[ROCm] Include hsa headers for rocm-triton whl (#129235)

* Include hsa headers for rocm-triton whl

* Update triton pin to release/3.0.x tip

* Update .ci/docker/ci_commit_pins/triton-rocm.txt

---------

Co-authored-by: Andrey Talman <atalman@fb.com>

.ci/docker/ci_commit_pins/triton-rocm.txt

@@ -1 +1 @@
-01cbe5045a6898c9a925f01435c8277b2fe6afcc
+21eae954efa5bf584da70324b640288c3ee7aede

.ci/docker/ci_commit_pins/triton-xpu.txt

@@ -1 +1 @@
-b8c64f64c18d8cac598b3adb355c21e7439c21de
+aac14a3b93f11d781d1d5ebc5400b15ae8df5185

.ci/pytorch/common_utils.sh

@@ -178,7 +178,7 @@ function install_torchrec_and_fbgemm() {
 
 function clone_pytorch_xla() {
   if [[ ! -d ./xla ]]; then
-    git clone --recursive --quiet https://github.com/pytorch/xla.git
+    git clone --recursive -b r2.4 https://github.com/pytorch/xla.git
     pushd xla
     # pin the xla hash so that we don't get broken by changes to xla
     git checkout "$(cat ../.github/ci_commit_pins/xla.txt)"

.circleci/scripts/binary_populate_env.sh

@@ -75,10 +75,9 @@ export PYTORCH_BUILD_NUMBER=1
 TRITON_VERSION=$(cat $PYTORCH_ROOT/.ci/docker/triton_version.txt)
 
 # Here PYTORCH_EXTRA_INSTALL_REQUIREMENTS is already set for the all the wheel builds hence append TRITON_CONSTRAINT
+TRITON_CONSTRAINT="platform_system == 'Linux' and platform_machine == 'x86_64' and python_version < '3.13'"
 if [[ "$PACKAGE_TYPE" =~ .*wheel.* &&  -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]]; then
-  # Only linux Python < 3.13 are supported wheels for triton
+  TRITON_REQUIREMENT="pytorch-triton==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"
-  TRITON_CONSTRAINT="platform_system == 'Linux' and platform_machine == 'x86_64' and python_version < '3.13'"
-  TRITON_REQUIREMENT="triton==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"
   if [[ -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*dev.* ]]; then
       TRITON_SHORTHASH=$(cut -c1-10 $PYTORCH_ROOT/.ci/docker/ci_commit_pins/triton.txt)
       TRITON_REQUIREMENT="pytorch-triton==${TRITON_VERSION}+${TRITON_SHORTHASH}; ${TRITON_CONSTRAINT}"
@@ -87,11 +86,11 @@ if [[ "$PACKAGE_TYPE" =~ .*wheel.* &&  -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:
 fi
 
 # Set triton via PYTORCH_EXTRA_INSTALL_REQUIREMENTS for triton rocm package
-if [[ "$PACKAGE_TYPE" =~ .*wheel.* && -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*rocm.* && $(uname) == "Linux" && "$DESIRED_PYTHON" != "3.12" ]]; then
+if [[ "$PACKAGE_TYPE" =~ .*wheel.* && -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*rocm.* && $(uname) == "Linux" ]]; then
-    TRITON_REQUIREMENT="pytorch-triton-rocm==${TRITON_VERSION}"
+    TRITON_REQUIREMENT="pytorch-triton-rocm==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"
     if [[ -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*dev.* ]]; then
         TRITON_SHORTHASH=$(cut -c1-10 $PYTORCH_ROOT/.ci/docker/ci_commit_pins/triton-rocm.txt)
-        TRITON_REQUIREMENT="pytorch-triton-rocm==${TRITON_VERSION}+${TRITON_SHORTHASH}"
+        TRITON_REQUIREMENT="pytorch-triton-rocm==${TRITON_VERSION}+${TRITON_SHORTHASH}; ${TRITON_CONSTRAINT}"
     fi
     if [[ -z "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]]; then
         export PYTORCH_EXTRA_INSTALL_REQUIREMENTS="${TRITON_REQUIREMENT}"

.github/ci_commit_pins/torchbench.txt

@@ -1 +1 @@
-d6015d42d9a1834bc7595c4bd6852562fb80b30b
+23512dbebd44a11eb84afbf53c3c071dd105297e

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-6f0b61e5d782913a0fc7743812f2a8e522189111
+r2.4

.github/scripts/amd/package_triton_wheel.sh

@@ -93,6 +93,8 @@ done
 
 # Copy Include Files
 cp -r $ROCM_HOME/include/hip $TRITON_ROCM_DIR/include
+cp -r $ROCM_HOME/include/roctracer $TRITON_ROCM_DIR/include
+cp -r $ROCM_HOME/include/hsa $TRITON_ROCM_DIR/include
 
 # Copy linker
 mkdir -p $TRITON_ROCM_DIR/llvm/bin

.github/scripts/filter_test_configs.py

@@ -38,9 +38,9 @@ SUPPORTED_PERIODICAL_MODES: Dict[str, Callable[[Optional[str]], bool]] = {
 }
 
 # The link to the published list of disabled jobs
-DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json"
+DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json?versionId=tIl0Qo224T_NDVw0dtG4hU1cZJM97inV"
 # and unstable jobs
-UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json"
+UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json?versionId=GPyRZRsOo26Gfk_WjAoNNxEMGXkIxIes"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"

.github/scripts/generate_binary_build_matrix.py

@@ -347,10 +347,6 @@ def generate_wheels_matrix(
     for python_version in python_versions:
         for arch_version in arches:
             gpu_arch_type = arch_type(arch_version)
-            # Disable py3.12 builds for ROCm because of triton dependency
-            # on llnl-hatchet, which doesn't have py3.12 wheels available
-            if gpu_arch_type == "rocm" and python_version == "3.12":
-                continue
             gpu_arch_version = (
                 ""
                 if arch_version == "cpu"

.github/templates/common.yml.j2

@@ -8,7 +8,7 @@
 # NOTE: If testing pytorch/builder changes you can change this variable to change what pytorch/builder reference
 #       the binary builds will check out
 {%- set builder_repo = "pytorch/builder" -%}
-{%- set builder_branch = "main" -%}
+{%- set builder_branch = "release/2.4" -%}
 
 {%- macro concurrency(build_environment) -%}
 concurrency:

.github/templates/linux_binary_build_workflow.yml.j2

@@ -113,8 +113,8 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: ROCm set GPU_FLAG
         run: |
           echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"

.github/templates/macos_binary_build_workflow.yml.j2

@@ -81,8 +81,8 @@ jobs:
           elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
             echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
           fi
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: Install sccache (only for non-forked PRs, and pushes to trunk)
         uses: nick-fields/retry@v2.8.2
         if: ${{ github.event_name == 'push' || github.event.pull_request.head.repo.full_name == github.repository }}

.github/templates/windows_binary_build_workflow.yml.j2

@@ -65,8 +65,8 @@ jobs:
     steps:
       !{{ common.setup_ec2_windows() }}
       !{{ set_runner_specific_vars() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: Populate binary env
         shell: bash
         run: |
@@ -105,8 +105,8 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: Populate binary env
         shell: bash
         run: |

.github/workflows/_android-build-test.yml

@@ -37,7 +37,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -59,25 +59,25 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -141,5 +141,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_android-full-build-test.yml

@@ -37,7 +37,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -59,25 +59,25 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -186,5 +186,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_bazel-build-test.yml

@@ -42,7 +42,7 @@ jobs:
       reenabled-issues: ${{ steps.filter.outputs.reenabled-issues }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -64,25 +64,25 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -92,7 +92,7 @@ jobs:
         run: echo "IN_ARC_RUNNER=$([ -f /.inarc ] && echo true || echo false)" >> "$GITHUB_OUTPUT"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.4
         if: ${{ inputs.cuda-version != 'cpu' && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
       - name: Output disk space left
@@ -201,5 +201,5 @@ jobs:
           file-suffix: bazel-${{ github.job }}_${{ steps.get-job-id.outputs.job-id }}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_binary-build-linux.yml

@@ -145,13 +145,13 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         if: inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -181,7 +181,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -195,7 +194,7 @@ jobs:
       - name: Checkout pytorch/builder to builder dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -221,7 +220,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ inputs.DOCKER_IMAGE }}
 
@@ -278,7 +277,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

.github/workflows/_binary-test-linux.yml

@@ -128,14 +128,14 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         if: inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
         # Setup the environment
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -158,7 +158,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
 
@@ -171,7 +170,7 @@ jobs:
       - name: Checkout pytorch/builder to builder dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -202,12 +201,12 @@ jobs:
           path: "${{ runner.temp }}/artifacts/"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.4
         if: ${{ inputs.GPU_ARCH_TYPE == 'cuda' && steps.filter.outputs.is-test-matrix-empty == 'False' }}
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ inputs.DOCKER_IMAGE }}
 
@@ -217,7 +216,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

.github/workflows/_binary-upload.yml

@@ -95,7 +95,7 @@ jobs:
       SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 

.github/workflows/_buck-build-test.yml

@@ -23,7 +23,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -44,7 +44,7 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Set up JDK 8
         uses: actions/setup-java@v3
@@ -53,7 +53,7 @@ jobs:
           distribution: 'temurin'
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: 3.8
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}

.github/workflows/_docs.yml

@@ -80,7 +80,7 @@ jobs:
     name: build-docs-${{ matrix.docs_type }}-${{ inputs.push }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -91,7 +91,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -106,12 +106,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -218,5 +218,5 @@ jobs:
           s3-prefix: pytorch/pytorch/${{ github.event.pull_request.number }}/functorchdocs
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_ios-build-test.yml

@@ -46,7 +46,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -80,7 +80,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Populate CI build options
         shell: bash
@@ -102,7 +102,7 @@ jobs:
             brew install libtool
 
       - name: Setup miniconda for iOS
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: "3.9"
           environment-file: .github/requirements/conda-env-iOS.txt

.github/workflows/_linux-build-label.yml

@@ -81,7 +81,7 @@ jobs:
       test-matrix: ${{ steps.linux-build.outputs.test-matrix }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
@@ -90,7 +90,7 @@ jobs:
       # checkout because when we run this action we don't *have* a local
       # checkout. In other cases you should prefer a local checkout.
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Linux Build
         id: linux-build

.github/workflows/_linux-build-rg.yml

@@ -86,7 +86,7 @@ jobs:
       # checkout because when we run this action we don't *have* a local
       # checkout. In other cases you should prefer a local checkout.
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Linux Build
         id: linux-build

.github/workflows/_linux-build.yml

@@ -90,7 +90,7 @@ jobs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
@@ -99,7 +99,7 @@ jobs:
       # checkout because when we run this action we don't *have* a local
       # checkout. In other cases you should prefer a local checkout.
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -114,7 +114,7 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
@@ -128,7 +128,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${tag/:/-}"
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -238,5 +238,5 @@ jobs:
           s3-bucket: ${{ inputs.s3-bucket }}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_linux-test-label.yml

@@ -67,7 +67,7 @@ jobs:
     timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Linux Test
         id: linux-test

.github/workflows/_linux-test-rg.yml

@@ -68,7 +68,7 @@ jobs:
     timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Linux Test
         id: linux-test

.github/workflows/_linux-test.yml

@@ -67,7 +67,7 @@ jobs:
     timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         if: ${{ !contains(matrix.runner, 'gcp.a100') }}
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -76,7 +76,7 @@ jobs:
               docker exec -it $(docker container ps --format '{{.ID}}') bash
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -91,7 +91,7 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
@@ -105,7 +105,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${tag/:/-}"
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -116,7 +116,7 @@ jobs:
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
         id: install-nvidia-driver
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.4
         if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
       - name: Lock NVIDIA A100 40GB Frequency
@@ -333,7 +333,7 @@ jobs:
           path: ./**/core.[1-9]*
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()
 
       # NB: We are currently having an intermittent GPU-related issue on G5 runners with

.github/workflows/_mac-build.yml

@@ -71,11 +71,11 @@ jobs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
     steps:
       - name: Clean up disk space before running MacOS workflow
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Set xcode version
         env:
@@ -87,7 +87,7 @@ jobs:
 
       - name: Setup miniconda
         if: inputs.environment-file == ''
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -97,7 +97,7 @@ jobs:
       # environment even though the arch is x86-64
       - name: Setup miniconda using the provided environment file
         if: inputs.environment-file != ''
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: ${{ inputs.environment-file }}
@@ -207,4 +207,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4

.github/workflows/_mac-test-mps.yml

@@ -40,7 +40,7 @@ jobs:
       reenabled-issues: ${{ steps.filter.outputs.reenabled-issues }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
 
@@ -81,7 +81,7 @@ jobs:
           use-gha: true
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -159,4 +159,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4

.github/workflows/_mac-test.yml

@@ -74,11 +74,11 @@ jobs:
           done
 
       - name: Clean up disk space before running MacOS workflow
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Download build artifacts
         uses: ./.github/actions/download-build-artifacts
@@ -93,7 +93,7 @@ jobs:
           use-gha: true
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -216,4 +216,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4

.github/workflows/_rocm-test.yml

@@ -58,7 +58,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 
@@ -80,12 +80,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 

.github/workflows/_run_android_tests.yml

@@ -23,7 +23,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -54,10 +54,10 @@ jobs:
       SUPPORT_ABI: '${{ matrix.support_abi }}'
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: 3.8
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}.txt

.github/workflows/_runner-determinator.yml

@@ -32,7 +32,7 @@ jobs:
       USERNAME: ${{ inputs.user_name }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: true

.github/workflows/_win-build.yml

@@ -60,10 +60,10 @@ jobs:
           git config --global core.fsmonitor false
 
       - name: Clean up leftover processes on non-ephemeral Windows runner
-        uses: pytorch/test-infra/.github/actions/cleanup-runner@main
+        uses: pytorch/test-infra/.github/actions/cleanup-runner@release/2.4
 
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -78,7 +78,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 

.github/workflows/_win-test.yml

@@ -54,10 +54,10 @@ jobs:
           git config --global core.fsmonitor false
 
       - name: Clean up leftover processes on non-ephemeral Windows runner
-        uses: pytorch/test-infra/.github/actions/cleanup-runner@main
+        uses: pytorch/test-infra/.github/actions/cleanup-runner@release/2.4
 
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -73,7 +73,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 

.github/workflows/_xpu-test.yml

@@ -54,7 +54,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup XPU
         uses: ./.github/actions/setup-xpu
@@ -72,12 +72,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 

.github/workflows/build-triton-wheel.yml

@@ -3,7 +3,7 @@ name: Build Triton wheels
 on:
   push:
     branches:
-      - main
+      - release/2.4
     tags:
       # NOTE: Binary build pipelines should only get triggered on release candidate builds
       # Release candidate tags look like: v1.11.0-rc1
@@ -47,12 +47,12 @@ jobs:
       BUILD_DEVICE: ${{ matrix.device }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
 
@@ -60,7 +60,7 @@ jobs:
         uses: ./.github/actions/setup-linux
 
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
@@ -124,7 +124,7 @@ jobs:
           path: ${{ runner.temp }}/artifacts/*
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()
 
   upload-wheel:
@@ -209,12 +209,12 @@ jobs:
       PY_VERS: ${{ matrix.py_vers }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
 
@@ -222,7 +222,7 @@ jobs:
         uses: ./.github/actions/setup-linux
 
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
@@ -257,7 +257,7 @@ jobs:
           path: ${{ runner.temp }}/artifacts/*
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()
 
   upload-conda:

.github/workflows/check-labels.yml

@@ -31,7 +31,7 @@ jobs:
     runs-on: linux.20_04.4x
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
           fetch-depth: 1

.github/workflows/close-nonexistent-disable-issues.yml

@@ -11,7 +11,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Run close_nonexistent_disable_issues.py
         env:

.github/workflows/docker-builds.yml

@@ -78,21 +78,21 @@ jobs:
       # [see note: pytorch repo ref]
       # deep clone (fetch-depth 0) required for git merge-base
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
 
       - name: Build docker image
         id: build-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: ${{ matrix.docker-image-name }}
           always-rebuild: true
           push: true
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.build-docker-image.outputs.docker-image }}
 
@@ -124,5 +124,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/docker-release.yml

@@ -41,7 +41,7 @@ jobs:
       matrix: ${{ steps.generate-matrix.outputs.matrix }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: true
@@ -69,7 +69,7 @@ jobs:
       CUDNN_VERSION: ${{ matrix.cudnn_version }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
       # [see note: pytorch repo ref]
@@ -147,12 +147,12 @@ jobs:
           fi
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()
 
   validate:
     needs: build
-    uses: pytorch/builder/.github/workflows/validate-docker-images.yml@main
+    uses: pytorch/builder/.github/workflows/validate-docker-images.yml@release/2.4
     with:
       channel: nightly
       ref: main

.github/workflows/generated-linux-aarch64-binary-manywheel-nightly.yml

@@ -48,7 +48,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.8"
       runs_on: linux.arm64.m7g.4xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -69,7 +69,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -91,7 +91,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-aarch64
     secrets:
@@ -111,7 +111,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       runs_on: linux.arm64.m7g.4xlarge
@@ -135,7 +135,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda-aarch64
@@ -156,7 +156,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.9"
       runs_on: linux.arm64.m7g.4xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -177,7 +177,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -199,7 +199,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
     secrets:
@@ -219,7 +219,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       runs_on: linux.arm64.m7g.4xlarge
@@ -243,7 +243,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda-aarch64
@@ -264,7 +264,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.10"
       runs_on: linux.arm64.m7g.4xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -285,7 +285,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -307,7 +307,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
     secrets:
@@ -327,7 +327,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       runs_on: linux.arm64.m7g.4xlarge
@@ -351,7 +351,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda-aarch64
@@ -372,7 +372,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.11"
       runs_on: linux.arm64.m7g.4xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -393,7 +393,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -415,7 +415,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
     secrets:
@@ -435,7 +435,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       runs_on: linux.arm64.m7g.4xlarge
@@ -459,7 +459,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda-aarch64
@@ -480,7 +480,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.12"
       runs_on: linux.arm64.m7g.4xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -501,7 +501,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -523,7 +523,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.4
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-aarch64
     secrets:
@@ -543,7 +543,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.12"
       runs_on: linux.arm64.m7g.4xlarge
@@ -567,7 +567,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cu124
       GPU_ARCH_TYPE: cuda-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cuda12.4-2.4
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda-aarch64

.github/workflows/generated-linux-binary-conda-nightly.yml

@@ -48,7 +48,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       build_environment: linux-binary-conda
@@ -66,7 +66,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       build_environment: linux-binary-conda
@@ -87,7 +87,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
     secrets:
@@ -108,7 +108,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.8"
       runs_on: linux.24xlarge
       build_name: conda-py3_8-cuda11_8
@@ -128,7 +128,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda11_8
       build_environment: linux-binary-conda
@@ -150,7 +150,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda11_8
     secrets:
@@ -171,7 +171,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.8"
       runs_on: linux.24xlarge
       build_name: conda-py3_8-cuda12_1
@@ -191,7 +191,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_1
       build_environment: linux-binary-conda
@@ -213,7 +213,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_1
     secrets:
@@ -234,7 +234,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.8"
       runs_on: linux.24xlarge
       build_name: conda-py3_8-cuda12_4
@@ -254,7 +254,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_4
       build_environment: linux-binary-conda
@@ -276,7 +276,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_4
     secrets:
@@ -296,7 +296,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       build_environment: linux-binary-conda
@@ -314,7 +314,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       build_environment: linux-binary-conda
@@ -335,7 +335,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
     secrets:
@@ -356,7 +356,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.9"
       runs_on: linux.24xlarge
       build_name: conda-py3_9-cuda11_8
@@ -376,7 +376,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
       build_environment: linux-binary-conda
@@ -398,7 +398,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
     secrets:
@@ -419,7 +419,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.9"
       runs_on: linux.24xlarge
       build_name: conda-py3_9-cuda12_1
@@ -439,7 +439,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
       build_environment: linux-binary-conda
@@ -461,7 +461,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
     secrets:
@@ -482,7 +482,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.9"
       runs_on: linux.24xlarge
       build_name: conda-py3_9-cuda12_4
@@ -502,7 +502,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_4
       build_environment: linux-binary-conda
@@ -524,7 +524,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_4
     secrets:
@@ -544,7 +544,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       build_environment: linux-binary-conda
@@ -562,7 +562,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       build_environment: linux-binary-conda
@@ -583,7 +583,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
     secrets:
@@ -604,7 +604,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.10"
       runs_on: linux.24xlarge
       build_name: conda-py3_10-cuda11_8
@@ -624,7 +624,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
       build_environment: linux-binary-conda
@@ -646,7 +646,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
     secrets:
@@ -667,7 +667,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.10"
       runs_on: linux.24xlarge
       build_name: conda-py3_10-cuda12_1
@@ -687,7 +687,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
       build_environment: linux-binary-conda
@@ -709,7 +709,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
     secrets:
@@ -730,7 +730,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.10"
       runs_on: linux.24xlarge
       build_name: conda-py3_10-cuda12_4
@@ -750,7 +750,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_4
       build_environment: linux-binary-conda
@@ -772,7 +772,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_4
     secrets:
@@ -792,7 +792,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       build_environment: linux-binary-conda
@@ -810,7 +810,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       build_environment: linux-binary-conda
@@ -831,7 +831,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
     secrets:
@@ -852,7 +852,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.11"
       runs_on: linux.24xlarge
       build_name: conda-py3_11-cuda11_8
@@ -872,7 +872,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
       build_environment: linux-binary-conda
@@ -894,7 +894,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
     secrets:
@@ -915,7 +915,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.11"
       runs_on: linux.24xlarge
       build_name: conda-py3_11-cuda12_1
@@ -935,7 +935,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
       build_environment: linux-binary-conda
@@ -957,7 +957,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
     secrets:
@@ -978,7 +978,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.11"
       runs_on: linux.24xlarge
       build_name: conda-py3_11-cuda12_4
@@ -998,7 +998,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_4
       build_environment: linux-binary-conda
@@ -1020,7 +1020,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_4
     secrets:
@@ -1040,7 +1040,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
       build_environment: linux-binary-conda
@@ -1058,7 +1058,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
       build_environment: linux-binary-conda
@@ -1079,7 +1079,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
     secrets:
@@ -1100,7 +1100,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.12"
       runs_on: linux.24xlarge
       build_name: conda-py3_12-cuda11_8
@@ -1120,7 +1120,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda11_8
       build_environment: linux-binary-conda
@@ -1142,7 +1142,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda11_8
     secrets:
@@ -1163,7 +1163,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.12"
       runs_on: linux.24xlarge
       build_name: conda-py3_12-cuda12_1
@@ -1183,7 +1183,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda12_1
       build_environment: linux-binary-conda
@@ -1205,7 +1205,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda12_1
     secrets:
@@ -1226,7 +1226,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.12"
       runs_on: linux.24xlarge
       build_name: conda-py3_12-cuda12_4
@@ -1246,7 +1246,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda12_4
       build_environment: linux-binary-conda
@@ -1268,7 +1268,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda12_4
     secrets:

.github/workflows/generated-linux-binary-libtorch-cxx11-abi-main.yml

@@ -43,7 +43,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -62,7 +62,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi

.github/workflows/generated-linux-binary-libtorch-cxx11-abi-nightly.yml

@@ -48,7 +48,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -67,7 +67,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -89,7 +89,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -111,7 +111,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-with-deps-cxx11-abi
@@ -131,7 +131,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-with-deps-cxx11-abi
@@ -154,7 +154,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-with-deps-cxx11-abi
@@ -176,7 +176,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-with-deps-cxx11-abi
@@ -196,7 +196,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-with-deps-cxx11-abi
@@ -219,7 +219,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-with-deps-cxx11-abi
@@ -241,7 +241,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.4-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_4-shared-with-deps-cxx11-abi
@@ -261,7 +261,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.4-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_4-shared-with-deps-cxx11-abi
@@ -284,7 +284,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.4-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_4-shared-with-deps-cxx11-abi
@@ -306,7 +306,7 @@ jobs:
       DESIRED_CUDA: rocm6.0
       GPU_ARCH_VERSION: 6.0
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.0-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.0-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm6_0-shared-with-deps-cxx11-abi
@@ -328,7 +328,7 @@ jobs:
       GPU_ARCH_VERSION: 6.0
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.0-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.0-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -342,7 +342,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -354,7 +353,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -370,7 +369,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm6.0-main
+          docker-image: pytorch/libtorch-cxx11-builder:rocm6.0-2.4
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -390,7 +389,7 @@ jobs:
       DESIRED_CUDA: rocm6.0
       GPU_ARCH_VERSION: 6.0
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.0-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.0-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm6_0-shared-with-deps-cxx11-abi
@@ -412,7 +411,7 @@ jobs:
       DESIRED_CUDA: rocm6.1
       GPU_ARCH_VERSION: 6.1
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.1-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm6_1-shared-with-deps-cxx11-abi
@@ -434,7 +433,7 @@ jobs:
       GPU_ARCH_VERSION: 6.1
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.1-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -448,7 +447,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -460,7 +458,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -476,7 +474,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm6.1-main
+          docker-image: pytorch/libtorch-cxx11-builder:rocm6.1-2.4
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -496,7 +494,7 @@ jobs:
       DESIRED_CUDA: rocm6.1
       GPU_ARCH_VERSION: 6.1
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.1-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm6.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm6_1-shared-with-deps-cxx11-abi

.github/workflows/generated-linux-binary-libtorch-pre-cxx11-main.yml

@@ -43,7 +43,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -62,7 +62,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11

.github/workflows/generated-linux-binary-libtorch-pre-cxx11-nightly.yml

@@ -48,7 +48,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -67,7 +67,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -89,7 +89,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -111,7 +111,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-with-deps-pre-cxx11
@@ -131,7 +131,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-with-deps-pre-cxx11
@@ -154,7 +154,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-with-deps-pre-cxx11
@@ -176,7 +176,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-with-deps-pre-cxx11
@@ -196,7 +196,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-with-deps-pre-cxx11
@@ -219,7 +219,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-with-deps-pre-cxx11
@@ -241,7 +241,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_4-shared-with-deps-pre-cxx11
@@ -261,7 +261,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_4-shared-with-deps-pre-cxx11
@@ -284,7 +284,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_4-shared-with-deps-pre-cxx11
@@ -306,7 +306,7 @@ jobs:
       DESIRED_CUDA: rocm6.0
       GPU_ARCH_VERSION: 6.0
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.0-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.0-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm6_0-shared-with-deps-pre-cxx11
@@ -328,7 +328,7 @@ jobs:
       GPU_ARCH_VERSION: 6.0
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.0-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.0-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -342,7 +342,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -354,7 +353,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -370,7 +369,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm6.0-main
+          docker-image: pytorch/manylinux-builder:rocm6.0-2.4
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -390,7 +389,7 @@ jobs:
       DESIRED_CUDA: rocm6.0
       GPU_ARCH_VERSION: 6.0
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.0-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.0-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm6_0-shared-with-deps-pre-cxx11
@@ -412,7 +411,7 @@ jobs:
       DESIRED_CUDA: rocm6.1
       GPU_ARCH_VERSION: 6.1
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm6_1-shared-with-deps-pre-cxx11
@@ -434,7 +433,7 @@ jobs:
       GPU_ARCH_VERSION: 6.1
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -448,7 +447,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -460,7 +458,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -476,7 +474,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm6.1-main
+          docker-image: pytorch/manylinux-builder:rocm6.1-2.4
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -496,7 +494,7 @@ jobs:
       DESIRED_CUDA: rocm6.1
       GPU_ARCH_VERSION: 6.1
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm6.1-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm6_1-shared-with-deps-pre-cxx11

.github/workflows/generated-linux-binary-manywheel-main.yml

@@ -44,7 +44,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -64,7 +64,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -84,7 +84,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel
@@ -104,7 +104,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel
@@ -124,7 +124,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_4
       build_environment: linux-binary-manywheel
@@ -144,7 +144,7 @@ jobs:
       DESIRED_CUDA: cu124
       GPU_ARCH_VERSION: 12.4
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.4-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_4
       build_environment: linux-binary-manywheel

.github/workflows/generated-linux-s390x-binary-manywheel-nightly.yml

@@ -48,7 +48,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.8"
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
@@ -69,7 +69,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-s390x
       build_environment: linux-s390x-binary-manywheel
@@ -91,7 +91,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-s390x
     secrets:
@@ -111,7 +111,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.9"
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
@@ -132,7 +132,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-s390x
       build_environment: linux-s390x-binary-manywheel
@@ -154,7 +154,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-s390x
     secrets:
@@ -174,7 +174,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.10"
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
@@ -195,7 +195,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-s390x
       build_environment: linux-s390x-binary-manywheel
@@ -217,7 +217,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-s390x
     secrets:
@@ -237,7 +237,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.11"
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
@@ -258,7 +258,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-s390x
       build_environment: linux-s390x-binary-manywheel
@@ -280,7 +280,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-s390x
     secrets:
@@ -300,7 +300,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.12"
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
@@ -321,7 +321,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-s390x
       build_environment: linux-s390x-binary-manywheel
@@ -343,7 +343,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-s390x
-      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-main
+      DOCKER_IMAGE: pytorch/manylinuxs390x-builder:cpu-s390x-2.4
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-s390x
     secrets:

.github/workflows/generated-macos-arm64-binary-conda-nightly.yml

@@ -77,7 +77,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -89,7 +88,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -141,7 +140,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       use_s3: False
@@ -195,7 +194,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -207,7 +205,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -259,7 +257,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       use_s3: False
@@ -313,7 +311,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -325,7 +322,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -377,7 +374,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       use_s3: False
@@ -431,7 +428,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -443,7 +439,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -495,7 +491,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       use_s3: False
@@ -549,7 +545,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -561,7 +556,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -613,7 +608,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu-main
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.4
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
       use_s3: False

.github/workflows/generated-macos-arm64-binary-libtorch-cxx11-abi-nightly.yml

@@ -81,7 +81,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -93,7 +92,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -145,7 +144,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.4
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi

.github/workflows/generated-macos-arm64-binary-wheel-nightly.yml

@@ -78,7 +78,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -90,7 +89,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -142,7 +141,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       DESIRED_PYTHON: "3.8"
       build_name: wheel-py3_8-cpu
       use_s3: False
@@ -197,7 +196,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -209,7 +207,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -261,7 +259,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       DESIRED_PYTHON: "3.9"
       build_name: wheel-py3_9-cpu
       use_s3: False
@@ -316,7 +314,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -328,7 +325,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -380,7 +377,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       DESIRED_PYTHON: "3.10"
       build_name: wheel-py3_10-cpu
       use_s3: False
@@ -435,7 +432,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -447,7 +443,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -499,7 +495,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       DESIRED_PYTHON: "3.11"
       build_name: wheel-py3_11-cpu
       use_s3: False
@@ -554,7 +550,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -566,7 +561,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -618,7 +613,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.4
       DESIRED_PYTHON: "3.12"
       build_name: wheel-py3_12-cpu
       use_s3: False

.github/workflows/generated-windows-binary-conda-nightly.yml

@@ -93,7 +93,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -105,7 +104,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -210,7 +209,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -222,7 +220,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -336,7 +334,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -348,7 +345,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -454,7 +451,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -466,7 +462,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -581,7 +577,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -593,7 +588,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -699,7 +694,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -711,7 +705,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -826,7 +820,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -838,7 +831,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -944,7 +937,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -956,7 +948,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1070,7 +1062,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1082,7 +1073,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1187,7 +1178,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1199,7 +1189,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1313,7 +1303,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1325,7 +1314,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1431,7 +1420,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1443,7 +1431,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1558,7 +1546,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1570,7 +1557,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1676,7 +1663,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1688,7 +1674,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1803,7 +1789,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1815,7 +1800,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1921,7 +1906,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1933,7 +1917,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2047,7 +2031,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2059,7 +2042,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2164,7 +2147,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2176,7 +2158,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2290,7 +2272,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2302,7 +2283,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2408,7 +2389,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2420,7 +2400,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2535,7 +2515,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2547,7 +2526,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2653,7 +2632,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2665,7 +2643,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2780,7 +2758,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2792,7 +2769,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2898,7 +2875,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2910,7 +2886,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3024,7 +3000,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3036,7 +3011,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3141,7 +3116,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3153,7 +3127,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3267,7 +3241,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3279,7 +3252,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3385,7 +3358,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3397,7 +3369,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3512,7 +3484,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3524,7 +3495,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3630,7 +3601,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3642,7 +3612,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3757,7 +3727,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3769,7 +3738,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3875,7 +3844,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3887,7 +3855,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4001,7 +3969,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4013,7 +3980,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4118,7 +4085,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4130,7 +4096,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4244,7 +4210,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4256,7 +4221,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4362,7 +4327,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4374,7 +4338,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4489,7 +4453,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4501,7 +4464,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4607,7 +4570,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4619,7 +4581,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4734,7 +4696,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4746,7 +4707,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4852,7 +4813,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4864,7 +4824,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/generated-windows-binary-libtorch-debug-main.yml

@@ -90,7 +90,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -102,7 +101,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -211,7 +210,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -223,7 +221,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/generated-windows-binary-libtorch-debug-nightly.yml

@@ -97,7 +97,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -109,7 +108,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -218,7 +217,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -230,7 +228,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -352,7 +350,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -364,7 +361,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -474,7 +471,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -486,7 +482,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -609,7 +605,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -621,7 +616,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -731,7 +726,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -743,7 +737,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -866,7 +860,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -878,7 +871,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -988,7 +981,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1000,7 +992,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/generated-windows-binary-libtorch-release-main.yml

@@ -90,7 +90,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -102,7 +101,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -211,7 +210,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -223,7 +221,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/generated-windows-binary-libtorch-release-nightly.yml

@@ -97,7 +97,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -109,7 +108,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -218,7 +217,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -230,7 +228,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -352,7 +350,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -364,7 +361,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -474,7 +471,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -486,7 +482,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -609,7 +605,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -621,7 +616,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -731,7 +726,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -743,7 +737,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -866,7 +860,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -878,7 +871,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -988,7 +981,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1000,7 +992,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/generated-windows-binary-wheel-nightly.yml

@@ -94,7 +94,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -106,7 +105,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -211,7 +210,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -223,7 +221,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -338,7 +336,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -350,7 +347,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -456,7 +453,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -468,7 +464,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -584,7 +580,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -596,7 +591,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -702,7 +697,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -714,7 +708,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -830,7 +824,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -842,7 +835,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -948,7 +941,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -960,7 +952,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1075,7 +1067,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1087,7 +1078,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1192,7 +1183,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1204,7 +1194,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1319,7 +1309,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1331,7 +1320,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1437,7 +1426,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1449,7 +1437,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1565,7 +1553,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1577,7 +1564,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1683,7 +1670,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1695,7 +1681,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1811,7 +1797,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1823,7 +1808,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1929,7 +1914,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -1941,7 +1925,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2056,7 +2040,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2068,7 +2051,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2173,7 +2156,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2185,7 +2167,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2300,7 +2282,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2312,7 +2293,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2418,7 +2399,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2430,7 +2410,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2546,7 +2526,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2558,7 +2537,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2664,7 +2643,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2676,7 +2654,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2792,7 +2770,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2804,7 +2781,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2910,7 +2887,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -2922,7 +2898,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3037,7 +3013,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3049,7 +3024,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3154,7 +3129,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3166,7 +3140,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3281,7 +3255,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3293,7 +3266,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3399,7 +3372,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3411,7 +3383,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3527,7 +3499,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3539,7 +3510,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3645,7 +3616,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3657,7 +3627,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3773,7 +3743,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3785,7 +3754,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3891,7 +3860,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -3903,7 +3871,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4018,7 +3986,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4030,7 +3997,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4135,7 +4102,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4147,7 +4113,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4262,7 +4228,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4274,7 +4239,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4380,7 +4345,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4392,7 +4356,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4508,7 +4472,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4520,7 +4483,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4626,7 +4589,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4638,7 +4600,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4754,7 +4716,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4766,7 +4727,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -4872,7 +4833,6 @@ jobs:
       - name: Checkout PyTorch
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -4884,7 +4844,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/lint-bc.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Run BC Lint Action
-        uses: pytorch/test-infra/.github/actions/bc-lint@main
+        uses: pytorch/test-infra/.github/actions/bc-lint@release/2.4
         with:
           repo: ${{ github.event.pull_request.head.repo.full_name }}
           base_sha: ${{ github.event.pull_request.base.sha }}

.github/workflows/lint.yml

@@ -16,7 +16,7 @@ permissions: read-all
 # When any other step fails, it's job will be retried once by retryBot.
 jobs:
   lintrunner-clang:
-    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@release/2.4
     with:
       timeout: 120
       runner: linux.2xlarge
@@ -32,7 +32,7 @@ jobs:
         .github/scripts/lintrunner.sh
 
   lintrunner-noclang:
-    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@release/2.4
     with:
       timeout: 120
       runner: linux.2xlarge
@@ -47,7 +47,7 @@ jobs:
         .github/scripts/lintrunner.sh
 
   quick-checks:
-    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@release/2.4
     with:
       runner: linux.2xlarge
       docker-image: pytorch-linux-focal-linter
@@ -88,7 +88,7 @@ jobs:
     if: github.event_name == 'pull_request' && !contains(github.event.pull_request.labels.*.name, 'skip-pr-sanity-checks')
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
           fetch-depth: -1
@@ -101,7 +101,7 @@ jobs:
           bash .github/scripts/pr-sanity-check.sh
 
   workflow-checks:
-    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@release/2.4
     with:
       runner: linux.2xlarge
       docker-image: pytorch-linux-focal-linter
@@ -112,6 +112,7 @@ jobs:
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
+        export RELEASE_VERSION_TAG="2.4"
 
         # Regenerate workflows
         .github/scripts/generate_ci_workflows.py
@@ -137,7 +138,7 @@ jobs:
         exit $RC
 
   toc:
-    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@release/2.4
     with:
       runner: linux.2xlarge
       docker-image: pytorch-linux-focal-linter
@@ -175,7 +176,7 @@ jobs:
   test-tools:
     name: Test tools
     if: ${{ github.repository == 'pytorch/pytorch' }}
-    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@release/2.4
     with:
       runner: linux.2xlarge
       docker-image: pytorch-linux-focal-linter
@@ -196,7 +197,7 @@ jobs:
     runs-on: linux.20_04.4x
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
           fetch-depth: 1
@@ -226,7 +227,7 @@ jobs:
       # [see note: pytorch repo ref]
       # deep clone (fetch-depth 0) required, to allow us to use git log
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
           fetch-depth: 1

.github/workflows/llm_td_retrieval.yml

@@ -116,5 +116,5 @@ jobs:
           AWS_REGION: ""
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/nightly-rockset-uploads.yml

@@ -21,7 +21,7 @@ jobs:
     environment: upload-stats
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false

.github/workflows/nightly.yml

@@ -41,7 +41,7 @@ jobs:
     environment: update-commit-hash
     steps:
       - name: update-vision-commit-hash
-        uses: pytorch/test-infra/.github/actions/update-commit-hash@main
+        uses: pytorch/test-infra/.github/actions/update-commit-hash@release/2.4
         if: ${{ github.event_name == 'schedule' }}
         with:
           repo-name: vision
@@ -56,7 +56,7 @@ jobs:
     environment: update-commit-hash
     steps:
       - name: update-audio-commit-hash
-        uses: pytorch/test-infra/.github/actions/update-commit-hash@main
+        uses: pytorch/test-infra/.github/actions/update-commit-hash@release/2.4
         if: ${{ github.event_name == 'schedule' }}
         with:
           repo-name: audio
@@ -71,7 +71,7 @@ jobs:
     environment: update-commit-hash
     steps:
       - name: update-executorch-commit-hash
-        uses: pytorch/test-infra/.github/actions/update-commit-hash@main
+        uses: pytorch/test-infra/.github/actions/update-commit-hash@release/2.4
         if: ${{ github.event_name == 'schedule' }}
         with:
           repo-name: executorch

.github/workflows/target-determination-indexer.yml

@@ -24,7 +24,7 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         with:
           docker-image-name: pytorch-linux-focal-cuda12.1-cudnn9-py3-gcc9
           working-directory: pytorch
@@ -39,13 +39,13 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${tag/:/-}"
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
         id: install-nvidia-driver
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.4
 
       - name: Clone CodeLlama
         uses: actions/checkout@v3
@@ -136,7 +136,7 @@ jobs:
             "s3://target-determinator-assets/indexes/latest/${ZIP_NAME}"
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()
 
 concurrency:

.github/workflows/target_determination.yml

@@ -14,7 +14,7 @@ jobs:
       # checkout because when we run this action we don't *have* a local
       # checkout. In other cases you should prefer a local checkout.
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
 

.github/workflows/update-viablestrict.yml

@@ -16,7 +16,7 @@ jobs:
     environment: ${{ (github.event_name == 'schedule') && 'mergebot' || '' }}
     steps:
       - name: Update viable/strict
-        uses: pytorch/test-infra/.github/actions/update-viablestrict@main
+        uses: pytorch/test-infra/.github/actions/update-viablestrict@release/2.4
         with:
           repository: pytorch/pytorch
           stable-branch: viable/strict

.github/workflows/update_pytorch_labels.yml

@@ -17,7 +17,7 @@ jobs:
       contents: read
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false

.github/workflows/upload-alerts.yml

@@ -44,7 +44,7 @@ jobs:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
           AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-        uses: pytorch/test-infra/.github/actions/upload-alerts@main
+        uses: pytorch/test-infra/.github/actions/upload-alerts@release/2.4
         with:
           alerts: '${{ steps.alert_creation_step.outputs.script-output }}'
           organization: "pytorch"

.github/workflows/upload-test-stats.yml

@@ -39,7 +39,7 @@ jobs:
         run: echo "${TRIGGERING_WORKFLOW}"
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - uses: actions/setup-python@v4
         with:

.github/workflows/upload-torch-dynamo-perf-stats.yml

@@ -29,7 +29,7 @@ jobs:
     name: Upload dynamo performance stats for ${{ github.event.workflow_run.id }}, attempt ${{ github.event.workflow_run.run_attempt }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
           fetch-depth: 1

.github/workflows/upload_test_stats_intermediate.yml

@@ -17,7 +17,7 @@ jobs:
     environment: upload-stats
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false

.github/workflows/weekly.yml

@@ -21,7 +21,7 @@ jobs:
           fetch-depth: 0
       - name: update-xla-commit-hash
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/update-commit-hash@main
+        uses: pytorch/test-infra/.github/actions/update-commit-hash@release/2.4
         with:
           repo-name: xla
           branch: master
@@ -30,7 +30,7 @@ jobs:
           updatebot-token: ${{ secrets.UPDATEBOT_TOKEN }}
           pytorchbot-token: ${{ secrets.GH_PYTORCHBOT_TOKEN }}
       - name: update-triton-commit-hash
-        uses: pytorch/test-infra/.github/actions/update-commit-hash@main
+        uses: pytorch/test-infra/.github/actions/update-commit-hash@release/2.4
         with:
           repo-owner: openai
           repo-name: triton

aten/src/ATen/Context.h

@@ -364,7 +364,7 @@ class TORCH_API Context {
   bool enabled_flashSDP = true;
   bool enabled_mem_efficientSDP = true;
   bool enabled_mathSDP = true;
-  bool enabled_cudnnSDP = true;
+  bool enabled_cudnnSDP = false;
 #ifdef USE_ROCM
   bool benchmark_cudnn = true;
 #else

aten/src/ATen/core/MetaFallbackKernel.cpp

@@ -17,7 +17,7 @@ static void metaFallback(
       "while using an operator with PT2 compilation APIs (torch.compile/torch.export); "
       "in order to use this operator with those APIs you'll need to add a fake impl. "
       "Please see the following for next steps:  "
-      "https://pytorch.org/docs/main/notes/custom_operators.html");
+      "https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html");
 }
 
 TORCH_LIBRARY_IMPL(_, Meta, m) {

aten/src/ATen/native/cudnn/MHA.cpp

@@ -614,13 +614,6 @@ void run_cudnn_SDP_bprop(
     Tensor& dV,
     const Tensor& dropoutseed,
     const Tensor& dropoutoffset) {
-  Tensor dO_ = dO;
-  if (!dO.strides()[dO.strides().size() - 1]) {
-    TORCH_WARN(
-        "cuDNN SDPA backward got an innermost stride of 0 in grad_out, which is unsupported. Materializing a contiguous\
-        tensor which will increase memory usage...");
-    dO_ = dO.contiguous();
-  }
   cudnnHandle_t handle = getCudnnHandle();
   auto key = MHACacheKeyWrapper(
       b, h, s_q, s_kv, d, q, k, v, dropout_probability, is_causal, true);
@@ -642,7 +635,7 @@ void run_cudnn_SDP_bprop(
         k,
         v,
         o,
-        dO_,
+        dO,
         softmaxstats,
         dQ,
         dK,

aten/src/ATen/native/native_functions.yaml

@@ -14728,12 +14728,12 @@
     CUDA: _scaled_dot_product_efficient_attention_backward_cuda
   tags: nondeterministic_seeded
 
-- func: _scaled_dot_product_cudnn_attention(Tensor query, Tensor key, Tensor value, bool compute_log_sumexp, float dropout_p=0.0, bool is_causal=False, *, float? scale=None) -> (Tensor output, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset)
+- func: _scaled_dot_product_cudnn_attention(Tensor query, Tensor key, Tensor value, float dropout_p=0.0, bool is_causal=False, bool return_debug_mask=False, *, float? scale=None) -> (Tensor output, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, Tensor philox_seed, Tensor philox_offset, Tensor debug_attn_mask)
   dispatch:
     CUDA: _scaled_dot_product_cudnn_attention_cuda
   tags: nondeterministic_seeded
 
-- func: _scaled_dot_product_cudnn_attention_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, Tensor philox_seed, Tensor philox_offset, float dropout_p, bool is_causal, *, float? scale=None) -> (Tensor, Tensor, Tensor)
+- func: _scaled_dot_product_cudnn_attention_backward(Tensor grad_out, Tensor query, Tensor key, Tensor value, Tensor out, Tensor logsumexp, Tensor cum_seq_q, Tensor cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, Tensor philox_seed, Tensor philox_offset, *, float? scale=None) -> (Tensor, Tensor, Tensor)
   dispatch:
     CUDA: _scaled_dot_product_cudnn_attention_backward_cuda
   tags: nondeterministic_seeded

aten/src/ATen/native/transformers/attention.cpp

@@ -666,7 +666,7 @@ Tensor scaled_dot_product_attention(
     case sdp::SDPBackend::cudnn_attention: {
       bool compute_logsumexp = should_compute_logsumexp(query_, key, value);
       auto out_lse_softmax = at::_scaled_dot_product_cudnn_attention(
-          query_, key, value, compute_logsumexp, dropout_p, is_causal, scale);
+          query_, key, value, dropout_p, is_causal, compute_logsumexp, scale);
       return std::get<0>(out_lse_softmax);
     }
     case sdp::SDPBackend::flash_attention: {

aten/src/ATen/native/transformers/cuda/attention.cu

@@ -735,27 +735,14 @@ std::tuple<Tensor, Tensor, Tensor, Tensor, c10::SymInt, c10::SymInt, Tensor, Ten
   return std::make_tuple(attention, logsumexp, Tensor(), Tensor(), max_seqlen_batch_q, max_seqlen_batch_k, philox_seed, philox_offset, debug_attn_mask);
 }
 
-// Adapted from TE
+std::tuple<Tensor, Tensor, Tensor, Tensor, c10::SymInt, c10::SymInt, Tensor, Tensor, Tensor> _scaled_dot_product_cudnn_attention_cuda(
-// extract seed and offset from PhiloxCudaState
-__global__ void unpack_cudnn(at::PhiloxCudaState arg, int64_t* seed_ptr, int64_t* offset_ptr) {
-  if (arg.captured_) {
-    *seed_ptr = static_cast<int64_t>(*arg.seed_.ptr);
-    *offset_ptr = static_cast<int64_t>(
-                    *(arg.offset_.ptr) + static_cast<int64_t>(arg.offset_intragraph_));
-  } else {
-    *seed_ptr = static_cast<int64_t>(arg.seed_.val);
-    *offset_ptr = static_cast<int64_t>(arg.offset_.val);
-  }
-}
-
-std::tuple<Tensor, Tensor, Tensor, Tensor> _scaled_dot_product_cudnn_attention_cuda(
     const Tensor& query,
     const Tensor& key,
     const Tensor& value,
-    bool compute_logsumexp,
     double dropout_p,
     bool is_causal,
-    c10::optional<double> scale) {
+    bool training,
+    std::optional<double> scale) {
   // Used for tracking usage statistics
   C10_LOG_API_USAGE_ONCE("torch.sdpa.flash_attention_cudnn");
   // Query (Batch x Num_heads x Q_seq_len  x Dim_per_head)
@@ -774,33 +761,9 @@ std::tuple<Tensor, Tensor, Tensor, Tensor> _scaled_dot_product_cudnn_attention_c
 
   Tensor attention, log_sumexp;
 
-  at::Tensor cudnn_seed, cudnn_offset;
+  auto cudnn_seed = at::zeros({1}, query.options().dtype(kLong));
-  cudnn_seed = at::empty({}, at::dtype(at::kLong).device(at::kCUDA));
+  auto cudnn_offset = at::zeros({1}, query.options().dtype(kLong));
-  cudnn_offset = at::empty({}, at::dtype(at::kLong).device(at::kCUDA));
-
-  const bool use_dropout = std::fpclassify(dropout_p) != FP_ZERO;
-
-  // See Note [Seed and Offset Device] in _efficient_attention_forward
-  at::PhiloxCudaState philox_state;
-  const bool in_capture_stream =
-      at::cuda::currentStreamCaptureStatus() != at::cuda::CaptureStatus::None;
-  if (use_dropout) {
-    // Device
-    auto gen = at::get_generator_or_default<at::CUDAGeneratorImpl>(
-        c10::nullopt, at::cuda::detail::getDefaultCUDAGenerator());
-
-    // See Note [Acquire lock when using random generators]
-    std::lock_guard<std::mutex> lock(gen->mutex_);
-    // if using dropout, we produce 1 random number for each element of the
-    // attention tensor
-    // TODO(eqy): should state be advanced per thread (local) amount or per call/launch (global) amount
-    philox_state = gen->philox_cuda_state(batch_size * num_heads * max_seqlen_batch_q * max_seqlen_batch_k);
-    unpack_cudnn<<<1, 1, 0, at::cuda::getCurrentCUDAStream()>>>(
-                                      philox_state, static_cast<int64_t*>(cudnn_seed.data_ptr()), static_cast<int64_t*>(cudnn_offset.data_ptr()));
-  }
-
   const auto softmax_scale = sdp::calculate_scale(query, scale).as_float_unchecked();
-  Tensor debugmask;
 
   run_cudnn_SDP_fprop(batch_size/*int64_t b*/,
                       num_heads/*int64_t h*/,
@@ -808,7 +771,7 @@ std::tuple<Tensor, Tensor, Tensor, Tensor> _scaled_dot_product_cudnn_attention_c
                       max_seqlen_batch_k/*int64_t s_kv*/,
                       head_dim/*int64_t d*/,
                       softmax_scale/*float scaling_factor*/,
-                      compute_logsumexp/* bool */,
+                      training/* bool */,
                       is_causal/* bool */,
                       dropout_p/*double dropout_probability*/,
                       query/* Tensor q*/,
@@ -819,7 +782,7 @@ std::tuple<Tensor, Tensor, Tensor, Tensor> _scaled_dot_product_cudnn_attention_c
                       cudnn_seed/*Tensor dropoutseed*/,
                       cudnn_offset/*Tensor dropoutoffset*/);
 
-  return std::make_tuple(attention, log_sumexp, cudnn_seed, cudnn_offset);
+  return std::make_tuple(attention, log_sumexp, Tensor(), Tensor(), max_seqlen_batch_q, max_seqlen_batch_k, cudnn_seed, cudnn_offset, Tensor());
 }
 
 std::tuple<Tensor, Tensor, Tensor, Tensor> _scaled_dot_product_efficient_attention_cuda(

aten/src/ATen/native/transformers/cuda/attention_backward.cu

@@ -171,32 +171,18 @@ std::tuple<Tensor, Tensor, Tensor> _scaled_dot_product_cudnn_attention_backward_
     const Tensor& value,
     const Tensor& out,
     const Tensor& logsumexp,
-    const Tensor& philox_seed,
+    const Tensor& cumulative_sequence_length_q,
-    const Tensor& philox_offset,
+    const Tensor& cumulative_sequence_length_k,
-//    const Tensor& cumulative_sequence_length_q,
+    const int64_t max_seqlen_batch_q,
-//    const Tensor& cumulative_sequence_length_k,
+    const int64_t max_seqlen_batch_k,
-//    const int64_t max_seqlen_batch_q,
-//    const int64_t max_seqlen_batch_k,
     double dropout_p,
     bool is_causal,
-    c10::optional<double> scale) {
+    const Tensor& philox_seed,
-
+    const Tensor& philox_offset,
-
+    std::optional<double> scale) {
-    auto& ctx = at::globalContext();
-    if (ctx.deterministicAlgorithms()) {
-      if (ctx.deterministicAlgorithmsWarnOnly()) {
-        TORCH_WARN_ONCE(
-            "cuDNN Attention defaults to a non-deterministic algorithm. ",
-            "To explicitly enable determinism call torch.use_deterministic_algorithms(True, warn_only=False).");
-      }
-    }
-
-
     const int64_t batch_size = query.size(0);
     const int64_t num_heads = query.size(1);
     const int64_t head_dim = query.size(3);
-    const int64_t max_seqlen_batch_q = query.size(1);
-    const int64_t max_seqlen_batch_k = key.size(1);
 
     const auto softmax_scale = sdp::calculate_scale(query, scale).as_float_unchecked();
 

aten/src/ATen/native/transformers/cuda/sdp_utils.cpp

@@ -6,7 +6,6 @@
 #include <ATen/core/Tensor.h>
 #include <ATen/core/grad_mode.h>
 #include <ATen/cuda/CUDAContext.h>
-#include <ATen/cuda/CUDAConfig.h>
 #include <ATen/detail/CUDAHooksInterface.h>
 #include <ATen/native/DispatchStub.h>
 #include <ATen/native/transformers/cuda/sdp_utils.h>
@@ -45,28 +44,14 @@
 
 namespace sdp {
 namespace {
-
-// TODO(eqy): more benchmarking to determine whether this should include sm86/89
-// Needs to be kept in-sync with test_fused_chocie in test_transformers.py
-bool check_prefer_cudnn_attention() {
-  auto dprops = at::cuda::getCurrentDeviceProperties();
-  return dprops->major >= 9;
-}
-
 // flash_attention V2 is universally faster than efficient_attention and Math
 std::array<SDPBackend, num_backends> priority_order(sdp_params const& params) {
   constexpr std::array<SDPBackend, num_backends> default_order{
-      SDPBackend::flash_attention,
-      SDPBackend::cudnn_attention,
-      SDPBackend::efficient_attention,
-      SDPBackend::math};
-  constexpr std::array<SDPBackend, num_backends> cudnn_order{
       SDPBackend::cudnn_attention,
       SDPBackend::flash_attention,
       SDPBackend::efficient_attention,
       SDPBackend::math};
-  static const bool prefer_cudnn = check_prefer_cudnn_attention();
+  return default_order;
-  return prefer_cudnn ? cudnn_order : default_order;
 }
 
 bool use_tensor_cores(sdp_params const& params, cudaDeviceProp* dprops, bool is_half) {
@@ -466,6 +451,17 @@ bool check_cudnn_hardware_support(sdp_params const& params, bool debug) {
   return true;
 }
 
+bool check_is_causal(sdp_params const& params, bool debug) {
+  // Check that the input is causal
+  if (!params.is_causal) {
+    if (debug) {
+      TORCH_WARN("CuDNN requires is_causal=True.");
+    }
+    return false;
+  }
+  return true;
+}
+
 bool check_for_nested_inputs(sdp_params const& params, bool debug) {
   // Check that the input is nested
   if (has_for_nested_inputs(params)) {
@@ -489,6 +485,22 @@ bool check_dtypes_low_precision(sdp_params const& params, bool debug) {
   }
 }
 
+bool check_runtime_enabled_cudnn(sdp_params const& params, bool debug) {
+  static c10::once_flag supported_flag;
+  static bool supported = false;
+  c10::call_once(supported_flag, []() {
+    supported = (c10::utils::check_env("TORCH_CUDNN_SDPA_ENABLED") == true);
+  });
+  if (!supported) {
+    if (debug) {
+      TORCH_WARN(
+          "The CuDNN backend needs to be enabled by setting the enviornment variable`TORCH_CUDNN_SDPA_ENABLED=1`");
+    }
+    return false;
+  }
+  return true;
+}
+
 bool check_runtime_disabled_cudnn(sdp_params const& params, bool debug) {
   // We check the global context to see if user has explicitly turned of cudnn
   // sdp kernels
@@ -501,15 +513,13 @@ bool check_runtime_disabled_cudnn(sdp_params const& params, bool debug) {
   return true;
 }
 
-bool check_cudnn_deterministic(const sdp_params& params, bool debug) {
+bool check_cudnn_requires_grad(sdp_params const& params, bool debug) {
-  auto& ctx = at::globalContext();
+  // Check that the input is causal
-  if (ctx.deterministicAlgorithms()) {
+  if (input_requires_grad(params)) {
-    if (!ctx.deterministicAlgorithmsWarnOnly()) {
+    if (debug) {
-      if (debug) {
+      TORCH_WARN("CuDNN does not currently support inputs with requires_grad=True.");
-        TORCH_WARN("cuDNN SDPA is not deterministic.");
-      }
-      return false;
     }
+    return false;
   }
   return true;
 }
@@ -517,29 +527,21 @@ bool check_cudnn_deterministic(const sdp_params& params, bool debug) {
 } // namespace
 
 bool can_use_cudnn_attention(const sdp_params& params, bool debug) {
-#if defined(USE_ROCM) || !AT_CUDNN_ENABLED() || \
+
-    (defined(CUDNN_VERSION) && CUDNN_VERSION < 8900)
-  TORCH_WARN_ONCE(!debug, "Torch was not compiled with cuDNN attention.");
-  return false;
-#endif
   // Define gate functions that determine if a flash kernel can be ran
   // Replace with std::to_array when we migrate to c++20
   constexpr auto general_constraints =
       array_of<bool (*)(sdp_params const&, bool)>(
-          check_for_nested_inputs,
+          check_runtime_enabled_cudnn,
-          check_nonzero_sequence_lengths_dense,
-          check_last_dim_stride_equals_1_dense<true /*ignore_singleton_dim>*/>,
-          check_all_tensors_on_device,
-          check_tensor_shapes,
-          check_cudnn_tensor_shapes,
           check_runtime_disabled_cudnn,
-          check_cudnn_deterministic,
+          check_cudnn_hardware_support,
-          // check_cudnn_layout,
+          check_all_tensors_on_device,
+          check_cudnn_tensor_shapes,
+          check_cudnn_layout,
           // check_is_causal,
-          check_dtypes_low_precision,
+          check_for_nested_inputs,
-          check_for_attn_mask_cudnn,
+          check_cudnn_requires_grad,
-          check_cudnn_hardware_support
+          check_dtypes_low_precision);
-          );
   for (auto& constraint : general_constraints) {
     if (!constraint(params, debug)) {
       return false;
@@ -683,7 +685,6 @@ SDPBackend select_sdp_backend(sdp_params const& kernel_params) {
     switch (backend) {
       case SDPBackend::cudnn_attention:
         if (sdp::can_use_cudnn_attention(kernel_params, print_debug)) {
-              TORCH_WARN("USING CUDNN SDPA");
               return SDPBackend::cudnn_attention;
         }
         break;

aten/src/ATen/native/transformers/sdp_utils_cpp.h

@@ -266,18 +266,7 @@ inline bool check_requires_grad_and_nested(sdp_params const& params, bool debug)
 inline bool check_for_attn_mask(sdp_params const& params, bool debug) {
   if (params.attn_mask.has_value()) {
     if (debug) {
-      TORCH_WARN("Flash Attention do not support non-null attn_mask.");
+      TORCH_WARN("Flash Attention does not support non-null attn_mask.");
-    }
-    return false;
-  }
-  return true;
-}
-
-// TODO(eqy): remove this once support is added
-inline bool check_for_attn_mask_cudnn(sdp_params const& params, bool debug) {
-  if (params.attn_mask.has_value()) {
-    if (debug) {
-      TORCH_WARN("cuDNN Attention does not support non-null attn_mask.");
     }
     return false;
   }
@@ -324,7 +313,7 @@ inline bool check_tensor_shapes(sdp_params const& params, bool debug) {
         (query_dim == 4))) {
     if (debug) {
       TORCH_WARN(
-          "All fused kernels requires query, key and value to be 4 dimensional, but got Query dim: ",
+          "Both fused kernels requires query, key and value to be 4 dimensional, but got Query dim: ",
           query_dim,
           ", Key dim: ",
           params.key.dim(),
@@ -436,7 +425,7 @@ inline bool check_nonzero_sequence_lengths_dense(sdp_params const& params, bool
   if (zero_seq_len_q || zero_seq_len_k) {
     if (debug) {
       TORCH_WARN(
-          "All fused kernels do not support zero seq_len_q or seq_len_kv.");
+          "Both fused kernels do not support zero seq_len_q or seq_len_kv.");
     }
     return false;
   }
@@ -471,7 +460,7 @@ inline bool check_last_dim_stride_equals_1_dense(sdp_params const& params, bool
       }
       epilogue_message << " instead.";
       TORCH_WARN(
-          "All fused kernels require the last dimension of the input to have stride 1. ",
+          "Both fused kernels require the last dimension of the input to have stride 1. ",
           "Got Query.stride(-1): ",
           params.query.sym_stride(-1),
           ", Key.stride(-1): ",

benchmarks/dynamo/common.py

@@ -1184,12 +1184,14 @@ class AOTInductorModelCache:
             else:
                 _register_dataclass_output_as_pytree(example_outputs)
 
-            gm = torch.export._trace._export(
+            # TODO(angelayi): change this to predispatch
+            # https://github.com/pytorch/pytorch/issues/127513 needs to be fixed before changing
+            # to predispatch to avoid performance regressions
+            gm = torch.export._trace._export_to_torch_ir(
                 model,
                 example_args,
                 example_kwargs,
-                pre_dispatch=True,
+            )
-            ).module()
             with torch.no_grad():
                 so_path = torch._inductor.aot_compile(
                     gm, example_args, example_kwargs

c10/core/StorageImpl.cpp

@@ -18,7 +18,7 @@ void throwNullDataPtrError() {
       "If you're using torch.compile/export/fx, it is likely that we are erroneously "
       "tracing into a custom kernel. To fix this, please wrap the custom kernel into "
       "an opaque custom op. Please see the following for details: "
-      "https://pytorch.org/docs/main/notes/custom_operators.html");
+      "https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html");
 }
 
 // NOTE: [FakeTensor.data_ptr deprecation]

c10/core/TensorImpl.h

@@ -1580,7 +1580,7 @@ struct C10_API TensorImpl : public c10::intrusive_ptr_target {
         "If you're using torch.compile/export/fx, it is likely that we are erroneously "
         "tracing into a custom kernel. To fix this, please wrap the custom kernel into "
         "an opaque custom op. Please see the following for details: "
-        "https://pytorch.org/docs/main/notes/custom_operators.html\n"
+        "https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html\n"
         "If you're using Caffe2, Caffe2 uses a lazy allocation, so you will need to call "
         "mutable_data() or raw_mutable_data() to actually allocate memory.");
     // Caller does the type check.

docs/source/backends.rst

@@ -92,8 +92,6 @@ torch.backends.cuda
 
 .. autofunction:: torch.backends.cuda.can_use_efficient_attention
 
-.. autofunction:: torch.backends.cuda.can_use_cudnn_attention
-
 .. autofunction:: torch.backends.cuda.sdp_kernel
 
 torch.backends.cudnn

docs/source/notes/custom_operators.rst

@@ -3,54 +3,4 @@
 PyTorch Custom Operators Landing Page
 =====================================
 
-PyTorch offers a large library of operators that work on Tensors (e.g. :func:`torch.add`,
+`This page has moved. Click here for the new page. <https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html>`_
-:func:`torch.sum`, etc). However, you may wish to bring a new custom operation to PyTorch
-and get it to work with subsystems like :func:`torch.compile`, autograd, and :func:`torch.vmap`.
-In order to do so, you must register the custom operation with PyTorch via the Python
-:ref:`torch-library-docs` or C++ TORCH_LIBRARY APIs.
-
-TL;DR
------
-
-How do I author a custom op from Python?
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-..
-  [comment] TODO(rzou): The following will be a link to a tutorial on the PyTorch tutorials site in 2.4
-
-Please see the `Python Custom Operators tutorial <https://colab.research.google.com/drive/1xCh5BNHxGnutqGLMHaHwm47cbDL9CB1g#scrollTo=gg6WorNtKzeh>`_
-
-
-How do I integrate custom C++ and/or CUDA code with PyTorch?
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-..
-  [comment] TODO(rzou): The following will be a link to a tutorial on the PyTorch tutorials site in 2.4
-
-Please see the `Custom C++ and CUDA Operators tutorial <https://docs.google.com/document/d/1-LdJZBzlxiF0Tm-8NfbyFvRJaofdwRgLcycXGmlIpS0>`_
-
-
-For more details
-^^^^^^^^^^^^^^^^
-
-Please see `The Custom Operators Manual (gdoc) <https://docs.google.com/document/d/1_W62p8WJOQQUzPsJYa7s701JXt0qf2OfLub2sbkHOaU>`_
-(we're working on moving the information to our docs site). We recommend that you
-first read one of the tutorials above and then use the Custom Operators Manual as a reference;
-it is not meant to be read head to toe.
-
-When should I create a Custom Operator?
----------------------------------------
-If your operation is expressible as a composition of built-in PyTorch operators
-then please write it as a Python function and call it instead of creating a
-custom operator. Use the operator registration APIs to create a custom op if you
-are calling into some library that PyTorch doesn't understand (e.g. custom C/C++ code,
-a custom CUDA kernel, or Python bindings to C/C++/CUDA extensions).
-
-Why should I create a Custom Operator?
---------------------------------------
-
-It is possible to use a C/C++/CUDA kernel by grabbing a Tensor's data pointer
-and passing it to a pybind'ed kernel. However, this approach doesn't compose with
-PyTorch subsystems like autograd, torch.compile, vmap, and more. In order
-for an operation to compose with PyTorch subsystems, it must be registered
-via the operator registration APIs.

test/distributed/tensor/parallel/test_tp_style.py

@@ -317,6 +317,18 @@ class TensorParallelStyleTest(DTensorTestBase):
         self.assertEqual(comm_mode.get_total_counts(), 2)
         self.assertEqual(output.shape, (1 * self.world_size, 8))
 
+        # test the case where x is a DTensor
+        x_dt = DTensor.from_local(
+            torch.randn(1, 8, device=self.device_type), mesh, [Shard(0)]
+        )
+        with comm_mode:
+            output = test_kwonly_mod(
+                x=x_dt, z=torch.ones(1, 8, device=self.device_type)
+            )
+
+        self.assertEqual(comm_mode.get_total_counts(), 2)
+        self.assertEqual(output.shape, (1 * self.world_size, 8))
+
     @with_comms
     def test_prepare_module_output(self):
         mesh = init_device_mesh(self.device_type, (self.world_size,))

test/dynamo/test_misc.py

@@ -246,16 +246,49 @@ class MiscTests(torch._inductor.test_case.TestCase):
             return module.foobar(x)
 
         with self.assertWarnsOnceRegex(
-            UserWarning, ".*https://pytorch.org/docs/main/notes/custom_operators.html.*"
+            UserWarning,
+            ".*https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html.*",
         ):
             f(x)
         self.assertEqual(len(counters["graph_break"]), 1)
         first_graph_break = list(counters["graph_break"].keys())[0]
         self.assertExpectedInline(
             first_graph_break,
-            """Graph break due to unsupported builtin mylib.PyCapsule.foobar. This function is either a Python builtin (e.g. _warnings.warn) or a third-party C/C++ Python extension (perhaps created with pybind). If it is a Python builtin, please file an issue on GitHub so the PyTorch team can add support for it and see the next case for a workaround. If it is a third-party C/C++ Python extension, please either wrap it into a PyTorch-understood custom operator (see https://pytorch.org/docs/main/notes/custom_operators.html for more details) or, if it is traceable, use torch.compiler.allow_in_graph.""",
+            """Graph break due to unsupported builtin mylib.PyCapsule.foobar. This function is either a Python builtin (e.g. _warnings.warn) or a third-party C/C++ Python extension (perhaps created with pybind). If it is a Python builtin, please file an issue on GitHub so the PyTorch team can add support for it and see the next case for a workaround. If it is a third-party C/C++ Python extension, please either wrap it into a PyTorch-understood custom operator (see https://pytorch.org/tutorials/advanced/custom_ops_landing_page.html for more details) or, if it is traceable, use torch.compiler.allow_in_graph.""",
         )
 
+        cpp_source = """
+        #include <torch/extension.h>
+        at::Tensor baz(const at::Tensor& x) {
+            return x.clone();
+        }
+        """
+        module2 = torch.utils.cpp_extension.load_inline(
+            name="mylib2",
+            cpp_sources=cpp_source,
+            functions="baz",
+            verbose=True,
+        )
+
+        torch._dynamo.reset()
+
+        # Test that each warning only happens once
+        @torch.compile(backend="eager")
+        def f(x):
+            module2.baz(x)
+            module.foobar(x)
+            module.foobar(x)
+            module2.baz(x)
+            module.foobar(x)
+            module2.baz(x)
+            return x.clone()
+
+        with warnings.catch_warnings(record=True) as ws:
+            warnings.simplefilter("always")
+            f(x)
+            f(x)
+        self.assertEqual(len(ws), 2)
+
     def test_callpacked(self):
         def call_packed(args):
             a, b, c = args

test/higher_order_ops/test_with_effects.py

@@ -198,6 +198,33 @@ def forward(self, arg0_1, arg1_1, arg2_1):
         res = torch.compile(f, backend="inductor")(*inputs)
         self.assertTrue(torch.allclose(res, f(*inputs)))
 
+    @unittest.skipIf(IS_WINDOWS, "Skipped on Windows!")
+    @skipIfNoDynamoSupport
+    def test_compile_inductor_external_op_return_none(self):
+        with torch.library._scoped_library("mylib", "FRAGMENT") as lib:
+            torch.library.define(
+                "mylib::inplace_add",
+                "(Tensor input, Tensor(a!) output) -> ()",
+                lib=lib,
+            )
+
+            def inplace_add(input: torch.Tensor, output: torch.Tensor) -> None:
+                assert input.device == output.device
+                output.add_(input)
+
+            lib.impl("inplace_add", inplace_add, "CompositeExplicitAutograd")
+
+            def f(x):
+                out = torch.empty(3)
+                out = torch.zeros_like(out)
+                torch.ops.mylib.inplace_add(x, out)
+                return out
+
+            inputs = (torch.randn(3),)
+
+            res = torch.compile(f, backend="inductor")(*inputs)
+            self.assertTrue(torch.allclose(res, f(*inputs)))
+
     def test_compile_aot_eager_requires_grad(self):
         def f(x):
             torch.ops.aten._print("moo")

test/inductor/test_cpu_repro.py

@@ -4,7 +4,6 @@ import copy
 import functools
 import itertools
 import math
-import os
 import platform
 import sys
 import unittest
@@ -67,13 +66,12 @@ aten = torch.ops.aten
 check_model = test_torchinductor.check_model
 
 requires_vectorization = unittest.skipUnless(
-    codecache.valid_vec_isa_list() and os.getenv("ATEN_CPU_CAPABILITY") != "default",
+    codecache.valid_vec_isa_list(), "Does not support vectorization"
-    "Does not support vectorization",
 )
 
 
 def check_metrics_vec_kernel_count(num_expected_vec_kernels):
-    if codecache.valid_vec_isa_list() and os.getenv("ATEN_CPU_CAPABILITY") != "default":
+    if codecache.valid_vec_isa_list():
         assert metrics.generated_cpp_vec_kernel_count == num_expected_vec_kernels
 
 
@@ -1582,71 +1580,6 @@ class CPUReproTests(TestCase):
             metrics.reset()
             self.common(fn, (value,))
 
-    @unittest.skipIf(
-        not codecache.valid_vec_isa_list()
-        or "avx2" in [str(vec_isa) for vec_isa in codecache.valid_vec_isa_list()],
-        "Does not support vectorization or not s390x/neon machine",
-    )
-    @patch("torch.cuda.is_available", lambda: False)
-    def test_auto_zvec_neon_simd(self):
-        vec_zvec_neon = codecache.valid_vec_isa_list()[0]
-        self.assertTrue(vec_zvec_neon.bit_width() == 256)
-
-        with config.patch({"cpp.simdlen": 0}):
-            isa = codecache.pick_vec_isa()
-            self.assertFalse(isa)
-
-        with config.patch({"cpp.simdlen": 1}):
-            isa = codecache.pick_vec_isa()
-            self.assertFalse(isa)
-
-        with config.patch({"cpp.simdlen": 257}):
-            isa = codecache.pick_vec_isa()
-            self.assertFalse(isa)
-
-        with config.patch({"cpp.simdlen": 256}):
-            isa = codecache.pick_vec_isa()
-            self.assertTrue(isa == vec_zvec_neon)
-
-        pre_var = os.getenv("ATEN_CPU_CAPABILITY")
-        if pre_var:
-            os.environ.pop("ATEN_CPU_CAPABILITY")
-
-        try:
-            with config.patch({"cpp.simdlen": None}):
-                isa = codecache.pick_vec_isa()
-                self.assertTrue(isa == vec_zvec_neon)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "avx2"
-                isa = codecache.pick_vec_isa()
-                self.assertTrue(isa == vec_zvec_neon)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "avx512"
-                isa = codecache.pick_vec_isa()
-                self.assertTrue(isa == vec_zvec_neon)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "default"
-                isa = codecache.pick_vec_isa()
-                self.assertFalse(isa)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "neon"
-                isa = codecache.pick_vec_isa()
-                self.assertTrue(isa == vec_zvec_neon)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "zvector"
-                isa = codecache.pick_vec_isa()
-                self.assertTrue(isa == vec_zvec_neon)
-        finally:
-            if pre_var:
-                os.environ["ATEN_CPU_CAPABILITY"] = pre_var
-            elif os.getenv("ATEN_CPU_CAPABILITY"):
-                os.environ.pop("ATEN_CPU_CAPABILITY")
-
     @unittest.skipIf(
         platform.machine() != "x86_64" or not codecache.valid_vec_isa_list(),
         "Does not support vectorization or not x86_64 machine",
@@ -1662,6 +1595,13 @@ class CPUReproTests(TestCase):
         self.assertTrue(vec_avx512.nelements(torch.bfloat16) == 32)
         self.assertTrue(vec_avx2.nelements(torch.bfloat16) == 16)
 
+        with config.patch({"cpp.simdlen": None}):
+            isa = codecache.pick_vec_isa()
+            if vec_avx512 in codecache.valid_vec_isa_list():
+                self.assertTrue(isa == vec_avx512)
+            else:
+                self.assertTrue(isa == vec_avx2)
+
         with config.patch({"cpp.simdlen": 0}):
             isa = codecache.pick_vec_isa()
             self.assertFalse(isa)
@@ -1691,60 +1631,6 @@ class CPUReproTests(TestCase):
                 isa = codecache.pick_vec_isa()
                 self.assertTrue(isa == vec_avx2)
 
-        pre_var = os.getenv("ATEN_CPU_CAPABILITY")
-        if pre_var:
-            os.environ.pop("ATEN_CPU_CAPABILITY")
-
-        try:
-            with config.patch({"cpp.simdlen": None}):
-                isa = codecache.pick_vec_isa()
-                if vec_avx512 in codecache.valid_vec_isa_list():
-                    self.assertTrue(isa == vec_avx512)
-                else:
-                    self.assertTrue(isa == vec_avx2)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "avx2"
-                isa = codecache.pick_vec_isa()
-                if vec_avx512 in codecache.valid_vec_isa_list():
-                    self.assertTrue(isa == vec_avx2)
-                elif vec_avx2 in codecache.valid_vec_isa_list():
-                    self.assertTrue(isa == vec_avx2)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "avx512"
-                isa = codecache.pick_vec_isa()
-                if vec_avx512 in codecache.valid_vec_isa_list():
-                    self.assertTrue(isa == vec_avx512)
-                else:
-                    self.assertTrue(isa == vec_avx2)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "default"
-                isa = codecache.pick_vec_isa()
-                self.assertFalse(isa)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "neon"
-                isa = codecache.pick_vec_isa()
-                if vec_avx512 in codecache.valid_vec_isa_list():
-                    self.assertTrue(isa == vec_avx512)
-                else:
-                    self.assertTrue(isa == vec_avx2)
-
-            with config.patch({"cpp.simdlen": None}):
-                os.environ["ATEN_CPU_CAPABILITY"] = "zvector"
-                isa = codecache.pick_vec_isa()
-                if vec_avx512 in codecache.valid_vec_isa_list():
-                    self.assertTrue(isa == vec_avx512)
-                else:
-                    self.assertTrue(isa == vec_avx2)
-        finally:
-            if pre_var:
-                os.environ["ATEN_CPU_CAPABILITY"] = pre_var
-            elif os.getenv("ATEN_CPU_CAPABILITY"):
-                os.environ.pop("ATEN_CPU_CAPABILITY")
-
     @requires_vectorization
     @patch("torch.cuda.is_available", lambda: False)
     def test_masked_fill_softmax(self):
@@ -3485,7 +3371,6 @@ class CPUReproTests(TestCase):
             self.common(m, (idx, x))
             check_metrics_vec_kernel_count(1)
 
-    @requires_vectorization
     def test_embedding_vec_bf16(self):
         class M(torch.nn.Module):
             def __init__(self):
@@ -3770,7 +3655,7 @@ class CPUReproTests(TestCase):
         x = torch.randint(0, 100, (819,), dtype=torch.int64)
         metrics.reset()
         self.common(fn, (x,))
-        check_metrics_vec_kernel_count(1)
+        assert metrics.generated_cpp_vec_kernel_count == 1
 
     def test_reduction_float_to_int64(self):
         # https://github.com/pytorch/pytorch/issues/124821
@@ -3780,7 +3665,7 @@ class CPUReproTests(TestCase):
         x = torch.randint(0, 100, (22, 51), dtype=torch.int64)
         metrics.reset()
         self.common(fn, (x,))
-        check_metrics_vec_kernel_count(1)
+        assert metrics.generated_cpp_vec_kernel_count == 1
 
     @config.patch({"cpp.dynamic_threads": True})
     def test_reduction_with_dynamic_threads(self):

test/inductor/test_extension_backend.py

@@ -8,7 +8,6 @@ import torch
 import torch._dynamo
 import torch.utils.cpp_extension
 from torch._C import FileCheck
-from torch._dynamo.testing import expectedFailureScalar
 
 try:
     from extension_backends.cpp.extension_codegen_backend import (
@@ -104,9 +103,6 @@ class ExtensionBackendTests(TestCase):
         # return the working directory (see setUp)
         os.chdir(self.old_working_dir)
 
-    # Fails when testing the scalar version
-    # See https://github.com/pytorch/pytorch/issues/126372.
-    @expectedFailureScalar
     def test_open_device_registration(self):
         torch.utils.rename_privateuse1_backend("extension_device")
         torch._register_device_module("extension_device", self.module)

test/inductor/test_flex_attention.py

@@ -776,11 +776,13 @@ def forward(self, arg0_1, arg1_1, arg2_1, arg3_1, arg4_1):
         metrics.reset()
         f(q, k, v)
         accessed_bytes = 1 * 8 * 1024 * 64 * torch.float32.itemsize
-        logsumexp_bytes = 1 * 8 * 1024 * torch.float32.itemsize
         num_accesses = 4  # q, k, v reads, one output.
-        self.assertEqual(
+        # TODO: Get rid of this fudge factor
-            metrics.num_bytes_accessed, accessed_bytes * num_accesses + logsumexp_bytes
+        # We need this fudge factor for now, since
-        )
+        # 1. For some reason we materialize the output of the attention unnecessarily (it's related to the mutation somehow)
+        # 2. We also write the extraneous logsumexp
+        num_accesses += 2
+        self.assertLess(metrics.num_bytes_accessed, accessed_bytes * num_accesses)
 
     @supported_platform
     @skip("Triton bug ")  # https://github.com/pytorch/pytorch/issues/124571

test/inductor/test_mkldnn_pattern_matcher.py

@@ -233,6 +233,7 @@ class TestPatternMatcherBase(TestCase):
         rtol=1.3e-6,
         check_quantization=False,
         check_dynamic=None,
+        num_include_ops=None,
     ):
         with torch.no_grad():
             clone_inputs = self._clone_inputs(inputs)
@@ -245,6 +246,12 @@ class TestPatternMatcherBase(TestCase):
             )
             for op in include_ops:
                 self.assertIn(op, source_code)
+            if num_include_ops is not None:
+                assert len(include_ops) == len(num_include_ops)
+                for i in range(len(include_ops)):
+                    self.assertEqual(
+                        source_code.count(include_ops[i]), num_include_ops[i]
+                    )
             for op in exclude_ops:
                 self.assertNotIn(op, source_code)
             if check_dynamic is not None:
@@ -400,13 +407,16 @@ class TestPatternMatcher(TestPatternMatcherBase):
         class M(torch.nn.Module):
             def __init__(self, dtype, unary_fn):
                 super().__init__()
-                self.linear = torch.nn.Linear(10, 64, bias=False)
+                self.linear1 = torch.nn.Linear(10, 64, bias=False)
-                self.bias = torch.randn(64).to(dtype=dtype)
+                self.bias1 = torch.randn(64).to(dtype=dtype)
+                self.linear2 = torch.nn.Linear(10, 64, bias=False)
+                self.bias2 = torch.randn(64).to(dtype=dtype)
                 self.unary_fn = unary_fn
 
             def forward(self, x):
-                x = self.linear(x) + self.bias
+                a = self.linear1(x) + self.bias1
-                return self.unary_fn(x)
+                b = self.linear2(x) + self.bias2
+                return self.unary_fn(a), self.unary_fn(b)
 
         dtypes = []
         if torch.ops.mkldnn._is_mkldnn_bf16_supported():
@@ -419,13 +429,14 @@ class TestPatternMatcher(TestPatternMatcherBase):
             mod = M(dtype, unary_fn).eval()
             v = torch.randn(2, 10)
             matcher_count = 3
-            # Add 1 for weight packing pass, add 2 for bias folding pass.
+            # Add 1 for weight packing pass, add 2 for bias folding pass per linear.
             matcher_nodes = unary_list[unary_fn] + 3
             if self._check_unary_is_decomposed(unary_fn):
                 # Has extra dtype conversion nodes for autocast.
                 matcher_nodes += 2
+            # we have 2 linears, so we double the matcher_count/nodes
             self._test_common(
-                mod, (v,), matcher_count, matcher_nodes, check_autocast=dtype
+                mod, (v,), matcher_count * 2, matcher_nodes * 2, check_autocast=dtype
             )
             self.assertEqual(metrics.generated_kernel_count, 1)
 
@@ -1808,6 +1819,32 @@ class TestPatternMatcher(TestPatternMatcherBase):
                     matcher_check_fn=matcher_check_fn,
                     is_qat=is_qat,
                 )
+                if torch._inductor.config.cpp_wrapper:
+                    # For CPP wrapper
+                    self._test_code_common(
+                        mod,
+                        (v,),
+                        [
+                            "op_qlinear_pointwise.call",
+                            "op_qlinear_pointwise_binary.call",
+                        ],
+                        [],
+                        check_quantization=True,
+                        num_include_ops=[2, 2],
+                    )
+                else:
+                    # For python wrapper
+                    self._test_code_common(
+                        mod,
+                        (v,),
+                        [
+                            "torch.ops.onednn.qlinear_pointwise.default",
+                            "torch.ops.onednn.qlinear_pointwise.binary",
+                        ],
+                        [],
+                        check_quantization=True,
+                        num_include_ops=[2, 2],
+                    )
 
     @skipIfNoDynamoSupport
     @skipIfNoONEDNN

test/inductor/test_torchinductor.py

@@ -34,7 +34,6 @@ from torch._dynamo.debug_utils import aot_graph_input_parser
 from torch._dynamo.testing import (
     CompileCounterWithBackend,
     expectedFailureCodegenDynamic,
-    expectedFailureScalar,
     rand_strided,
     same,
     skipIfPy312,
@@ -1316,9 +1315,6 @@ class CommonTemplate:
 
         self.common(fn, (torch.randn(1024),))
 
-    # Fails when testing the scalar version
-    # See https://github.com/pytorch/pytorch/issues/128029.
-    @expectedFailureScalar
     @skipIfRocm
     @config.patch(debug_index_asserts=False)
     def test_neg_index(self):
@@ -1581,40 +1577,16 @@ class CommonTemplate:
         def fn(a):
             return torch.var(a)
 
-        atol = None
+        self.common(fn, ((torch.rand((10, 3, 352, 352), dtype=torch.float32),)))
-        rtol = None
+        self.common(fn, ((torch.rand((14923), dtype=torch.float32),)))
-        if self.device == "cpu" and os.getenv("ATEN_CPU_CAPABILITY") == "default":
-            atol = 1e-4
-            rtol = 1e-4
-        self.common(
-            fn,
-            ((torch.rand((10, 3, 352, 352), dtype=torch.float32),)),
-            rtol=rtol,
-            atol=atol,
-        )
-        self.common(
-            fn, ((torch.rand((14923), dtype=torch.float32),)), rtol=rtol, atol=atol
-        )
 
     @skipCPUIf(IS_MACOS, "fails on macos")
     def test_multilayer_var_lowp(self):
         def fn(a):
             return torch.var(a)
 
-        atol = None
+        self.common(fn, (torch.rand((16, 16, 352, 352), dtype=torch.float16),))
-        rtol = None
+        self.common(fn, (torch.rand((14923), dtype=torch.float16),))
-        if self.device == "cpu" and os.getenv("ATEN_CPU_CAPABILITY") == "default":
-            atol = 1e-3
-            rtol = 1e-3
-        self.common(
-            fn,
-            (torch.rand((16, 16, 352, 352), dtype=torch.float16),),
-            rtol=rtol,
-            atol=atol,
-        )
-        self.common(
-            fn, (torch.rand((14923), dtype=torch.float16),), rtol=rtol, atol=atol
-        )
 
     def test_split_cumsum(self):
         def fn(a):
@@ -8227,7 +8199,7 @@ class CommonTemplate:
             rand_strided(shape, stride, dtype).requires_grad_(True).add(1)
             for shape, stride, dtype in args
         ]
-        self.common(forward, args, atol=1e-05, rtol=1e-05)
+        self.common(forward, args)
 
     @requires_gpu()
     def test_tmp_not_defined_issue3(self):
@@ -9309,7 +9281,6 @@ class CommonTemplate:
     # To support this behavior, we need to allow const-propping tensors that store symint data.
     # For now, dynamo will explicitly graph break when it encounters user code with this behavior.
     @expectedFailureCodegenDynamic
-    @expectedFailureScalar
     def test_AllenaiLongformerBase_repro(self):
         def fn(query, scores, window_overlap):
             batch_size, seq_len, num_heads, _ = query.size()
@@ -9345,9 +9316,6 @@ class CommonTemplate:
             opt_fn = torch._dynamo.optimize("inductor")(fn)
             _, code = run_and_get_cpp_code(opt_fn, *args)
             print(code)
-            # When testing the scalar version, i.e., ATEN_CPU_CAPABILITY=default,
-            # static_cast<int>(256) is not found, but static_cast<int64_t>(256).
-            # See https://github.com/pytorch/pytorch/issues/126262.
             FileCheck().check_count(
                 "static_cast<int32_t>(256)",
                 1,

test/test_custom_ops.py

@@ -2291,10 +2291,13 @@ class TestCustomOpAPI(TestCase):
         class Stack(torch.autograd.Function):
             @staticmethod
             def forward(ctx, xs):
+                ctx.num_xs = len(xs)
                 return torch.stack(xs)
 
             @staticmethod
             def backward(ctx, grad):
+                expected = ([True] * ctx.num_xs,)
+                self.assertEqual(ctx.needs_input_grad, expected)
                 return list(grad.unbind(0))
 
         # call two applys, do a backward on the first
@@ -2327,19 +2330,21 @@ class TestCustomOpAPI(TestCase):
         class Foo(torch.autograd.Function):
             @staticmethod
             def forward(ctx, xs):
-                if len(xs) > 0:
+                if len(xs) > 1:
                     return Foo.apply(xs[1:])
                 ctx.len_xs = len(xs)
-                return x.sin()
+                return xs[0].sin()
 
             @staticmethod
             def backward(ctx, grad):
-                result = [None] * len_xs
+                result = [None] * ctx.len_xs
                 result[-1] = grad.cos()
                 return result
 
-        with self.assertRaisesRegex(NotImplementedError, "Recursive call"):
+        # should work
-            Foo.apply(xs)
+        result = Foo.apply(xs)
+        expected = xs[-1].sin()
+        self.assertEqual(result, expected)
 
         # recursive on backward
         @torch._library.autograd.supports_tensorlist

test/test_flop_counter.py

@@ -9,7 +9,6 @@ import torch.utils.flop_counter
 from torch.testing._internal.common_cuda import (
     PLATFORM_SUPPORTS_FLASH_ATTENTION,
     PLATFORM_SUPPORTS_MEM_EFF_ATTENTION,
-    PLATFORM_SUPPORTS_CUDNN_ATTENTION
 )
 from torch.testing._internal.common_utils import (
     run_tests,
@@ -301,8 +300,7 @@ class TestFlopCounter(TestCase):
     @unittest.skipIf(not HAS_CUDA, "CUDA not available")
     @unittest.skipIf(
         not PLATFORM_SUPPORTS_FLASH_ATTENTION
-        or not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION
+        or not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION,
-        or not PLATFORM_SUPPORTS_CUDNN_ATTENTION,
         "Does not support all SDPA backends (pre-SM80 hardware on CUDA)",
     )
     def test_sdpa(self):
@@ -357,31 +355,15 @@ class TestFlopCounter(TestCase):
 
             if backend == "math":
                 backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=False,
+                    enable_flash=False, enable_math=True, enable_mem_efficient=False
-                    enable_math=True,
-                    enable_mem_efficient=False,
-                    enable_cudnn=False,
                 )
             elif backend == "flash":
                 backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=True,
+                    enable_flash=True, enable_math=False, enable_mem_efficient=False
-                    enable_math=False,
-                    enable_mem_efficient=False,
-                    enable_cudnn=False,
                 )
             elif backend == "mem_efficient":
                 backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=False,
+                    enable_flash=False, enable_math=False, enable_mem_efficient=True
-                    enable_math=False,
-                    enable_mem_efficient=True,
-                    enable_cudnn=False,
-                )
-            elif backend == "cudnn":
-                backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=False,
-                    enable_math=False,
-                    enable_mem_efficient=False,
-                    enable_cudnn=True,
                 )
 
             mode = FlopCounterMode()
@@ -407,24 +389,22 @@ class TestFlopCounter(TestCase):
 
         flops = [
             run_uniform_flops(backend, with_backward=False)
-            for backend in ["math", "flash", "mem_efficient", "cudnn"]
+            for backend in ["math", "flash", "mem_efficient"]
         ]
-        flops_fw_math, flops_fw_flash, flops_fw_efficient, flops_fw_cudnn = flops
+        flops_fw_math, flops_fw_flash, flops_fw_efficient = flops
         self.assertEqual(flops_fw_math, flops_fw_flash)
         self.assertEqual(flops_fw_math, flops_fw_efficient)
-        self.assertEqual(flops_fw_math, flops_fw_cudnn)
 
         self.assertExpectedInline(str(flops_fw_math), """134217728""")
 
         flops = [
             run_uniform_flops(backend, with_backward=True)
-            for backend in ["math", "flash", "mem_efficient", "cudnn"]
+            for backend in ["math", "flash", "mem_efficient"]
         ]
-        flops_fw_bw_math, flops_fw_bw_flash, flops_fw_bw_efficient, flops_fw_bw_cudnn = flops
+        flops_fw_bw_math, flops_fw_bw_flash, flops_fw_bw_efficient = flops
         self.assertEqual(flops_fw_math * 3, flops_fw_bw_math)
         self.assertEqual(flops_fw_math * 7 // 2, flops_fw_bw_flash)
         self.assertEqual(flops_fw_bw_flash, flops_fw_bw_efficient)
-        self.assertEqual(flops_fw_bw_flash, flops_fw_bw_cudnn)
 
         run_nonuniform_flops = functools.partial(
             get_flops,
@@ -468,24 +448,15 @@ class TestFlopCounter(TestCase):
 
             if backend == "math":
                 backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=False,
+                    enable_flash=False, enable_math=True, enable_mem_efficient=False
-                    enable_math=True,
-                    enable_mem_efficient=False,
-                    enable_cudnn=False,
                 )
             elif backend == "flash":
                 backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=True,
+                    enable_flash=True, enable_math=False, enable_mem_efficient=False
-                    enable_math=False,
-                    enable_mem_efficient=False,
-                    enable_cudnn=False,
                 )
             elif backend == "mem_efficient":
                 backend = torch.backends.cuda.sdp_kernel(
-                    enable_flash=False,
+                    enable_flash=False, enable_math=False, enable_mem_efficient=True
-                    enable_math=False,
-                    enable_mem_efficient=True,
-                    enable_cudnn=False,
                 )
 
             with backend, mode:

test/test_optim.py

@@ -287,7 +287,7 @@ class TestOptimRenewed(TestCase):
             inpt = torch.randn(5, device=device, dtype=dtype)
 
             # avoid endless recompiles by wrapping LR in a tensor if we're compiling
-            lr = torch.tensor(0.01) if torch.compiler.is_compiling() else 0.01
+            lr = torch.tensor(0.01) if torch._utils.is_compiling() else 0.01
             optimizer = optim_cls([{"params": [weight]}, {"params": [bias], "lr": lr}])
             schedulers = [scheduler_c(optimizer) for scheduler_c in schedulers_c]
 

test/test_transformers.py

@@ -43,8 +43,7 @@ from torch.testing._internal.common_cuda import (
     IS_JETSON, SM80OrLater, PLATFORM_SUPPORTS_FLASH_ATTENTION,
     PLATFORM_SUPPORTS_MEM_EFF_ATTENTION,
     PLATFORM_SUPPORTS_FUSED_ATTENTION,
-    PLATFORM_SUPPORTS_CUDNN_ATTENTION,
+    PLATFORM_SUPPORTS_CUDNN_ATTENTION
-    tf32_on_and_off
 )
 
 if TEST_FAIRSEQ:
@@ -316,7 +315,6 @@ class TestTransformers(NNTestCase):
         with torch.no_grad():
             model(src, src_mask=src_mask)
 
-    @tf32_on_and_off(0.001)
     @parametrize("use_torchscript", [False])
     @parametrize("enable_nested_tensor", [True, False])
     @parametrize("use_autocast", [True, False])
@@ -407,9 +405,8 @@ class TestTransformers(NNTestCase):
                 # no garauntees on output corresponding to masked tokens, so they may vary between slow/fast path. set all to 0.
                 fastpath_output_expanded = fastpath_output_expanded.masked_fill(src_key_padding_mask.unsqueeze(-1), 0)
                 slowpath_output = slowpath_output.masked_fill(src_key_padding_mask.unsqueeze(-1), 0)
-                self.assertEqual(fastpath_output_expanded, slowpath_output)
+                torch.testing.assert_close(fastpath_output_expanded, slowpath_output, rtol=1e-7, atol=1e-5)
 
-    @tf32_on_and_off(0.001)
     @parametrize("with_no_grad", [True, False])
     @parametrize("training", [True, False])
     @parametrize("enable_nested_tensor", [False])
@@ -453,7 +450,7 @@ class TestTransformers(NNTestCase):
                                     [2.419836044311523, 0.017548924311996, -0.608187675476074, -0.085347734391689]]]
                                   ).to(device)
         self.assertEqual(tuple(result.shape), tuple(ref_output.shape))
-        self.assertEqual(result, ref_output)
+        torch.testing.assert_close(result, ref_output, rtol=1e-7, atol=1e-5)
 
     @parametrize("batch_first", [True, False])
     @parametrize("training", [True, False])
@@ -1400,7 +1397,7 @@ class TestSDPAFailureModes(NNTestCase):
             q = torch.randn(size, device=device, dtype=dtype)
             k = torch.randn(size, device=device, dtype=dtype)
             v = torch.randn(size, device=device, dtype=dtype)
-            with self.assertWarnsRegex(UserWarning, "All fused kernels requires query, key and value to be 4 dimensional"):
+            with self.assertWarnsRegex(UserWarning, "Both fused kernels requires query, key and value to be 4 dimensional"):
                 self.assertRaises(RuntimeError, lambda: torch.nn.functional.scaled_dot_product_attention(
                     q, k, v, None, 0.0, False))
 
@@ -1432,7 +1429,7 @@ class TestSDPAFailureModes(NNTestCase):
             make_tensor = partial(torch.rand, device=device, dtype=dtype)
             size = SdpaShape(2, 2, 0, 8)
             q, k, v = make_tensor(size), make_tensor(size), make_tensor(size)
-            with self.assertWarnsRegex(UserWarning, "All fused kernels do not support zero seq_len_q or seq_len_kv."):
+            with self.assertWarnsRegex(UserWarning, "Both fused kernels do not support zero seq_len_q or seq_len_kv."):
                 self.assertRaises(RuntimeError, lambda: torch.nn.functional.scaled_dot_product_attention(
                     q, k, v, None, 0.0, False))
 
@@ -1447,7 +1444,7 @@ class TestSDPAFailureModes(NNTestCase):
             size = SdpaShape(2, 2, 8, 8)
             q, k, v = make_tensor(size), make_tensor(size), make_tensor(size)
             q.as_strided_(size, [2, 2, 2, 2])
-            with self.assertWarnsRegex(UserWarning, "All fused kernels require the last dimension of the input to have stride 1."):
+            with self.assertWarnsRegex(UserWarning, "Both fused kernels require the last dimension of the input to have stride 1."):
                 self.assertRaises(RuntimeError, lambda: torch.nn.functional.scaled_dot_product_attention(
                     q, k, v, None, 0.0, False))
 
@@ -2356,7 +2353,7 @@ class TestSDPACudaOnly(NNTestCase):
         math_ref_lp_test = math_ref_lp_test.to(dtype=torch.float32).contiguous()
 
         self.assertEqual(math_ref_test, math_ref_lp_test, atol=7e-3, rtol=7e-3)
-        self.assertEqual(actual_test, math_ref_test, atol=7e-3, rtol=7e-3)
+        self.assertEqual(actual_test, math_ref_test, atol=5e-3, rtol=5e-3)
 
     @unittest.skipIf(not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Efficient Attention was not built for this system")
     @parametrize("contiguous_inputs", [True, False])
@@ -2474,12 +2471,7 @@ class TestSDPACudaOnly(NNTestCase):
         value = value.view(batch_size, -1, num_heads, head_dim).transpose(1, 2)
         key = key.view(batch_size, -1, num_heads, head_dim).transpose(1, 2)
 
-        major, minor = torch.cuda.get_device_capability(device)
+        if PLATFORM_SUPPORTS_FLASH_ATTENTION:
-        is_sm90_or_newer = major >= 9
-
-        if type != "nested" and PLATFORM_SUPPORTS_CUDNN_ATTENTION and is_sm90_or_newer:
-            assert torch._fused_sdp_choice(query, key, value) == SDPBackend.CUDNN_ATTENTION.value
-        elif PLATFORM_SUPPORTS_FLASH_ATTENTION:
             assert torch._fused_sdp_choice(query, key, value) == SDPBackend.FLASH_ATTENTION.value
         else:
             assert torch._fused_sdp_choice(query, key, value) == SDPBackend.EFFICIENT_ATTENTION.value
@@ -2519,8 +2511,7 @@ class TestSDPACudaOnly(NNTestCase):
         make_tensor = partial(rand_sdpa_tensor, type="dense", device=device, dtype=torch.float16, packed=False, requires_grad=True)
         query, key, value = make_tensor(shape), make_tensor(shape), make_tensor(shape)
 
-        kernel_name = "Memory Efficient attention" if fused_kernel == SDPBackend.EFFICIENT_ATTENTION else \
+        kernel_name = "Memory Efficient attention" if fused_kernel == SDPBackend.EFFICIENT_ATTENTION else "Flash Attention"
-            "Flash Attention" if fused_kernel == SDPBackend.FLASH_ATTENTION else "cuDNN Attention"
         warning_context = (
             self.assertWarnsRegex(
                 UserWarning,
@@ -2532,12 +2523,7 @@ class TestSDPACudaOnly(NNTestCase):
         with use_deterministic_algorithims(True, warn_only=warn_only):
             with sdpa_kernel(backends=[fused_kernel]):
                 with warning_context:
-                    if warn_only or fused_kernel != SDPBackend.CUDNN_ATTENTION:
+                    torch.nn.functional.scaled_dot_product_attention(query, key, value).sum().backward()
-                        torch.nn.functional.scaled_dot_product_attention(query, key, value).sum().backward()
-                    else:
-                        # cuDNN attention has no deterministic fallback
-                        self.assertRaises(RuntimeError, lambda:
-                                          torch.nn.functional.scaled_dot_product_attention(query, key, value).sum().backward())
 
     @unittest.skip("This test is not behaving deterministaclly non-deterministaclly on CI/CD")
     @unittest.skipIf(not PLATFORM_SUPPORTS_FLASH_ATTENTION, "Platform does not support fused SDPA")
@@ -2677,7 +2663,7 @@ class TestSDPACudaOnly(NNTestCase):
         output_ref_atol, output_ref_rtol = get_tolerances(out_ref, out_lp_ref)
 
         # Fudge Factor when dropout is enabled
-        dropout_fudge_factor = 1.5 if dropout_p == 0.0 else 2.0
+        dropout_fudge_factor = 1.0 if dropout_p == 0.0 else 2.0
 
         query_fudge_factor = dropout_fudge_factor
         grad_q_ref_atol, grad_q_ref_rtol = get_tolerances(query_ref.grad, query_ref_lp.grad, query_fudge_factor)
@@ -2800,8 +2786,8 @@ class TestSDPACudaOnly(NNTestCase):
         # Fudge Factor when dropout is enabled
         dropout_fudge_factor = 1.0 if dropout_p == 0.0 else 1.75
         mask_fudge_factor = 1.0 if attn_mask is None else 1.5
-        query_fudge_factor = 2.0
 
+        query_fudge_factor = dropout_fudge_factor
         grad_q_ref_atol, grad_q_ref_rtol = get_tolerances(query_ref.grad, query_ref_lp.grad, query_fudge_factor)
 
         # TODO: Investigate why grad_k needs larger tolerances
@@ -3003,8 +2989,7 @@ class TestSDPACudaOnly(NNTestCase):
                            device=device, dtype=dtype, requires_grad=True)
 
         fused_op = (torch.ops.aten._scaled_dot_product_efficient_attention
-                    if fused_kernel == SDPBackend.EFFICIENT_ATTENTION else torch.ops.aten._scaled_dot_product_flash_attention
+                    if fused_kernel == SDPBackend.EFFICIENT_ATTENTION else torch.ops.aten._scaled_dot_product_flash_attention)
-                    if fused_kernel == SDPBackend.FLASH_ATTENTION else torch.ops.aten._scaled_dot_product_cudnn_attention)
         # Run the math kernel on low precision references
         query_ref_lp, key_ref_lp, value_ref_lp = query_key_value_clones(query, key, value, dtype=dtype)
 
@@ -3022,10 +3007,6 @@ class TestSDPACudaOnly(NNTestCase):
             kwargs["attn_bias"] = None
         if fused_kernel == SDPBackend.FLASH_ATTENTION:
             kwargs['return_debug_mask'] = dropout_p > 0.0
-        if fused_kernel == SDPBackend.CUDNN_ATTENTION:
-            kwargs["compute_log_sumexp"] = True
-            if "return_debug_mask" in kwargs:
-                kwargs.pop("return_debug_mask")
         with torch.cuda.stream(s):
             # Create real output
             output_tuple = fused_op(query, key, value, **kwargs)
@@ -3063,8 +3044,7 @@ class TestSDPACudaOnly(NNTestCase):
                 # Low Precision Math Reference
                 out_lp_ref = F.scaled_dot_product_attention(query_ref_lp, key_ref_lp, value_ref_lp,
                                                             dropout_p=dropout_p, is_causal=is_causal, scale=scale)
-            # cuDNN attention doesn't support returning dropout mask
+            else:
-            elif fused_kernel != SDPBackend.CUDNN_ATTENTION:
                 # Create the dropout_mask
                 dropout_mask = get_dropout_mask(output_tuple, fused_kernel, batch_size,
                                                 n_heads, seq_len_q, seq_len_k, dropout_p, device)
@@ -3082,38 +3062,37 @@ class TestSDPACudaOnly(NNTestCase):
         with torch.cuda.graph(g1):
             out.backward(upstream_grad)
         g1.replay()
-        if fused_kernel != SDPBackend.CUDNN_ATTENTION or dropout_p == 0.0:
+        out_ref.backward(upstream_grad.to(out_ref.dtype))
-            out_ref.backward(upstream_grad.to(out_ref.dtype))
+        out_lp_ref.backward(upstream_grad.to(out_lp_ref.dtype))
-            out_lp_ref.backward(upstream_grad.to(out_lp_ref.dtype))
 
-            # [Note] Fused Tolerances
+        # [Note] Fused Tolerances
-            # Establish the numerical error between the "true" high precision math output
+        # Establish the numerical error between the "true" high precision math output
-            # and the low precision math reference. We use this reference for the atol
+        # and the low precision math reference. We use this reference for the atol
-            # And we use the default rtol for the low precision type.
+        # And we use the default rtol for the low precision type.
-            # We then provide a fudge factor for gradients respectively to account
+        # We then provide a fudge factor for gradients respectively to account
-            # for the use of the fused kernel rather than the eager implemntation.
+        # for the use of the fused kernel rather than the eager implemntation.
-            output_ref_atol, output_ref_rtol = get_tolerances(out_ref, out_lp_ref)
+        output_ref_atol, output_ref_rtol = get_tolerances(out_ref, out_lp_ref)
 
-            # Fudge Factor when dropout is enabled
+        # Fudge Factor when dropout is enabled
-            dropout_fudge_factor = 1.0 if dropout_p == 0.0 else 1.5
+        dropout_fudge_factor = 1.0 if dropout_p == 0.0 else 1.5
 
-            query_fudge_factor = dropout_fudge_factor
+        query_fudge_factor = dropout_fudge_factor
-            grad_q_ref_atol, grad_q_ref_rtol = get_tolerances(query_ref.grad, query_ref_lp.grad, query_fudge_factor)
+        grad_q_ref_atol, grad_q_ref_rtol = get_tolerances(query_ref.grad, query_ref_lp.grad, query_fudge_factor)
 
-            # TODO: Investigate why grad_k needs larger tolerances
+        # TODO: Investigate why grad_k needs larger tolerances
-            key_fudge_factor = 8 * dropout_fudge_factor
+        key_fudge_factor = 8 * dropout_fudge_factor
-            grad_k_ref_atol, grad_k_ref_rtol = get_tolerances(key_ref.grad, key_ref_lp.grad, key_fudge_factor)
+        grad_k_ref_atol, grad_k_ref_rtol = get_tolerances(key_ref.grad, key_ref_lp.grad, key_fudge_factor)
 
-            value_fudge_factor = 7 if not SM80OrLater and dtype == torch.float16 else 1.0
+        value_fudge_factor = 7 if not SM80OrLater and dtype == torch.float16 else 1.0
-            grad_v_ref_atol, grad_v_ref_rtol = get_tolerances(value_ref.grad, value_ref_lp.grad, value_fudge_factor)
+        grad_v_ref_atol, grad_v_ref_rtol = get_tolerances(value_ref.grad, value_ref_lp.grad, value_fudge_factor)
 
-            self.assertEqual(out, out_ref.to(out.dtype), atol=output_ref_atol, rtol=output_ref_rtol)
+        self.assertEqual(out, out_ref.to(out.dtype), atol=output_ref_atol, rtol=output_ref_rtol)
-            self.assertEqual(query.grad, query_ref.grad.to(query.grad.dtype),
+        self.assertEqual(query.grad, query_ref.grad.to(query.grad.dtype),
-                             atol=grad_q_ref_atol, rtol=grad_q_ref_rtol)
+                         atol=grad_q_ref_atol, rtol=grad_q_ref_rtol)
-            self.assertEqual(key.grad, key_ref.grad.to(key.grad.dtype),
+        self.assertEqual(key.grad, key_ref.grad.to(key.grad.dtype),
-                             atol=grad_k_ref_atol, rtol=grad_k_ref_rtol)
+                         atol=grad_k_ref_atol, rtol=grad_k_ref_rtol)
-            self.assertEqual(value.grad, value_ref.grad.to(value.grad.dtype),
+        self.assertEqual(value.grad, value_ref.grad.to(value.grad.dtype),
-                             atol=grad_v_ref_atol, rtol=grad_v_ref_rtol)
+                         atol=grad_v_ref_atol, rtol=grad_v_ref_rtol)
 
     @skipIfRocm  # Nested Tensor
     @unittest.skipIf(not PLATFORM_SUPPORTS_FUSED_ATTENTION, "Fused SDPA was not built for this system")
@@ -3237,7 +3216,7 @@ class TestSDPACudaOnly(NNTestCase):
                 query_expanded.contiguous(), key_expanded.contiguous(), value_expanded.contiguous(),
                 attn_mask=None, dropout_p=0.0, is_causal=False)
 
-        self.assertEqual(actual.contiguous(), math_ref.contiguous().to(dtype), atol=1.5e-3, rtol=1e-2)
+        self.assertEqual(actual.contiguous(), math_ref.contiguous().to(dtype), atol=1e-3, rtol=1e-2)
 
     @skipIfRocm  # Nested tensor
     @unittest.skipIf(not PLATFORM_SUPPORTS_MEM_EFF_ATTENTION, "Fused SDPA was not built for this system")
@@ -3400,7 +3379,6 @@ class TestAttnBias(NNTestCase):
         forw_tolerances: Optional[Tolerances] = None,
         grad_tolerances: Optional[Tolerances] = None,
         backend=None,
-        causal_variant=None,
     ):
         if backend is not None:
             torch._dynamo.reset()
@@ -3468,11 +3446,9 @@ class TestAttnBias(NNTestCase):
         if causal_variant == CausalVariant.UPPER_LEFT:
             attn_bias = causal_upper_left(seq_len_q, seq_len_kv)
         else:
-            print(seq_len_q, seq_len_kv)
             attn_bias = causal_lower_right(seq_len_q, seq_len_kv)
 
-        with sdpa_kernel(backends=[SDPBackend.EFFICIENT_ATTENTION, SDPBackend.FLASH_ATTENTION, SDPBackend.MATH]):
+        self.run_test(device, make_q_tensor, make_kv_tensor, attn_bias, forw_tol, grad_tol, backend=None)
-            self.run_test(device, make_q_tensor, make_kv_tensor, attn_bias, forw_tol, grad_tol, backend=None)
 
     @skipIfRocm  # CausalVariant
     @parametrize("causal_variant", [CausalVariant.UPPER_LEFT, CausalVariant.LOWER_RIGHT])
@@ -3503,8 +3479,7 @@ class TestAttnBias(NNTestCase):
         else:
             attn_bias = causal_lower_right(seq_len_q, seq_len_kv)
 
-        with sdpa_kernel(backends=[SDPBackend.EFFICIENT_ATTENTION, SDPBackend.FLASH_ATTENTION, SDPBackend.MATH]):
+        self.run_test(device, make_q_tensor, make_kv_tensor, attn_bias, forw_tol, grad_tol, backend=cnts)
-            self.run_test(device, make_q_tensor, make_kv_tensor, attn_bias, forw_tol, grad_tol, backend=cnts)
         self.assertEqual(cnts.frame_count, 1, "Compiled graph should have 1 frame!")
 
     @parametrize("shape", [(16, 16, 128, 128, 16), (16, 16, 128, 256, 32), (16, 16, 256, 128, 32), (1, 1, 23, 56, 15)])