Put back "[Release only] Release 2.3 start using triton package from pypi"" (#127290)

Revert "Revert "[Release only] Release 2.3 start using triton package from py…"

This reverts commit 194698a4ac82c485acbe6d2f40b7de063e28aa8f.

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

@@ -1 +1 @@
-0a22a91d04c2b4a029a69a198eac390089c3e891
+c8ad905211f45e162102823149f0d7f2cfaa4418

.ci/docker/ci_commit_pins/triton.txt

@@ -1 +1 @@
-a9bc1a36470eefafe0e2ab2503b8698f1e89e7e3
+958fccea74da58e7e0595ab88ae6cd3f6795a173

.ci/docker/common/install_conda.sh

@@ -57,8 +57,21 @@ fi
   # Uncomment the below when resolved to track the latest conda update
   # as_jenkins conda update -y -n base conda
 
+  if [[ $(uname -m) == "aarch64" ]]; then
+    export SYSROOT_DEP="sysroot_linux-aarch64=2.17"
+  else
+    export SYSROOT_DEP="sysroot_linux-64=2.17"
+  fi
+
   # Install correct Python version
-  as_jenkins conda create -n py_$ANACONDA_PYTHON_VERSION -y python="$ANACONDA_PYTHON_VERSION"
+  # Also ensure sysroot is using a modern GLIBC to match system compilers
+  as_jenkins conda create -n py_$ANACONDA_PYTHON_VERSION -y\
+             python="$ANACONDA_PYTHON_VERSION" \
+             ${SYSROOT_DEP}
+
+  # libstdcxx from conda default channels are too old, we need GLIBCXX_3.4.30
+  # which is provided in libstdcxx 12 and up.
+  conda_install libstdcxx-ng=12.3.0 -c conda-forge
 
   # Install PyTorch conda deps, as per https://github.com/pytorch/pytorch README
   if [[ $(uname -m) == "aarch64" ]]; then
@@ -110,14 +123,5 @@ fi
     pip_install -r /opt/conda/requirements-docs.txt
   fi
 
-  # HACK HACK HACK
-  # gcc-9 for ubuntu-18.04 from http://ppa.launchpad.net/ubuntu-toolchain-r/test/ubuntu
-  # Pulls llibstdc++6 13.1.0-8ubuntu1~18.04 which is too new for conda
-  # So remove libstdc++6.so.3.29 installed by https://anaconda.org/anaconda/libstdcxx-ng/files?version=11.2.0
-  # Same is true for gcc-12 from Ubuntu-22.04
-  if grep -e [12][82].04.[623] /etc/issue >/dev/null; then
-    rm /opt/conda/envs/py_$ANACONDA_PYTHON_VERSION/lib/libstdc++.so.6
-  fi
-
   popd
 fi

.ci/docker/common/install_rocm.sh

@@ -84,7 +84,11 @@ install_ubuntu() {
     if [[ $(ver $ROCM_VERSION) -ge $(ver 6.0) ]]; then
         for kdb in /opt/rocm/share/miopen/db/*.kdb
         do
-            sqlite3 $kdb "PRAGMA journal_mode=off; PRAGMA VACUUM;"
+          # journal_mode=delete seems to work on some kdbs that have "wal" as initial journal_mode
+          sqlite3 $kdb "PRAGMA journal_mode=delete; PRAGMA VACUUM;"
+          JOURNAL_MODE=$(sqlite3 $kdb "PRAGMA journal_mode;")
+          # Both "delete and "off" work in cases where user doesn't have write permissions to directory where kdbs are installed
+          if [[ $JOURNAL_MODE != "delete" ]] && [[ $JOURNAL_MODE != "off" ]]; then echo "kdb journal_mode change failed" && exit 1; fi
         done
     fi
 
@@ -163,7 +167,11 @@ install_centos() {
   if [[ $(ver $ROCM_VERSION) -ge $(ver 6.0) ]]; then
       for kdb in /opt/rocm/share/miopen/db/*.kdb
       do
-          sqlite3 $kdb "PRAGMA journal_mode=off; PRAGMA VACUUM;"
+        # journal_mode=delete seems to work on some kdbs that have "wal" as initial journal_mode
+        sqlite3 $kdb "PRAGMA journal_mode=delete; PRAGMA VACUUM;"
+        JOURNAL_MODE=$(sqlite3 $kdb "PRAGMA journal_mode;")
+        # Both "delete" and "off" work in cases where user doesn't have write permissions to directory where kdbs are installed
+        if [[ $JOURNAL_MODE != "delete" ]] && [[ $JOURNAL_MODE != "off" ]]; then echo "kdb journal_mode change failed" && exit 1; fi
       done
   fi
 

.ci/docker/triton_version.txt

@@ -1 +1 @@
-3.0.0
+2.3.1

.ci/pytorch/build.sh

@@ -255,6 +255,11 @@ else
     # or building non-XLA tests.
     if [[ "$BUILD_ENVIRONMENT" != *rocm*  &&
           "$BUILD_ENVIRONMENT" != *xla* ]]; then
+      if [[ "$BUILD_ENVIRONMENT" != *py3.8* ]]; then
+        # Install numpy-2.0 release candidate for builds
+        # Which should be backward compatible with Numpy-1.X
+        python -mpip install --pre numpy==2.0.0rc1
+      fi
       WERROR=1 python setup.py bdist_wheel
     else
       python setup.py bdist_wheel

.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 r2.3 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)"

.ci/pytorch/python_doc_push_script.sh

@@ -26,8 +26,8 @@ echo "error: python_doc_push_script.sh: version (arg2) not specified"
 fi
 
 # Argument 1: Where to copy the built documentation to
-# (pytorch.github.io/$install_path)
-install_path="${1:-${DOCS_INSTALL_PATH:-docs/${DOCS_VERSION}}}"
+# (pytorch_docs/$install_path)
+install_path="${1:-${DOCS_INSTALL_PATH:-${DOCS_VERSION}}}"
 if [ -z "$install_path" ]; then
 echo "error: python_doc_push_script.sh: install_path (arg1) not specified"
   exit 1
@@ -68,8 +68,8 @@ build_docs () {
 }
 
 
-git clone https://github.com/pytorch/pytorch.github.io -b "$branch" --depth 1
-pushd pytorch.github.io
+git clone https://github.com/pytorch/docs pytorch_docs -b "$branch" --depth 1
+pushd pytorch_docs
 
 export LC_ALL=C
 export PATH=/opt/conda/bin:$PATH

.ci/pytorch/test.sh

@@ -302,6 +302,7 @@ test_inductor_distributed() {
   pytest test/distributed/_composable/fsdp/test_fully_shard_frozen.py
   pytest test/distributed/_composable/fsdp/test_fully_shard_mixed_precision.py -k test_compute_dtype
   pytest test/distributed/_composable/fsdp/test_fully_shard_mixed_precision.py -k test_reduce_dtype
+  pytest test/distributed/fsdp/test_fsdp_tp_integration.py -k test_fsdp_tp_integration
 
   # this runs on both single-gpu and multi-gpu instance. It should be smart about skipping tests that aren't supported
   # with if required # gpus aren't available

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-707a632930bfde19ffb361cdf5c31a7682af4e67
+r2.3

.github/requirements/pip-requirements-macOS.txt

@@ -27,3 +27,6 @@ rockset==1.0.3
 z3-solver==4.12.2.0
 tensorboard==2.13.0
 optree==0.9.1
+# NB: test_hparams_* from test_tensorboard is failing with protobuf 5.26.0 in
+# which the stringify metadata is wrong when escaping double quote
+protobuf==3.20.2

.github/scripts/build_triton_wheel.py

@@ -10,9 +10,6 @@ from typing import Optional
 SCRIPT_DIR = Path(__file__).parent
 REPO_DIR = SCRIPT_DIR.parent.parent
 
-# TODO: Remove me once Triton version is again in sync for vanilla and ROCm
-ROCM_TRITION_VERSION = "2.1.0"
-
 
 def read_triton_pin(rocm_hash: bool = False) -> str:
     triton_file = "triton.txt" if not rocm_hash else "triton-rocm.txt"
@@ -99,7 +96,14 @@ def build_triton(
             triton_repo = "https://github.com/openai/triton"
             triton_pkg_name = "pytorch-triton"
         check_call(["git", "clone", triton_repo], cwd=tmpdir)
-        check_call(["git", "checkout", commit_hash], cwd=triton_basedir)
+        if release:
+            ver, rev, patch = version.split(".")
+            check_call(
+                ["git", "checkout", f"release/{ver}.{rev}.x"], cwd=triton_basedir
+            )
+        else:
+            check_call(["git", "checkout", commit_hash], cwd=triton_basedir)
+
         if build_conda:
             with open(triton_basedir / "meta.yaml", "w") as meta:
                 print(
@@ -109,7 +113,7 @@ def build_triton(
                 print("source:\n  path: .\n", file=meta)
                 print(
                     "build:\n  string: py{{py}}\n  number: 1\n  script: cd python; "
-                    "python setup.py install --record=record.txt\n",
+                    "python setup.py install --single-version-externally-managed --record=record.txt\n",
                     " script_env:\n   - MAX_JOBS\n",
                     file=meta,
                 )
@@ -155,7 +159,7 @@ def build_triton(
         patch_init_py(
             triton_pythondir / "triton" / "__init__.py",
             version=f"{version}",
-            expected_version=ROCM_TRITION_VERSION if build_rocm else None,
+            expected_version=None,
         )
 
         if build_rocm:
@@ -164,7 +168,7 @@ def build_triton(
                 triton_pythondir / "setup.py",
                 name=triton_pkg_name,
                 version=f"{version}",
-                expected_version=ROCM_TRITION_VERSION,
+                expected_version=None,
             )
             check_call("scripts/amd/setup_rocm_libs.sh", cwd=triton_basedir, shell=True)
             print("ROCm libraries setup for triton installation...")

.github/scripts/filter_test_configs.py

@@ -62,9 +62,9 @@ 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=qO7aEr.Og33PtLXfNq0j0yj.bbLC7SzR"
 # 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=7NhgpqKTtGXVUnL1C79KboTW_5qQx8y5"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"

.github/scripts/generate_docker_release_matrix.py

@@ -21,6 +21,8 @@ DOCKER_IMAGE_TYPES = ["runtime", "devel"]
 
 def generate_docker_matrix() -> Dict[str, List[Dict[str, str]]]:
     ret: List[Dict[str, str]] = []
+    # CUDA amd64 Docker images are available as both runtime and devel while
+    # CPU arm64 image is only available as runtime.
     for cuda, version in generate_binary_build_matrix.CUDA_ARCHES_FULL_VERSION.items():
         for image in DOCKER_IMAGE_TYPES:
             ret.append(
@@ -31,9 +33,19 @@ def generate_docker_matrix() -> Dict[str, List[Dict[str, str]]]:
                         cuda
                     ],
                     "image_type": image,
-                    "platform": "linux/arm64,linux/amd64",
+                    "platform": "linux/amd64",
                 }
             )
+    ret.append(
+        {
+            "cuda": "cpu",
+            "cuda_full_version": "",
+            "cudnn_version": "",
+            "image_type": "runtime",
+            "platform": "linux/arm64",
+        }
+    )
+
     return {"include": ret}
 
 

.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/2.3" -%}
 
 {%- macro concurrency(build_environment) -%}
 concurrency:

.github/templates/linux_binary_build_workflow.yml.j2

@@ -100,8 +100,8 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: ROCm set GPU_FLAG
         run: |
           echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${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="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ 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, 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="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ 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, 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="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ 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, 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/2.3
         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/2.3
         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/2.3
 
       - 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/2.3
         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/2.3
         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/2.3
         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/2.3
         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/2.3
         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/2.3
 
       - 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/2.3
         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/2.3
         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/2.3
         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/2.3
         with:
           fetch-depth: 1
           submodules: false
@@ -64,30 +64,30 @@ 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/2.3
         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/2.3
 
       - 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/2.3
         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/2.3
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
       - 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/2.3
         if: ${{ inputs.cuda-version != 'cpu' }}
 
       - name: Output disk space left
@@ -196,5 +196,5 @@ 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/2.3
         if: always()

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

@@ -78,7 +78,7 @@ on:
 jobs:
   build:
     runs-on: ${{ inputs.runs_on }}
-    timeout-minutes: 180
+    timeout-minutes: 210
     env:
       PYTORCH_ROOT: ${{ inputs.PYTORCH_ROOT }}
       BUILDER_ROOT: ${{ inputs.BUILDER_ROOT }}
@@ -139,13 +139,13 @@ jobs:
         run: env
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.3
         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/2.3
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' }}
 
@@ -173,7 +173,6 @@ 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
@@ -187,7 +186,7 @@ jobs:
       - name: Checkout pytorch/builder to builder dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -213,7 +212,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.3
         with:
           docker-image: ${{ inputs.DOCKER_IMAGE }}
 
@@ -270,7 +269,7 @@ jobs:
 
       - name: Teardown Linux
         if: always()
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.3
 
       - name: Chown workspace
         if: always()

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

@@ -127,14 +127,14 @@ jobs:
           } >> "${GITHUB_ENV} }}"
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.3
         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/2.3
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' }}
 
@@ -155,7 +155,6 @@ 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
 
@@ -168,7 +167,7 @@ jobs:
       - name: Checkout pytorch/builder to builder dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -199,12 +198,12 @@ 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/2.3
         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' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.3
         with:
           docker-image: ${{ inputs.DOCKER_IMAGE }}
 
@@ -214,7 +213,7 @@ jobs:
 
       - name: Teardown Linux
         if: always()
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.3
 
       - name: Chown workspace
         if: always()

.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/2.3
         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/2.3
         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/2.3
 
       - 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/2.3
         with:
           python-version: 3.8
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}

.github/workflows/_docs.yml

@@ -66,7 +66,7 @@ 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/2.3
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -77,19 +77,19 @@ 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/2.3
 
       - 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/2.3
         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/2.3
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -163,7 +163,7 @@ jobs:
           retention-days: 14
           s3-bucket: doc-previews
           if-no-files-found: error
-          path: pytorch.github.io/docs/main/
+          path: pytorch_docs/main/
           s3-prefix: pytorch/pytorch/${{ github.event.pull_request.number }}
 
       - name: Upload C++ Docs Preview
@@ -187,5 +187,5 @@ 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/2.3
         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/2.3
         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/2.3
 
       - 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/2.3
         with:
           python-version: "3.9"
           environment-file: .github/requirements/conda-env-iOS.txt

.github/workflows/_linux-build.yml

@@ -73,7 +73,7 @@ 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/2.3
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
@@ -82,14 +82,14 @@ 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/2.3
 
       - 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/2.3
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
@@ -103,7 +103,7 @@ 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/2.3
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -209,5 +209,5 @@ jobs:
           path: sccache-stats-*.json
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.3
         if: always()

.github/workflows/_linux-test.yml

@@ -57,7 +57,7 @@ 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/2.3
         if: ${{ !contains(matrix.runner, 'gcp.a100') }}
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -66,14 +66,14 @@ 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/2.3
 
       - 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/2.3
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
@@ -87,13 +87,13 @@ 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/2.3
         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/2.3
         if: contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu')
 
       - name: Lock NVIDIA A100 40GB Frequency
@@ -307,7 +307,7 @@ 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/2.3
         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/2.3
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.3
 
       - 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/2.3
         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/2.3
         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/2.3

.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/2.3
         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/2.3
         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/2.3

.github/workflows/_mac-test.yml

@@ -79,11 +79,11 @@ 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/2.3
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.3
 
       - name: Download build artifacts
         uses: ./.github/actions/download-build-artifacts
@@ -98,7 +98,7 @@ 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/2.3
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -227,4 +227,4 @@ 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/2.3

.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/2.3
         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/2.3
         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/2.3
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 

.github/workflows/_run_android_tests.yml

@@ -23,7 +23,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.3
         with:
           fetch-depth: 1
           submodules: false
@@ -54,10 +54,10 @@ jobs:
       SUPPORT_ABI: '${{ matrix.support_abi }}'
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.3
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.3
         with:
           python-version: 3.8
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}.txt

.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/2.3
 
       - 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/2.3
         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/2.3
         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/2.3
 
       - 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/2.3
         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/2.3
         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/2.3
 
       - 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/2.3
         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/2.3
         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/2.3
     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/2.3
         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/2.3
         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/2.3
         with:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
@@ -125,7 +125,7 @@ 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/2.3
         if: always()
 
   upload-wheel:
@@ -203,12 +203,12 @@ 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/2.3
         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/2.3
         with:
           submodules: false
 
@@ -216,7 +216,7 @@ 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/2.3
         with:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
@@ -252,7 +252,7 @@ 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/2.3
         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/2.3
         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/2.3
 
       - name: Run close_nonexistent_disable_issues.py
         env:

.github/workflows/docker-builds.yml

@@ -74,21 +74,21 @@ 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/2.3
 
       - 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/2.3
         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/2.3
         with:
           docker-image: ${{ steps.build-docker-image.outputs.docker-image }}
 
@@ -120,5 +120,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.3
         if: always()

.github/workflows/docker-release.yml

@@ -7,6 +7,7 @@ on:
       - Dockerfile
       - docker.Makefile
       - .github/workflows/docker-release.yml
+      - .github/scripts/generate_docker_release_matrix.py
   push:
     branches:
       - nightly
@@ -40,7 +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/2.3
         with:
           fetch-depth: 1
           submodules: true
@@ -68,7 +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/2.3
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
       # [see note: pytorch repo ref]
@@ -129,17 +130,27 @@ jobs:
         if: ${{ github.event.ref == 'refs/heads/nightly' && matrix.image_type == 'runtime' }}
         run: |
           PYTORCH_DOCKER_TAG="${PYTORCH_VERSION}-cuda${CUDA_VERSION_SHORT}-cudnn${CUDNN_VERSION}-runtime"
+          CUDA_SUFFIX="-cu${CUDA_VERSION}"
+          if [[ ${CUDA_VERSION_SHORT} == "cpu" ]]; then
+            PYTORCH_DOCKER_TAG="${PYTORCH_VERSION}-runtime"
+            CUDA_SUFFIX=""
+          fi
 
           PYTORCH_NIGHTLY_COMMIT=$(docker run ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_DOCKER_TAG}" \
                                           python -c 'import torch; print(torch.version.git_version[:7],end="")')
 
           docker tag ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_DOCKER_TAG}" \
-                 ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}-cu${CUDA_VERSION}"
-          docker push ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}-cu${CUDA_VERSION}"
+                 ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}${CUDA_SUFFIX}"
+
+          docker push ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}${CUDA_SUFFIX}"
+
+          # Please note, here we ned to pin specific verison of CUDA as with latest label
+          if [[ ${CUDA_VERSION_SHORT} == "12.1" ]]; then
+            docker tag ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}${CUDA_SUFFIX}" \
+                    ghcr.io/pytorch/pytorch-nightly:latest
+            docker push ghcr.io/pytorch/pytorch-nightly:latest
+          fi
 
-          docker tag ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}-cu${CUDA_VERSION}" \
-                 ghcr.io/pytorch/pytorch-nightly:latest
-          docker push ghcr.io/pytorch/pytorch-nightly:latest
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.3
         if: always()

.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-aarch64-2.3
       DESIRED_PYTHON: "3.8"
       runs_on: linux.arm64.2xlarge
       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-aarch64-2.3
       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-aarch64-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-aarch64
     secrets:
@@ -111,7 +111,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.9"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -132,7 +132,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -154,7 +154,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
     secrets:
@@ -174,7 +174,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.10"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -195,7 +195,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -217,7 +217,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
     secrets:
@@ -237,7 +237,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.11"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -258,7 +258,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -280,7 +280,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
     secrets:
@@ -300,7 +300,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.12"
       runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
@@ -321,7 +321,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -343,7 +343,7 @@ 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-aarch64-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-aarch64
     secrets:

.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-2.3
       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-2.3
       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-2.3
       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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_1
     secrets:
@@ -233,7 +233,7 @@ 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-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       build_environment: linux-binary-conda
@@ -251,7 +251,7 @@ 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-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       build_environment: linux-binary-conda
@@ -272,7 +272,7 @@ 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-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
     secrets:
@@ -293,7 +293,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.9"
       runs_on: linux.24xlarge
       build_name: conda-py3_9-cuda11_8
@@ -313,7 +313,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
       build_environment: linux-binary-conda
@@ -335,7 +335,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
     secrets:
@@ -356,7 +356,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.9"
       runs_on: linux.24xlarge
       build_name: conda-py3_9-cuda12_1
@@ -376,7 +376,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
       build_environment: linux-binary-conda
@@ -398,7 +398,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
     secrets:
@@ -418,7 +418,7 @@ 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-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       build_environment: linux-binary-conda
@@ -436,7 +436,7 @@ 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-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       build_environment: linux-binary-conda
@@ -457,7 +457,7 @@ 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-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
     secrets:
@@ -478,7 +478,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.10"
       runs_on: linux.24xlarge
       build_name: conda-py3_10-cuda11_8
@@ -498,7 +498,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
       build_environment: linux-binary-conda
@@ -520,7 +520,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
     secrets:
@@ -541,7 +541,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.10"
       runs_on: linux.24xlarge
       build_name: conda-py3_10-cuda12_1
@@ -561,7 +561,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
       build_environment: linux-binary-conda
@@ -583,7 +583,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
     secrets:
@@ -603,7 +603,7 @@ 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-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       build_environment: linux-binary-conda
@@ -621,7 +621,7 @@ 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-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       build_environment: linux-binary-conda
@@ -642,7 +642,7 @@ 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-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
     secrets:
@@ -663,7 +663,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.11"
       runs_on: linux.24xlarge
       build_name: conda-py3_11-cuda11_8
@@ -683,7 +683,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
       build_environment: linux-binary-conda
@@ -705,7 +705,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
     secrets:
@@ -726,7 +726,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.11"
       runs_on: linux.24xlarge
       build_name: conda-py3_11-cuda12_1
@@ -746,7 +746,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
       build_environment: linux-binary-conda
@@ -768,7 +768,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
     secrets:
@@ -788,7 +788,7 @@ 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-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
       build_environment: linux-binary-conda
@@ -806,7 +806,7 @@ 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-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
       build_environment: linux-binary-conda
@@ -827,7 +827,7 @@ 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-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cpu
     secrets:
@@ -848,7 +848,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.12"
       runs_on: linux.24xlarge
       build_name: conda-py3_12-cuda11_8
@@ -868,7 +868,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda11_8
       build_environment: linux-binary-conda
@@ -890,7 +890,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda11_8
     secrets:
@@ -911,7 +911,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.12"
       runs_on: linux.24xlarge
       build_name: conda-py3_12-cuda12_1
@@ -931,7 +931,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda12_1
       build_environment: linux-binary-conda
@@ -953,7 +953,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.12"
       build_name: conda-py3_12-cuda12_1
     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-2.3
       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-2.3
       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-2.3
       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-2.3
       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-2.3
       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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       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: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.7-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_7-shared-with-deps-cxx11-abi
@@ -263,7 +263,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.7-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -277,7 +277,6 @@ 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
@@ -289,7 +288,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -305,7 +304,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm5.7-main
+          docker-image: pytorch/libtorch-cxx11-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -325,7 +324,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.7-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_7-shared-with-deps-cxx11-abi
@@ -347,7 +346,7 @@ 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:rocm6.0-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm6_0-shared-with-deps-cxx11-abi
@@ -369,7 +368,7 @@ 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:rocm6.0-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -383,7 +382,6 @@ 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
@@ -395,7 +393,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -411,7 +409,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -431,7 +429,7 @@ 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:rocm6.0-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm6_0-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-2.3
       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-2.3
       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-2.3
       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-2.3
       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-2.3
       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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       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: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_7-shared-with-deps-pre-cxx11
@@ -263,7 +263,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -277,7 +277,6 @@ 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
@@ -289,7 +288,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -305,7 +304,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.7-main
+          docker-image: pytorch/manylinux-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -325,7 +324,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_7-shared-with-deps-pre-cxx11
@@ -347,7 +346,7 @@ 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:rocm6.0-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm6_0-shared-with-deps-pre-cxx11
@@ -369,7 +368,7 @@ 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:rocm6.0-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -383,7 +382,6 @@ 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
@@ -395,7 +393,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -411,7 +409,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -431,7 +429,7 @@ 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:rocm6.0-2.3
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm6_0-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:cuda11.8-2.3
       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:cuda11.8-2.3
       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:cuda12.1-2.3
       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:cuda12.1-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel

.github/workflows/generated-linux-binary-manywheel-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-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu
       build_environment: linux-binary-manywheel
@@ -66,7 +66,7 @@ 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-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu
       build_environment: linux-binary-manywheel
@@ -87,7 +87,7 @@ 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-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu
     secrets:
@@ -107,7 +107,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-cxx11-abi
@@ -126,7 +126,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-cxx11-abi
@@ -148,7 +148,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-cxx11-abi
@@ -170,7 +170,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -190,7 +190,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -212,7 +212,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
     secrets:
@@ -233,7 +233,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel
@@ -253,7 +253,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel
@@ -275,7 +275,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
     secrets:
@@ -296,7 +296,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm5_7
       build_environment: linux-binary-manywheel
@@ -317,7 +317,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.8"
     steps:
       - name: Setup ROCm
@@ -330,7 +330,6 @@ 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
@@ -342,7 +341,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -358,7 +357,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.7-main
+          docker-image: pytorch/manylinux-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -378,7 +377,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm5_7
     secrets:
@@ -399,7 +398,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm6_0
       build_environment: linux-binary-manywheel
@@ -420,7 +419,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.8"
     steps:
       - name: Setup ROCm
@@ -433,7 +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
@@ -445,7 +443,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -461,7 +459,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -481,7 +479,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm6_0
     secrets:
@@ -501,7 +499,7 @@ 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-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu
       build_environment: linux-binary-manywheel
@@ -519,7 +517,7 @@ 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-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu
       build_environment: linux-binary-manywheel
@@ -540,7 +538,7 @@ 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-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu
     secrets:
@@ -560,7 +558,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-cxx11-abi
@@ -579,7 +577,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-cxx11-abi
@@ -601,7 +599,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-cxx11-abi
@@ -623,7 +621,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda11_8
       build_environment: linux-binary-manywheel
@@ -643,7 +641,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda11_8
       build_environment: linux-binary-manywheel
@@ -665,7 +663,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda11_8
     secrets:
@@ -686,7 +684,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1
       build_environment: linux-binary-manywheel
@@ -706,7 +704,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1
       build_environment: linux-binary-manywheel
@@ -728,7 +726,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1
     secrets:
@@ -749,7 +747,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm5_7
       build_environment: linux-binary-manywheel
@@ -770,7 +768,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.9"
     steps:
       - name: Setup ROCm
@@ -783,7 +781,6 @@ 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
@@ -795,7 +792,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -811,7 +808,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.7-main
+          docker-image: pytorch/manylinux-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -831,7 +828,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm5_7
     secrets:
@@ -852,7 +849,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm6_0
       build_environment: linux-binary-manywheel
@@ -873,7 +870,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.9"
     steps:
       - name: Setup ROCm
@@ -886,7 +883,6 @@ 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
@@ -898,7 +894,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -914,7 +910,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -934,7 +930,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm6_0
     secrets:
@@ -954,7 +950,7 @@ 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-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu
       build_environment: linux-binary-manywheel
@@ -972,7 +968,7 @@ 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-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu
       build_environment: linux-binary-manywheel
@@ -993,7 +989,7 @@ 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-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu
     secrets:
@@ -1013,7 +1009,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-cxx11-abi
@@ -1032,7 +1028,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-cxx11-abi
@@ -1054,7 +1050,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-cxx11-abi
@@ -1076,7 +1072,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1096,7 +1092,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1118,7 +1114,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda11_8
     secrets:
@@ -1139,7 +1135,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1159,7 +1155,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1181,7 +1177,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1
     secrets:
@@ -1202,7 +1198,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm5_7
       build_environment: linux-binary-manywheel
@@ -1223,7 +1219,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.10"
     steps:
       - name: Setup ROCm
@@ -1236,7 +1232,6 @@ 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
@@ -1248,7 +1243,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1264,7 +1259,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.7-main
+          docker-image: pytorch/manylinux-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1284,7 +1279,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm5_7
     secrets:
@@ -1305,7 +1300,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm6_0
       build_environment: linux-binary-manywheel
@@ -1326,7 +1321,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.10"
     steps:
       - name: Setup ROCm
@@ -1339,7 +1334,6 @@ 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
@@ -1351,7 +1345,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1367,7 +1361,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1387,7 +1381,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm6_0
     secrets:
@@ -1407,7 +1401,7 @@ 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-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu
       build_environment: linux-binary-manywheel
@@ -1425,7 +1419,7 @@ 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-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu
       build_environment: linux-binary-manywheel
@@ -1446,7 +1440,7 @@ 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-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu
     secrets:
@@ -1466,7 +1460,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-cxx11-abi
@@ -1485,7 +1479,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-cxx11-abi
@@ -1507,7 +1501,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-cxx11-abi
@@ -1529,7 +1523,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1549,7 +1543,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1571,7 +1565,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda11_8
     secrets:
@@ -1592,7 +1586,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1612,7 +1606,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1634,7 +1628,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1
     secrets:
@@ -1655,7 +1649,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm5_7
       build_environment: linux-binary-manywheel
@@ -1676,7 +1670,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.11"
     steps:
       - name: Setup ROCm
@@ -1689,7 +1683,6 @@ 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
@@ -1701,7 +1694,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1717,7 +1710,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.7-main
+          docker-image: pytorch/manylinux-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1737,7 +1730,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm5_7
     secrets:
@@ -1758,7 +1751,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm6_0
       build_environment: linux-binary-manywheel
@@ -1779,7 +1772,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.11"
     steps:
       - name: Setup ROCm
@@ -1792,7 +1785,6 @@ 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
@@ -1804,7 +1796,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1820,7 +1812,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1840,7 +1832,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm6_0
     secrets:
@@ -1860,7 +1852,7 @@ 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-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu
       build_environment: linux-binary-manywheel
@@ -1878,7 +1870,7 @@ 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-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu
       build_environment: linux-binary-manywheel
@@ -1899,7 +1891,7 @@ 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-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu
     secrets:
@@ -1919,7 +1911,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-cxx11-abi
@@ -1938,7 +1930,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-cxx11-abi
@@ -1960,7 +1952,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-main
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.3
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cpu-cxx11-abi
@@ -1982,7 +1974,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda11_8
       build_environment: linux-binary-manywheel
@@ -2002,7 +1994,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda11_8
       build_environment: linux-binary-manywheel
@@ -2024,7 +2016,7 @@ 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:cuda11.8-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda11_8
     secrets:
@@ -2045,7 +2037,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda12_1
       build_environment: linux-binary-manywheel
@@ -2065,7 +2057,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda12_1
       build_environment: linux-binary-manywheel
@@ -2087,7 +2079,7 @@ 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:cuda12.1-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-cuda12_1
     secrets:
@@ -2108,7 +2100,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-rocm5_7
       build_environment: linux-binary-manywheel
@@ -2129,7 +2121,7 @@ jobs:
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.12"
     steps:
       - name: Setup ROCm
@@ -2142,7 +2134,6 @@ 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
@@ -2154,7 +2145,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2170,7 +2161,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.7-main
+          docker-image: pytorch/manylinux-builder:rocm5.7-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -2190,7 +2181,7 @@ jobs:
       DESIRED_CUDA: rocm5.7
       GPU_ARCH_VERSION: 5.7
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-main
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.7-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-rocm5_7
     secrets:
@@ -2211,7 +2202,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-rocm6_0
       build_environment: linux-binary-manywheel
@@ -2232,7 +2223,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.12"
     steps:
       - name: Setup ROCm
@@ -2245,7 +2236,6 @@ 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
@@ -2257,7 +2247,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2273,7 +2263,7 @@ 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:rocm6.0-2.3
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -2293,7 +2283,7 @@ 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:rocm6.0-2.3
       DESIRED_PYTHON: "3.12"
       build_name: manywheel-py3_12-rocm6_0
     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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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-2.3
       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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -825,7 +819,6 @@ 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
@@ -837,7 +830,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -942,7 +935,6 @@ 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
@@ -954,7 +946,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1068,7 +1060,6 @@ 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
@@ -1080,7 +1071,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1186,7 +1177,6 @@ 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
@@ -1198,7 +1188,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           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/2.3
           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/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1557,7 +1545,6 @@ 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
@@ -1569,7 +1556,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1674,7 +1661,6 @@ 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
@@ -1686,7 +1672,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1800,7 +1786,6 @@ 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
@@ -1812,7 +1797,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1918,7 +1903,6 @@ 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
@@ -1930,7 +1914,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2045,7 +2029,6 @@ 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
@@ -2057,7 +2040,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2163,7 +2146,6 @@ 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
@@ -2175,7 +2157,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2289,7 +2271,6 @@ 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
@@ -2301,7 +2282,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2406,7 +2387,6 @@ 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
@@ -2418,7 +2398,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2532,7 +2512,6 @@ 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
@@ -2544,7 +2523,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2650,7 +2629,6 @@ 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
@@ -2662,7 +2640,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2777,7 +2755,6 @@ 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
@@ -2789,7 +2766,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2895,7 +2872,6 @@ 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
@@ -2907,7 +2883,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3021,7 +2997,6 @@ 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
@@ -3033,7 +3008,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3138,7 +3113,6 @@ 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
@@ -3150,7 +3124,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3264,7 +3238,6 @@ 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
@@ -3276,7 +3249,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3382,7 +3355,6 @@ 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
@@ -3394,7 +3366,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3509,7 +3481,6 @@ 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
@@ -3521,7 +3492,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3627,7 +3598,6 @@ 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
@@ -3639,7 +3609,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           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/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -829,7 +823,6 @@ 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
@@ -841,7 +834,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -946,7 +939,6 @@ 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
@@ -958,7 +950,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1073,7 +1065,6 @@ 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
@@ -1085,7 +1076,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1191,7 +1182,6 @@ 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
@@ -1203,7 +1193,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           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/2.3
           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/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1564,7 +1552,6 @@ 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
@@ -1576,7 +1563,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1681,7 +1668,6 @@ 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
@@ -1693,7 +1679,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1808,7 +1794,6 @@ 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
@@ -1820,7 +1805,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -1926,7 +1911,6 @@ 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
@@ -1938,7 +1922,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2054,7 +2038,6 @@ 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
@@ -2066,7 +2049,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2172,7 +2155,6 @@ 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
@@ -2184,7 +2166,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2299,7 +2281,6 @@ 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
@@ -2311,7 +2292,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2416,7 +2397,6 @@ 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
@@ -2428,7 +2408,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2543,7 +2523,6 @@ 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
@@ -2555,7 +2534,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2661,7 +2640,6 @@ 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
@@ -2673,7 +2651,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2789,7 +2767,6 @@ 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
@@ -2801,7 +2778,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -2907,7 +2884,6 @@ 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
@@ -2919,7 +2895,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3034,7 +3010,6 @@ 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
@@ -3046,7 +3021,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3151,7 +3126,6 @@ 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
@@ -3163,7 +3137,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3278,7 +3252,6 @@ 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
@@ -3290,7 +3263,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3396,7 +3369,6 @@ 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
@@ -3408,7 +3380,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3524,7 +3496,6 @@ 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
@@ -3536,7 +3507,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -3642,7 +3613,6 @@ 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
@@ -3654,7 +3624,7 @@ jobs:
       - name: Checkout pytorch/builder
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.3
           submodules: recursive
           repository: pytorch/builder
           path: builder

.github/workflows/inductor.yml

@@ -111,7 +111,7 @@ jobs:
     name: linux-jammy-cpu-py3.8-gcc11-inductor
     uses: ./.github/workflows/_linux-build.yml
     with:
-      build-environment: linux-jammy-py3_8-gcc11-build
+      build-environment: linux-jammy-py3.8-gcc11-build
       docker-image-name: pytorch-linux-jammy-py3.8-gcc11-inductor-benchmarks
       test-matrix: |
         { include: [
@@ -135,7 +135,7 @@ jobs:
     uses: ./.github/workflows/_linux-test.yml
     needs: linux-jammy-cpu-py3_8-gcc11-inductor-build
     with:
-      build-environment: linux-jammy-py3_8-gcc11-build
+      build-environment: linux-jammy-py3.8-gcc11-build
       docker-image: ${{ needs.linux-jammy-cpu-py3_8-gcc11-inductor-build.outputs.docker-image }}
       test-matrix: ${{ needs.linux-jammy-cpu-py3_8-gcc11-inductor-build.outputs.test-matrix }}
     secrets:

.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/2.3
         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/2.3
     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/2.3
     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/2.3
     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/2.3
         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/2.3
     with:
       runner: linux.2xlarge
       docker-image: pytorch-linux-focal-linter
@@ -113,6 +113,7 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
+        export RELEASE_VERSION_TAG="2.3"
         # 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/2.3
     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/2.3
     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/2.3
         with:
           submodules: false
           fetch-depth: 1
@@ -226,15 +227,15 @@ 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/2.3
         with:
           submodules: false
           fetch-depth: 1
-      - name: Setup Python 3.5
+      - name: Setup Python 3.6
         if: matrix.test_type == 'older_python_version'
         uses: actions/setup-python@v4
         with:
-          python-version: '3.5'
+          python-version: '3.6'
           architecture: x64
           check-latest: false
           cache: pip

.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/2.3
         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/2.3
         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/2.3
         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/2.3
         if: ${{ github.event_name == 'schedule' }}
         with:
           repo-name: executorch

.github/workflows/pull.yml

@@ -311,7 +311,7 @@ jobs:
     name: linux-focal-py3_8-clang9-xla
     uses: ./.github/workflows/_linux-build.yml
     with:
-      build-environment: linux-focal-py3_8-clang9-xla
+      build-environment: linux-focal-py3.8-clang9-xla
       docker-image-name: 308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/xla_base:v1.1-lite
       test-matrix: |
         { include: [
@@ -323,7 +323,7 @@ jobs:
     uses: ./.github/workflows/_linux-test.yml
     needs: linux-focal-py3_8-clang9-xla-build
     with:
-      build-environment: linux-focal-py3_8-clang9-xla
+      build-environment: linux-focal-py3.8-clang9-xla
       docker-image: ${{ needs.linux-focal-py3_8-clang9-xla-build.outputs.docker-image }}
       test-matrix: ${{ needs.linux-focal-py3_8-clang9-xla-build.outputs.test-matrix }}
 

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

@@ -18,7 +18,7 @@ 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/2.3
         with:
           docker-image-name: pytorch-linux-focal-cuda12.1-cudnn8-py3-gcc9
 
@@ -32,13 +32,13 @@ 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/2.3
         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/2.3
 
       - name: Clone PyTorch
         uses: actions/checkout@v3

.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/2.3
         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/2.3
         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/2.3
         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: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-        uses: pytorch/test-infra/.github/actions/upload-alerts@main
+        uses: pytorch/test-infra/.github/actions/upload-alerts@release/2.3
         with:
           alerts: '${{ 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/2.3
 
       - 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/2.3
         with:
           submodules: false
           fetch-depth: 1

.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/2.3
         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/2.3
         with:
           repo-owner: openai
           repo-name: triton

RELEASE.md

@@ -218,7 +218,7 @@ Validate the release jobs for pytorch and domain libraries should be green. Vali
   * [TorchVision](https://hud.pytorch.org/hud/pytorch/vision/release%2F1.12)
   * [TorchAudio](https://hud.pytorch.org/hud/pytorch/audio/release%2F1.12)
 
-Validate that the documentation build has completed and generated entry corresponding to the release in  [docs folder](https://github.com/pytorch/pytorch.github.io/tree/site/docs/) of pytorch.github.io repository
+Validate that the documentation build has completed and generated entry corresponding to the release in  [docs repository](https://github.com/pytorch/docs/tree/main/).
 
 ### Cherry Picking Fixes
 

aten/src/ATen/DLConvertor.h

@@ -13,6 +13,10 @@ namespace at {
 TORCH_API ScalarType toScalarType(const DLDataType& dtype);
 TORCH_API DLManagedTensor* toDLPack(const Tensor& src);
 TORCH_API Tensor fromDLPack(DLManagedTensor* src);
+C10_DEPRECATED_MESSAGE("Please migrate to a non-const variant")
+inline Tensor fromDLPack(const DLManagedTensor* src) {
+  return fromDLPack(const_cast<DLManagedTensor*>(src));
+}
 TORCH_API Tensor
 fromDLPack(DLManagedTensor* src, std::function<void(void*)> deleter);
 TORCH_API DLDataType getDLDataType(const Tensor& t);

aten/src/ATen/FunctionalInverses.cpp

@@ -303,6 +303,29 @@ Tensor FunctionalInverses::_nested_view_from_buffer_inverse(const Tensor& base,
     return Tensor();
 }
 
+Tensor FunctionalInverses::_nested_view_from_jagged_inverse(const Tensor& base, const Tensor& mutated_view, InverseReturnMode inverse_return_mode, const Tensor& offsets, const Tensor& dummy, const std::optional<Tensor>& lengths, int64_t ragged_idx) {
+  auto values = at::_nested_get_values(mutated_view);
+  if (inverse_return_mode != InverseReturnMode::NeverView) {
+    return values;
+  } else {
+    return values.clone(/*memory_format=*/at::MemoryFormat::Contiguous);
+  }
+}
+
+Tensor FunctionalInverses::_nested_get_values_inverse(const Tensor& base, const Tensor& mutated_view, InverseReturnMode inverse_return_mode) {
+  auto offsets = at::_nested_get_offsets(base);
+  auto lengths = at::_nested_get_lengths(base);
+  auto ragged_idx = at::_nested_get_ragged_idx(base);
+  auto dummy = at::_nested_get_jagged_dummy(base);
+  auto nt = at::_nested_view_from_jagged(mutated_view, offsets, dummy, lengths, ragged_idx);
+
+  if (inverse_return_mode != InverseReturnMode::NeverView) {
+    return nt;
+  } else {
+    return nt.clone(/*memory_format=*/at::MemoryFormat::Contiguous);
+  }
+}
+
 Tensor FunctionalInverses::unsqueeze_inverse(const Tensor& base, const Tensor& mutated_view, InverseReturnMode inverse_return_mode, int64_t dim) {
     if (inverse_return_mode != InverseReturnMode::NeverView) {
       return at::squeeze(mutated_view, dim);

aten/src/ATen/mps/MPSStream.mm

@@ -173,11 +173,22 @@ void MPSStream::copy(id<MTLBuffer> srcBuffer,
       endKernelCoalescing();
       id<MTLBlitCommandEncoder> blitEncoder = [commandBuffer() blitCommandEncoder];
 
-      [blitEncoder copyFromBuffer:srcBuffer
-                     sourceOffset:(NSUInteger)srcOffset
-                         toBuffer:dstBuffer
-                destinationOffset:(NSUInteger)dstOffset
-                             size:(NSUInteger)length];
+      // For some reason copyFromBuffer for 4Gb fails without returning an error
+      // See https://github.com/pytorch/pytorch/issues/124335
+      // Workaround by batching copy commands into 2Gb chunks
+      constexpr size_t max_copy_size = 0x80000000; // 2GB
+      size_t bytes_copied = 0;
+      size_t bytes_remains = length;
+      while (bytes_remains > 0) {
+        NSUInteger bytes_to_copy = std::min(max_copy_size, bytes_remains);
+        [blitEncoder copyFromBuffer:srcBuffer
+                       sourceOffset:(NSUInteger)srcOffset + bytes_copied
+                           toBuffer:dstBuffer
+                  destinationOffset:(NSUInteger)dstOffset + bytes_copied
+                               size:bytes_to_copy];
+        bytes_copied += bytes_to_copy;
+        bytes_remains -= bytes_to_copy;
+      }
       [blitEncoder endEncoding];
 
       // profilerId has a value only if copy profiling is enabled

aten/src/ATen/native/cudnn/Conv_v8.cpp

@@ -850,19 +850,13 @@ void try_plans(
       benchmark_cache.update(key, plan);
       return;
     } catch (cudnn_frontend::cudnnException& e) {
-      TORCH_WARN("Plan failed with a cudnnException: ", e.what());
     } catch (CuDNNError& e) {
-      TORCH_WARN("Plan failed with a CuDNNError: ", e.what());
     } catch (c10::OutOfMemoryError& e) {
       (void)cudaGetLastError(); // clear CUDA error
-      TORCH_WARN("Plan failed with an OutOfMemoryError: ", e.what());
     }
   }
   TORCH_CHECK(
-      false,
-      "FIND was unable to find an engine to execute this computation after trying ",
-      plans.size(),
-      " plans.");
+      false, "FIND was unable to find an engine to execute this computation");
 }
 
 void try_plans_fused(
@@ -880,19 +874,13 @@ void try_plans_fused(
       benchmark_cache_fused.update(key, plan);
       return;
     } catch (cudnn_frontend::cudnnException& e) {
-      TORCH_WARN("Plan failed with a cudnnException: ", e.what());
     } catch (CuDNNError& e) {
-      TORCH_WARN("Plan failed with a CuDNNError: ", e.what());
     } catch (c10::OutOfMemoryError& e) {
       (void)cudaGetLastError(); // clear CUDA error
-      TORCH_WARN("Plan failed with an OutOfMemoryError: ", e.what());
     }
   }
   TORCH_CHECK(
-      false,
-      "FIND was unable to find an engine to execute this computation after trying ",
-      plans.size(),
-      " plans.");
+      false, "FIND was unable to find an engine to execute this computation");
 }
 
 bool try_configs(
@@ -916,12 +904,9 @@ bool try_configs(
       benchmark_cache.update(key, plan);
       return true;
     } catch (cudnn_frontend::cudnnException& e) {
-      TORCH_WARN("Plan failed with a cudnnException: ", e.what());
     } catch (CuDNNError& e) {
-      TORCH_WARN("Plan failed with a CuDNNError: ", e.what());
     } catch (c10::OutOfMemoryError& e) {
       (void)cudaGetLastError(); // clear CUDA error
-      TORCH_WARN("Plan failed with an OutOfMemoryError: ", e.what());
     }
   }
   return false;
@@ -950,12 +935,9 @@ bool try_configs_fused(
       benchmark_cache_fused.update(key, plan);
       return true;
     } catch (cudnn_frontend::cudnnException& e) {
-      TORCH_WARN("Plan failed with a cudnnException: ", e.what());
     } catch (CuDNNError& e) {
-      TORCH_WARN("Plan failed with a CuDNNError: ", e.what());
     } catch (c10::OutOfMemoryError& e) {
       (void)cudaGetLastError(); // clear CUDA error
-      TORCH_WARN("Plan failed with an OutOfMemoryError: ", e.what());
     }
   }
   return false;

aten/src/ATen/native/mps/MPSGraphSonomaOps.h

@@ -25,6 +25,10 @@ typedef NS_ENUM(NSUInteger, MPSGraphFFTScalingMode)
 -(MPSGraphTensor * _Nonnull) conjugateWithTensor:(MPSGraphTensor * _Nonnull) tensor
                                             name:(NSString * _Nullable) name;
 
+-(MPSGraphTensor * _Nonnull) realPartOfTensor:(MPSGraphTensor * _Nonnull) tensor
+                                         name:(NSString * _Nullable) name;
+
+
 -(MPSGraphTensor * _Nonnull) fastFourierTransformWithTensor:(MPSGraphTensor * _Nonnull) tensor
                                                        axes:(NSArray<NSNumber *> * _Nonnull) axes
                                                  descriptor:(MPSGraphFFTDescriptor * _Nonnull) descriptor

aten/src/ATen/native/mps/OperationUtils.mm

@@ -210,6 +210,9 @@ std::string scalarToMetalTypeString(const c10::ScalarType& scalar_type) {
       return "float";
     case ScalarType::Half:
       return "half";
+    case ScalarType::BFloat16:
+      checkSupportsBFloat16();
+      return "bfloat";
     case ScalarType::Int:
       return "int";
     case ScalarType::Long:

aten/src/ATen/native/mps/operations/ConstantOps.mm

@@ -28,6 +28,7 @@ static Tensor& fill_scalar_mps_impl(Tensor& self, const Scalar& value) {
 
   struct CachedGraph : public MPSCachedGraph {
     CachedGraph(MPSGraph* graph) : MPSCachedGraph(graph) {}
+    MPSGraphTensor* inputTensor_ = nil;
     MPSGraphTensor* outputTensor_ = nil;
   };
 
@@ -35,36 +36,23 @@ static Tensor& fill_scalar_mps_impl(Tensor& self, const Scalar& value) {
     string key = "fill_scalar_mps_impl" + getTensorsStringKey(self) + ":" + to_string(value.toDouble());
 
     auto cachedGraph = LookUpOrCreateCachedGraph<CachedGraph>(key, [&](auto mpsGraph, auto newCachedGraph) {
-      auto isBool = self.scalar_type() == c10::ScalarType::Bool;
-      auto isUInt8 = self.scalar_type() == c10::ScalarType::Byte;
-      auto dataType = !isUInt8 ? !isBool ? getMPSScalarType(self.scalar_type()) : MPSDataTypeInt8 : MPSDataTypeUInt32;
-      // constantWithScalar does not work for boolTypes on MacOS-12.[34]
-      // workaround by filing it as int8 tensor and than casting to bool
-      // See https://github.com/pytorch/pytorch/issues/82427
-      // constantWithScalar does not work for UInt8 Types on MacOS-12.[34]/Ventura preview
-      // workaround by filing it as uint32 tensor and than casting to uint8
-      // See https://github.com/pytorch/pytorch/issues/83692
-      MPSGraphTensor* inputTensor = [mpsGraph constantWithScalar:value.toDouble()
-                                                           shape:getMPSShape(self)
-                                                        dataType:dataType];
+      MPSGraphTensor* inputTensor = mpsGraphScalarPlaceHolder(mpsGraph, getMPSDataType(self.scalar_type()));
       MPSGraphTensor* outputTensor = [mpsGraph identityWithTensor:inputTensor name:nil];
-      if (isBool) {
-        outputTensor = [mpsGraph castTensor:outputTensor toType:MPSDataTypeBool name:@"constWithBool-workaround"];
-      }
-      if (isUInt8) {
-        outputTensor = [mpsGraph castTensor:outputTensor toType:MPSDataTypeUInt8 name:@"constWithUInt8-workaround"];
-      }
-
+      newCachedGraph->inputTensor_ = inputTensor;
       newCachedGraph->outputTensor_ = outputTensor;
     });
 
+    auto mpsScalar = getMPSScalar(value, self.scalar_type());
+    auto mpsScalarData = getMPSGraphTensorFromScalar(getCurrentMPSStream(), mpsScalar);
+    NSDictionary<MPSGraphTensor*, MPSGraphTensorData*>* feeds = @{cachedGraph->inputTensor_ : mpsScalarData};
+
     Placeholder outputPlaceholder =
         Placeholder(cachedGraph->outputTensor_, needsCopyToOutput ? output : self, nullptr, !needsCopyToOutput);
 
     NSDictionary<MPSGraphTensor*, MPSGraphTensorData*>* results =
         @{outputPlaceholder.getMPSGraphTensor() : outputPlaceholder.getMPSGraphTensorData()};
 
-    runMPSGraph(getCurrentMPSStream(), cachedGraph->graph(), /*feeds*/ nil, results);
+    runMPSGraph(getCurrentMPSStream(), cachedGraph->graph(), feeds, results);
 
     if (needsCopyToOutput) {
       self.copy_(output);

aten/src/ATen/native/mps/operations/Indexing.mm

@@ -240,14 +240,20 @@ static void index_put_kernel_mps(TensorIterator& iter,
 } // namespace mps
 
 static Tensor nonzero_fallback(const Tensor& self) {
-  TORCH_WARN_ONCE("MPS: nonzero op is supported natively starting from macOS 13.0. ",
-                  "Falling back on CPU. This may have performance implications.");
-
   return at::nonzero(self.to("cpu")).clone().to("mps");
 }
 
 Tensor& nonzero_out_mps(const Tensor& self, Tensor& out_) {
-  if (!is_macos_13_or_newer()) {
+  if (!is_macos_13_or_newer(MacOSVersion::MACOS_VER_14_0_PLUS)) {
+    TORCH_WARN_ONCE("MPS: nonzero op is supported natively starting from macOS 13.0. ",
+                    "Falling back on CPU. This may have performance implications.");
+    Tensor out_fallback = nonzero_fallback(self);
+    at::native::resize_output(out_, out_fallback.sizes());
+    out_.copy_(out_fallback.to("mps"));
+    return out_;
+  } else if (self.is_complex()) {
+    TORCH_WARN_ONCE("MPS: nonzero op is not supported for complex datatypes. ",
+                    "Falling back on CPU. This may have performance implications.");
     Tensor out_fallback = nonzero_fallback(self);
     at::native::resize_output(out_, out_fallback.sizes());
     out_.copy_(out_fallback.to("mps"));
@@ -281,7 +287,6 @@ Tensor& nonzero_out_mps(const Tensor& self, Tensor& out_) {
     CachedGraph(MPSGraph* graph) : MPSCachedGraph(graph) {}
     MPSGraphTensor* inputTensor_ = nil;
     MPSGraphTensor* outputTensor_ = nil;
-    MPSGraphTensor* scatterDataTensor_ = nil;
   };
 
   dispatch_sync(stream->queue(), ^() {
@@ -299,93 +304,20 @@ Tensor& nonzero_out_mps(const Tensor& self, Tensor& out_) {
     out = at::empty(out_.sizes(), out_.scalar_type(), c10::nullopt, kMPS, c10::nullopt, c10::nullopt);
   }
 
-  int64_t _apparentInputShape = 1;
-  for (auto dim : self.sizes()) {
-    _apparentInputShape *= dim;
-  }
-  MPSShape* apparentOutputShape = @[ @(total_nonzero * nDim) ];
-  MPSShape* apparentInputShape = @[ @(_apparentInputShape) ];
-
-  // Pseudocode:
-  //
-  // inputTensor     = [1,  0,  0,  3]
-  // inputNonZero    = [1,  0,  0,  1]
-  // indices         = [1,  1,  1,  2]
-  // maskedIndices   = [0, -1, -1,  1]
-  // coordinates     = [0,  1,  2,  3]
-  // scatterResult   = [0,  3]
-
   @autoreleasepool {
     string key = "nonzero_out_mps" + getTensorsStringKey(self);
     auto cachedGraph = LookUpOrCreateCachedGraph<CachedGraph>(key, [&](auto mpsGraph, auto newCachedGraph) {
-      MPSDataType inputDataType = getMPSDataType(self);
-      MPSShape* inputShape = getMPSShape(self);
+      MPSGraphTensor* inputTensor = mpsGraphRankedPlaceHolder(mpsGraph, getMPSDataType(self), getMPSShape(self));
 
-      MPSGraphTensor* inputTensor =
-          mpsGraphRankedPlaceHolder(mpsGraph, getMPSScalarType(self.scalar_type()), apparentInputShape);
-      MPSGraphTensor* scatterDataTensor = mpsGraphUnrankedPlaceHolder(mpsGraph, getMPSScalarType(out.scalar_type()));
-      MPSGraphTensor* zeroTensor = [mpsGraph constantWithScalar:0.0 dataType:inputDataType];
-      MPSGraphTensor* oneTensor = [mpsGraph constantWithScalar:1.0 dataType:MPSDataTypeInt32];
-      MPSGraphTensor* minusMaxDimTensor = [mpsGraph constantWithScalar:-maxDimensions dataType:MPSDataTypeInt32];
-      MPSGraphTensor* inputNotEqualToZeroTensor = [mpsGraph notEqualWithPrimaryTensor:inputTensor
-                                                                      secondaryTensor:zeroTensor
-                                                                                 name:nil];
-      MPSGraphTensor* maskTensor = [mpsGraph castTensor:inputNotEqualToZeroTensor
-                                                 toType:MPSDataTypeInt32
-                                                   name:@"castToInt32"];
-      MPSGraphTensor* indicesTensor = [mpsGraph cumulativeSumWithTensor:maskTensor axis:0 name:nil];
-      MPSGraphTensor* indicesMinusOneTensor = [mpsGraph subtractionWithPrimaryTensor:indicesTensor
-                                                                     secondaryTensor:oneTensor
-                                                                                name:nil];
-      MPSGraphTensor* maskedIndicesTensor = [mpsGraph selectWithPredicateTensor:inputNotEqualToZeroTensor
-                                                            truePredicateTensor:indicesMinusOneTensor
-                                                           falsePredicateTensor:minusMaxDimTensor
-                                                                           name:nil];
-      MPSGraphTensor* coordinatesTensor = [mpsGraph reshapeTensor:[mpsGraph coordinateAlongAxis:0
-                                                                                      withShape:inputShape
-                                                                                           name:nil]
-                                                        withShape:@[ @-1 ]
-                                                             name:nil];
-      if (nDim > 1) {
-        NSMutableArray<MPSGraphTensor*>* maskedIndicesTensorArray = [NSMutableArray arrayWithCapacity:nDim];
-        NSMutableArray<MPSGraphTensor*>* coordinatesTensorArray = [NSMutableArray arrayWithCapacity:nDim];
-
-        MPSGraphTensor* constantRankTensor = [mpsGraph constantWithScalar:nDim dataType:MPSDataTypeInt32];
-        maskedIndicesTensorArray[0] = [mpsGraph multiplicationWithPrimaryTensor:maskedIndicesTensor
-                                                                secondaryTensor:constantRankTensor
-                                                                           name:nil];
-        coordinatesTensorArray[0] = coordinatesTensor;
-        for (int i = 1; i < nDim; i++) {
-          maskedIndicesTensorArray[i] = [mpsGraph additionWithPrimaryTensor:maskedIndicesTensorArray[i - 1]
-                                                            secondaryTensor:oneTensor
-                                                                       name:nil];
-          coordinatesTensorArray[i] = [mpsGraph reshapeTensor:[mpsGraph coordinateAlongAxis:i
-                                                                                  withShape:inputShape
-                                                                                       name:nil]
-                                                    withShape:@[ @-1 ]
-                                                         name:nil];
-        }
-        maskedIndicesTensor = [mpsGraph concatTensors:maskedIndicesTensorArray dimension:0 interleave:YES name:nil];
-        coordinatesTensor = [mpsGraph concatTensors:coordinatesTensorArray dimension:0 interleave:YES name:nil];
-      }
-
-      MPSGraphTensor* outputTensor = [mpsGraph scatterWithDataTensor:scatterDataTensor
-                                                       updatesTensor:coordinatesTensor
-                                                       indicesTensor:maskedIndicesTensor
-                                                                axis:0
-                                                                mode:MPSGraphScatterModeSet
-                                                                name:nil];
+      MPSGraphTensor* outputTensor = [mpsGraph nonZeroIndicesOfTensor:inputTensor name:nil];
 
       newCachedGraph->inputTensor_ = inputTensor;
-      newCachedGraph->scatterDataTensor_ = scatterDataTensor;
       newCachedGraph->outputTensor_ = outputTensor;
     });
 
-    Placeholder selfPlaceholder = Placeholder(cachedGraph->inputTensor_, self, apparentInputShape);
-    Placeholder outputPlaceholder = Placeholder(cachedGraph->outputTensor_, out, apparentOutputShape);
-    Placeholder scatterPlaceholder = Placeholder(cachedGraph->scatterDataTensor_, out, apparentOutputShape);
-
-    auto feeds = dictionaryFromPlaceholders(selfPlaceholder, scatterPlaceholder);
+    Placeholder selfPlaceholder = Placeholder(cachedGraph->inputTensor_, self);
+    Placeholder outputPlaceholder = Placeholder(cachedGraph->outputTensor_, out);
+    auto feeds = dictionaryFromPlaceholders(selfPlaceholder);
     runMPSGraph(stream, cachedGraph->graph(), feeds, outputPlaceholder);
   }
 
@@ -397,7 +329,13 @@ Tensor& nonzero_out_mps(const Tensor& self, Tensor& out_) {
 }
 
 Tensor nonzero_mps(const Tensor& self) {
-  if (!is_macos_13_or_newer()) {
+  if (!is_macos_13_or_newer(MacOSVersion::MACOS_VER_14_0_PLUS)) {
+    TORCH_WARN_ONCE("MPS: nonzero op is supported natively starting from macOS 13.0. ",
+                    "Falling back on CPU. This may have performance implications.");
+    return nonzero_fallback(self);
+  } else if (self.is_complex()) {
+    TORCH_WARN_ONCE("MPS: nonzero op is not supported for complex datatypes ",
+                    "Falling back on CPU. This may have performance implications.");
     return nonzero_fallback(self);
   }
 

aten/src/ATen/native/mps/operations/LinearAlgebra.mm

@@ -4,6 +4,8 @@
 #include <ATen/mps/MPSProfiler.h>
 #include <ATen/native/LinearAlgebraUtils.h>
 #include <ATen/native/Resize.h>
+// For MTLLanguageVersion_3_1
+#include <ATen/native/mps/MPSGraphSonomaOps.h>
 #include <ATen/native/mps/OperationUtils.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
@@ -29,7 +31,7 @@ static const char* METAL_LINALG = R"MATMUL_METAL(
 using namespace metal;
 template<typename T>
 T dot_product(constant T *v1, constant T* v2, ulong2 strides, uint32_t size) {
-  T rc = 0.0;
+  T rc = T(0.0);
   for (uint32_t i = 0; i < size; ++i) {
     rc += v1[i * strides.x] * v2[i * strides.y];
   }
@@ -69,6 +71,9 @@ kernel void naive_matmul<DTYPE>(                                           \
 
 INSTANTIATE_NAIVE_MM(float);
 INSTANTIATE_NAIVE_MM(half);
+#if __METAL_VERSION__ >= 310
+INSTANTIATE_NAIVE_MM(bfloat);
+#endif
 )MATMUL_METAL";
 
 id<MTLLibrary> compileLinalgOpLibrary(id<MTLDevice> device) {
@@ -79,7 +84,8 @@ id<MTLLibrary> compileLinalgOpLibrary(id<MTLDevice> device) {
 
   NSError* error = nil;
   MTLCompileOptions* options = [[MTLCompileOptions new] autorelease];
-  [options setLanguageVersion:MTLLanguageVersion2_3];
+  [options setLanguageVersion:is_macos_13_or_newer(MacOSVersion::MACOS_VER_14_0_PLUS) ? MTLLanguageVersion3_1
+                                                                                      : MTLLanguageVersion2_3];
   linalgLibrary = [device newLibraryWithSource:[NSString stringWithCString:METAL_LINALG encoding:NSASCIIStringEncoding]
                                        options:options
                                          error:&error];

aten/src/ATen/native/mps/operations/TensorCompare.mm

@@ -30,41 +30,53 @@ static void clamp_mps_graph(CachedGraph* cachedGraph,
                             const Tensor& min_tensor,
                             const Tensor& max_tensor) {
   auto input_dtype = input_tensor.scalar_type();
-  auto min_dtype = input_dtype;
-  auto max_dtype = input_dtype;
-  if (cachedGraph->minTensor) {
-    min_dtype = min_tensor.scalar_type();
-  }
-  if (cachedGraph->maxTensor) {
-    max_dtype = max_tensor.scalar_type();
-  }
+  auto min_dtype = cachedGraph->minTensor ? min_tensor.scalar_type() : input_dtype;
+  auto max_dtype = cachedGraph->maxTensor ? max_tensor.scalar_type() : input_dtype;
 
   MPSGraph* mpsGraph = cachedGraph->graph();
 
   cachedGraph->inputTensor = mpsGraphRankedPlaceHolder(mpsGraph, input_tensor);
 
-  MPSGraphTensor* minTensor = cachedGraph->minTensor;
-  MPSGraphTensor* maxTensor = cachedGraph->maxTensor;
+  auto minTensor = cachedGraph->minTensor;
+  auto maxTensor = cachedGraph->maxTensor;
+
   if (input_dtype != min_dtype) {
     minTensor = castMPSTensor(mpsGraph, cachedGraph->minTensor, input_dtype);
   }
   if (input_dtype != max_dtype) {
     maxTensor = castMPSTensor(mpsGraph, cachedGraph->maxTensor, input_dtype);
   }
-  if (cachedGraph->minTensor && cachedGraph->maxTensor) {
-    cachedGraph->outputTensor = [mpsGraph clampWithTensor:cachedGraph->inputTensor
-                                           minValueTensor:minTensor
-                                           maxValueTensor:maxTensor
-                                                     name:nil];
-  } else if (cachedGraph->maxTensor) {
-    cachedGraph->outputTensor = [mpsGraph minimumWithPrimaryTensor:cachedGraph->inputTensor
-                                                   secondaryTensor:maxTensor
-                                                              name:nil];
-  } else if (cachedGraph->minTensor) {
-    cachedGraph->outputTensor = [mpsGraph maximumWithPrimaryTensor:cachedGraph->inputTensor
-                                                   secondaryTensor:minTensor
-                                                              name:nil];
+  if (c10::isIntegralType(input_dtype, /*includeBool=*/true)) {
+    if (minTensor && maxTensor) {
+      cachedGraph->outputTensor = [mpsGraph clampWithTensor:cachedGraph->inputTensor
+                                             minValueTensor:minTensor
+                                             maxValueTensor:maxTensor
+                                                       name:nil];
+    } else if (maxTensor) {
+      cachedGraph->outputTensor = [mpsGraph minimumWithPrimaryTensor:cachedGraph->inputTensor
+                                                     secondaryTensor:maxTensor
+                                                                name:nil];
+    } else if (minTensor) {
+      cachedGraph->outputTensor = [mpsGraph maximumWithPrimaryTensor:cachedGraph->inputTensor
+                                                     secondaryTensor:minTensor
+                                                                name:nil];
+    }
+    return;
   }
+  // clampWithTensor doesn't propagate NaN through so simulate it as composition of
+  // maximumWithNaNPropagationWithPrimaryTensor and minimumWithNaNPropagationWithPrimaryTensor
+  auto outputTensor = cachedGraph->inputTensor;
+  if (minTensor) {
+    outputTensor = [mpsGraph maximumWithNaNPropagationWithPrimaryTensor:outputTensor
+                                                        secondaryTensor:minTensor
+                                                                   name:nil];
+  }
+  if (maxTensor) {
+    outputTensor = [mpsGraph minimumWithNaNPropagationWithPrimaryTensor:outputTensor
+                                                        secondaryTensor:maxTensor
+                                                                   name:nil];
+  }
+  cachedGraph->outputTensor = outputTensor;
 }
 
 static void check_min_max_dims(const OptionalTensorRef clamp_opt, const Tensor& input_t, string op_name) {

aten/src/ATen/native/mps/operations/UnaryOps.mm

@@ -75,23 +75,10 @@ static bool is_empty_tensor(const Tensor& self) {
   return self.numel() == 0;
 }
 
-static void unary_op(const Tensor& self,
-                     const Tensor& output_,
-                     std::string op_name,
-                     UnaryOpBlock unaryBlock,
-                     is_noop_p is_noop = is_empty_tensor) {
+static void unary_op_noresize(const Tensor& self, const Tensor& output_, std::string op_name, UnaryOpBlock unaryBlock) {
   TORCH_CHECK(!(!is_macos_13_or_newer() && self.scalar_type() == ScalarType::Byte),
               "MPS support unary op with uint8 natively starting from macOS 13.0");
 
-  if (!output_.is_same_size(self)) {
-    output_.resize_(self.sizes());
-  }
-
-  if (is_noop(self)) {
-    output_.copy_(self);
-    return;
-  }
-
   auto output = output_;
   bool needsCopyToOutput = false;
   if (output.storage_offset() || !output.is_contiguous()) {
@@ -139,6 +126,23 @@ static void unary_op(const Tensor& self,
   }
 }
 
+static void unary_op(const Tensor& self,
+                     const Tensor& output_,
+                     std::string op_name,
+                     UnaryOpBlock unaryBlock,
+                     is_noop_p is_noop = is_empty_tensor) {
+  if (!output_.is_same_size(self)) {
+    output_.resize_(self.sizes());
+  }
+
+  if (is_noop(self)) {
+    output_.copy_(self);
+    return;
+  }
+
+  unary_op_noresize(self, output_, op_name, unaryBlock);
+}
+
 MPSGraphTensor* trunc_tensor(MPSGraph* mpsGraph, MPSGraphTensor* inputTensor) {
   // Rounding is a no-op for integral types, and also a reasonable workaround
   // For MPSGraph bug on Apple Silicon, that throws `Function floorOp_i64 was not found in the library`
@@ -168,6 +172,12 @@ MPSGraphTensor* log1p(MPSGraph* mpsGraph, MPSGraphTensor* inputTensor) {
   return [mpsGraph logarithmWithTensor:addedTensor name:nil];
 }
 
+static MPSGraphTensor* lengthOfComplexAsReal(MPSGraph* mpsGraph, MPSGraphTensor* inputTensor) {
+  auto squares = [mpsGraph squareWithTensor:inputTensor name:nil];
+  auto sumSquares = [mpsGraph reductionSumWithTensor:squares axis:-1 name:nil];
+  return [mpsGraph squareRootWithTensor:sumSquares name:nil];
+}
+
 } // namespace mps
 
 TORCH_IMPL_FUNC(trunc_out_mps)(const Tensor& self, const Tensor& output) {
@@ -226,14 +236,6 @@ CREATE_MPS_STRUCTURED_UNARY_ROUNDING_TORCH_IMPL_FUNC(round_out_mps, round)
     });                                                                                                          \
   }
 
-#define CREATE_MPS_UNARY_TORCH_IMPL_FUNC(func_out, func_stub)                                                    \
-  Tensor& func_out(const Tensor& self, Tensor& output) {                                                         \
-    mps::unary_op(self, output, #func_out, ^MPSGraphTensor*(MPSGraph * mpsGraph, MPSGraphTensor * inputTensor) { \
-      return [mpsGraph func_stub##WithTensor:inputTensor name:nil];                                              \
-    });                                                                                                          \
-    return output;                                                                                               \
-  }
-
 CREATE_MPS_STRUCTURED_UNARY_TORCH_IMPL_FUNC(exp_out_mps, exponent)
 CREATE_MPS_STRUCTURED_UNARY_TORCH_IMPL_FUNC(exp2_out_mps, exponentBase2)
 CREATE_MPS_STRUCTURED_UNARY_TORCH_IMPL_FUNC(reciprocal_out_mps, reciprocal)
@@ -257,7 +259,35 @@ CREATE_MPS_STRUCTURED_UNARY_TORCH_IMPL_FUNC(asinh_out_mps, asinh)
 CREATE_MPS_STRUCTURED_UNARY_TORCH_IMPL_FUNC(acosh_out_mps, acosh)
 CREATE_MPS_STRUCTURED_UNARY_TORCH_IMPL_FUNC(atanh_out_mps, atanh)
 
-CREATE_MPS_UNARY_TORCH_IMPL_FUNC(abs_out_mps, absolute)
+Tensor& abs_out_mps(const Tensor& self, Tensor& output) {
+  using namespace mps;
+
+  if (!output.is_same_size(self)) {
+    output.resize_(self.sizes());
+  }
+
+  if (self.numel() == 0) {
+    return output;
+  }
+
+  if (supportsComplex() || !self.is_complex()) {
+    unary_op_noresize(self, output, "abs_out_mps", ^MPSGraphTensor*(MPSGraph* mpsGraph, MPSGraphTensor* inputTensor) {
+      auto rc = [mpsGraph absoluteWithTensor:inputTensor name:nil];
+      if (self.is_complex()) {
+        rc = [mpsGraph realPartOfTensor:rc name:nil];
+      }
+      return rc;
+    });
+  } else {
+    Tensor realInput = at::view_as_real(self);
+    unary_op_noresize(
+        realInput, output, "abs_out_mps", ^MPSGraphTensor*(MPSGraph* mpsGraph, MPSGraphTensor* inputTensor) {
+          auto rc = lengthOfComplexAsReal(mpsGraph, inputTensor);
+          return [mpsGraph reshapeTensor:rc withShape:getMPSShape(output) name:nil];
+        });
+  }
+  return output;
+}
 
 Tensor& logical_not_out_mps(const Tensor& self, Tensor& output) {
   auto bool_self = self.to(ScalarType::Bool);
@@ -484,9 +514,7 @@ TORCH_IMPL_FUNC(sgn_out_mps)(const Tensor& self, const Tensor& output) {
   Tensor realOutput = at::view_as_real(output);
 
   auto complex_sgn_op = [&](MPSGraph* mpsGraph, MPSGraphTensor* inputTensor) -> MPSGraphTensor* {
-    MPSGraphTensor* squares = [mpsGraph squareWithTensor:inputTensor name:nil];
-    MPSGraphTensor* sumSquares = [mpsGraph reductionSumWithTensor:squares axis:-1 name:nil];
-    MPSGraphTensor* norm = [mpsGraph squareRootWithTensor:sumSquares name:nil];
+    MPSGraphTensor* norm = mps::lengthOfComplexAsReal(mpsGraph, inputTensor);
     MPSGraphTensor* zero = [mpsGraph constantWithScalar:0.0 dataType:norm.dataType];
     MPSGraphTensor* isZero = [mpsGraph equalWithPrimaryTensor:norm secondaryTensor:zero name:nil];
     MPSGraphTensor* sgnTensor = [mpsGraph divisionWithPrimaryTensor:inputTensor secondaryTensor:norm name:nil];

aten/src/ATen/native/native_functions.yaml

@@ -6154,6 +6154,52 @@
     CompositeExplicitAutogradNonFunctional: _nested_view_from_buffer_copy
   autogen: _nested_view_from_buffer_copy.out
 
+- func: _nested_view_from_jagged(Tensor(a) self, Tensor offsets, Tensor dummy, Tensor? lengths=None, int ragged_idx=1) -> Tensor(a)
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_view_from_jagged_copy(Tensor self, Tensor offsets, Tensor dummy, Tensor? lengths=None, int ragged_idx=1) -> Tensor
+  variants: function
+  device_check: NoCheck
+  tags: view_copy
+  dispatch:
+    CompositeExplicitAutogradNonFunctional: _nested_view_from_jagged_copy
+  autogen: _nested_view_from_jagged_copy.out
+
+- func: _nested_get_values(Tensor(a) self) -> Tensor(a)
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_get_values_copy(Tensor self) -> Tensor
+  variants: function
+  device_check: NoCheck
+  tags: view_copy
+  dispatch:
+    CompositeExplicitAutogradNonFunctional: _nested_get_values_copy
+  autogen: _nested_get_values_copy.out
+
+- func: _nested_get_offsets(Tensor self) -> Tensor
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+# returns undefined Tensor if no lengths present
+- func: _nested_get_lengths(Tensor self) -> Tensor
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_get_ragged_idx(Tensor self) -> int
+  variants: function
+  device_check: NoCheck
+  dispatch: {}
+
+- func: _nested_get_jagged_dummy(Tensor any) -> Tensor
+  category_override: dummy
+  dispatch: {}
+
 - func: _trilinear(Tensor i1, Tensor i2, Tensor i3, int[] expand1, int[] expand2, int[] expand3, int[] sumdim, int unroll_dim=1) -> Tensor
   dispatch:
     # calls unsqueeze

aten/src/ATen/native/transformers/hip/flash_attn/flash_api.hip

@@ -157,43 +157,6 @@ aotriton::TensorView<Rank> mk_aotensor(const at::Tensor& q, c10::string_view ten
                                     cast_dtype(q.dtype()));
 }
 
-template<bool COPY_FROM_INPUT,  // For Input Tensor
-         bool COPY_BACK>        // For Output Tensor
-class TensorStorageSanitizer {
-public:
-  TensorStorageSanitizer(const at::Tensor& ref,
-                         at::Tensor& to_sanitize)
-    : ref_(ref), to_sanitize_(to_sanitize)
-  {
-    need_sanitize = ref_.strides() !=  to_sanitize_.strides();
-    if (!need_sanitize)
-      return;
-
-    temp_ = at::empty_like(ref_);
-    if (COPY_FROM_INPUT) {
-      temp_.copy_(to_sanitize_);
-    }
-  }
-
-  ~TensorStorageSanitizer()
-  {
-    if (need_sanitize && COPY_BACK)
-      to_sanitize_.copy_(temp_);
-  }
-
-  at::Tensor& sanitized_tensor()
-  {
-    if (need_sanitize)
-      return temp_;
-    return to_sanitize_;
-  }
-private:
-  const at::Tensor& ref_;
-  at::Tensor& to_sanitize_;
-  at::Tensor temp_;
-  bool need_sanitize = false;
-};
-
 }
 
 #define CHECK_DEVICE(x) TORCH_CHECK(x.is_cuda(), #x " must be on CUDA")
@@ -531,9 +494,6 @@ mha_bwd(const at::Tensor &dout,  // batch_size x seqlen_q x num_heads, x head_si
   int d_head = head_size_og;
   hipError_t err; // TODO: Error handling
   {
-    TensorStorageSanitizer<true, false> dq_s(q_t, dq_t);
-    TensorStorageSanitizer<true, false> dk_s(k_t, dk_t);
-    TensorStorageSanitizer<true, false> dv_s(v_t, dv_t);
     using aotriton::v2::flash::attn_bwd;
     err = attn_bwd(mk_aotensor(q_t, "q"),
                    mk_aotensor(k_t, "k"),
@@ -541,9 +501,9 @@ mha_bwd(const at::Tensor &dout,  // batch_size x seqlen_q x num_heads, x head_si
                    softmax_scale,
                    mk_aotensor(out_t, "out"),
                    mk_aotensor(dout_t, "dout"),
-                   mk_aotensor(dq_s.sanitized_tensor(), "dq"),
-                   mk_aotensor(dk_s.sanitized_tensor(), "dk"),
-                   mk_aotensor(dv_s.sanitized_tensor(), "dv"),
+                   mk_aotensor(dq_t, "dq"),
+                   mk_aotensor(dk_t, "dk"),
+                   mk_aotensor(dv_t, "dv"),
                    mk_aotensor<2>(softmax_lse_cont, "L"),
                    mk_aotensor<2>(delta, "delta"),
                    p_dropout,

cmake/External/aotriton.cmake

@@ -6,7 +6,7 @@ if(NOT __AOTRITON_INCLUDED)
   set(__AOTRITON_INSTALL_DIR "${PROJECT_SOURCE_DIR}/torch")
   ExternalProject_Add(aotriton_external
     GIT_REPOSITORY https://github.com/ROCm/aotriton.git
-    GIT_TAG 9044fe5eb16130e49a0a1f781ea15037353ad542
+    GIT_TAG 24a3fe9cb57e5cda3c923df29743f9767194cc27
     SOURCE_DIR ${__AOTRITON_SOURCE_DIR}
     BINARY_DIR ${__AOTRITON_BUILD_DIR}
     PREFIX ${__AOTRITON_INSTALL_DIR}

docker.Makefile

@@ -83,6 +83,22 @@ devel-push: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(CU
 devel-push:
 	$(DOCKER_PUSH)
 
+ifeq ("$(CUDA_VERSION_SHORT)","cpu")
+
+.PHONY: runtime-image
+runtime-image: BASE_IMAGE := $(BASE_RUNTIME)
+runtime-image: DOCKER_TAG := $(PYTORCH_VERSION)-runtime
+runtime-image:
+	$(DOCKER_BUILD)
+
+.PHONY: runtime-push
+runtime-push: BASE_IMAGE := $(BASE_RUNTIME)
+runtime-push: DOCKER_TAG := $(PYTORCH_VERSION)-runtime
+runtime-push:
+	$(DOCKER_PUSH)
+
+else
+
 .PHONY: runtime-image
 runtime-image: BASE_IMAGE := $(BASE_RUNTIME)
 runtime-image: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-runtime
@@ -95,6 +111,8 @@ runtime-push: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(
 runtime-push:
 	$(DOCKER_PUSH)
 
+endif
+
 .PHONY: clean
 clean:
 	-docker rmi -f $(shell docker images -q $(DOCKER_FULL_NAME))

setup.py

@@ -548,7 +548,9 @@ class build_ext(setuptools.command.build_ext.build_ext):
         omp_lib_name = (
             "libomp.dylib" if os.uname().machine == "arm64" else "libiomp5.dylib"
         )
-        if os.path.join("@rpath", omp_lib_name) not in libs:
+        omp_rpath_lib_path = os.path.join("@rpath", omp_lib_name)
+        omp_loader_lib_path = os.path.join("@loader_path", omp_lib_name)
+        if omp_rpath_lib_path not in libs:
             return
 
         # Copy libomp/libiomp5 from rpath locations
@@ -558,6 +560,20 @@ class build_ext(setuptools.command.build_ext.build_ext):
                 continue
             target_lib = os.path.join(self.build_lib, "torch", "lib", omp_lib_name)
             self.copy_file(source_lib, target_lib)
+            # Change OMP library load path to loader_path and delete old rpath
+            # This should prevent delocate from attempting to package another instance
+            # of OpenMP library in torch wheel
+            subprocess.check_call(
+                [
+                    "install_name_tool",
+                    "-change",
+                    omp_rpath_lib_path,
+                    omp_loader_lib_path,
+                    "-delete_rpath",
+                    rpath,
+                    libtorch_cpu_path,
+                ]
+            )
             break
 
         # Copy omp.h from OpenMP_C_FLAGS and copy it into include folder

test/distributed/_tensor/test_api.py

@@ -250,6 +250,44 @@ class DTensorAPITest(DTensorTestBase):
         self.assertIsInstance(local_out, torch.Tensor)
         self.assertNotIsInstance(local_out, DTensor)
 
+    @with_comms
+    def test_distribute_module_casting(self):
+        device_mesh = DeviceMesh(self.device_type, list(range(self.world_size)))
+
+        # check DTensor casting
+        dt = DTensor.from_local(torch.rand(10), device_mesh, [Replicate()])
+        dt = dt.to(torch.bfloat16)
+        self.assertEqual(dt.dtype, torch.bfloat16)
+        self.assertEqual(dt._local_tensor.dtype, torch.bfloat16)
+
+        # check distribute_tensor casting
+        dt = distribute_tensor(torch.rand(10), device_mesh, [Replicate()])
+        dt = dt.to(torch.bfloat16)
+        self.assertEqual(dt.dtype, torch.bfloat16)
+        self.assertEqual(dt._local_tensor.dtype, torch.bfloat16)
+
+        # check distribute_module casting
+        model = MyModel(10, 10, device=self.device_type)
+        replica_model = distribute_module(
+            model,
+            device_mesh,
+        )
+        replica_model = replica_model.to(torch.bfloat16)
+        self.assertEqual(replica_model.seq[0].weight.dtype, torch.bfloat16)
+        self.assertEqual(
+            replica_model.seq[0].weight._local_tensor.dtype, torch.bfloat16
+        )
+
+        # check autocast
+        dt = distribute_tensor(torch.rand(10), device_mesh, [Replicate()])
+        replica_model = distribute_module(
+            model,
+            device_mesh,
+        )
+        with torch.autocast(device_type=self.device_type, dtype=torch.bfloat16):
+            output = replica_model(dt)
+        self.assertEqual(output.dtype, torch.bfloat16)
+
     @with_comms
     def test_distribute_module_meta(self):
         # If  the model is too big, the user may first the create entire model on the meta device and then initialize

test/distributed/checkpoint/test_state_dict.py

@@ -22,6 +22,7 @@ from torch.distributed.checkpoint.state_dict import (
     _patch_model_state_dict,
     _patch_optimizer_state_dict,
     get_model_state_dict,
+    get_optimizer_state_dict,
     get_state_dict,
     set_model_state_dict,
     set_state_dict,
@@ -506,7 +507,7 @@ class TestStateDict(DTensorTestBase, VerifyStateDictMixin):
 
     @with_comms
     @skip_if_lt_x_gpu(1)
-    def test_activation_ckpt_fqns(self) -> None:
+    def test_activation_ckpt_fqns_ddp(self) -> None:
         """Tests that activation checkpointing prefixes are removed from module names"""
         model = CompositeParamModel(device=torch.device("cuda"))
         original_keys = get_model_state_dict(model).keys()
@@ -517,6 +518,68 @@ class TestStateDict(DTensorTestBase, VerifyStateDictMixin):
 
         self.assertEqual(original_keys, new_keys)
 
+    @with_comms
+    @skip_if_lt_x_gpu(1)
+    def test_extra_state(self) -> None:
+        model = CompositeParamModel(device=torch.device("cuda"))
+
+        def get_extra_state(self):
+            return "MyState"
+
+        def set_extra_state(self, state):
+            return
+
+        UnitModule.get_extra_state = get_extra_state
+        UnitModule.set_extra_state = set_extra_state
+
+        ddp_model = DDP(copy.deepcopy(model))
+        set_model_state_dict(ddp_model, get_model_state_dict(ddp_model))
+        self.assertEqual(model.state_dict()["u1._extra_state"], "MyState")
+        self.assertEqual(model.state_dict(), get_model_state_dict(ddp_model))
+
+    @with_comms
+    @skip_if_lt_x_gpu(1)
+    def test_activation_ckpt_fqns_fsdp1(self) -> None:
+        self.run_subtests(
+            {"use_orig_params": [True, False]},
+            self._test_activation_ckpt_fqns_fsdp1,
+        )
+
+    def _test_activation_ckpt_fqns_fsdp1(self, use_orig_params: bool) -> None:
+        """Tests that activation checkpointing prefixes are removed from module names"""
+        model = CompositeParamModel(device=torch.device("cuda"))
+        original_keys = get_model_state_dict(model).keys()
+
+        apply_activation_checkpointing(model)
+        model = FSDP(model, use_orig_params=use_orig_params)
+        new_keys = get_model_state_dict(model).keys()
+
+        self.assertEqual(original_keys, new_keys)
+
+    @with_comms
+    @skip_if_lt_x_gpu(1)
+    def test_non_persistent_buffers(self) -> None:
+        model = CompositeParamModel(device=torch.device("cuda"))
+        model.register_buffer(
+            "dont_save_me", torch.rand(100, device="cuda"), persistent=False
+        )
+        ddp_model = DDP(copy.deepcopy(model))
+        set_model_state_dict(ddp_model, get_model_state_dict(ddp_model))
+        self.assertEqual(model.state_dict(), get_model_state_dict(ddp_model))
+
+    @with_comms
+    @skip_if_lt_x_gpu(2)
+    def test_fsdp_root_not_initialized(self) -> None:
+        # This test verifies that FSDP root is not initialized but we should
+        # still be able to  get the state_dict without errors because
+        # fsdp_model.state_dict() will trigger the FSDP initialization.
+        device_mesh = init_device_mesh("cuda", (self.world_size,))
+        model = CompositeParamModel(device=torch.device("cuda"))
+        fsdp_model = FSDP(copy.deepcopy(model), device_mesh=device_mesh)
+        fsdp_optim = torch.optim.Adam(fsdp_model.parameters())
+        get_model_state_dict(fsdp_model)
+        get_optimizer_state_dict(fsdp_model, fsdp_optim)
+
 
 if __name__ == "__main__":
     run_tests()

test/distributed/fsdp/test_fsdp_tp_integration.py

@@ -18,6 +18,7 @@ from torch.distributed._tensor.debug import CommDebugMode
 from torch.distributed.fsdp.fully_sharded_data_parallel import (
     CPUOffload,
     FullyShardedDataParallel as FSDP,
+    ShardingStrategy,
 )
 from torch.distributed.tensor.parallel import (
     ColwiseParallel,
@@ -28,7 +29,6 @@ from torch.testing._internal.common_distributed import skip_if_lt_x_gpu
 from torch.testing._internal.common_fsdp import FSDPTest
 from torch.testing._internal.common_utils import (
     instantiate_parametrized_tests,
-    parametrize,
     run_tests,
     TEST_WITH_DEV_DBG_ASAN,
 )
@@ -141,31 +141,36 @@ class TestTPFSDPIntegration(FSDPTest):
         tp_world_size = tp_pg.size()
         fsdp_world_size = self.world_size // tp_world_size
         assert (
-            type(tp_fsdp_model) is FSDP and len(list(tp_fsdp_model.parameters())) == 1
+            type(tp_fsdp_model) is FSDP
+            and len([m for m in tp_fsdp_model.modules() if type(m) is FSDP]) == 1
         ), (
             "The following logic assumes a single top-level-only FSDP wrapping "
             "the model with TP already applied"
         )
-        flat_param = tp_fsdp_model.params[0]
-        splits = tuple(param_name_to_numel.values())
-        # Create a mask over the gradient elements to manually reduce
-        unsharded_size = torch.Size([flat_param.numel() * fsdp_world_size])
-        unsharded_zeros = torch.zeros(unsharded_size, device=flat_param.device)
-        per_param_masks = unsharded_zeros.split(splits)
-        for param_idx, param_name in enumerate(
-            param_name_to_numel.keys()
-        ):  # assumes fixed order
-            if param_name not in non_sharded_param_names:
-                per_param_masks[param_idx][:] = 1
-        unsharded_mask = torch.cat(per_param_masks).contiguous().type(torch.BoolTensor)
-        sharded_mask = unsharded_mask.chunk(fsdp_world_size)[self.rank // tp_world_size]
-        grad_device = flat_param.grad.device
-        grad = flat_param.grad.detach().clone().cuda(self.rank)
-        dist.all_reduce(grad, op=dist.ReduceOp.SUM, group=tp_pg)
-        grad = grad.to(grad_device)
-        flat_param.grad[~sharded_mask] = grad[~sharded_mask]
-        # Average *all* gradient elements to match the FSDP only semantics
-        flat_param.grad /= tp_world_size
+        for flat_param in tp_fsdp_model.params:
+            splits = tuple(param_name_to_numel.values())
+            # Create a mask over the gradient elements to manually reduce
+            unsharded_size = torch.Size([flat_param.numel() * fsdp_world_size])
+            unsharded_zeros = torch.zeros(unsharded_size, device=flat_param.device)
+            per_param_masks = unsharded_zeros.split(splits)
+            for param_idx, param_name in enumerate(
+                param_name_to_numel.keys()
+            ):  # assumes fixed order
+                if param_name not in non_sharded_param_names:
+                    per_param_masks[param_idx][:] = 1
+            unsharded_mask = (
+                torch.cat(per_param_masks).contiguous().type(torch.BoolTensor)
+            )
+            sharded_mask = unsharded_mask.chunk(fsdp_world_size)[
+                self.rank // tp_world_size
+            ]
+            grad_device = flat_param.grad.device
+            grad = flat_param.grad.detach().clone().cuda(self.rank)
+            dist.all_reduce(grad, op=dist.ReduceOp.SUM, group=tp_pg)
+            grad = grad.to(grad_device)
+            flat_param.grad[~sharded_mask] = grad[~sharded_mask]
+            # Average *all* gradient elements to match the FSDP only semantics
+            flat_param.grad /= tp_world_size
 
     def _get_grads_as_flattened(
         self,
@@ -182,7 +187,14 @@ class TestTPFSDPIntegration(FSDPTest):
         returns the same value on all ranks.
         """
         local_grads_as_flattened = (
-            torch.cat([torch.flatten(param.grad) for param in model.parameters()])
+            torch.cat(
+                [
+                    torch.flatten(param.grad)
+                    if param.grad is not None
+                    else torch.zeros_like(torch.flatten(param))
+                    for param in model.parameters()
+                ]
+            )
             .contiguous()
             .cuda(self.rank)
         )
@@ -214,16 +226,27 @@ class TestTPFSDPIntegration(FSDPTest):
         return torch.cat(all_grads_per_param).contiguous()
 
     @skip_if_lt_x_gpu(4)
-    @parametrize("tensor_parallel_size", [2, 4])
-    @parametrize(
-        "cpu_offload",
-        [CPUOffload(offload_params=False), CPUOffload(offload_params=True)],
-    )
-    def test_fsdp_tp_integration(self, tensor_parallel_size, cpu_offload):
+    def test_fsdp_tp_integration(self):
+        self.run_subtests(
+            {
+                "cpu_offload": [
+                    CPUOffload(offload_params=False),
+                    CPUOffload(offload_params=True),
+                ],
+                "sharding_strategy": [None, ShardingStrategy.SHARD_GRAD_OP],
+                "use_orig_params": [False, True],
+            },
+            self._test_fsdp_tp_integration,
+        )
+
+    def _test_fsdp_tp_integration(
+        self, cpu_offload, sharding_strategy, use_orig_params
+    ):
         """
         Tests training for TP + FSDP integration by comparing an FSDP-only
         model with a TP + FSDP model.
         """
+        tensor_parallel_size = 2
         LR = 3e-5
         torch.manual_seed(0)
         model = SimpleModel().cuda(self.rank)
@@ -246,7 +269,13 @@ class TestTPFSDPIntegration(FSDPTest):
         self.assertEqual(model(inp), tp_fsdp_model(inp))  # sanity check
 
         mesh_1d = init_device_mesh("cuda", (self.world_size,))
-        fsdp_model = FSDP(model, cpu_offload=cpu_offload, device_mesh=mesh_1d)
+        fsdp_model = FSDP(
+            model,
+            cpu_offload=cpu_offload,
+            device_mesh=mesh_1d,
+            sharding_strategy=sharding_strategy,
+            use_orig_params=use_orig_params,
+        )
         mesh_2d = init_device_mesh(
             "cuda",
             (self.world_size // tensor_parallel_size, tensor_parallel_size),
@@ -269,6 +298,8 @@ class TestTPFSDPIntegration(FSDPTest):
             tp_fsdp_model,
             cpu_offload=cpu_offload,
             device_mesh=mesh_2d["dp"],
+            sharding_strategy=sharding_strategy,
+            use_orig_params=use_orig_params,
         )
         fsdp_pg = mesh_2d["dp"].get_group(mesh_dim=0)
 

test/distributed/test_device_mesh.py

@@ -14,6 +14,7 @@ from torch.distributed._tensor.placement_types import _Partial, Shard
 from torch.distributed.device_mesh import _mesh_resources, DeviceMesh, init_device_mesh
 
 from torch.distributed.distributed_c10d import (
+    _world,
     get_global_rank,
     get_world_size,
     init_process_group,
@@ -28,7 +29,6 @@ from torch.testing._internal.common_utils import (
 )
 from torch.testing._internal.distributed._tensor.common_dtensor import (
     DTensorTestBase,
-    skip_if_lt_x_gpu,
     skip_unless_torch_gpu,
     with_comms,
 )
@@ -79,10 +79,7 @@ class DeviceMeshTest(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_group(self):
-        # TODO: `test_get_group` still periodically timeout on cpu
-        # remove `@skip_unless_torch_gpu` after the problem is fixed.
         mesh_shape = (2, self.world_size // 2)
         mesh_2d = init_device_mesh(
             self.device_type, mesh_shape, mesh_dim_names=("dp", "tp")
@@ -103,10 +100,7 @@ class DeviceMeshTest(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_local_rank_raises_exception(self):
-        # TODO: `test_get_local_rank_raises_exception` still periodically timeout on cpu
-        # remove `@skip_unless_torch_gpu` after the problem is fixed.
         mesh_shape = (2, self.world_size // 2)
         mesh_2d = init_device_mesh(
             self.device_type, mesh_shape, mesh_dim_names=("dp", "tp")
@@ -120,10 +114,7 @@ class DeviceMeshTest(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_local_rank(self):
-        # TODO: `test_get_local_rank_raises_exception` still periodically timeout on cpu
-        # remove `@skip_unless_torch_gpu` after the problem is fixed.
         mesh_shape = (2, self.world_size // 2)
         mesh_2d = init_device_mesh(
             self.device_type, mesh_shape, mesh_dim_names=("dp", "tp")
@@ -162,7 +153,17 @@ class DeviceMeshTest(DTensorTestBase):
             )
             self.assertEqual(global_ranks, current_rank_expected_group_ranks)
 
-    @run_with_both_funcol_impls
+    @with_comms
+    def test_device_mesh_init_backend(self):
+        mesh = DeviceMesh(self.device_type, [1], _init_backend=False)
+
+        with self.assertRaisesRegex(RuntimeError, "process groups not initialized!"):
+            mesh.get_group()
+
+        # coordinates should always been populated when init_backend is False, as whenever
+        # we call init_backend we should make sure the default pg already created
+        mesh.get_coordinate()
+
     def test_fake_pg_device_mesh(self):
         fake_store = FakeStore()
         init_process_group("fake", store=fake_store, rank=0, world_size=self.world_size)
@@ -276,71 +277,47 @@ class TestDeviceMeshGetItem(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    def test_raises_invalid_mesh_dim_names(self):
-        error_msg = "Invalid mesh_dim_name"
-        # Case 1: the DeviceMesh does not have a mesh_dim_names attribute
-        with self.assertRaisesRegex(
-            RuntimeError, "Cannot slice a DeviceMesh without mesh_dim_names."
-        ):
+    def test_raises_no_mesh_dim_found(self):
+        with self.assertRaisesRegex(KeyError, "No `mesh_dim_names` found."):
             mesh = init_device_mesh(self.device_type, (2, 4))
             child_mesh = mesh["DP"]
 
-        child_mesh_dim_names = "PP"
-        with self.assertRaisesRegex(ValueError, error_msg):
+    @with_comms
+    @run_with_both_funcol_impls
+    def test_raises_invalid_mesh_dim_name(self):
+        child_mesh_dim_name = "PP"
+        with self.assertRaisesRegex(
+            KeyError, f"Mesh dimension '{child_mesh_dim_name}' does not exist."
+        ):
             mesh_dim_names = ("DP", "TP")
             mesh = init_device_mesh(
                 self.device_type, (2, 4), mesh_dim_names=mesh_dim_names
             )
-            child_mesh = mesh[child_mesh_dim_names]
-
-        # Case 2
-        child_mesh_dim_names = ["PP", "CP"]
-        with self.assertRaisesRegex(ValueError, error_msg):
-            mesh_dim_names = ("DP", "TP")
-            mesh = init_device_mesh(
-                self.device_type, (2, 4), mesh_dim_names=mesh_dim_names
-            )
-            child_mesh = mesh[child_mesh_dim_names]
-
-        # Case 3: a given child_mesh_dim_name is not a contiguous subset of the parent mesh's mesh_dim_names.
-        child_mesh_dim_names = ("TP", "DP")
-        with self.assertRaisesRegex(ValueError, error_msg):
-            mesh_dim_names = ("DP", "TP")
-            mesh = init_device_mesh(
-                self.device_type, (2, 4), mesh_dim_names=mesh_dim_names
-            )
-            child_mesh = mesh[child_mesh_dim_names]
-
-        # Case 3
-        child_mesh_dim_names = ("PP", "TP")
-        with self.assertRaisesRegex(ValueError, error_msg):
-            mesh_dim_names = ("PP", "DP", "TP")
-            mesh = init_device_mesh(
-                self.device_type, (2, 2, 2), mesh_dim_names=mesh_dim_names
-            )
-            child_mesh = mesh[child_mesh_dim_names]
+            child_mesh = mesh[child_mesh_dim_name]
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_if_lt_x_gpu(8)
-    def test_get_item_2d(self):
-        # TODO: `test_get_item_2d` still periodically timeout on cpu
-        # remove `@skip_if_lt_x_gpu` after the problem is fixed.
+    def test_get_item(self):
         mesh_shape = (2, 4)
         mesh_dim_names = ("DP", "TP")
         mesh_2d = init_device_mesh(
             self.device_type, mesh_shape, mesh_dim_names=mesh_dim_names
         )
 
+        pg_ranks_by_dim_name = {}
+        for mesh_dim_name in mesh_dim_names:
+            mesh_dim = mesh_dim_names.index(mesh_dim_name)
+            pg_ranks_by_dim_name[mesh_dim_name] = mesh_2d.mesh.swapdims(
+                -1, mesh_dim
+            ).reshape(-1, mesh_2d.mesh.size(mesh_dim))
+
         tp_mesh = mesh_2d["TP"]
-        tp_group = [[0, 1, 2, 3], [4, 5, 6, 7]]
         tp_group_idx = self.rank // 4
-        self.assertEqual(tp_mesh.mesh.tolist(), tp_group[tp_group_idx])
+        self.assertEqual(tp_mesh.mesh, pg_ranks_by_dim_name["TP"][tp_group_idx])
 
         dp_mesh = mesh_2d["DP"]
-        dp_group = [[0, 4], [1, 5], [2, 6], [3, 7]]
         dp_group_idx = self.rank % 4
-        self.assertEqual(dp_mesh.mesh.tolist(), dp_group[dp_group_idx])
+        self.assertEqual(mesh_2d["DP"].mesh, pg_ranks_by_dim_name["DP"][dp_group_idx])
 
     @with_comms
     @run_with_both_funcol_impls
@@ -351,50 +328,32 @@ class TestDeviceMeshGetItem(DTensorTestBase):
         dp_mesh = mesh["dp"]
         self.assertEqual(dp_mesh, mesh)
 
-        with self.assertRaisesRegex(ValueError, "Invalid mesh_dim_name"):
+        with self.assertRaisesRegex(RuntimeError, "Invalid mesh_dim_name"):
             dp_mesh = mesh["dim0"]
 
     @with_comms
-    @skip_if_lt_x_gpu(8)
-    def test_get_item_3d(self):
-        # TODO: `test_get_item_3d` still periodically timeout on cpu
-        # remove `@skip_if_lt_x_gpu` after the problem is fixed.
-        mesh_shape = (2, 2, 2)
-        mesh_dim_names = ("Replicate", "Shard", "TP")
-        mesh_3d = init_device_mesh(
-            self.device_type, mesh_shape, mesh_dim_names=mesh_dim_names
-        )
+    @run_with_both_funcol_impls
+    def test_cache_and_reuse_submesh_slice_result(self):
+        mesh = init_device_mesh(self.device_type, (2, 4), mesh_dim_names=("dp", "tp"))
 
-        tp_group = [[0, 1], [2, 3], [4, 5], [6, 7]]
-        tp_group_idx = int(self.rank / 2)
-        self.assertEqual(mesh_3d["TP"].mesh.tolist(), tp_group[tp_group_idx])
+        dp_mesh = mesh["dp"]
+        ref_pg_count = _world.group_count
 
-        shard_group = [[0, 2], [1, 3], [4, 6], [5, 7]]
-        shard_group_idx = self.rank % 2 + self.rank // 4 * 2
-        self.assertEqual(mesh_3d["Shard"].mesh.tolist(), shard_group[shard_group_idx])
+        # When we call the "dp" slice second time, it should not create any new pg.
+        # As we are just using the cached result so the pg count should be the same.
+        dp_mesh_2 = mesh["dp"]
+        self.assertEqual(ref_pg_count, _world.group_count)
 
-        replicate_group = [[0, 4], [1, 5], [2, 6], [3, 7]]
-        replicate_group_idx = self.rank % 4
-        self.assertEqual(
-            mesh_3d["Replicate"].mesh.tolist(), replicate_group[replicate_group_idx]
-        )
-
-        # We support both UX for nD slicing.
-        # mesh_3d[["Replicate", "Shard"]] or mesh_3d["Replicate", "Shard"]
-        hsdp_mesh_1 = mesh_3d[["Replicate", "Shard"]]
-        hsdp_mesh_2 = mesh_3d["Replicate", "Shard"]
-        hsdp_group = [[[0, 2], [4, 6]], [[1, 3], [5, 7]]]
-        hsdp_group_idx = self.rank % 2
-        self.assertEqual(hsdp_mesh_1.mesh.tolist(), hsdp_group[hsdp_group_idx])
-        self.assertEqual(hsdp_mesh_2.mesh.tolist(), hsdp_group[hsdp_group_idx])
-        self.assertEqual(hsdp_mesh_1, hsdp_mesh_2)
+        # When we call the "tp" slice, it should not create a new pg, as the "tp" slice would
+        # just reuse the parent mesh pg.
+        tp_mesh = mesh["tp"]
+        self.assertEqual(_world.group_count, ref_pg_count)
 
 
 @instantiate_parametrized_tests
 class TestMeshEnv(DTensorTestBase):
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_parent_mesh(self):
         mesh_shape = (2, self.world_size // 2)
         mesh_dim_names = ("DP", "TP")
@@ -415,7 +374,6 @@ class TestMeshEnv(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_parent_mesh_dim_exist(self):
         mesh_shape = (2, self.world_size // 2)
         mesh_dim_names = ("DP", "TP")
@@ -428,7 +386,6 @@ class TestMeshEnv(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_parent_mesh_dim_not_exist(self):
         mesh_shape = (self.world_size,)
         mesh = init_device_mesh(self.device_type, mesh_shape)
@@ -437,7 +394,6 @@ class TestMeshEnv(DTensorTestBase):
 
     @with_comms
     @run_with_both_funcol_impls
-    @skip_unless_torch_gpu
     def test_get_mesh_dim_by_name(self):
         mesh_shape = (2, self.world_size // 2)
         mesh_dim_names = ("DP", "TP")
@@ -555,9 +511,11 @@ class DeviceMeshCollectiveTest(DTensorTestBase):
                 torch.chunk(big_tensor, device_mesh.size(), dim=shard_dim)
             )
             unpadded_list = [
-                shard_placement._unpad_tensor(big_tensor_chunks[i], pad_sizes[i])
-                if pad_sizes[i] > 0
-                else big_tensor_chunks[i]
+                (
+                    shard_placement._unpad_tensor(big_tensor_chunks[i], pad_sizes[i])
+                    if pad_sizes[i] > 0
+                    else big_tensor_chunks[i]
+                )
                 for i, big_tensor in enumerate(big_tensor_chunks)
             ]
             all_gathered_tensor = torch.cat(unpadded_list, dim=shard_dim)

test/dynamo/test_dynamic_shapes.py

@@ -87,10 +87,7 @@ if TEST_Z3:
 
     # TODO model is somehow not being freed when z3 is available
     unittest.expectedFailure(
-        DynamicShapesMiscTests.test_custom_module_free_dynamic_shapes  # noqa: F821
-    )
-    unittest.expectedFailure(
-        DynamicShapesMiscTests.test_sequential_module_free_dynamic_shapes  # noqa: F821
+        DynamicShapesMiscTests.test_parameter_free_dynamic_shapes  # noqa: F821
     )
 
 unittest.expectedFailure(

test/dynamo/test_misc.py

@@ -48,7 +48,6 @@ from torch._dynamo.testing import (
     same,
     skipIfNotPy311,
     unsupported,
-    xfailIfPy311,
 )
 from torch._dynamo.utils import CompileProfiler, counters, ifdynstaticdefault
 from torch._inductor.utils import run_and_get_code
@@ -9740,7 +9739,6 @@ fn
             lambda mod: mod.fc,
         )
 
-    @xfailIfPy311
     def test_sequential_module_free(self):
         self._test_compile_model_free(
             lambda: (
@@ -9753,13 +9751,48 @@ fn
             lambda mod: mod[0],
         )
 
-    @unittest.expectedFailure
     def test_linear_module_free(self):
         self._test_compile_model_free(
             lambda: (torch.nn.Linear(100, 100), torch.randn(100, 100)),
             lambda mod: mod,
         )
 
+    def test_outside_linear_module_free(self):
+        # Compared to test_linear_module_free, the linear
+        # layer is not the code object that is directly compiled.
+        def model_inp_ctr():
+            fc = torch.nn.Linear(100, 100)
+
+            class Mod(torch.nn.Module):
+                def __init__(self):
+                    super().__init__()
+                    self.fc_ref = fc
+
+                def forward(self, x):
+                    return self.fc_ref(x[0])
+
+            # return fc to keep it alive in _test_compile_model_free
+            return Mod(), (torch.randn(100, 100), fc)
+
+        self._test_compile_model_free(model_inp_ctr, lambda mod: mod.fc_ref)
+
+    def test_parameter_free(self):
+        def model_inp_ctr():
+            param = torch.nn.Parameter(torch.randn(100, 100))
+
+            class Mod(torch.nn.Module):
+                def __init__(self):
+                    super().__init__()
+                    self.param = param
+
+                def forward(self, x):
+                    return self.param * x[0]
+
+            # return param to keep it alive in _test_compile_model_free
+            return Mod(), (torch.randn(100, 100), param)
+
+        self._test_compile_model_free(model_inp_ctr, lambda mod: mod.param)
+
     def test_dynamo_cache_move_to_front(self):
         class Mod(torch.nn.Module):
             def __init__(self):

test/dynamo/test_subclasses.py

@@ -19,10 +19,10 @@ from torch.fx.experimental.symbolic_shapes import (
     StatelessSymbolicContext,
 )
 from torch.nested._internal.nested_tensor import (
-    buffer_from_jagged,
     jagged_from_list,
     jagged_from_tensor_and_lengths,
-    ViewBufferFromNested,
+    nested_view_from_values_offsets,
+    NestedTensor,
 )
 from torch.testing._internal.common_utils import (
     instantiate_parametrized_tests,
@@ -1166,19 +1166,20 @@ class TestNestedTensor(torch._dynamo.test_case.TestCase):
         a = torch.randn(2, 3, requires_grad=True, dtype=torch.float64)
         b = torch.randn(3, 3, requires_grad=True, dtype=torch.float64)
         c = torch.randn(4, 3, requires_grad=True, dtype=torch.float64)
-        nt, offsets = jagged_from_list([a, b, c], None)
-        nt2, _ = jagged_from_list([a, b, c], offsets)
+        nt = torch.nested.as_nested_tensor([a, b, c], layout=torch.jagged)
+        # TODO: Switch to public API when it exists
+        nt2, _ = jagged_from_list([a, b, c], nt.offsets())
 
         def fn1(nt1, nt2):
             return (nt1 + nt2).sin().cos()
 
         compiled_f = torch.compile(fn1, fullgraph=True, backend=backend, dynamic=True)
         out = compiled_f(nt, nt2)
-        out_buffer = ViewBufferFromNested.apply(out)
+        out_buffer = out.values()
         ga, gb, gc = torch.autograd.grad(out_buffer.sum(), (a, b, c))
 
         out_ref = fn1(nt, nt2)
-        out_buffer_ref = ViewBufferFromNested.apply(out_ref)
+        out_buffer_ref = out_ref.values()
         ga_ref, gb_ref, gc_ref = torch.autograd.grad(out_buffer_ref.sum(), (a, b, c))
 
         self.assertTrue(torch.allclose(ga, ga_ref))
@@ -1218,10 +1219,10 @@ class TestNestedTensor(torch._dynamo.test_case.TestCase):
         ret = fn_c(nt, y)[0]
         ref = fn(nt_copy, y_copy)[0]
 
-        self.assertEqual(buffer_from_jagged(ret), buffer_from_jagged(ref))
+        self.assertEqual(ret.values(), ref.values())
 
-        buffer_from_jagged(ret).sum().backward()
-        buffer_from_jagged(ref).sum().backward()
+        ret.values().sum().backward()
+        ref.values().sum().backward()
         for ref_v, res_v in zip(values_copy, values):
             self.assertEqual(ref_v.grad, res_v.grad)
 
@@ -1254,83 +1255,112 @@ class TestNestedTensor(torch._dynamo.test_case.TestCase):
         self._check_recompiles(fn, (nt,), (nt3,), True)
 
     def _get_views(self):
-        # There are three cases to consider here based on the logic in
-        # meta_utils.py
-        #
-        # (1) basic case:
-        # view is not a leaf and has the same requires grad as its basic case
-        x, _ = self._get_jagged_tensor(((2, 3, 4), 3), None, requires_grad=True)
-        self.assertEqual(x.is_leaf, False)
-        yield x.unsqueeze(-1)
+        # Test all cases with both an NT base and a dense base
+        # Subclass -> Subclass
+        # Dense -> Subclass
+        for base_is_nt in [False, True]:
+            # There are three cases to consider here based on the logic in
+            # meta_utils.py
+            #
+            # (1) basic case:
+            # view is not a leaf and has the same requires grad as its basic case
+            x, _ = self._get_jagged_tensor(((2, 3, 4), 3), None, requires_grad=True)
+            x = x.clone() if base_is_nt else x
+            self.assertEqual(x.is_leaf, False)
+            yield x.unsqueeze(-1)
 
-        # (2) leaf view case:
-        # the view has to be a leaf (w/ requires_grad True or requires_grad False)
-        # base w/ requires_grad True or requires_grad False
-        for requires_grad_1, requires_grad_2 in itertools.product(
-            [True, False], repeat=2
-        ):
-            x, _ = self._get_jagged_tensor(
-                ((2, 3, 4), 3), None, requires_grad=requires_grad_1
-            )
+            # (2) leaf view case:
+            # the view has to be a leaf (w/ requires_grad True or requires_grad False)
+            # base w/ requires_grad True or requires_grad False
+            for requires_grad_1, requires_grad_2 in itertools.product(
+                [True, False], repeat=2
+            ):
+                x, _ = self._get_jagged_tensor(
+                    ((2, 3, 4), 3), None, requires_grad=requires_grad_1
+                )
+                x = x.clone() if base_is_nt else x
+                with torch.no_grad():
+                    x_view = x.unsqueeze(-1)
+                    # The issue is this doesn't quite work
+                    x_view.requires_grad_(requires_grad_2)
+                yield x_view
+
+            # (3) obscure case:
+            # view is not a leaf (implies requires_grad True)
+            # base w/ requires_grad False)
+            x, _ = self._get_jagged_tensor(((2, 3, 4), 3), None, requires_grad=False)
+            x = x.clone() if base_is_nt else x
+            # intermediate leaf view
             with torch.no_grad():
                 x_view = x.unsqueeze(-1)
-                # The issue is this doesn't quite work
-                x_view.requires_grad_(requires_grad_2)
-            yield x_view
+            x_view.requires_grad_(True)
+            x_view_view = x_view.unsqueeze(-1)
+            yield x_view_view
 
-        # (3) obscure case:
-        # view is not a leaf (implies requires_grad True)
-        # base w/ requires_grad False)
-        x, _ = self._get_jagged_tensor(((2, 3, 4), 3), None, requires_grad=False)
-        # intermediate leaf view
-        with torch.no_grad():
-            x_view = x.unsqueeze(-1)
-        x_view.requires_grad_(True)
-        x_view_view = x_view.unsqueeze(-1)
-        yield x_view_view
+        # Subclass -> Dense
+        x = self._get_jagged_tensor(((2, 3, 4), 3), None, requires_grad=True)[0].clone()
+        yield x.values()
+
+        # Dense -> Subclass -> Dense -> Subclass
+        values = torch.randn(10, 5)
+        offsets = torch.tensor([0, 3, 6, 10])
+        offsets2 = offsets.clone().detach()
+        yield nested_view_from_values_offsets(
+            nested_view_from_values_offsets(values, offsets).values(), offsets
+        )
+
+    def _input_view_test(self, nt_view):
+        def fn(x):
+            return x.sin()
+
+        out_ref = fn(nt_view)
+        torch._dynamo.reset()
+        compile_fn = torch.compile(
+            fn, fullgraph=True, backend="aot_eager", dynamic=True
+        )
+        out = compile_fn(nt_view)
+
+        # Check metadata and values are correct
+        self.assertTrue(out.size() == out_ref.size())
+        self.assertTrue(out.stride() == out_ref.stride())
+        if out.is_nested:
+            self.assertTrue(torch.allclose(out.values(), out_ref.values()))
+        else:
+            self.assertTrue(torch.allclose(out, out_ref))
+
+        # Check that no upper/lower bound guards are incurred
+        def backend(gm, args):
+            context = torch._guards.TracingContext.get()
+            guards = [str(g.expr) for g in context.fake_mode.shape_env.guards]
+
+            # varies based on the type of view
+            guard_str = "\n".join(guards)
+            if isinstance(nt_view._base, NestedTensor):
+                self.assertExpectedInline(guard_str, """Eq(s3 - 1, s0)""")
+            else:
+                self.assertExpectedInline(guard_str, """""")
+            return gm
+
+        torch._dynamo.reset()
+        compile_fn = torch.compile(fn, fullgraph=True, backend=backend, dynamic=True)
+        out = compile_fn(nt_view)
 
     def test_inputs_to_compiled_fn_are_views(self):
         for nt_view in self._get_views():
+            self._input_view_test(nt_view)
 
-            def fn(x):
-                return x.sin()
-
-            out_ref = fn(nt_view)
-            torch._dynamo.reset()
-            compile_fn = torch.compile(
-                fn, fullgraph=True, backend="aot_eager", dynamic=True
-            )
-            out = compile_fn(nt_view)
-
-            # Check metadata and values are correct
-            self.assertTrue(out.size() == out_ref.size())
-            self.assertTrue(out.stride() == out_ref.stride())
-            self.assertTrue(torch.allclose(out.values(), out_ref.values()))
-
-            # Check that no upper/lower bound guards are incurred
-            def backend(gm, args):
-                context = torch._guards.TracingContext.get()
-                guards = [str(g.expr) for g in context.fake_mode.shape_env.guards]
-                ranges = [
-                    f"{s}: [{vr.lower}, {vr.upper}]"
-                    for s, vr in context.fake_mode.shape_env.var_to_range.items()
-                ]
-                self.assertExpectedInline("\n".join(guards), """Eq(s3 - 1, s0)""")
-                self.assertExpectedInline(
-                    "\n".join(ranges),
-                    """\
-s0: [2, 9223372036854775805]
-s2: [2, 9223372036854775806]
-s3: [3, 9223372036854775806]
-s5: [2, 9223372036854775806]""",
-                )
-                return gm
-
-            torch._dynamo.reset()
-            compile_fn = torch.compile(
-                fn, fullgraph=True, backend=backend, dynamic=True
-            )
-            out = compile_fn(nt_view)
+    # NJT1 -> Dense -> NJT2 -> Dense view
+    # During view replay, the Dense -> NJT2 part will construct an intermediate,
+    # symbolically-sized NJT that is immediately deconstructed to return the final dense
+    # view. To construct this intermediate properly, we need the associated nested int
+    # to be symbolic. This view is expected to fail compilation until symbolic nested ints
+    # are cached onto fake offsets to solve this problem.
+    @unittest.expectedFailure
+    def test_subclass_dense_subclass_dense_view(self):
+        x = self._get_jagged_tensor(((2, 3, 4), 3), None, requires_grad=True)[0].clone()
+        offsets2 = x.offsets().clone().detach()
+        nt_view = nested_view_from_values_offsets(x.values(), offsets2).values()
+        self._input_view_test(nt_view)
 
 
 if __name__ == "__main__":

test/expect/HasDecompTest.test_has_decomposition.expect

@@ -435,6 +435,13 @@ aten::_nested_from_padded
 aten::_nested_from_padded.out
 aten::_nested_from_padded_and_nested_example
 aten::_nested_from_padded_and_nested_example.out
+aten::_nested_get_jagged_dummy
+aten::_nested_get_lengths
+aten::_nested_get_offsets
+aten::_nested_get_ragged_idx
+aten::_nested_get_values
+aten::_nested_get_values_copy
+aten::_nested_get_values_copy.out
 aten::_nested_select_backward
 aten::_nested_sum_backward
 aten::_nested_tensor_from_mask
@@ -452,6 +459,9 @@ aten::_nested_tensor_strides.out
 aten::_nested_view_from_buffer
 aten::_nested_view_from_buffer_copy
 aten::_nested_view_from_buffer_copy.out
+aten::_nested_view_from_jagged
+aten::_nested_view_from_jagged_copy
+aten::_nested_view_from_jagged_copy.out
 aten::_new_zeros_with_same_feature_meta
 aten::_new_zeros_with_same_feature_meta.out
 aten::_nnpack_spatial_convolution

test/export/test_export.py

@@ -72,12 +72,28 @@ from torch.export import export
 
 
 torch.library.define("testlib::returns_tensor_symint", "(Tensor x) -> (Tensor, SymInt)")
+torch.library.define(
+    "testlib::foo",
+    "(Tensor(a!) x, Tensor(b!) z) -> (Tensor, Tensor, Tensor)",
+    tags=torch.Tag.pt2_compliant_tag,
+)
 
 @torch.library.impl("testlib::returns_tensor_symint", "cpu")
 @torch.library.impl_abstract("testlib::returns_tensor_symint")
 def returns_tensor_symint_impl(x):
     return x, x.shape[0]
 
+@torch.library.impl("testlib::foo", "cpu")
+@torch._dynamo.disable
+def foo_impl(x, z):
+    x.add_(5)
+    z.add_(5)
+    return x, z, x + z
+
+@torch.library.impl_abstract("testlib::foo")
+def foo_abstract(x, z):
+    return x, z, x + z
+
 
 @unittest.skipIf(not torchdynamo.is_dynamo_supported(), "dynamo isn't support")
 class TestDynamismExpression(TestCase):
@@ -3551,6 +3567,30 @@ def forward(self, arg0_1, arg1_1, arg2_1):
 
         self._test_export_same_as_eager(Module(), (torch.randn(4, 4),))
 
+    def test_custom_op_auto_functionalize(self):
+        class M(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+
+            def forward(self, x, z):
+                return torch.ops.testlib.foo(x, z)
+
+        inps = (torch.ones(5), torch.ones(5))
+        inps_for_export = (torch.ones(5), torch.ones(5))
+        inps_for_export_with_decomp = (torch.ones(5), torch.ones(5))
+
+        ep = torch.export.export(M(), inps_for_export)
+        x_new_eager, z_new_eager, legit_eager = M()(*inps)
+        x_new_export, z_new_export, legit_export = ep.module()(*inps_for_export)
+        self.assertTrue(torch.allclose(x_new_eager, x_new_export))
+        self.assertTrue(torch.allclose(z_new_eager, z_new_export))
+        self.assertTrue(torch.allclose(legit_eager, legit_export))
+
+        ep = ep.run_decompositions()
+        x_new_export, z_new_export, legit_export = ep.module()(*inps_for_export_with_decomp)
+        self.assertTrue(torch.allclose(x_new_eager, x_new_export))
+        self.assertTrue(torch.allclose(z_new_eager, z_new_export))
+        self.assertTrue(torch.allclose(legit_eager, legit_export))
 
 @unittest.skipIf(not torchdynamo.is_dynamo_supported(), "dynamo isn't support")
 class TestOneOffModelExportResult(TestCase):

test/export/test_passes.py

@@ -7,8 +7,8 @@ with test_functionalization_with_native_python_assertion)
 import math
 import operator
 import unittest
-from typing import List, Set
 from re import escape
+from typing import List, Set
 
 import torch
 from functorch.experimental.control_flow import cond
@@ -17,22 +17,38 @@ from torch._export.pass_base import _ExportPassBaseDeprecatedDoNotUse
 from torch._export.passes.functionalize_side_effectful_ops_pass import (
     _FunctionalizeSideEffectfulOpsPass,
 )
+from torch._export.passes.replace_set_grad_with_hop_pass import (
+    _is_set_grad_enabled_node,
+    _is_set_grad_enabled_sub_mod,
+)
 from torch._export.passes.replace_view_ops_with_view_copy_ops_pass import (
     get_view_copy_of_view_op,
     is_view_op,
     ReplaceViewOpsWithViewCopyOpsPass,
 )
+from torch._export.utils import (
+    node_inline_,
+    nodes_count,
+    nodes_filter,
+    nodes_map,
+    sequential_split,
+)
+from torch._higher_order_ops.auto_functionalize import auto_functionalized
 from torch.export import export
+from torch.export._remove_auto_functionalized_pass import (
+    unsafe_remove_auto_functionalized_pass,
+)
 from torch.fx.passes.infra.partitioner import Partition
 from torch.fx.passes.operator_support import OperatorSupport
+from torch.library import impl, _scoped_library
 from torch.testing import FileCheck
-from torch.testing._internal.common_utils import run_tests, TestCase, skipIfTorchDynamo, IS_WINDOWS
-from torch.utils import _pytree as pytree
-from torch._export.utils import sequential_split, nodes_filter, nodes_map, node_inline_, nodes_count
-from torch._export.passes.replace_set_grad_with_hop_pass import (
-    _is_set_grad_enabled_node, _is_set_grad_enabled_sub_mod
+from torch.testing._internal.common_utils import (
+    IS_WINDOWS,
+    run_tests,
+    skipIfTorchDynamo,
+    TestCase,
 )
-
+from torch.utils import _pytree as pytree
 
 def count_call_function(graph: torch.fx.Graph, target: torch.ops.OpOverload) -> int:
     count = 0
@@ -620,5 +636,96 @@ def forward(self, sin, cos):
             self.assertEqual(before_str, after_inline_str)
             self.assertEqual(gm(*args), new_gm(*args))
 
+    def test_remove_auto_functionalized_pass(self) -> None:
+        with _scoped_library("DO_NOT_USE_TEST_ONLY", "DEF") as lib:
+
+            lib.define("custom_mutator(Tensor x, Tensor(a!) y) -> Tensor")
+
+            @impl(lib, "custom_mutator", "Meta")
+            def custom_mutator_meta(
+                x: torch.Tensor,
+                y: torch.Tensor,
+            ) -> torch.Tensor:
+                return torch.empty_like(x)
+
+
+            @impl(lib, "custom_mutator", "CompositeExplicitAutograd")
+            def custom_mutator(
+                x: torch.Tensor,
+                y: torch.Tensor,
+            ) -> torch.Tensor:
+                return x + y.add_(1)
+
+            class M(torch.nn.Module):
+                def __init__(self):
+                    super().__init__()
+                    self.register_buffer("state", torch.zeros(1))
+
+                def forward(self, x):
+                    return torch.ops.DO_NOT_USE_TEST_ONLY.custom_mutator(x, self.state)
+
+            mod = M()
+            x = torch.randn([3, 3])
+            ep = export(mod, (x,))
+            inplace_ep = unsafe_remove_auto_functionalized_pass(ep)
+            nodes = inplace_ep.graph.nodes
+            for node in nodes:
+                if node.op == "call_function":
+                    self.assertFalse(node.target is auto_functionalized)
+                    self.assertFalse(node.target is operator.getitem)
+
+            for spec in inplace_ep.graph_signature.output_specs:
+                self.assertFalse("getitem" in spec.arg.name)
+
+    def test_remove_auto_functionalized_pass_tuple(self) -> None:
+        with _scoped_library("DO_NOT_USE_TEST_ONLY", "DEF") as lib:
+
+            lib.define("custom_mutator_tuple(Tensor x, Tensor(a!) y) -> (Tensor, Tensor)")
+
+            @impl(lib, "custom_mutator_tuple", "Meta")
+            def custom_mutator_tuple_meta(
+                x: torch.Tensor,
+                y: torch.Tensor,
+            ):
+                return (torch.empty_like(x), torch.empty_like(x))
+
+
+            @impl(lib, "custom_mutator_tuple", "CompositeExplicitAutograd")
+            def custom_mutator_tuple(
+                x: torch.Tensor,
+                y: torch.Tensor,
+            ):
+                return (x, x + y.add_(1))
+
+            class M(torch.nn.Module):
+                def __init__(self):
+                    super().__init__()
+                    self.register_buffer("state", torch.zeros(1))
+
+                def forward(self, x):
+                    return torch.ops.DO_NOT_USE_TEST_ONLY.custom_mutator_tuple(
+                        x, self.state
+                    )
+
+            mod = M()
+            x = torch.randn([3, 3])
+            ep = export(mod, (x,))
+            inplace_ep = unsafe_remove_auto_functionalized_pass(ep)
+
+            nodes = inplace_ep.graph.nodes
+            getitems = 0
+            for node in nodes:
+                if node.op == "call_function":
+                    self.assertFalse(node.target is auto_functionalized)
+                    if node.target is operator.getitem:
+                        getitems += 1
+            self.assertEqual(getitems, 2)  # tuple return of len 2
+
+            out_specs = inplace_ep.graph_signature.output_specs
+            self.assertEqual(out_specs[0].arg.name, "arg0_1")  # state
+            self.assertEqual(out_specs[1].arg.name, "getitem")  # tuple return 1
+            self.assertEqual(out_specs[2].arg.name, "getitem_1")  # tuple return 2
+
+
 if __name__ == '__main__':
     run_tests()