build_name: wheel-py3_10-xpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_11-cpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_11-cpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_11-cpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_11-cpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_11-cpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_11-cpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_11-cpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
DESIRED_PYTHON: "3.11"
build_name: wheel-py3_11-cpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_11-cuda11_8-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_11-cuda11_8
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_11-cuda11_8-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_11-cuda11_8-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_11-cuda11_8
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_11-cuda11_8-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_11-cuda11_8-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.11"
build_name: wheel-py3_11-cuda11_8
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_11-cuda12_4-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_11-cuda12_4
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_11-cuda12_4-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_11-cuda12_4-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_11-cuda12_4
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_11-cuda12_4-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_11-cuda12_4-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.11"
build_name: wheel-py3_11-cuda12_4
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_11-cuda12_6-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_11-cuda12_6
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_11-cuda12_6-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_11-cuda12_6-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_11-cuda12_6
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_11-cuda12_6-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_11-cuda12_6-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.11"
build_name: wheel-py3_11-cuda12_6
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_11-xpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: intel-cmplr-lib-rt==2025.0.2 | intel-cmplr-lib-ur==2025.0.2 | intel-cmplr-lic-rt==2025.0.2 | intel-sycl-rt==2025.0.2 | tcmlib==1.2.0 | umf==0.9.1 | intel-pti==0.10.0
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_11-xpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_11-xpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_11-xpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.11"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_11-xpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_11-xpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_11-xpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
DESIRED_PYTHON: "3.11"
build_name: wheel-py3_11-xpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_12-cpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_12-cpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_12-cpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_12-cpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_12-cpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_12-cpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_12-cpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
DESIRED_PYTHON: "3.12"
build_name: wheel-py3_12-cpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_12-cuda11_8-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_12-cuda11_8
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_12-cuda11_8-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_12-cuda11_8-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_12-cuda11_8
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_12-cuda11_8-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_12-cuda11_8-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.12"
build_name: wheel-py3_12-cuda11_8
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_12-cuda12_4-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_12-cuda12_4
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_12-cuda12_4-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_12-cuda12_4-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_12-cuda12_4
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_12-cuda12_4-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_12-cuda12_4-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.12"
build_name: wheel-py3_12-cuda12_4
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_12-cuda12_6-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_12-cuda12_6
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_12-cuda12_6-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_12-cuda12_6-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_12-cuda12_6
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_12-cuda12_6-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_12-cuda12_6-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.12"
build_name: wheel-py3_12-cuda12_6
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_12-xpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: intel-cmplr-lib-rt==2025.0.2 | intel-cmplr-lib-ur==2025.0.2 | intel-cmplr-lic-rt==2025.0.2 | intel-sycl-rt==2025.0.2 | tcmlib==1.2.0 | umf==0.9.1 | intel-pti==0.10.0
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_12-xpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_12-xpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_12-xpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.12"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_12-xpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_12-xpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_12-xpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
DESIRED_PYTHON: "3.12"
build_name: wheel-py3_12-xpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13-cpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13-cpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13-cpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13-cpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13-cpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13-cpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13-cpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
DESIRED_PYTHON: "3.13"
build_name: wheel-py3_13-cpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13-cuda11_8-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13-cuda11_8
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13-cuda11_8-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13-cuda11_8-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13-cuda11_8
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13-cuda11_8-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13-cuda11_8-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.13"
build_name: wheel-py3_13-cuda11_8
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13-cuda12_4-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13-cuda12_4
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13-cuda12_4-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13-cuda12_4-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13-cuda12_4
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13-cuda12_4-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13-cuda12_4-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.13"
build_name: wheel-py3_13-cuda12_4
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13-cuda12_6-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13-cuda12_6
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13-cuda12_6-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13-cuda12_6-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13-cuda12_6
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13-cuda12_6-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13-cuda12_6-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.13"
build_name: wheel-py3_13-cuda12_6
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13-xpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: intel-cmplr-lib-rt==2025.0.2 | intel-cmplr-lib-ur==2025.0.2 | intel-cmplr-lic-rt==2025.0.2 | intel-sycl-rt==2025.0.2 | tcmlib==1.2.0 | umf==0.9.1 | intel-pti==0.10.0
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13-xpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13-xpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13-xpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13-xpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13-xpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13-xpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
DESIRED_PYTHON: "3.13"
build_name: wheel-py3_13-xpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13t-cpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13t-cpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13t-cpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13t-cpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13t-cpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13t-cpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13t-cpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cpu
GPU_ARCH_TYPE: cpu
DESIRED_PYTHON: "3.13t"
build_name: wheel-py3_13t-cpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13t-cuda11_8-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13t-cuda11_8
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13t-cuda11_8-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13t-cuda11_8-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13t-cuda11_8
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13t-cuda11_8-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13t-cuda11_8-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu118
GPU_ARCH_VERSION: 11.8
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.13t"
build_name: wheel-py3_13t-cuda11_8
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13t-cuda12_4-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13t-cuda12_4
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13t-cuda12_4-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13t-cuda12_4-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13t-cuda12_4
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13t-cuda12_4-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13t-cuda12_4-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu124
GPU_ARCH_VERSION: 12.4
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.13t"
build_name: wheel-py3_13t-cuda12_4
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13t-cuda12_6-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.4.5.8; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.2.1.3; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.5.147; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.6.1.9; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.3.1.170; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.6.2; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.25.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.4.127; platform_system == 'Linux' and platform_machine == 'x86_64'
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13t-cuda12_6
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13t-cuda12_6-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13t-cuda12_6-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13t-cuda12_6
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13t-cuda12_6-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13t-cuda12_6-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: cu126
GPU_ARCH_VERSION: 12.6
GPU_ARCH_TYPE: cuda
DESIRED_PYTHON: "3.13t"
build_name: wheel-py3_13t-cuda12_6
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml
wheel-py3_13t-xpu-build:
if: ${{ github.repository_owner == 'pytorch' }}
needs: get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
PYTORCH_EXTRA_INSTALL_REQUIREMENTS: intel-cmplr-lib-rt==2025.0.2 | intel-cmplr-lib-ur==2025.0.2 | intel-cmplr-lic-rt==2025.0.2 | intel-sycl-rt==2025.0.2 | tcmlib==1.2.0 | umf==0.9.1 | intel-pti==0.10.0
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
- name: 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: |
"${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
- uses: actions/upload-artifact@v4.4.0
if: always()
with:
name: wheel-py3_13t-xpu
retention-days: 14
if-no-files-found: error
path: "${{ 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
wheel-py3_13t-xpu-test: # Testing
if: ${{ github.repository_owner == 'pytorch' }}
needs:
- wheel-py3_13t-xpu-build
- get-label-type
runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge.nonephemeral"
timeout-minutes: 300
env:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
SKIP_ALL_TESTS: 1
DESIRED_PYTHON: "3.13t"
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 $True -ErrorAction Ignore
# NOTE: These environment variables are put here so that they can be applied on every job equally
# They are also here because setting them at a workflow level doesn't give us access to the
# runner.temp variable, which we need.
- name: Populate binary env
shell: bash
run: |
echo "BINARY_ENV_FILE=${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: wheel-py3_13t-xpu
path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
- 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: 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
wheel-py3_13t-xpu-upload: # Uploading
if: ${{ github.repository_owner == 'pytorch' }}
permissions:
id-token: write
contents: read
needs: wheel-py3_13t-xpu-test
with:
PYTORCH_ROOT: ${{ github.workspace }}/pytorch
PACKAGE_TYPE: wheel
# TODO: This is a legacy variable that we eventually want to get rid of in
# favor of GPU_ARCH_VERSION
DESIRED_CUDA: xpu
GPU_ARCH_TYPE: xpu
DESIRED_PYTHON: "3.13t"
build_name: wheel-py3_13t-xpu
secrets:
github-token: ${{ secrets.GITHUB_TOKEN }}
uses: ./.github/workflows/_binary-upload.yml