Cherrypick: Add the RunLLM widget to the website (#159592)

Add the RunLLM widget to the website (#152055)

Fixes #ISSUE_NUMBER

Pull Request resolved: https://github.com/pytorch/pytorch/pull/152055
Approved by: https://github.com/albanD

(cherry picked from commit e4e2701429c17078c3c475382a8b1fa4c8a8cefc)

.bazelrc

@@ -2,7 +2,7 @@ build --cxxopt=--std=c++17
 build --copt=-I.
 # Bazel does not support including its cc_library targets as system
 # headers. We work around this for generated code
-# (e.g. torch/headeronly/macros/cmake_macros.h) by making the generated directory a
+# (e.g. c10/macros/cmake_macros.h) by making the generated directory a
 # system include path.
 build --copt=-isystem --copt bazel-out/k8-fastbuild/bin
 build --copt=-isystem --copt bazel-out/darwin-fastbuild/bin

.ci/aarch64_linux/aarch64_wheel_ci_build.py

@@ -79,7 +79,6 @@ def package_cuda_wheel(wheel_path, desired_cuda) -> None:
     os.system(f"unzip {wheel_path} -d {folder}/tmp")
     libs_to_copy = [
         "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12",
-        "/usr/local/cuda/extras/CUPTI/lib64/libnvperf_host.so",
         "/usr/local/cuda/lib64/libcudnn.so.9",
         "/usr/local/cuda/lib64/libcublas.so.12",
         "/usr/local/cuda/lib64/libcublasLt.so.12",

.ci/caffe2/test.sh

@@ -5,7 +5,7 @@ source "$(dirname "${BASH_SOURCE[0]}")/common.sh"
 
 if [[ ${BUILD_ENVIRONMENT} == *onnx* ]]; then
   pip install click mock tabulate networkx==2.0
-  pip -q install "file:///var/lib/jenkins/workspace/third_party/onnx#egg=onnx"
+  pip -q install --user "file:///var/lib/jenkins/workspace/third_party/onnx#egg=onnx"
 fi
 
 # Skip tests in environments where they are not built/applicable
@@ -147,8 +147,8 @@ export DNNL_MAX_CPU_ISA=AVX2
 if [[ "${SHARD_NUMBER:-1}" == "1" ]]; then
   # TODO(sdym@meta.com) remove this when the linked issue resolved.
   # py is temporary until https://github.com/Teemu/pytest-sugar/issues/241 is fixed
-  pip install py==1.11.0
-  pip install pytest-sugar
+  pip install --user py==1.11.0
+  pip install --user pytest-sugar
   # NB: Warnings are disabled because they make it harder to see what
   # the actual erroring test is
   "$PYTHON" \

.ci/docker/README.md

@@ -36,104 +36,3 @@ See `build.sh` for valid build environments (it's the giant switch).
 # Set flags (see build.sh) and build image
 sudo bash -c 'TRITON=1 ./build.sh pytorch-linux-bionic-py3.8-gcc9 -t myimage:latest
 ```
-
-## [Guidance] Adding a New Base Docker Image
-
-### Background
-
-The base Docker images in directory `.ci/docker/` are built by the `docker-builds.yml` workflow. Those images are used throughout the PyTorch CI/CD pipeline. You should only create or modify a base Docker image if you need specific environment changes or dependencies before building PyTorch on CI.
-
-1. **Automatic Rebuilding**:
-   - The Docker image building process is triggered automatically when changes are made to files in the `.ci/docker/*` directory
-   - This ensures all images stay up-to-date with the latest dependencies and configurations
-
-2. **Image Reuse in PyTorch Build Workflows** (example: linux-build):
-   - The images generated by `docker-builds.yml` are reused in `_linux-build.yml` through the `calculate-docker-image` step
-   - The `_linux-build.yml` workflow:
-     - Pulls the Docker image determined by the `calculate-docker-image` step
-     - Runs a Docker container with that image
-     - Executes `.ci/pytorch/build.sh` inside the container to build PyTorch
-
-3. **Usage in Test Workflows** (example: linux-test):
-   - The same Docker images are also used in `_linux-test.yml` for running tests
-   - The `_linux-test.yml` workflow follows a similar pattern:
-     - It uses the `calculate-docker-image` step to determine which Docker image to use
-     - It pulls the Docker image and runs a container with that image
-     - It installs the wheels from the artifacts generated by PyTorch build jobs
-     - It executes test scripts (like `.ci/pytorch/test.sh` or `.ci/pytorch/multigpu-test.sh`) inside the container
-
-### Understanding File Purposes
-
-#### `.ci/docker/build.sh` vs `.ci/pytorch/build.sh`
-- **`.ci/docker/build.sh`**:
-  - Used for building base Docker images
-  - Executed by the `docker-builds.yml` workflow to pre-build Docker images for CI
-  - Contains configurations for different Docker build environments
-
-- **`.ci/pytorch/build.sh`**:
-  - Used for building PyTorch inside a Docker container
-  - Called by workflows like `_linux-build.yml` after the Docker container is started
-  - Builds PyTorch wheels and other artifacts
-
-#### `.ci/docker/ci_commit_pins/` vs `.github/ci_commit_pins`
-- **`.ci/docker/ci_commit_pins/`**:
-  - Used for pinning dependency versions during base Docker image building
-  - Ensures consistent environments for building PyTorch
-  - Changes here trigger base Docker image rebuilds
-
-- **`.github/ci_commit_pins`**:
-  - Used for pinning dependency versions during PyTorch building and tests
-  - Ensures consistent dependencies for PyTorch across different builds
-  - Used by build scripts running inside Docker containers
-
-### Step-by-Step Guide for Adding a New Base Docker Image
-
-#### 1. Add Pinned Commits (If Applicable)
-
-We use pinned commits for build stability. The `nightly.yml` workflow checks and updates pinned commits for certain repository dependencies daily.
-
-If your new Docker image needs a library installed from a specific pinned commit or built from source:
-
-1. Add the repository you want to track in `nightly.yml` and `merge-rules.yml`
-2. Add the initial pinned commit in `.ci/docker/ci_commit_pins/`. The text filename should match the one defined in step 1
-
-#### 2. Configure the Base Docker Image
-1. **Add new Base Docker image configuration** (if applicable):
-
-   Add the configuration in `.ci/docker/build.sh`. For example:
-   ```bash
-   pytorch-linux-jammy-cuda12.8-cudnn9-py3.12-gcc11-new1)
-     CUDA_VERSION=12.8.1
-     ANACONDA_PYTHON_VERSION=3.12
-     GCC_VERSION=11
-     VISION=yes
-     KATEX=yes
-     UCX_COMMIT=${_UCX_COMMIT}
-     UCC_COMMIT=${_UCC_COMMIT}
-     TRITON=yes
-     NEW_ARG_1=yes
-     ;;
-   ```
-
-2. **Add build arguments to Docker build command**:
-
-   If you're introducing a new argument to the Docker build, make sure to add it in the Docker build step in `.ci/docker/build.sh`:
-   ```bash
-   docker build \
-      ....
-      --build-arg "NEW_ARG_1=${NEW_ARG_1}"
-   ```
-
-3. **Update Dockerfile logic**:
-
-   Update the Dockerfile to use the new argument. For example, in `ubuntu/Dockerfile`:
-   ```dockerfile
-   ARG NEW_ARG_1
-   # Set up environment for NEW_ARG_1
-   RUN if [ -n "${NEW_ARG_1}" ]; then bash ./do_something.sh; fi
-   ```
-
-4. **Add the Docker configuration** in `.github/workflows/docker-builds.yml`:
-
-   The `docker-builds.yml` workflow pre-builds the Docker images whenever changes occur in the `.ci/docker/` directory. This includes the
-   pinned commit updates.

.ci/docker/build.sh

@@ -52,8 +52,6 @@ fi
 
 if [[ "$image" == *-jammy* ]]; then
   UBUNTU_VERSION=22.04
-elif [[ "$image" == *-noble* ]]; then
-  UBUNTU_VERSION=24.04
 elif [[ "$image" == *ubuntu* ]]; then
   extract_version_from_image_name ubuntu UBUNTU_VERSION
 fi
@@ -91,18 +89,9 @@ tag=$(echo $image | awk -F':' '{print $2}')
 # configuration, so we hardcode everything here rather than do it
 # from scratch
 case "$tag" in
-  pytorch-linux-jammy-cuda12.4-cudnn9-py3-gcc11)
-    CUDA_VERSION=12.4
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=11
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    TRITON=yes
-    ;;
   pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11)
     CUDA_VERSION=12.8.1
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=11
     VISION=yes
@@ -113,6 +102,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.8.1
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=9
     VISION=yes
@@ -124,6 +114,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.8-cudnn9-py3.12-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.8.1
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.12
     GCC_VERSION=9
     VISION=yes
@@ -135,6 +126,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.8-cudnn9-py3.13-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.8.1
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.13
     GCC_VERSION=9
     VISION=yes
@@ -146,6 +138,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.6-cudnn9-py3-gcc9)
     CUDA_VERSION=12.6.3
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=9
     VISION=yes
@@ -154,18 +147,9 @@ case "$tag" in
     UCC_COMMIT=${_UCC_COMMIT}
     TRITON=yes
     ;;
-  pytorch-linux-jammy-cuda12.8-cudnn9-py3.12-gcc11-vllm)
-    CUDA_VERSION=12.8.1
-    ANACONDA_PYTHON_VERSION=3.12
-    GCC_VERSION=11
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    TRITON=yes
-    ;;
   pytorch-linux-jammy-cuda12.6-cudnn9-py3-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.6
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=9
     VISION=yes
@@ -177,6 +161,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.6-cudnn9-py3.12-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.6
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.12
     GCC_VERSION=9
     VISION=yes
@@ -188,6 +173,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.6-cudnn9-py3.13-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.6
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.13
     GCC_VERSION=9
     VISION=yes
@@ -199,6 +185,7 @@ case "$tag" in
     ;;
   pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9)
     CUDA_VERSION=12.8.1
+    CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=9
     VISION=yes
@@ -231,12 +218,20 @@ case "$tag" in
     VISION=yes
     TRITON=yes
     ;;
-  pytorch-linux-jammy-rocm-n-py3 | pytorch-linux-noble-rocm-n-py3)
-    if [[ $tag =~ "jammy" ]]; then
-      ANACONDA_PYTHON_VERSION=3.10
-    else
-      ANACONDA_PYTHON_VERSION=3.12
-    fi
+  pytorch-linux-jammy-rocm-n-1-py3)
+    ANACONDA_PYTHON_VERSION=3.10
+    GCC_VERSION=11
+    VISION=yes
+    ROCM_VERSION=6.3
+    NINJA_VERSION=1.9.0
+    TRITON=yes
+    KATEX=yes
+    UCX_COMMIT=${_UCX_COMMIT}
+    UCC_COMMIT=${_UCC_COMMIT}
+    INDUCTOR_BENCHMARKS=yes
+    ;;
+  pytorch-linux-jammy-rocm-n-py3)
+    ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=11
     VISION=yes
     ROCM_VERSION=6.4
@@ -247,19 +242,6 @@ case "$tag" in
     UCC_COMMIT=${_UCC_COMMIT}
     INDUCTOR_BENCHMARKS=yes
     ;;
-  pytorch-linux-noble-rocm-alpha-py3)
-    ANACONDA_PYTHON_VERSION=3.12
-    GCC_VERSION=11
-    VISION=yes
-    ROCM_VERSION=7.0
-    NINJA_VERSION=1.9.0
-    TRITON=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    INDUCTOR_BENCHMARKS=yes
-    PYTORCH_ROCM_ARCH="gfx90a;gfx942;gfx950"
-    ;;
   pytorch-linux-jammy-xpu-2025.0-py3)
     ANACONDA_PYTHON_VERSION=3.9
     GCC_VERSION=11
@@ -276,7 +258,7 @@ case "$tag" in
     NINJA_VERSION=1.9.0
     TRITON=yes
     ;;
-  pytorch-linux-jammy-py3.9-gcc11-inductor-benchmarks)
+    pytorch-linux-jammy-py3.9-gcc11-inductor-benchmarks)
     ANACONDA_PYTHON_VERSION=3.9
     GCC_VERSION=11
     VISION=yes
@@ -288,10 +270,22 @@ case "$tag" in
   pytorch-linux-jammy-cuda12.8-cudnn9-py3.9-clang12)
     ANACONDA_PYTHON_VERSION=3.9
     CUDA_VERSION=12.8.1
+    CUDNN_VERSION=9
     CLANG_VERSION=12
     VISION=yes
     TRITON=yes
     ;;
+  pytorch-linux-jammy-py3-clang12-asan)
+    ANACONDA_PYTHON_VERSION=3.9
+    CLANG_VERSION=12
+    VISION=yes
+    TRITON=yes
+    ;;
+  pytorch-linux-jammy-py3-clang15-asan)
+    ANACONDA_PYTHON_VERSION=3.10
+    CLANG_VERSION=15
+    VISION=yes
+    ;;
   pytorch-linux-jammy-py3-clang18-asan)
     ANACONDA_PYTHON_VERSION=3.10
     CLANG_VERSION=18
@@ -366,6 +360,7 @@ case "$tag" in
     fi
     if [[ "$image" == *cuda* ]]; then
       extract_version_from_image_name cuda CUDA_VERSION
+      extract_version_from_image_name cudnn CUDNN_VERSION
     fi
     if [[ "$image" == *rocm* ]]; then
       extract_version_from_image_name rocm ROCM_VERSION
@@ -417,6 +412,9 @@ docker build \
        --build-arg "PYTHON_VERSION=${PYTHON_VERSION}" \
        --build-arg "GCC_VERSION=${GCC_VERSION}" \
        --build-arg "CUDA_VERSION=${CUDA_VERSION}" \
+       --build-arg "CUDNN_VERSION=${CUDNN_VERSION}" \
+       --build-arg "TENSORRT_VERSION=${TENSORRT_VERSION}" \
+       --build-arg "GRADLE_VERSION=${GRADLE_VERSION}" \
        --build-arg "NINJA_VERSION=${NINJA_VERSION:-}" \
        --build-arg "KATEX=${KATEX:-}" \
        --build-arg "ROCM_VERSION=${ROCM_VERSION:-}" \

.ci/docker/ci_commit_pins/nccl-cu12.txt

@@ -1 +1 @@
-v2.27.5-1
+v2.27.3-1

.ci/docker/common/common_utils.sh

@@ -23,10 +23,6 @@ conda_install() {
   as_jenkins conda install -q -n py_$ANACONDA_PYTHON_VERSION -y python="$ANACONDA_PYTHON_VERSION" $*
 }
 
-conda_install_through_forge() {
-  as_jenkins conda install -c conda-forge -q -n py_$ANACONDA_PYTHON_VERSION -y python="$ANACONDA_PYTHON_VERSION" $*
-}
-
 conda_run() {
   as_jenkins conda run -n py_$ANACONDA_PYTHON_VERSION --no-capture-output $*
 }

.ci/docker/common/install_base.sh

@@ -15,9 +15,6 @@ install_ubuntu() {
   elif [[ "$UBUNTU_VERSION" == "22.04"* ]]; then
     cmake3="cmake=3.22*"
     maybe_libiomp_dev=""
-  elif [[ "$UBUNTU_VERSION" == "24.04"* ]]; then
-    cmake3="cmake=3.28*"
-    maybe_libiomp_dev=""
   else
     cmake3="cmake=3.5*"
     maybe_libiomp_dev="libiomp-dev"

.ci/docker/common/install_conda.sh

@@ -65,10 +65,10 @@ if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
   fi
 
   # Install PyTorch conda deps, as per https://github.com/pytorch/pytorch README
-  if [[ $(uname -m) != "aarch64" ]]; then
-    pip_install mkl==2024.2.0
-    pip_install mkl-static==2024.2.0
-    pip_install mkl-include==2024.2.0
+  if [[ $(uname -m) == "aarch64" ]]; then
+    conda_install "openblas==0.3.29=*openmp*"
+  else
+    conda_install "mkl=2021.4.0 mkl-include=2021.4.0"
   fi
 
   # Install llvm-8 as it is required to compile llvmlite-0.30.0 from source
@@ -82,10 +82,6 @@ if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
     conda_run ${SCRIPT_FOLDER}/install_magma_conda.sh $(cut -f1-2 -d'.' <<< ${CUDA_VERSION})
   fi
 
-  if [[ "$UBUNTU_VERSION" == "24.04"* ]] ; then
-    conda_install_through_forge libstdcxx-ng=14
-  fi
-
   # Install some other packages, including those needed for Python test reporting
   pip_install -r /opt/conda/requirements-ci.txt
 

.ci/docker/common/install_cpython.sh

@@ -3,10 +3,11 @@
 set -uex -o pipefail
 
 PYTHON_DOWNLOAD_URL=https://www.python.org/ftp/python
+PYTHON_DOWNLOAD_GITHUB_BRANCH=https://github.com/python/cpython/archive/refs/heads  # @lint-ignore
 GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py
 
 # Python versions to be installed in /opt/$VERSION_NO
-CPYTHON_VERSIONS=${CPYTHON_VERSIONS:-"3.9.0 3.10.1 3.11.0 3.12.0 3.13.0 3.13.0t 3.14.0 3.14.0t"}
+CPYTHON_VERSIONS=${CPYTHON_VERSIONS:-"3.9.0 3.10.1 3.11.0 3.12.0 3.13.0 3.13.0t"}
 
 function check_var {
     if [ -z "$1" ]; then
@@ -23,8 +24,9 @@ function do_cpython_build {
     tar -xzf Python-$py_ver.tgz
 
     local additional_flags=""
-    if [[ "$py_ver" == *"t" ]]; then
+    if [ "$py_ver" == "3.13.0t" ]; then
         additional_flags=" --disable-gil"
+        mv cpython-3.13/ cpython-3.13t/
     fi
 
     pushd $py_folder
@@ -66,7 +68,7 @@ function do_cpython_build {
         ln -s pip3 ${prefix}/bin/pip
     fi
     # install setuptools since python 3.12 is required to use distutils
-    ${prefix}/bin/pip install wheel==0.45.1 setuptools==80.9.0
+    ${prefix}/bin/pip install wheel==0.34.2 setuptools==68.2.2
     local abi_tag=$(${prefix}/bin/python -c "from wheel.pep425tags import get_abbr_impl, get_impl_ver, get_abi_tag; print('{0}{1}-{2}'.format(get_abbr_impl(), get_impl_ver(), get_abi_tag()))")
     ln -sf ${prefix} /opt/python/${abi_tag}
 }
@@ -74,20 +76,24 @@ function do_cpython_build {
 function build_cpython {
     local py_ver=$1
     check_var $py_ver
-    local py_suffix=$py_ver
-    local py_folder=$py_ver
+    check_var $PYTHON_DOWNLOAD_URL
+    local py_ver_folder=$py_ver
 
-    # Special handling for nogil
-    if [[ "${py_ver}" == *"t" ]]; then
-        py_suffix=${py_ver::-1}
-        py_folder=$py_suffix
+    if [ "$py_ver" = "3.13.0t" ]; then
+        PY_VER_SHORT="3.13"
+        PYT_VER_SHORT="3.13t"
+        check_var $PYTHON_DOWNLOAD_GITHUB_BRANCH
+        wget $PYTHON_DOWNLOAD_GITHUB_BRANCH/$PY_VER_SHORT.tar.gz -O Python-$py_ver.tgz
+        do_cpython_build $py_ver cpython-$PYT_VER_SHORT
+    elif [ "$py_ver" = "3.13.0" ]; then
+        PY_VER_SHORT="3.13"
+        check_var $PYTHON_DOWNLOAD_GITHUB_BRANCH
+        wget $PYTHON_DOWNLOAD_GITHUB_BRANCH/$PY_VER_SHORT.tar.gz -O Python-$py_ver.tgz
+        do_cpython_build $py_ver cpython-$PY_VER_SHORT
+    else
+        wget -q $PYTHON_DOWNLOAD_URL/$py_ver_folder/Python-$py_ver.tgz
+        do_cpython_build $py_ver Python-$py_ver
     fi
-    # Only b3 is available now
-    if [ "$py_suffix" == "3.14.0" ]; then
-        py_suffix="3.14.0b3"
-    fi
-    wget -q $PYTHON_DOWNLOAD_URL/$py_folder/Python-$py_suffix.tgz -O Python-$py_ver.tgz
-    do_cpython_build $py_ver Python-$py_suffix
 
     rm -f Python-$py_ver.tgz
 }

.ci/docker/common/install_cuda.sh

@@ -10,8 +10,6 @@ else
   arch_path='sbsa'
 fi
 
-NVSHMEM_VERSION=3.3.9
-
 function install_cuda {
   version=$1
   runfile=$2
@@ -42,65 +40,13 @@ function install_cudnn {
   rm -rf tmp_cudnn
 }
 
-function install_nvshmem {
-  cuda_major_version=$1      # e.g. "12"
-  nvshmem_version=$2         # e.g. "3.3.9"
-
-  case "${arch_path}" in
-    sbsa)
-      dl_arch="aarch64"
-      ;;
-    x86_64)
-      dl_arch="x64"
-      ;;
-    *)
-      dl_arch="${arch}"
-      ;;
-  esac
-
-  tmpdir="tmp_nvshmem"
-  mkdir -p "${tmpdir}" && cd "${tmpdir}"
-
-  # nvSHMEM license: https://docs.nvidia.com/nvshmem/api/sla.html
-  filename="libnvshmem_cuda${cuda_major_version}-linux-${arch_path}-${nvshmem_version}"
-  url="https://developer.download.nvidia.com/compute/redist/nvshmem/${nvshmem_version}/builds/cuda${cuda_major_version}/txz/agnostic/${dl_arch}/${filename}.tar.gz"
-
-  # download, unpack, install
-  wget -q "${url}"
-  tar xf "${filename}.tar.gz"
-  cp -a "libnvshmem/include/"* /usr/local/include/
-  cp -a "libnvshmem/lib/"*     /usr/local/lib/
-
-  # cleanup
-  cd ..
-  rm -rf "${tmpdir}"
-
-  echo "nvSHMEM ${nvshmem_version} for CUDA ${cuda_major_version} (${arch_path}) installed."
-}
-
-function install_124 {
-  CUDNN_VERSION=9.1.0.70
-  echo "Installing CUDA 12.4.1 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.6.2"
-  install_cuda 12.4.1 cuda_12.4.1_550.54.15_linux
-
-  install_cudnn 12 $CUDNN_VERSION
-
-  CUDA_VERSION=12.4 bash install_nccl.sh
-
-  CUDA_VERSION=12.4 bash install_cusparselt.sh
-
-  ldconfig
-}
-
 function install_126 {
   CUDNN_VERSION=9.10.2.21
-  echo "Installing CUDA 12.6.3 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
+  echo "Installing CUDA 12.6.3 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.7.1"
   install_cuda 12.6.3 cuda_12.6.3_560.35.05_linux
 
   install_cudnn 12 $CUDNN_VERSION
 
-  install_nvshmem 12 $NVSHMEM_VERSION
-
   CUDA_VERSION=12.6 bash install_nccl.sh
 
   CUDA_VERSION=12.6 bash install_cusparselt.sh
@@ -110,15 +56,13 @@ function install_126 {
 
 function install_129 {
   CUDNN_VERSION=9.10.2.21
-  echo "Installing CUDA 12.9.1 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
+  echo "Installing CUDA 12.9.1 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.7.1"
   # install CUDA 12.9.1 in the same container
   install_cuda 12.9.1 cuda_12.9.1_575.57.08_linux
 
   # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
   install_cudnn 12 $CUDNN_VERSION
 
-  install_nvshmem 12 $NVSHMEM_VERSION
-
   CUDA_VERSION=12.9 bash install_nccl.sh
 
   CUDA_VERSION=12.9 bash install_cusparselt.sh
@@ -126,40 +70,6 @@ function install_129 {
   ldconfig
 }
 
-function prune_124 {
-  echo "Pruning CUDA 12.4"
-  #####################################################################################
-  # CUDA 12.4 prune static libs
-  #####################################################################################
-  export NVPRUNE="/usr/local/cuda-12.4/bin/nvprune"
-  export CUDA_LIB_DIR="/usr/local/cuda-12.4/lib64"
-
-  export GENCODE="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
-  export GENCODE_CUDNN="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
-
-  if [[ -n "$OVERRIDE_GENCODE" ]]; then
-      export GENCODE=$OVERRIDE_GENCODE
-  fi
-  if [[ -n "$OVERRIDE_GENCODE_CUDNN" ]]; then
-      export GENCODE_CUDNN=$OVERRIDE_GENCODE_CUDNN
-  fi
-
-  # all CUDA libs except CuDNN and CuBLAS
-  ls $CUDA_LIB_DIR/ | grep "\.a" | grep -v "culibos" | grep -v "cudart" | grep -v "cudnn" | grep -v "cublas" | grep -v "metis"  \
-      | xargs -I {} bash -c \
-                "echo {} && $NVPRUNE $GENCODE $CUDA_LIB_DIR/{} -o $CUDA_LIB_DIR/{}"
-
-  # prune CuDNN and CuBLAS
-  $NVPRUNE $GENCODE_CUDNN $CUDA_LIB_DIR/libcublas_static.a -o $CUDA_LIB_DIR/libcublas_static.a
-  $NVPRUNE $GENCODE_CUDNN $CUDA_LIB_DIR/libcublasLt_static.a -o $CUDA_LIB_DIR/libcublasLt_static.a
-
-  #####################################################################################
-  # CUDA 12.4 prune visual tools
-  #####################################################################################
-  export CUDA_BASE="/usr/local/cuda-12.4/"
-  rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2024.1.0 $CUDA_BASE/nsight-systems-2023.4.4/
-}
-
 function prune_126 {
   echo "Pruning CUDA 12.6"
   #####################################################################################
@@ -196,15 +106,13 @@ function prune_126 {
 
 function install_128 {
   CUDNN_VERSION=9.8.0.87
-  echo "Installing CUDA 12.8.1 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
+  echo "Installing CUDA 12.8.1 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.7.1"
   # install CUDA 12.8.1 in the same container
   install_cuda 12.8.1 cuda_12.8.1_570.124.06_linux
 
   # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
   install_cudnn 12 $CUDNN_VERSION
 
-  install_nvshmem 12 $NVSHMEM_VERSION
-
   CUDA_VERSION=12.8 bash install_nccl.sh
 
   CUDA_VERSION=12.8 bash install_cusparselt.sh
@@ -216,8 +124,6 @@ function install_128 {
 while test $# -gt 0
 do
     case "$1" in
-    12.4) install_124; prune_124
-        ;;
     12.6|12.6.*) install_126; prune_126
         ;;
     12.8|12.8.*) install_128;

.ci/docker/common/install_cudnn.sh

@@ -0,0 +1,24 @@
+#!/bin/bash
+
+if [[ -n "${CUDNN_VERSION}" ]]; then
+    # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
+    mkdir tmp_cudnn
+    pushd tmp_cudnn
+    if [[ ${CUDA_VERSION:0:4} == "12.9" || ${CUDA_VERSION:0:4} == "12.8" ]]; then
+        CUDNN_NAME="cudnn-linux-x86_64-9.10.2.21_cuda12-archive"
+    elif [[ ${CUDA_VERSION:0:4} == "12.6" ]]; then
+        CUDNN_NAME="cudnn-linux-x86_64-9.10.2.21_cuda12-archive"
+    elif [[ ${CUDA_VERSION:0:2} == "11" ]]; then
+        CUDNN_NAME="cudnn-linux-x86_64-9.1.0.70_cuda11-archive"
+    else
+        print "Unsupported CUDA version ${CUDA_VERSION}"
+        exit 1
+    fi
+    curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/${CUDNN_NAME}.tar.xz
+    tar xf ${CUDNN_NAME}.tar.xz
+    cp -a ${CUDNN_NAME}/include/* /usr/local/cuda/include/
+    cp -a ${CUDNN_NAME}/lib/* /usr/local/cuda/lib64/
+    popd
+    rm -rf tmp_cudnn
+    ldconfig
+fi

.ci/docker/common/install_cusparselt.sh

@@ -13,14 +13,6 @@ if [[ ${CUDA_VERSION:0:4} =~ ^12\.[5-9]$ ]]; then
     fi
     CUSPARSELT_NAME="libcusparse_lt-linux-${arch_path}-0.7.1.0-archive"
     curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-${arch_path}/${CUSPARSELT_NAME}.tar.xz
-elif [[ ${CUDA_VERSION:0:4} == "12.4" ]]; then
-    arch_path='sbsa'
-    export TARGETARCH=${TARGETARCH:-$(uname -m)}
-    if [ ${TARGETARCH} = 'amd64' ] || [ "${TARGETARCH}" = 'x86_64' ]; then
-        arch_path='x86_64'
-    fi
-    CUSPARSELT_NAME="libcusparse_lt-linux-${arch_path}-0.6.2.3-archive"
-    curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-${arch_path}/${CUSPARSELT_NAME}.tar.xz
 else
     echo "Not sure which libcusparselt version to install for this ${CUDA_VERSION}"
 fi

.ci/docker/common/install_rocm.sh

@@ -8,11 +8,9 @@ ver() {
 
 install_ubuntu() {
     apt-get update
-    # gpg-agent is not available by default
-    apt-get install -y --no-install-recommends gpg-agent
-    if [[ $(ver $UBUNTU_VERSION) -ge $(ver 22.04) ]]; then
-        echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' \
-            | sudo tee /etc/apt/preferences.d/rocm-pin-600
+    if [[ $UBUNTU_VERSION == 20.04 ]]; then
+      # gpg-agent is not available by default on 20.04
+      apt-get install -y --no-install-recommends gpg-agent
     fi
     apt-get install -y kmod
     apt-get install -y wget
@@ -30,25 +28,16 @@ EOF
 
     # we want the patch version of 6.4 instead
     if [[ $(ver $ROCM_VERSION) -eq $(ver 6.4) ]]; then
-        ROCM_VERSION="${ROCM_VERSION}.2"
-    fi
-
-    # Default url values
-    rocm_baseurl="http://repo.radeon.com/rocm/apt/${ROCM_VERSION}"
-    amdgpu_baseurl="https://repo.radeon.com/amdgpu/${ROCM_VERSION}/ubuntu"
-
-    # Special case for ROCM_VERSION == 7.0
-    if [[ $(ver "$ROCM_VERSION") -eq $(ver 7.0) ]]; then
-        rocm_baseurl="https://repo.radeon.com/rocm/apt/7.0_alpha2"
-        amdgpu_baseurl="https://repo.radeon.com/amdgpu/30.10_alpha2/ubuntu"
+        ROCM_VERSION="${ROCM_VERSION}.1"
     fi
 
     # Add amdgpu repository
     UBUNTU_VERSION_NAME=`cat /etc/os-release | grep UBUNTU_CODENAME | awk -F= '{print $2}'`
-    echo "deb [arch=amd64] ${amdgpu_baseurl} ${UBUNTU_VERSION_NAME} main" > /etc/apt/sources.list.d/amdgpu.list
+    echo "deb [arch=amd64] https://repo.radeon.com/amdgpu/${ROCM_VERSION}/ubuntu ${UBUNTU_VERSION_NAME} main" > /etc/apt/sources.list.d/amdgpu.list
 
     # Add rocm repository
     wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add -
+    local rocm_baseurl="http://repo.radeon.com/rocm/apt/${ROCM_VERSION}"
     echo "deb [arch=amd64] ${rocm_baseurl} ${UBUNTU_VERSION_NAME} main" > /etc/apt/sources.list.d/rocm.list
     apt-get update --allow-insecure-repositories
 
@@ -82,33 +71,29 @@ EOF
     done
 
     # ROCm 6.3 had a regression where initializing static code objects had significant overhead
-    # CI no longer builds for ROCm 6.3, but
     # ROCm 6.4 did not yet fix the regression, also HIP branch names are different
-    if [[ $(ver $ROCM_VERSION) -ge $(ver 6.4) ]] && [[ $(ver $ROCM_VERSION) -lt $(ver 7.0) ]]; then
-        if [[ $(ver $ROCM_VERSION) -eq $(ver 6.4.2) ]]; then
-            HIP_TAG=rocm-6.4.2
-            CLR_HASH=74d78ba3ac4bac235d02bcb48511c30b5cfdd457  # branch release/rocm-rel-6.4.2-statco-hotfix
-        elif [[ $(ver $ROCM_VERSION) -eq $(ver 6.4.1) ]]; then
-            HIP_TAG=rocm-6.4.1
-            CLR_HASH=efe6c35790b9206923bfeed1209902feff37f386  # branch release/rocm-rel-6.4.1-statco-hotfix
+    if [[ $(ver $ROCM_VERSION) -ge $(ver 6.3) ]] && [[ $(ver $ROCM_VERSION) -lt $(ver 7.0) ]]; then
+        if [[ $(ver $ROCM_VERSION) -eq $(ver 6.4.1) ]]; then
+            HIP_BRANCH=release/rocm-rel-6.4
+            VER_STR=6.4
+            VER_PATCH=.1
         elif [[ $(ver $ROCM_VERSION) -eq $(ver 6.4) ]]; then
-            HIP_TAG=rocm-6.4.0
-            CLR_HASH=600f5b0d2baed94d5121e2174a9de0851b040b0c  # branch release/rocm-rel-6.4-statco-hotfix
+            HIP_BRANCH=release/rocm-rel-6.4
+            VER_STR=6.4
+        elif [[ $(ver $ROCM_VERSION) -eq $(ver 6.3) ]]; then
+            HIP_BRANCH=rocm-6.3.x
+            VER_STR=6.3
         fi
         # clr build needs CppHeaderParser but can only find it using conda's python
-        python -m pip install CppHeaderParser
-        git clone https://github.com/ROCm/HIP -b $HIP_TAG
+        /opt/conda/bin/python -m pip install CppHeaderParser
+        git clone https://github.com/ROCm/HIP -b $HIP_BRANCH
         HIP_COMMON_DIR=$(readlink -f HIP)
-        git clone https://github.com/jeffdaily/clr
-        pushd clr
-        git checkout $CLR_HASH
-        popd
+        git clone https://github.com/jeffdaily/clr -b release/rocm-rel-${VER_STR}${VER_PATCH}-statco-hotfix
         mkdir -p clr/build
         pushd clr/build
-        # Need to point CMake to the correct python installation to find CppHeaderParser
-        cmake .. -DPython3_EXECUTABLE=/opt/conda/envs/py_${ANACONDA_PYTHON_VERSION}/bin/python3 -DCLR_BUILD_HIP=ON -DHIP_COMMON_DIR=$HIP_COMMON_DIR
+        cmake .. -DCLR_BUILD_HIP=ON -DHIP_COMMON_DIR=$HIP_COMMON_DIR
         make -j
-        cp hipamd/lib/libamdhip64.so.6.4.* /opt/rocm/lib/libamdhip64.so.6.4.*
+        cp hipamd/lib/libamdhip64.so.${VER_STR}.* /opt/rocm/lib/libamdhip64.so.${VER_STR}.*
         popd
         rm -rf HIP clr
     fi

.ci/docker/common/install_rocm_magma.sh

@@ -5,12 +5,7 @@ set -eou pipefail
 
 function do_install() {
     rocm_version=$1
-    if [[ ${rocm_version} =~ ^[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
-        # chop off any patch version
-        rocm_version="${rocm_version%.*}"
-    fi
-
-    rocm_version_nodot=${rocm_version//./}
+    rocm_version_nodot=${1//./}
 
     # Version 2.7.2 + ROCm related updates
     MAGMA_VERSION=a1625ff4d9bc362906bd01f805dbbe12612953f6

.ci/docker/common/install_triton.sh

@@ -98,10 +98,6 @@ fi
 if [ -n "${NUMPY_VERSION}" ]; then
   pip_install "numpy==${NUMPY_VERSION}"
 fi
-
-# IMPORTANT: helion needs to be installed without dependencies.
-# It depends on torch and triton. We don't want to install
-# triton and torch from production on Docker CI images
 if [[ "$ANACONDA_PYTHON_VERSION" != 3.9* ]]; then
-  pip_install helion==0.0.10 --no-deps
+  pip_install helion
 fi

.ci/docker/common/install_xpu.sh

@@ -56,10 +56,14 @@ function install_ubuntu() {
 
 function install_rhel() {
     . /etc/os-release
-
-    if [[ ! " 8.8 8.10 9.0 9.2 9.3 " =~ " ${VERSION_ID} " ]]; then
-        echo "RHEL version ${VERSION_ID} not supported"
-        exit
+    if [[ "${ID}" == "rhel" ]]; then
+        if [[ ! " 8.8 8.9 9.0 9.2 9.3 " =~ " ${VERSION_ID} " ]]; then
+            echo "RHEL version ${VERSION_ID} not supported"
+            exit
+        fi
+    elif [[ "${ID}" == "almalinux" ]]; then
+        # Workaround for almalinux8 which used by quay.io/pypa/manylinux_2_28_x86_64
+        VERSION_ID="8.8"
     fi
 
     dnf install -y 'dnf-command(config-manager)'

.ci/docker/libtorch/build.sh

@@ -39,10 +39,6 @@ case ${DOCKER_TAG_PREFIX} in
         DOCKER_GPU_BUILD_ARG=""
         ;;
     rocm*)
-        # we want the patch version of 6.4 instead
-        if [[ $(ver $GPU_ARCH_VERSION) -eq $(ver 6.4) ]]; then
-            GPU_ARCH_VERSION="${GPU_ARCH_VERSION}.2"
-        fi
         BASE_TARGET=rocm
         GPU_IMAGE=rocm/dev-ubuntu-22.04:${GPU_ARCH_VERSION}-complete
         PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"

.ci/docker/linter/Dockerfile

@@ -27,7 +27,5 @@ COPY ./common/install_linter.sh install_linter.sh
 RUN bash ./install_linter.sh
 RUN rm install_linter.sh
 
-RUN chown -R jenkins:jenkins /var/lib/jenkins/ci_env
-
 USER jenkins
 CMD ["bash"]

.ci/docker/manywheel/Dockerfile_s390x

@@ -131,8 +131,6 @@ RUN pip3 install flatbuffers && \
   git clone https://github.com/microsoft/onnxruntime && \
   cd onnxruntime && git checkout v1.21.0 && \
   git submodule update --init --recursive && \
-  wget https://github.com/microsoft/onnxruntime/commit/f57db79743c4d1a3553aa05cf95bcd10966030e6.patch && \
-  patch -p1 < f57db79743c4d1a3553aa05cf95bcd10966030e6.patch && \
   ./build.sh --config Release --parallel 0 --enable_pybind \
   --build_wheel --enable_training --enable_training_apis \
   --enable_training_ops --skip_tests --allow_running_as_root \

.ci/docker/manywheel/build.sh

@@ -75,10 +75,6 @@ case ${image} in
         DOCKERFILE_SUFFIX="_cuda_aarch64"
         ;;
     manylinux2_28-builder:rocm*)
-        # we want the patch version of 6.4 instead
-        if [[ $(ver $GPU_ARCH_VERSION) -eq $(ver 6.4) ]]; then
-            GPU_ARCH_VERSION="${GPU_ARCH_VERSION}.2"
-        fi
         TARGET=rocm_final
         MANY_LINUX_VERSION="2_28"
         DEVTOOLSET_VERSION="11"

.ci/docker/requirements-ci.txt

@@ -16,7 +16,6 @@ click
 #test that import:
 
 coremltools==5.0b5 ; python_version < "3.12"
-coremltools==8.3 ; python_version == "3.12"
 #Description: Apple framework for ML integration
 #Pinned versions: 5.0b5
 #test that import:
@@ -50,7 +49,7 @@ flatbuffers==24.12.23
 hypothesis==5.35.1
 # Pin hypothesis to avoid flakiness: https://github.com/pytorch/pytorch/issues/31136
 #Description: advanced library for generating parametrized tests
-#Pinned versions: 5.35.1
+#Pinned versions: 3.44.6, 4.53.2
 #test that import: test_xnnpack_integration.py, test_pruning_op.py, test_nn.py
 
 junitparser==2.1.1
@@ -64,7 +63,6 @@ lark==0.12.0
 #test that import:
 
 librosa>=0.6.2 ; python_version < "3.11"
-librosa==0.10.2 ; python_version == "3.12"
 #Description: A python package for music and audio analysis
 #Pinned versions: >=0.6.2
 #test that import: test_spectral_ops.py
@@ -113,7 +111,6 @@ ninja==1.11.1.3
 numba==0.49.0 ; python_version < "3.9"
 numba==0.55.2 ; python_version == "3.9"
 numba==0.55.2 ; python_version == "3.10"
-numba==0.60.0 ; python_version == "3.12"
 #Description: Just-In-Time Compiler for Numerical Functions
 #Pinned versions: 0.54.1, 0.49.0, <=0.49.1
 #test that import: test_numba_integration.py
@@ -221,9 +218,9 @@ pygments==2.15.0
 #Pinned versions: 2.12.0
 #test that import: the doctests
 
-#pyyaml
+#PyYAML
 #Description: data serialization format
-#Pinned versions: 6.0.2
+#Pinned versions:
 #test that import:
 
 #requests
@@ -233,7 +230,7 @@ pygments==2.15.0
 
 #rich
 #Description: rich text and beautiful formatting in the terminal
-#Pinned versions: 14.1.0
+#Pinned versions: 10.9.0
 #test that import:
 
 scikit-image==0.19.3 ; python_version < "3.10"
@@ -307,7 +304,7 @@ pytest-cpp==2.3.0
 #Pinned versions: 2.3.0
 #test that import:
 
-z3-solver==4.15.1.0
+z3-solver==4.12.6.0
 #Description: The Z3 Theorem Prover Project
 #Pinned versions:
 #test that import:
@@ -363,10 +360,10 @@ pwlf==2.2.1
 
 
 # To build PyTorch itself
-pyyaml
+astunparse
+PyYAML
 pyzstd
-setuptools>=70.1.0
-six
+setuptools
 
 scons==4.5.2 ; platform_machine == "aarch64"
 
@@ -386,12 +383,6 @@ cmake==4.0.0
 tlparse==0.3.30
 #Description: required for log parsing
 
-cuda-bindings>=12.0,<13.0 ; platform_machine != "s390x"
+cuda-bindings>=12.0,<13.0
 #Description: required for testing CUDAGraph::raw_cuda_graph(). See https://nvidia.github.io/cuda-python/cuda-bindings/latest/support.html for how this version was chosen. Note "Any fix in the latest bindings would be backported to the prior major version" means that only the newest version of cuda-bindings will get fixes. Depending on the latest version of 12.x is okay because all 12.y versions will be supported via "CUDA minor version compatibility". Pytorch builds against 13.z versions of cuda toolkit work with 12.x versions of cuda-bindings as well because newer drivers work with old toolkits.
 #test that import: test_cuda.py
-
-setuptools-git-versioning==2.1.0
-scikit-build==0.18.1
-pyre-extensions==0.0.32
-tabulate==0.9.0
-#Description: These package are needed to build FBGEMM and torchrec on PyTorch CI

.ci/docker/requirements-docs.txt

@@ -1,7 +1,7 @@
 sphinx==5.3.0
 #Description: This is used to generate PyTorch docs
 #Pinned versions: 5.3.0
--e git+https://github.com/pytorch/pytorch_sphinx_theme.git@pytorch_sphinx_theme2#egg=pytorch_sphinx_theme2
+-e git+https://github.com/pytorch/pytorch_sphinx_theme.git@722b7e6f9ca512fcc526ad07d62b3d28c50bb6cd#egg=pytorch_sphinx_theme2
 
 # TODO: sphinxcontrib.katex 0.9.0 adds a local KaTeX server to speed up pre-rendering
 # but it doesn't seem to work and hangs around idly. The initial thought that it is probably
@@ -19,10 +19,9 @@ sphinx_sitemap==2.6.0
 #Description: This is used to generate sitemap for PyTorch docs
 #Pinned versions: 2.6.0
 
-matplotlib==3.5.3 ; python_version < "3.13"
-matplotlib==3.6.3 ; python_version >= "3.13"
+matplotlib==3.5.3
 #Description: This is used to generate PyTorch docs
-#Pinned versions: 3.6.3 if python > 3.12. Otherwise 3.5.3.
+#Pinned versions: 3.5.3
 
 tensorboard==2.13.0 ; python_version < "3.13"
 tensorboard==2.18.0 ; python_version >= "3.13"
@@ -50,7 +49,7 @@ IPython==8.12.0
 #Pinned versions: 8.12.0
 
 myst-nb==0.17.2
-#Description: This is used to generate PyTorch functorch and torch.compile docs
+#Description: This is used to generate PyTorch functorch and torch.compile docs.
 #Pinned versions: 0.17.2
 
 # The following are required to build torch.distributed.elastic.rendezvous.etcd* docs

.ci/manywheel/build_common.sh

@@ -97,7 +97,7 @@ if [[ -z "$PYTORCH_ROOT" ]]; then
     exit 1
 fi
 pushd "$PYTORCH_ROOT"
-retry pip install -qUr requirements-build.txt
+retry pip install -q cmake
 python setup.py clean
 retry pip install -qr requirements.txt
 case ${DESIRED_PYTHON} in

.ci/manywheel/build_cuda.sh

@@ -134,8 +134,6 @@ if [[ $CUDA_VERSION == 12* ]]; then
             "/usr/local/cuda/lib64/libnvrtc-builtins.so"
             "/usr/local/cuda/lib64/libcufile.so.0"
             "/usr/local/cuda/lib64/libcufile_rdma.so.1"
-            "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12"
-            "/usr/local/cuda/extras/CUPTI/lib64/libnvperf_host.so"
         )
         DEPS_SONAME+=(
             "libcudnn_adv.so.9"
@@ -154,14 +152,7 @@ if [[ $CUDA_VERSION == 12* ]]; then
             "libnvrtc-builtins.so"
             "libcufile.so.0"
             "libcufile_rdma.so.1"
-            "libcupti.so.12"
-            "libnvperf_host.so"
         )
-        # Add libnvToolsExt only if CUDA version is not 12.9
-        if [[ $CUDA_VERSION != 12.9* ]]; then
-            DEPS_LIST+=("/usr/local/cuda/lib64/libnvToolsExt.so.1")
-            DEPS_SONAME+=("libnvToolsExt.so.1")
-        fi
     else
         echo "Using nvidia libs from pypi."
         CUDA_RPATHS=(
@@ -177,7 +168,6 @@ if [[ $CUDA_VERSION == 12* ]]; then
             '$ORIGIN/../../nvidia/cusparselt/lib'
             '$ORIGIN/../../cusparselt/lib'
             '$ORIGIN/../../nvidia/nccl/lib'
-            '$ORIGIN/../../nvidia/nvshmem/lib'
             '$ORIGIN/../../nvidia/nvtx/lib'
             '$ORIGIN/../../nvidia/cufile/lib'
         )

.ci/manywheel/build_libtorch.sh

@@ -92,7 +92,7 @@ if [[ -z "$PYTORCH_ROOT" ]]; then
     exit 1
 fi
 pushd "$PYTORCH_ROOT"
-retry pip install -qUr requirements-build.txt
+retry pip install -q cmake
 python setup.py clean
 retry pip install -qr requirements.txt
 retry pip install -q numpy==2.0.1
@@ -104,7 +104,7 @@ if [[ "$DESIRED_CUDA" == *"rocm"* ]]; then
     export ROCclr_DIR=/opt/rocm/rocclr/lib/cmake/rocclr
 fi
 
-echo "Calling 'python -m pip install .' at $(date)"
+echo "Calling setup.py install at $(date)"
 
 if [[ $LIBTORCH_VARIANT = *"static"* ]]; then
     STATIC_CMAKE_FLAG="-DTORCH_STATIC=1"
@@ -120,7 +120,7 @@ fi
         # TODO: Remove this flag once https://github.com/pytorch/pytorch/issues/55952 is closed
         CFLAGS='-Wno-deprecated-declarations' \
         BUILD_LIBTORCH_CPU_WITH_DEBUG=1 \
-        python -m pip install --no-build-isolation -v .
+        python setup.py install
 
     mkdir -p libtorch/{lib,bin,include,share}
 

.ci/onnx/test.sh

@@ -19,7 +19,7 @@ git config --global --add safe.directory /var/lib/jenkins/workspace
 
 if [[ "$BUILD_ENVIRONMENT" == *onnx* ]]; then
   # TODO: This can be removed later once vision is also part of the Docker image
-  pip install -q --no-use-pep517 "git+https://github.com/pytorch/vision.git@$(cat .github/ci_commit_pins/vision.txt)"
+  pip install -q --user --no-use-pep517 "git+https://github.com/pytorch/vision.git@$(cat .github/ci_commit_pins/vision.txt)"
   # JIT C++ extensions require ninja, so put it into PATH.
   export PATH="/var/lib/jenkins/.local/bin:$PATH"
   # NB: ONNX test is fast (~15m) so it's ok to retry it few more times to avoid any flaky issue, we

.ci/pytorch/build-mobile.sh

@@ -0,0 +1,34 @@
+#!/usr/bin/env bash
+# DO NOT ADD 'set -x' not to reveal CircleCI secret context environment variables
+set -eu -o pipefail
+
+# This script uses linux host toolchain + mobile build options in order to
+# build & test mobile libtorch without having to setup Android/iOS
+# toolchain/simulator.
+
+# shellcheck source=./common.sh
+source "$(dirname "${BASH_SOURCE[0]}")/common.sh"
+# shellcheck source=./common-build.sh
+source "$(dirname "${BASH_SOURCE[0]}")/common-build.sh"
+
+# Install torch & torchvision - used to download & trace test model.
+# Ideally we should use the libtorch built on the PR so that backward
+# incompatible changes won't break this script - but it will significantly slow
+# down mobile CI jobs.
+# Here we install nightly instead of stable so that we have an option to
+# temporarily skip mobile CI jobs on BC-breaking PRs until they are in nightly.
+retry pip install --pre torch torchvision \
+  -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html \
+  --progress-bar off
+
+# Run end-to-end process of building mobile library, linking into the predictor
+# binary, and running forward pass with a real model.
+if [[ "$BUILD_ENVIRONMENT" == *-mobile-custom-build-static* ]]; then
+  TEST_CUSTOM_BUILD_STATIC=1 test/mobile/custom_build/build.sh
+elif [[ "$BUILD_ENVIRONMENT" == *-mobile-lightweight-dispatch* ]]; then
+  test/mobile/lightweight_dispatch/build.sh
+else
+  TEST_DEFAULT_BUILD=1 test/mobile/custom_build/build.sh
+fi
+
+print_sccache_stats

.ci/pytorch/build.sh

@@ -11,6 +11,10 @@ source "$(dirname "${BASH_SOURCE[0]}")/common.sh"
 # shellcheck source=./common-build.sh
 source "$(dirname "${BASH_SOURCE[0]}")/common-build.sh"
 
+if [[ "$BUILD_ENVIRONMENT" == *-mobile-*build* ]]; then
+  exec "$(dirname "${BASH_SOURCE[0]}")/build-mobile.sh" "$@"
+fi
+
 echo "Python version:"
 python --version
 
@@ -120,8 +124,26 @@ if [[ "$BUILD_ENVIRONMENT" == *libtorch* ]]; then
 fi
 
 # Use special scripts for Android builds
+if [[ "${BUILD_ENVIRONMENT}" == *-android* ]]; then
+  export ANDROID_NDK=/opt/ndk
+  build_args=()
+  if [[ "${BUILD_ENVIRONMENT}" == *-arm-v7a* ]]; then
+    build_args+=("-DANDROID_ABI=armeabi-v7a")
+  elif [[ "${BUILD_ENVIRONMENT}" == *-arm-v8a* ]]; then
+    build_args+=("-DANDROID_ABI=arm64-v8a")
+  elif [[ "${BUILD_ENVIRONMENT}" == *-x86_32* ]]; then
+    build_args+=("-DANDROID_ABI=x86")
+  elif [[ "${BUILD_ENVIRONMENT}" == *-x86_64* ]]; then
+    build_args+=("-DANDROID_ABI=x86_64")
+  fi
+  if [[ "${BUILD_ENVIRONMENT}" == *vulkan* ]]; then
+    build_args+=("-DUSE_VULKAN=ON")
+  fi
+  build_args+=("-DUSE_LITE_INTERPRETER_PROFILER=OFF")
+  exec ./scripts/build_android.sh "${build_args[@]}" "$@"
+fi
 
-if [[ "$BUILD_ENVIRONMENT" == *vulkan* ]]; then
+if [[ "$BUILD_ENVIRONMENT" != *android* && "$BUILD_ENVIRONMENT" == *vulkan* ]]; then
   export USE_VULKAN=1
   # shellcheck disable=SC1091
   source /var/lib/jenkins/vulkansdk/setup-env.sh
@@ -176,8 +198,10 @@ fi
 
 # We only build FlashAttention files for CUDA 8.0+, and they require large amounts of
 # memory to build and will OOM
-if [[ "$BUILD_ENVIRONMENT" == *cuda* ]] && [[ 1 -eq $(echo "${TORCH_CUDA_ARCH_LIST} >= 8.0" | bc) ]]; then
-  export BUILD_CUSTOM_STEP="ninja -C build flash_attention -j 2"
+if [[ "$BUILD_ENVIRONMENT" == *cuda* ]] && [[ 1 -eq $(echo "${TORCH_CUDA_ARCH_LIST} >= 8.0" | bc) ]] && [ -z "$MAX_JOBS_OVERRIDE" ]; then
+  echo "WARNING: FlashAttention files require large amounts of memory to build and will OOM"
+  echo "Setting MAX_JOBS=(nproc-2)/3 to reduce memory usage"
+  export MAX_JOBS="$(( $(nproc --ignore=2) / 3 ))"
 fi
 
 if [[ "${BUILD_ENVIRONMENT}" == *clang* ]]; then
@@ -203,7 +227,7 @@ if [[ "${BUILD_ENVIRONMENT}" == *-pch* ]]; then
     export USE_PRECOMPILED_HEADERS=1
 fi
 
-if [[ "${BUILD_ENVIRONMENT}" != *cuda* ]]; then
+if [[ "${BUILD_ENVIRONMENT}" != *android* && "${BUILD_ENVIRONMENT}" != *cuda* ]]; then
   export BUILD_STATIC_RUNTIME_BENCHMARK=ON
 fi
 
@@ -284,22 +308,6 @@ else
     fi
     pip_install_whl "$(echo dist/*.whl)"
 
-    if [[ "${BUILD_ADDITIONAL_PACKAGES:-}" == *vision* ]]; then
-      install_torchvision
-    fi
-
-    if [[ "${BUILD_ADDITIONAL_PACKAGES:-}" == *audio* ]]; then
-      install_torchaudio
-    fi
-
-    if [[ "${BUILD_ADDITIONAL_PACKAGES:-}" == *torchrec* || "${BUILD_ADDITIONAL_PACKAGES:-}" == *fbgemm* ]]; then
-      install_torchrec_and_fbgemm
-    fi
-
-    if [[ "${BUILD_ADDITIONAL_PACKAGES:-}" == *torchao* ]]; then
-      install_torchao
-    fi
-
     if [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
       echo "Checking that xpu is compiled"
       pushd dist/
@@ -387,8 +395,10 @@ else
     # This is an attempt to mitigate flaky libtorch build OOM error. By default, the build parallelization
     # is set to be the number of CPU minus 2. So, let's try a more conservative value here. A 4xlarge has
     # 16 CPUs
-    MAX_JOBS=$(nproc --ignore=4)
-    export MAX_JOBS
+    if [ -z "$MAX_JOBS_OVERRIDE" ]; then
+      MAX_JOBS=$(nproc --ignore=4)
+      export MAX_JOBS
+    fi
 
     # NB: Install outside of source directory (at the same level as the root
     # pytorch folder) so that it doesn't get cleaned away prior to docker push.

.ci/pytorch/common-build.sh

@@ -13,13 +13,6 @@ if [[ "$BUILD_ENVIRONMENT" != *win-* ]]; then
     fi
 
     if which sccache > /dev/null; then
-        # Clear SCCACHE_BUCKET and SCCACHE_REGION if they are empty, otherwise
-        # sccache will complain about invalid bucket configuration
-        if [[ -z "${SCCACHE_BUCKET:-}" ]]; then
-          unset SCCACHE_BUCKET
-          unset SCCACHE_REGION
-        fi
-
         # Save sccache logs to file
         sccache --stop-server > /dev/null  2>&1 || true
         rm -f ~/sccache_error.log || true

.ci/pytorch/common_utils.sh

@@ -78,34 +78,6 @@ function pip_install_whl() {
   fi
 }
 
-function pip_build_and_install() {
-  local build_target=$1
-  local wheel_dir=$2
-
-  local found_whl=0
-  for file in "${wheel_dir}"/*.whl
-  do
-    if [[ -f "${file}" ]]; then
-      found_whl=1
-      break
-    fi
-  done
-
-  # Build the wheel if it doesn't exist
-  if [ "${found_whl}" == "0" ]; then
-    python3 -m pip wheel \
-      --no-build-isolation \
-      --no-deps \
-      --no-use-pep517 \
-      -w "${wheel_dir}" \
-      "${build_target}"
-  fi
-
-  for file in "${wheel_dir}"/*.whl
-  do
-    pip_install_whl "${file}"
-  done
-}
 
 function pip_install() {
   # retry 3 times
@@ -152,7 +124,14 @@ function get_pinned_commit() {
 function install_torchaudio() {
   local commit
   commit=$(get_pinned_commit audio)
-  pip_build_and_install "git+https://github.com/pytorch/audio.git@${commit}" dist/audio
+  if [[ "$1" == "cuda" ]]; then
+    # TODO: This is better to be passed as a parameter from _linux-test workflow
+    # so that it can be consistent with what is set in build
+    TORCH_CUDA_ARCH_LIST="8.0;8.6" pip_install --no-use-pep517 --user "git+https://github.com/pytorch/audio.git@${commit}"
+  else
+    pip_install --no-use-pep517 --user "git+https://github.com/pytorch/audio.git@${commit}"
+  fi
+
 }
 
 function install_torchtext() {
@@ -160,8 +139,8 @@ function install_torchtext() {
   local text_commit
   data_commit=$(get_pinned_commit data)
   text_commit=$(get_pinned_commit text)
-  pip_build_and_install "git+https://github.com/pytorch/data.git@${data_commit}" dist/data
-  pip_build_and_install "git+https://github.com/pytorch/text.git@${text_commit}" dist/text
+  pip_install --no-use-pep517 --user "git+https://github.com/pytorch/data.git@${data_commit}"
+  pip_install --no-use-pep517 --user "git+https://github.com/pytorch/text.git@${text_commit}"
 }
 
 function install_torchvision() {
@@ -174,14 +153,7 @@ function install_torchvision() {
     echo 'char* dlerror(void) { return "";}'|gcc -fpic -shared -o "${HOME}/dlerror.so" -x c -
     LD_PRELOAD=${orig_preload}:${HOME}/dlerror.so
   fi
-
-  if [[ "${BUILD_ENVIRONMENT}" == *cuda* ]]; then
-    # Not sure if both are needed, but why not
-    export FORCE_CUDA=1
-    export WITH_CUDA=1
-  fi
-  pip_build_and_install "git+https://github.com/pytorch/vision.git@${commit}" dist/vision
-
+  pip_install --no-use-pep517 --user "git+https://github.com/pytorch/vision.git@${commit}"
   if [ -n "${LD_PRELOAD}" ]; then
     LD_PRELOAD=${orig_preload}
   fi
@@ -201,79 +173,31 @@ function install_torchrec_and_fbgemm() {
 
   if [[ "$BUILD_ENVIRONMENT" == *rocm* ]] ; then
     # install torchrec first because it installs fbgemm nightly on top of rocm fbgemm
-    pip_build_and_install "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}" dist/torchrec
+    pip_install --no-use-pep517 --user "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}"
     pip_uninstall fbgemm-gpu-nightly
 
-    # Set ROCM_HOME isn't available, use ROCM_PATH if set or /opt/rocm
-    ROCM_HOME="${ROCM_HOME:-${ROCM_PATH:-/opt/rocm}}"
-
-    # Find rocm_version.h header file for ROCm version extract
-    rocm_version_h="${ROCM_HOME}/include/rocm-core/rocm_version.h"
-    if [ ! -f "$rocm_version_h" ]; then
-        rocm_version_h="${ROCM_HOME}/include/rocm_version.h"
-    fi
-
-    # Error out if rocm_version.h not found
-    if [ ! -f "$rocm_version_h" ]; then
-        echo "Error: rocm_version.h not found in expected locations." >&2
-        exit 1
-    fi
-
-    # Extract major, minor and patch ROCm version numbers
-    MAJOR_VERSION=$(grep 'ROCM_VERSION_MAJOR' "$rocm_version_h" | awk '{print $3}')
-    MINOR_VERSION=$(grep 'ROCM_VERSION_MINOR' "$rocm_version_h" | awk '{print $3}')
-    PATCH_VERSION=$(grep 'ROCM_VERSION_PATCH' "$rocm_version_h" | awk '{print $3}')
-    ROCM_INT=$((MAJOR_VERSION * 10000 + MINOR_VERSION * 100 + PATCH_VERSION))
-    echo "ROCm version: $ROCM_INT"
-    export BUILD_ROCM_VERSION="$MAJOR_VERSION.$MINOR_VERSION"
-
     pip_install tabulate  # needed for newer fbgemm
     pip_install patchelf  # needed for rocm fbgemm
-    pushd /tmp
-
-    local wheel_dir=dist/fbgemm_gpu
-    local found_whl=0
-    for file in "${wheel_dir}"/*.whl
-    do
-      if [[ -f "${file}" ]]; then
-        found_whl=1
-        break
-      fi
-    done
-
-    # Build the wheel if it doesn't exist
-    if [ "${found_whl}" == "0" ]; then
-      git clone --recursive https://github.com/pytorch/fbgemm
-      pushd fbgemm/fbgemm_gpu
-      git checkout "${fbgemm_commit}"
-      python setup.py bdist_wheel \
-        --build-variant=rocm \
-        -DHIP_ROOT_DIR="${ROCM_PATH}" \
-        -DCMAKE_C_FLAGS="-DTORCH_USE_HIP_DSA" \
-        -DCMAKE_CXX_FLAGS="-DTORCH_USE_HIP_DSA"
-      popd
-
-      # Save the wheel before cleaning up
-      mkdir -p dist/fbgemm_gpu
-      cp fbgemm/fbgemm_gpu/dist/*.whl dist/fbgemm_gpu
-    fi
-
-    for file in "${wheel_dir}"/*.whl
-    do
-      pip_install_whl "${file}"
-    done
-
-    rm -rf fbgemm
+    git clone --recursive https://github.com/pytorch/fbgemm
+    pushd fbgemm/fbgemm_gpu
+    git checkout "${fbgemm_commit}"
+    python setup.py install \
+      --package_variant=rocm \
+      -DHIP_ROOT_DIR="${ROCM_PATH}" \
+      -DCMAKE_C_FLAGS="-DTORCH_USE_HIP_DSA" \
+      -DCMAKE_CXX_FLAGS="-DTORCH_USE_HIP_DSA"
     popd
+    rm -rf fbgemm
   else
-    pip_build_and_install "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}" dist/torchrec
-    pip_build_and_install "git+https://github.com/pytorch/FBGEMM.git@${fbgemm_commit}#subdirectory=fbgemm_gpu" dist/fbgemm_gpu
+    # See https://github.com/pytorch/pytorch/issues/106971
+    CUDA_PATH=/usr/local/cuda-12.1 pip_install --no-use-pep517 --user "git+https://github.com/pytorch/FBGEMM.git@${fbgemm_commit}#egg=fbgemm-gpu&subdirectory=fbgemm_gpu"
+    pip_install --no-use-pep517 --user "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}"
   fi
 }
 
 function clone_pytorch_xla() {
   if [[ ! -d ./xla ]]; then
-    git clone --recursive --quiet https://github.com/pytorch/xla.git
+    git clone --recursive -b r2.8 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 "$(cat ../.github/ci_commit_pins/xla.txt)"
@@ -310,7 +234,7 @@ function checkout_install_torchbench() {
 function install_torchao() {
   local commit
   commit=$(get_pinned_commit torchao)
-  pip_build_and_install "git+https://github.com/pytorch/ao.git@${commit}" dist/ao
+  pip_install --no-use-pep517 --user "git+https://github.com/pytorch/ao.git@${commit}"
 }
 
 function print_sccache_stats() {

.ci/pytorch/create_test_cert.py

@@ -0,0 +1,123 @@
+from datetime import datetime, timedelta, timezone
+from tempfile import mkdtemp
+
+from cryptography import x509
+from cryptography.hazmat.primitives import hashes, serialization
+from cryptography.hazmat.primitives.asymmetric import rsa
+from cryptography.x509.oid import NameOID
+
+
+temp_dir = mkdtemp()
+print(temp_dir)
+
+
+def genrsa(path):
+    key = rsa.generate_private_key(
+        public_exponent=65537,
+        key_size=2048,
+    )
+    with open(path, "wb") as f:
+        f.write(
+            key.private_bytes(
+                encoding=serialization.Encoding.PEM,
+                format=serialization.PrivateFormat.TraditionalOpenSSL,
+                encryption_algorithm=serialization.NoEncryption(),
+            )
+        )
+    return key
+
+
+def create_cert(path, C, ST, L, O, key):
+    subject = issuer = x509.Name(
+        [
+            x509.NameAttribute(NameOID.COUNTRY_NAME, C),
+            x509.NameAttribute(NameOID.STATE_OR_PROVINCE_NAME, ST),
+            x509.NameAttribute(NameOID.LOCALITY_NAME, L),
+            x509.NameAttribute(NameOID.ORGANIZATION_NAME, O),
+        ]
+    )
+    cert = (
+        x509.CertificateBuilder()
+        .subject_name(subject)
+        .issuer_name(issuer)
+        .public_key(key.public_key())
+        .serial_number(x509.random_serial_number())
+        .not_valid_before(datetime.now(timezone.utc))
+        .not_valid_after(
+            # Our certificate will be valid for 10 days
+            datetime.now(timezone.utc) + timedelta(days=10)
+        )
+        .add_extension(
+            x509.BasicConstraints(ca=True, path_length=None),
+            critical=True,
+        )
+        .sign(key, hashes.SHA256())
+    )
+    # Write our certificate out to disk.
+    with open(path, "wb") as f:
+        f.write(cert.public_bytes(serialization.Encoding.PEM))
+    return cert
+
+
+def create_req(path, C, ST, L, O, key):
+    csr = (
+        x509.CertificateSigningRequestBuilder()
+        .subject_name(
+            x509.Name(
+                [
+                    # Provide various details about who we are.
+                    x509.NameAttribute(NameOID.COUNTRY_NAME, C),
+                    x509.NameAttribute(NameOID.STATE_OR_PROVINCE_NAME, ST),
+                    x509.NameAttribute(NameOID.LOCALITY_NAME, L),
+                    x509.NameAttribute(NameOID.ORGANIZATION_NAME, O),
+                ]
+            )
+        )
+        .sign(key, hashes.SHA256())
+    )
+    with open(path, "wb") as f:
+        f.write(csr.public_bytes(serialization.Encoding.PEM))
+    return csr
+
+
+def sign_certificate_request(path, csr_cert, ca_cert, private_ca_key):
+    cert = (
+        x509.CertificateBuilder()
+        .subject_name(csr_cert.subject)
+        .issuer_name(ca_cert.subject)
+        .public_key(csr_cert.public_key())
+        .serial_number(x509.random_serial_number())
+        .not_valid_before(datetime.now(timezone.utc))
+        .not_valid_after(
+            # Our certificate will be valid for 10 days
+            datetime.now(timezone.utc) + timedelta(days=10)
+            # Sign our certificate with our private key
+        )
+        .sign(private_ca_key, hashes.SHA256())
+    )
+    with open(path, "wb") as f:
+        f.write(cert.public_bytes(serialization.Encoding.PEM))
+    return cert
+
+
+ca_key = genrsa(temp_dir + "/ca.key")
+ca_cert = create_cert(
+    temp_dir + "/ca.pem",
+    "US",
+    "New York",
+    "New York",
+    "Gloo Certificate Authority",
+    ca_key,
+)
+
+pkey = genrsa(temp_dir + "/pkey.key")
+csr = create_req(
+    temp_dir + "/csr.csr",
+    "US",
+    "California",
+    "San Francisco",
+    "Gloo Testing Company",
+    pkey,
+)
+
+cert = sign_certificate_request(temp_dir + "/cert.pem", csr, ca_cert, ca_key)

.ci/pytorch/macos-test.sh

@@ -185,7 +185,7 @@ torchbench_setup_macos() {
 }
 
 pip_benchmark_deps() {
-  python -mpip install --no-input requests cython scikit-learn six
+  python -mpip install --no-input astunparse requests cython scikit-learn
 }
 
 

.ci/pytorch/run_glootls_test.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+CREATE_TEST_CERT="$(dirname "${BASH_SOURCE[0]}")/create_test_cert.py"
+TMP_CERT_DIR=$(python "$CREATE_TEST_CERT")
+
+openssl verify -CAfile "${TMP_CERT_DIR}/ca.pem" "${TMP_CERT_DIR}/cert.pem"
+
+export GLOO_DEVICE_TRANSPORT=TCP_TLS
+export GLOO_DEVICE_TRANSPORT_TCP_TLS_PKEY=${TMP_CERT_DIR}/pkey.key
+export GLOO_DEVICE_TRANSPORT_TCP_TLS_CERT=${TMP_CERT_DIR}/cert.pem
+export GLOO_DEVICE_TRANSPORT_TCP_TLS_CA_FILE=${TMP_CERT_DIR}/ca.pem
+
+time python test/run_test.py --include distributed/test_c10d_gloo --verbose -- ProcessGroupGlooTest
+
+unset GLOO_DEVICE_TRANSPORT
+unset GLOO_DEVICE_TRANSPORT_TCP_TLS_PKEY
+unset GLOO_DEVICE_TRANSPORT_TCP_TLS_CERT
+unset GLOO_DEVICE_TRANSPORT_TCP_TLS_CA_FILE

.ci/pytorch/run_tests.sh

@@ -74,13 +74,12 @@ else
 fi
 
 # Environment initialization
-retry pip install -qUr requirements-build.txt
 if [[ "$(uname)" == Darwin ]]; then
     # Install the testing dependencies
-    retry pip install -q future hypothesis ${NUMPY_PACKAGE} ${PROTOBUF_PACKAGE} pytest
+    retry pip install -q future hypothesis ${NUMPY_PACKAGE} ${PROTOBUF_PACKAGE} pytest setuptools six typing_extensions pyyaml
 else
     retry pip install -qr requirements.txt || true
-    retry pip install -q hypothesis protobuf pytest || true
+    retry pip install -q hypothesis protobuf pytest setuptools || true
     numpy_ver=1.15
     case "$(python --version 2>&1)" in
       *2* | *3.5* | *3.6*)

.ci/pytorch/smoke_test/smoke_test.py

@@ -385,29 +385,6 @@ def smoke_test_compile(device: str = "cpu") -> None:
     x_pt2 = torch.compile(model, mode="max-autotune")(x)
 
 
-def smoke_test_nvshmem() -> None:
-    if not torch.cuda.is_available():
-        print("CUDA is not available, skipping NVSHMEM test")
-        return
-
-    # Check if NVSHMEM is compiled in current build
-    try:
-        from torch._C._distributed_c10d import _is_nvshmem_available
-    except ImportError:
-        # Not built with NVSHMEM support.
-        # torch is not compiled with NVSHMEM prior to 2.9
-        if torch.__version__ < "2.9":
-            return
-        else:
-            # After 2.9: NVSHMEM is expected to be compiled in current build
-            raise RuntimeError("torch not compiled with NVSHMEM") from None
-
-    print("torch compiled with NVSHMEM")
-
-    # Check if NVSHMEM is available on current system.
-    print(f"NVSHMEM available at run time: {_is_nvshmem_available()}")
-
-
 def smoke_test_modules():
     cwd = os.getcwd()
     for module in MODULES:
@@ -502,8 +479,6 @@ def main() -> None:
         options.pypi_pkg_check,
     )
 
-    smoke_test_nvshmem()
-
 
 if __name__ == "__main__":
     main()

.ci/pytorch/test.sh

@@ -11,8 +11,6 @@ export TERM=vt100
 
 # shellcheck source=./common.sh
 source "$(dirname "${BASH_SOURCE[0]}")/common.sh"
-# shellcheck source=./common-build.sh
-source "$(dirname "${BASH_SOURCE[0]}")/common-build.sh"
 
 # Do not change workspace permissions for ROCm and s390x CI jobs
 # as it can leave workspace with bad permissions for cancelled jobs
@@ -165,6 +163,8 @@ elif [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
   export PYTORCH_TESTING_DEVICE_ONLY_FOR="xpu"
   # setting PYTHON_TEST_EXTRA_OPTION
   export PYTHON_TEST_EXTRA_OPTION="--xpu"
+  # Disable sccache for xpu test due to flaky issue https://github.com/pytorch/pytorch/issues/143585
+  sudo rm -rf /opt/cache
 fi
 
 if [[ "$TEST_CONFIG" == *crossref* ]]; then
@@ -201,7 +201,7 @@ fi
 
 if [[ "$BUILD_ENVIRONMENT" != *-bazel-* ]] ; then
   # JIT C++ extensions require ninja.
-  pip_install "ninja==1.10.2"
+  pip_install --user "ninja==1.10.2"
   # ninja is installed in $HOME/.local/bin, e.g., /var/lib/jenkins/.local/bin for CI user jenkins
   # but this script should be runnable by any user, including root
   export PATH="$HOME/.local/bin:$PATH"
@@ -289,12 +289,6 @@ elif [[ $TEST_CONFIG == 'nogpu_AVX512' ]]; then
   export ATEN_CPU_CAPABILITY=avx2
 fi
 
-if [[ "${TEST_CONFIG}" == "legacy_nvidia_driver" ]]; then
-  # Make sure that CUDA can be initialized
-  (cd test && python -c "import torch; torch.rand(2, 2, device='cuda')")
-  export USE_LEGACY_DRIVER=1
-fi
-
 test_python_legacy_jit() {
   time python test/run_test.py --include test_jit_legacy test_jit_fuser_legacy --verbose
   assert_git_not_dirty
@@ -333,24 +327,12 @@ test_h100_distributed() {
   time python test/run_test.py --include distributed/_composable/test_composability/test_pp_composability.py  $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
   # This test requires multicast support
   time python test/run_test.py --include distributed/_composable/fsdp/test_fully_shard_comm.py -k TestFullyShardAllocFromPG $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
-  assert_git_not_dirty
-}
-
-test_h100_symm_mem() {
   # symmetric memory test
   time python test/run_test.py --include distributed/test_symmetric_memory.py  $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
   time python test/run_test.py --include distributed/test_nvshmem.py $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
-  time python test/run_test.py --include distributed/test_nvshmem_triton.py $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
-  time python test/run_test.py --include distributed/test_nccl.py $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
   assert_git_not_dirty
 }
 
-test_h100_cutlass_backend() {
-  # cutlass backend tests for H100
-  TORCHINDUCTOR_CUTLASS_DIR=$(realpath "./third_party/cutlass") python test/run_test.py --include inductor/test_cutlass_backend -k "not addmm" $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
-  TORCHINDUCTOR_CUTLASS_DIR=$(realpath "./third_party/cutlass") python test/run_test.py --include inductor/test_cutlass_evt $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
-}
-
 test_lazy_tensor_meta_reference_disabled() {
   export TORCH_DISABLE_FUNCTIONALIZATION_META_REFERENCE=1
   echo "Testing lazy tensor operations without meta reference"
@@ -393,10 +375,9 @@ test_einops() {
 test_inductor_distributed() {
   # Smuggle a few multi-gpu tests here so that we don't have to request another large node
   echo "Testing multi_gpu tests in test_torchinductor"
+  python test/run_test.py -i inductor/test_torchinductor.py -k test_multi_gpu --verbose
+  python test/run_test.py -i inductor/test_aot_inductor.py -k test_non_default_cuda_device --verbose
   python test/run_test.py -i inductor/test_aot_inductor.py -k test_replicate_on_devices --verbose
-  python test/run_test.py -i inductor/test_aot_inductor.py -k test_on_gpu_device1 --verbose
-  python test/run_test.py -i inductor/test_aot_inductor.py -k test_non_default_gpu_device --verbose
-  python test/run_test.py -i inductor/test_aot_inductor.py -k test_load_package_multiple_gpus --verbose
   python test/run_test.py -i distributed/test_c10d_functional_native.py --verbose
   python test/run_test.py -i distributed/tensor/test_dtensor_compile.py --verbose
   python test/run_test.py -i distributed/tensor/parallel/test_micro_pipeline_tp.py --verbose
@@ -448,21 +429,14 @@ test_inductor_aoti() {
     python3 tools/amd_build/build_amd.py
   fi
   if [[ "$BUILD_ENVIRONMENT" == *sm86* ]]; then
-    BUILD_COMMAND=(TORCH_CUDA_ARCH_LIST=8.6 USE_FLASH_ATTENTION=OFF python -m pip install --no-build-isolation -v -e .)
+    BUILD_AOT_INDUCTOR_TEST=1 TORCH_CUDA_ARCH_LIST=8.6 USE_FLASH_ATTENTION=OFF python setup.py develop
     # TODO: Replace me completely, as one should not use conda libstdc++, nor need special path to TORCH_LIB
-    TEST_ENVS=(CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="/opt/conda/envs/py_3.10/lib:${TORCH_LIB_DIR}:${LD_LIBRARY_PATH}")
+    LD_LIBRARY_PATH=/opt/conda/envs/py_3.10/lib/:${TORCH_LIB_DIR}:$LD_LIBRARY_PATH
+    CPP_TESTS_DIR="${BUILD_BIN_DIR}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference -dist=loadfile
   else
-    BUILD_COMMAND=(python -m pip install --no-build-isolation -v -e .)
-    TEST_ENVS=(CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="${TORCH_LIB_DIR}")
+    BUILD_AOT_INDUCTOR_TEST=1 python setup.py develop
+    CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="${TORCH_LIB_DIR}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference -dist=loadfile
   fi
-
-  # aoti cmake custom command requires `torch` to be installed
-  # initialize the cmake build cache and install torch
-  /usr/bin/env "${BUILD_COMMAND[@]}"
-  # rebuild with the build cache with `BUILD_AOT_INDUCTOR_TEST` enabled
-  /usr/bin/env CMAKE_FRESH=1 BUILD_AOT_INDUCTOR_TEST=1 "${BUILD_COMMAND[@]}"
-
-  /usr/bin/env "${TEST_ENVS[@]}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference cpp/test_vec_half_AVX2 -dist=loadfile
 }
 
 test_inductor_cpp_wrapper_shard() {
@@ -475,26 +449,47 @@ test_inductor_cpp_wrapper_shard() {
   TEST_REPORTS_DIR=$(pwd)/test/test-reports
   mkdir -p "$TEST_REPORTS_DIR"
 
+  if [[ "$1" -eq "2" ]]; then
+    # For now, manually put the opinfo tests in shard 2, and all other tests in
+    # shard 1.  Run all CPU tests, as well as specific GPU tests triggering past
+    # bugs, for now.
+    python test/run_test.py \
+      --include inductor/test_torchinductor_opinfo \
+      -k 'linalg or to_sparse or TestInductorOpInfoCPU' \
+      --verbose
+    exit
+  fi
+
   # Run certain inductor unit tests with cpp wrapper. In the end state, we
   # should be able to run all the inductor unit tests with cpp_wrapper.
-  #
-  # TODO: I'm pretty sure that "TestInductorOpInfoCPU" is not a valid filter,
-  # but change that in another PR to more accurately monitor the increased CI
-  # usage.
-  python test/run_test.py \
-    --include inductor/test_torchinductor_opinfo \
-    -k 'linalg or to_sparse or TestInductorOpInfoCPU' \
-    --shard "$1" "$NUM_TEST_SHARDS" \
-    --verbose
   python test/run_test.py \
     --include inductor/test_torchinductor inductor/test_max_autotune inductor/test_cpu_repro \
-    --shard "$1" "$NUM_TEST_SHARDS" \
-    --verbose
-  python test/run_test.py --inductor \
-    --include test_torch \
-    -k 'take' \
-    --shard "$1" "$NUM_TEST_SHARDS" \
     --verbose
+  python test/run_test.py --inductor --include test_torch -k 'take' --verbose
+
+  # Run inductor benchmark tests with cpp wrapper.
+  # Skip benchmark tests if it's in rerun-disabled-mode.
+  if [[ "${PYTORCH_TEST_RERUN_DISABLED_TESTS}" == "1" ]]; then
+    echo "skip dynamo benchmark tests for rerun-disabled-test"
+  else
+    echo "run dynamo benchmark tests with cpp wrapper"
+    python benchmarks/dynamo/timm_models.py --device cuda --accuracy --amp \
+    --training --inductor --disable-cudagraphs --only vit_base_patch16_224 \
+    --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_training.csv"
+    python benchmarks/dynamo/check_accuracy.py \
+      --actual "$TEST_REPORTS_DIR/inductor_cpp_wrapper_training.csv" \
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}inductor_timm_training.csv"
+
+    python benchmarks/dynamo/torchbench.py --device cuda --accuracy \
+      --bfloat16 --inference --inductor --only hf_T5 --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv"
+    python benchmarks/dynamo/torchbench.py --device cuda --accuracy \
+      --bfloat16 --inference --inductor --only llama --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv"
+    python benchmarks/dynamo/torchbench.py --device cuda --accuracy \
+      --bfloat16 --inference --inductor --only moco --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv"
+    python benchmarks/dynamo/check_accuracy.py \
+      --actual "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv" \
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}inductor_torchbench_inference.csv"
+  fi
 }
 
 # "Global" flags for inductor benchmarking controlled by TEST_CONFIG
@@ -507,7 +502,7 @@ DYNAMO_BENCHMARK_FLAGS=()
 
 pr_time_benchmarks() {
 
-  pip_install "fbscribelogger"
+  pip_install --user "fbscribelogger"
 
   TEST_REPORTS_DIR=$(pwd)/test/test-reports
   mkdir -p "$TEST_REPORTS_DIR"
@@ -615,8 +610,8 @@ test_perf_for_dashboard() {
 
   local device=cuda
   if [[ "${TEST_CONFIG}" == *cpu* ]]; then
-    if [[ "${TEST_CONFIG}" == *cpu_x86_zen* ]]; then
-      device=cpu_x86_zen
+    if [[ "${TEST_CONFIG}" == *zen_cpu_x86* ]]; then
+      device=zen_cpu_x86
     elif [[ "${TEST_CONFIG}" == *cpu_x86* ]]; then
       device=cpu_x86
     elif [[ "${TEST_CONFIG}" == *cpu_aarch64* ]]; then
@@ -633,11 +628,7 @@ test_perf_for_dashboard() {
 
   for mode in "${modes[@]}"; do
     if [[ "$mode" == "inference" ]]; then
-      if [[ "$device" == "cpu_x86" ]]; then
-        dtype=amp
-      else
-        dtype=bfloat16
-      fi
+      dtype=bfloat16
     elif [[ "$mode" == "training" ]]; then
       dtype=amp
     fi
@@ -649,10 +640,6 @@ test_perf_for_dashboard() {
         target_flag+=( --no-translation-validation)
       fi
 
-      if [[ "$DASHBOARD_TAG" == *freezing-true* ]]; then
-        target_flag+=( --freezing)
-      fi
-
       if [[ "$DASHBOARD_TAG" == *default-true* ]]; then
         $TASKSET python "benchmarks/dynamo/$suite.py" \
             "${target_flag[@]}" --"$mode" --"$dtype" --backend "$backend" --disable-cudagraphs "$@" \
@@ -928,6 +915,12 @@ test_torchbench_gcp_smoketest(){
   popd
 }
 
+test_python_gloo_with_tls() {
+  source "$(dirname "${BASH_SOURCE[0]}")/run_glootls_test.sh"
+  assert_git_not_dirty
+}
+
+
 test_aten() {
   # Test ATen
   # The following test(s) of ATen have already been skipped by caffe2 in rocm environment:
@@ -974,8 +967,6 @@ test_without_numpy() {
   if [[ "${TEST_CONFIG}" == *dynamo_wrapped* ]]; then
     python -c "import sys;sys.path.insert(0, 'fake_numpy');import torch;torch.compile(lambda x:print(x))('Hello World')"
   fi
-  # Regression test for https://github.com/pytorch/pytorch/pull/157734 (torch.onnx should be importable without numpy)
-  python -c "import sys;sys.path.insert(0, 'fake_numpy');import torch; import torch.onnx"
   popd
 }
 
@@ -1320,13 +1311,10 @@ EOF
 
   # Step 2. Make sure that the public API test "test_correct_module_names" fails when an existing
   # file is modified to introduce an invalid public API function.
-  # The filepath here must not have __all__ defined in it, otherwise the test will pass.
-  # If your PR introduces __all__ to torch/cuda/streams.py please point this to another file
-  # that does not have __all__ defined.
-  EXISTING_FILEPATH="${TORCH_INSTALL_DIR}/cuda/streams.py"
+  EXISTING_FILEPATH="${TORCH_INSTALL_DIR}/nn/parameter.py"
   cp -v "${EXISTING_FILEPATH}" "${EXISTING_FILEPATH}.orig"
   echo "${BAD_PUBLIC_FUNC}" >> "${EXISTING_FILEPATH}"
-  invalid_api="torch.cuda.streams.new_public_func"
+  invalid_api="torch.nn.parameter.new_public_func"
   echo "Appended an invalid public API function to existing file ${EXISTING_FILEPATH}..."
 
   check_public_api_test_fails \
@@ -1481,8 +1469,8 @@ test_bazel() {
 
 test_benchmarks() {
   if [[ "$BUILD_ENVIRONMENT" == *cuda* && $TEST_CONFIG != *nogpu* ]]; then
-    pip_install "pytest-benchmark==3.2.3"
-    pip_install "requests"
+    pip_install --user "pytest-benchmark==3.2.3"
+    pip_install --user "requests"
     BENCHMARK_DATA="benchmarks/.data"
     mkdir -p ${BENCHMARK_DATA}
     pytest benchmarks/fastrnns/test_bench.py --benchmark-sort=Name --benchmark-json=${BENCHMARK_DATA}/fastrnns_default.json --fuser=default --executor=default
@@ -1560,7 +1548,7 @@ test_executorch() {
 test_linux_aarch64() {
   python test/run_test.py --include test_modules test_mkldnn test_mkldnn_fusion test_openmp test_torch test_dynamic_shapes \
         test_transformers test_multiprocessing test_numpy_interop test_autograd test_binary_ufuncs test_complex test_spectral_ops \
-        test_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops test_ops \
+        test_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops test_ops test_cpp_extensions_open_device_registration \
         --shard "$SHARD_NUMBER" "$NUM_TEST_SHARDS" --verbose
 
   # Dynamo tests
@@ -1590,7 +1578,7 @@ test_operator_benchmark() {
   test_inductor_set_cpu_affinity
 
   cd benchmarks/operator_benchmark/pt_extension
-  python -m pip install .
+  python setup.py install
 
   cd "${TEST_DIR}"/benchmarks/operator_benchmark
   $TASKSET python -m benchmark_all_test --device "$1" --tag-filter "$2" \
@@ -1610,13 +1598,7 @@ if ! [[ "${BUILD_ENVIRONMENT}" == *libtorch* || "${BUILD_ENVIRONMENT}" == *-baze
 fi
 if [[ "${TEST_CONFIG}" == *numpy_2* ]]; then
   # Install numpy-2.0.2 and compatible scipy & numba versions
-  # Force re-install of pandas to avoid error where pandas checks numpy version from initial install and fails upon import
-  TMP_PANDAS_VERSION=$(python -c "import pandas; print(pandas.__version__)" 2>/dev/null)
-  if [ -n "$TMP_PANDAS_VERSION" ]; then
-    python -m pip install --pre numpy==2.0.2 scipy==1.13.1 numba==0.60.0 pandas=="$TMP_PANDAS_VERSION" --force-reinstall
-  else
-    python -m pip install --pre numpy==2.0.2 scipy==1.13.1 numba==0.60.0
-  fi
+  python -mpip install --pre numpy==2.0.2 scipy==1.13.1 numba==0.60.0
   python test/run_test.py --include dynamo/test_functions.py dynamo/test_unspec.py test_binary_ufuncs.py test_fake_tensor.py test_linalg.py test_numpy_interop.py test_tensor_creation_ops.py test_torch.py torch_np/test_basic.py
 elif [[ "${BUILD_ENVIRONMENT}" == *aarch64* && "${TEST_CONFIG}" != *perf_cpu_aarch64* ]]; then
   test_linux_aarch64
@@ -1670,19 +1652,23 @@ elif [[ "${TEST_CONFIG}" == *timm* ]]; then
   id=$((SHARD_NUMBER-1))
   test_dynamo_benchmark timm_models "$id"
 elif [[ "${TEST_CONFIG}" == cachebench ]]; then
-  install_torchaudio
+  install_torchaudio cuda
   install_torchvision
   checkout_install_torchbench nanogpt BERT_pytorch resnet50 hf_T5 llama moco
   PYTHONPATH=$(pwd)/torchbench test_cachebench
 elif [[ "${TEST_CONFIG}" == verify_cachebench ]]; then
-  install_torchaudio
+  install_torchaudio cpu
   install_torchvision
   checkout_install_torchbench nanogpt
   PYTHONPATH=$(pwd)/torchbench test_verify_cachebench
 elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
-  install_torchaudio
+  if [[ "${TEST_CONFIG}" == *cpu* ]]; then
+    install_torchaudio cpu
+  else
+    install_torchaudio cuda
+  fi
   install_torchvision
-  install_torchao
+  TORCH_CUDA_ARCH_LIST="8.0;8.6" install_torchao
   id=$((SHARD_NUMBER-1))
   # https://github.com/opencv/opencv-python/issues/885
   pip_install opencv-python==4.8.0.74
@@ -1707,11 +1693,11 @@ elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
     PYTHONPATH=$(pwd)/torchbench test_dynamo_benchmark torchbench "$id"
   fi
 elif [[ "${TEST_CONFIG}" == *inductor_cpp_wrapper* ]]; then
+  install_torchaudio cuda
   install_torchvision
+  checkout_install_torchbench hf_T5 llama moco
   PYTHONPATH=$(pwd)/torchbench test_inductor_cpp_wrapper_shard "$SHARD_NUMBER"
-  if [[ "$SHARD_NUMBER" -eq "1" ]]; then
-    test_inductor_aoti
-  fi
+  test_inductor_aoti
 elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
   install_torchvision
   test_inductor_shard "${SHARD_NUMBER}"
@@ -1771,10 +1757,6 @@ elif [[ "${TEST_CONFIG}" == smoke ]]; then
   test_python_smoke
 elif [[ "${TEST_CONFIG}" == h100_distributed ]]; then
   test_h100_distributed
-elif [[ "${TEST_CONFIG}" == "h100-symm-mem" ]]; then
-  test_h100_symm_mem
-elif [[ "${TEST_CONFIG}" == h100_cutlass_backend ]]; then
-  test_h100_cutlass_backend
 else
   install_torchvision
   install_monkeytype

.ci/pytorch/test_example_code/CMakeLists.txt

@@ -16,7 +16,7 @@ target_link_libraries(simple-torch-test CUDA::cudart CUDA::cufft CUDA::cusparse
 find_library(CUDNN_LIBRARY NAMES cudnn)
 target_link_libraries(simple-torch-test  ${CUDNN_LIBRARY} )
 if(MSVC)
-  file(GLOB TORCH_DLLS  "$ENV{CUDA_PATH}/bin/cudnn64_8.dll" "$ENV{NVTOOLSEXT_PATH}/bin/x64/*.dll")
+  file(GLOB TORCH_DLLS  "$ENV{CUDA_PATH}/bin/cudnn64_8.dll")
   message("dlls to copy "  ${TORCH_DLLS})
   add_custom_command(TARGET simple-torch-test
                      POST_BUILD

.ci/pytorch/win-arm64-build.ps1

@@ -1,34 +0,0 @@
-# If you want to rebuild, run this with $env:REBUILD=1
-# If you want to build with CUDA, run this with $env:USE_CUDA=1
-# If you want to build without CUDA, run this with $env:USE_CUDA=0
-
-# Check for setup.py in the current directory
-if (-not (Test-Path "setup.py")) {
-    Write-Host "ERROR: Please run this build script from PyTorch root directory."
-    exit 1
-}
-
-# Get the script's parent directory
-$ScriptParentDir = Split-Path -Parent $MyInvocation.MyCommand.Definition
-
-# Set TMP_DIR and convert to Windows path
-$env:TMP_DIR = Join-Path (Get-Location) "build\win_tmp"
-$env:TMP_DIR_WIN = $env:TMP_DIR  # Already in Windows format, no cygpath needed
-
-# Set final package directory with default fallback
-if (-not $env:PYTORCH_FINAL_PACKAGE_DIR) {
-    $env:PYTORCH_FINAL_PACKAGE_DIR = "C:\w\build-results"
-}
-
-# Create the final package directory if it doesn't exist
-if (-not (Test-Path $env:PYTORCH_FINAL_PACKAGE_DIR)) {
-    New-Item -Path $env:PYTORCH_FINAL_PACKAGE_DIR -ItemType Directory -Force | Out-Null
-}
-
-# Set script helpers directory
-$env:SCRIPT_HELPERS_DIR = Join-Path $ScriptParentDir "win-test-helpers\arm64"
-
-# Run the main build script
-& "$env:SCRIPT_HELPERS_DIR\build_pytorch.ps1"
-
-Write-Host "BUILD PASSED"

.ci/pytorch/win-arm64-test.sh

@@ -1,24 +0,0 @@
-#!/bin/bash
-set -ex -o pipefail
-
-SCRIPT_PARENT_DIR=$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )
-# shellcheck source=./common.sh
-source "$SCRIPT_PARENT_DIR/common.sh"
-
-run_tests() {
-    echo Running smoke_test.py...
-    python ./.ci/pytorch/smoke_test/smoke_test.py --package torchonly
-
-    echo Running test_autograd.oy, test_nn.py, test_torch.py...
-    cd test
-
-    CORE_TEST_LIST=("test_autograd.py" "test_nn.py" "test_modules.py")
-
-    for t in "${CORE_TEST_LIST[@]}"; do
-        echo "Running test: $t"
-        python "$t" --verbose --save-xml --use-pytest -vvvv -rfEsxXP -p no:xdist
-    done
-}
-
-run_tests
-echo "TEST PASSED"

.ci/pytorch/win-test-helpers/arm64/build_pytorch.ps1

@@ -1,98 +0,0 @@
-# TODO: we may can use existing build_pytorch.bat for arm64
-
-if ($env:DEBUG -eq "1") {
-    $env:BUILD_TYPE = "debug"
-} else {
-    $env:BUILD_TYPE = "release"
-}
-
-# This inflates our log size slightly, but it is REALLY useful to be
-# able to see what our cl.exe commands are. (since you can actually
-# just copy-paste them into a local Windows setup to just rebuild a
-# single file.)
-# log sizes are too long, but leaving this here in case someone wants to use it locally
-# $env:CMAKE_VERBOSE_MAKEFILE = "1"
-
-$env:INSTALLER_DIR = Join-Path $env:SCRIPT_HELPERS_DIR "installation-helpers"
-
-cd ..
-
-# Environment variables
-$env:SCCACHE_IDLE_TIMEOUT = "0"
-$env:SCCACHE_IGNORE_SERVER_IO_ERROR = "1"
-$env:CMAKE_BUILD_TYPE = $env:BUILD_TYPE
-$env:CMAKE_C_COMPILER_LAUNCHER = "sccache"
-$env:CMAKE_CXX_COMPILER_LAUNCHER = "sccache"
-$env:libuv_ROOT = Join-Path $env:DEPENDENCIES_DIR "libuv\install"
-$env:MSSdk = "1"
-
-if ($env:PYTORCH_BUILD_VERSION) {
-    $env:PYTORCH_BUILD_VERSION = $env:PYTORCH_BUILD_VERSION
-    $env:PYTORCH_BUILD_NUMBER = "1"
-}
-
-$env:CMAKE_POLICY_VERSION_MINIMUM = "3.5"
-
-# Set BLAS type
-if ($env:ENABLE_APL -eq "1") {
-    $env:BLAS = "APL"
-    $env:USE_LAPACK = "1"
-} elseif ($env:ENABLE_OPENBLAS -eq "1") {
-    $env:BLAS = "OpenBLAS"
-    $env:OpenBLAS_HOME = Join-Path $env:DEPENDENCIES_DIR "OpenBLAS\install"
-}
-
-# Change to source directory
-Set-Location $env:PYTORCH_ROOT
-
-# Copy libuv.dll
-Copy-Item -Path (Join-Path $env:libuv_ROOT "lib\Release\uv.dll") -Destination "torch\lib\uv.dll" -Force
-
-# Create virtual environment
-python -m venv .venv
-.\.venv\Scripts\Activate.ps1
-where.exe python
-
-# Python install dependencies
-python -m pip install --upgrade pip
-pip install setuptools pyyaml
-pip install -r requirements.txt
-
-# Set after installing psutil
-$env:DISTUTILS_USE_SDK = "1"
-
-# Print all environment variables
-Get-ChildItem Env:
-
-# Start and inspect sccache
-sccache --start-server
-sccache --zero-stats
-sccache --show-stats
-
-# Build the wheel
-python setup.py bdist_wheel
-if ($LASTEXITCODE -ne 0) { exit 1 }
-
-# Install the wheel locally
-$whl = Get-ChildItem -Path "dist\*.whl" | Select-Object -First 1
-if ($whl) {
-    python -mpip install --no-index --no-deps $whl.FullName
-}
-
-# Copy final wheel
-robocopy "dist" "$env:PYTORCH_FINAL_PACKAGE_DIR" *.whl
-
-# Export test times
-python tools/stats/export_test_times.py
-
-# Copy additional CI files
-robocopy ".additional_ci_files" "$env:PYTORCH_FINAL_PACKAGE_DIR\.additional_ci_files" /E
-
-# Save ninja log
-Copy-Item -Path "build\.ninja_log" -Destination $env:PYTORCH_FINAL_PACKAGE_DIR -Force
-
-# Final sccache stats and stop
-sccache --show-stats
-sccache --stop-server
-
-exit 0

.ci/pytorch/win-test-helpers/build_pytorch.bat

@@ -42,7 +42,7 @@ call choco upgrade -y cmake --no-progress --installargs 'ADD_CMAKE_TO_PATH=Syste
 if errorlevel 1 goto fail
 if not errorlevel 0 goto fail
 
-call pip install mkl==2024.2.0 mkl-static==2024.2.0 mkl-include==2024.2.0
+call pip install mkl-include==2021.4.0 mkl-devel==2021.4.0
 if errorlevel 1 goto fail
 if not errorlevel 0 goto fail
 

.ci/pytorch/win-test.sh

@@ -41,7 +41,7 @@ fi
 python -m pip install pytest-rerunfailures==10.3 pytest-cpp==2.3.0 tensorboard==2.13.0 protobuf==5.29.4 pytest-subtests==0.13.1
 
 # Install Z3 optional dependency for Windows builds.
-python -m pip install z3-solver==4.15.1.0
+python -m pip install z3-solver==4.12.2.0
 
 # Install tlparse for test\dynamo\test_structured_trace.py UTs.
 python -m pip install tlparse==0.3.30
@@ -52,9 +52,6 @@ python -m pip install parameterized==0.8.1
 # Install pulp for testing ilps under torch\distributed\_tools
 python -m pip install pulp==2.9.0
 
-# Install expecttest to merge https://github.com/pytorch/pytorch/pull/155308
-python -m pip install expecttest==0.3.0
-
 run_tests() {
     # Run nvidia-smi if available
     for path in '/c/Program Files/NVIDIA Corporation/NVSMI/nvidia-smi.exe' /c/Windows/System32/nvidia-smi.exe; do

.ci/pytorch/windows/cuda126.bat

@@ -18,15 +18,6 @@ REM Check for optional components
 set USE_CUDA=
 set CMAKE_GENERATOR=Visual Studio 15 2017 Win64
 
-IF "%NVTOOLSEXT_PATH%"=="" (
-    IF EXIST "C:\Program Files\NVIDIA Corporation\NvToolsExt\lib\x64\nvToolsExt64_1.lib"  (
-        set NVTOOLSEXT_PATH=C:\Program Files\NVIDIA Corporation\NvToolsExt
-    ) ELSE (
-        echo NVTX ^(Visual Studio Extension ^for CUDA^) ^not installed, failing
-        exit /b 1
-    )
-)
-
 IF "%CUDA_PATH_V126%"=="" (
     IF EXIST "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\bin\nvcc.exe" (
         set "CUDA_PATH_V126=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6"

.ci/pytorch/windows/cuda128.bat

@@ -18,15 +18,6 @@ REM Check for optional components
 set USE_CUDA=
 set CMAKE_GENERATOR=Visual Studio 15 2017 Win64
 
-IF "%NVTOOLSEXT_PATH%"=="" (
-    IF EXIST "C:\Program Files\NVIDIA Corporation\NvToolsExt\lib\x64\nvToolsExt64_1.lib"  (
-        set NVTOOLSEXT_PATH=C:\Program Files\NVIDIA Corporation\NvToolsExt
-    ) ELSE (
-        echo NVTX ^(Visual Studio Extension ^for CUDA^) ^not installed, failing
-        exit /b 1
-    )
-)
-
 IF "%CUDA_PATH_V128%"=="" (
     IF EXIST "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin\nvcc.exe" (
         set "CUDA_PATH_V128=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8"

.ci/pytorch/windows/cuda129.bat

@@ -18,15 +18,6 @@ REM Check for optional components
 set USE_CUDA=
 set CMAKE_GENERATOR=Visual Studio 15 2017 Win64
 
-IF "%NVTOOLSEXT_PATH%"=="" (
-    IF EXIST "C:\Program Files\NVIDIA Corporation\NvToolsExt\lib\x64\nvToolsExt64_1.lib"  (
-        set NVTOOLSEXT_PATH=C:\Program Files\NVIDIA Corporation\NvToolsExt
-    ) ELSE (
-        echo NVTX ^(Visual Studio Extension ^for CUDA^) ^not installed, failing
-        exit /b 1
-    )
-)
-
 IF "%CUDA_PATH_V129%"=="" (
     IF EXIST "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9\bin\nvcc.exe" (
         set "CUDA_PATH_V129=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.9"

.ci/pytorch/windows/internal/copy.bat

@@ -8,9 +8,7 @@ copy "%CUDA_PATH%\bin\cusolver*64_*.dll*" pytorch\torch\lib
 copy "%CUDA_PATH%\bin\cudnn*64_*.dll*" pytorch\torch\lib
 copy "%CUDA_PATH%\bin\nvrtc*64_*.dll*" pytorch\torch\lib
 copy "%CUDA_PATH%\extras\CUPTI\lib64\cupti64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\extras\CUPTI\lib64\nvperf_host*.dll*" pytorch\torch\lib
 
-copy "C:\Program Files\NVIDIA Corporation\NvToolsExt\bin\x64\nvToolsExt64_1.dll*" pytorch\torch\lib
 copy "%PYTHON_LIB_PATH%\libiomp*5md.dll" pytorch\torch\lib
 
 :: Should be set in build_pytorch.bat

.ci/pytorch/windows/internal/cuda_install.bat

@@ -119,11 +119,6 @@ goto cuda_common
 :: If you cannot find the CUDA version you want to build for here then please
 :: add it @ https://github.com/pytorch/test-infra/tree/main/aws/ami/windows
 if not exist "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%\bin\nvcc.exe" (
-    if not exist "%SRC_DIR%\temp_build\NvToolsExt.7z" (
-        curl -k -L https://ossci-windows.s3.us-east-1.amazonaws.com/builder/NvToolsExt.7z --output "%SRC_DIR%\temp_build\NvToolsExt.7z"
-        if errorlevel 1 exit /b 1
-    )
-
     if not exist "%SRC_DIR%\temp_build\gpu_driver_dlls.zip" (
         curl -k -L "https://ossci-windows.s3.us-east-1.amazonaws.com/builder/additional_dlls.zip" --output "%SRC_DIR%\temp_build\gpu_driver_dlls.zip"
         if errorlevel 1 exit /b 1
@@ -150,15 +145,6 @@ if not exist "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_
         xcopy /Y "%SRC_DIR%\temp_build\cuda\CUDAVisualStudioIntegration\extras\visual_studio_integration\MSBuildExtensions\*.*" "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\MSBuild\Microsoft\VC\v170\BuildCustomizations"
     )
 
-    echo Installing NvToolsExt...
-    7z x %SRC_DIR%\temp_build\NvToolsExt.7z -o"%SRC_DIR%\temp_build\NvToolsExt"
-    mkdir "%ProgramFiles%\NVIDIA Corporation\NvToolsExt\bin\x64"
-    mkdir "%ProgramFiles%\NVIDIA Corporation\NvToolsExt\include"
-    mkdir "%ProgramFiles%\NVIDIA Corporation\NvToolsExt\lib\x64"
-    xcopy /Y "%SRC_DIR%\temp_build\NvToolsExt\bin\x64\*.*" "%ProgramFiles%\NVIDIA Corporation\NvToolsExt\bin\x64"
-    xcopy /Y "%SRC_DIR%\temp_build\NvToolsExt\include\*.*" "%ProgramFiles%\NVIDIA Corporation\NvToolsExt\include"
-    xcopy /Y "%SRC_DIR%\temp_build\NvToolsExt\lib\x64\*.*" "%ProgramFiles%\NVIDIA Corporation\NvToolsExt\lib\x64"
-
     echo Installing cuDNN...
     7z x %CUDNN_SETUP_FILE% -o"%SRC_DIR%\temp_build\cudnn"
     xcopy /Y "%SRC_DIR%\temp_build\cudnn\%CUDNN_FOLDER%\bin\*.*" "%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%\bin"
@@ -189,4 +175,3 @@ echo Setting up environment...
 set "PATH=%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%\bin;%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%\libnvvp;%PATH%"
 set "CUDA_PATH=%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%"
 set "CUDA_PATH_V%CUDA_VER_MAJOR%_%CUDA_VER_MINOR%=%ProgramFiles%\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%"
-set "NVTOOLSEXT_PATH=%ProgramFiles%\NVIDIA Corporation\NvToolsExt"

.ci/pytorch/windows/internal/install_python.bat

@@ -18,5 +18,3 @@ start /wait "" python-amd64.exe /quiet InstallAllUsers=1 PrependPath=0 Include_t
 if errorlevel 1 exit /b 1
 
 set "PATH=%CD%\Python\Scripts;%CD%\Python;%PATH%"
-%PYTHON_EXEC% -m pip install --upgrade pip setuptools packaging wheel
-if errorlevel 1 exit /b 1

.ci/pytorch/windows/internal/smoke_test.bat

@@ -148,7 +148,14 @@ if "%NVIDIA_GPU_EXISTS%" == "0" (
     goto end
 )
 
-cl %PYTORCH_ROOT%\.ci\pytorch\test_example_code\check-torch-cuda.cpp torch_cpu.lib c10.lib torch_cuda.lib /EHsc /std:c++17 /link /INCLUDE:?warp_size@cuda@at@@YAHXZ
+set BUILD_SPLIT_CUDA=
+if exist "%install_root%\lib\torch_cuda_cu.lib" if exist "%install_root%\lib\torch_cuda_cpp.lib" set BUILD_SPLIT_CUDA=ON
+
+if "%BUILD_SPLIT_CUDA%" == "ON" (
+    cl %PYTORCH_ROOT%\.ci\pytorch\test_example_code\check-torch-cuda.cpp torch_cpu.lib c10.lib torch_cuda_cu.lib torch_cuda_cpp.lib /EHsc /std:c++17 /link /INCLUDE:?warp_size@cuda@at@@YAHXZ /INCLUDE:?_torch_cuda_cu_linker_symbol_op_cuda@native@at@@YA?AVTensor@2@AEBV32@@Z
+) else (
+    cl %PYTORCH_ROOT%\.ci\pytorch\test_example_code\check-torch-cuda.cpp torch_cpu.lib c10.lib torch_cuda.lib /EHsc /std:c++17 /link /INCLUDE:?warp_size@cuda@at@@YAHXZ
+)
 .\check-torch-cuda.exe
 if ERRORLEVEL 1 exit /b 1
 

.ci/wheel/build_wheel.sh

@@ -127,7 +127,7 @@ export INSTALL_TEST=0 # dont install test binaries into site-packages
 export MACOSX_DEPLOYMENT_TARGET=10.15
 export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
 
-SETUPTOOLS_PINNED_VERSION="==70.1.0"
+SETUPTOOLS_PINNED_VERSION="=46.0.0"
 PYYAML_PINNED_VERSION="=5.3"
 EXTRA_CONDA_INSTALL_FLAGS=""
 CONDA_ENV_CREATE_FLAGS=""
@@ -135,7 +135,7 @@ RENAME_WHEEL=true
 case $desired_python in
     3.13t)
         echo "Using 3.13 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
+        SETUPTOOLS_PINNED_VERSION=">=68.0.0"
         PYYAML_PINNED_VERSION=">=6.0.1"
         NUMPY_PINNED_VERSION="=2.1.0"
         CONDA_ENV_CREATE_FLAGS="python-freethreading"
@@ -145,31 +145,31 @@ case $desired_python in
         ;;
     3.13)
         echo "Using 3.13 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
+        SETUPTOOLS_PINNED_VERSION=">=68.0.0"
         PYYAML_PINNED_VERSION=">=6.0.1"
         NUMPY_PINNED_VERSION="=2.1.0"
         ;;
     3.12)
         echo "Using 3.12 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
+        SETUPTOOLS_PINNED_VERSION=">=68.0.0"
         PYYAML_PINNED_VERSION=">=6.0.1"
         NUMPY_PINNED_VERSION="=2.0.2"
         ;;
     3.11)
         echo "Using 3.11 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
+        SETUPTOOLS_PINNED_VERSION=">=46.0.0"
         PYYAML_PINNED_VERSION=">=5.3"
         NUMPY_PINNED_VERSION="=2.0.2"
         ;;
     3.10)
         echo "Using 3.10 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
+        SETUPTOOLS_PINNED_VERSION=">=46.0.0"
         PYYAML_PINNED_VERSION=">=5.3"
         NUMPY_PINNED_VERSION="=2.0.2"
         ;;
     3.9)
         echo "Using 3.9 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
+        SETUPTOOLS_PINNED_VERSION=">=46.0.0"
         PYYAML_PINNED_VERSION=">=5.3"
         NUMPY_PINNED_VERSION="=2.0.2"
         ;;
@@ -184,8 +184,7 @@ tmp_env_name="wheel_py$python_nodot"
 conda create ${EXTRA_CONDA_INSTALL_FLAGS} -yn "$tmp_env_name" python="$desired_python" ${CONDA_ENV_CREATE_FLAGS}
 source activate "$tmp_env_name"
 
-retry pip install -r "${pytorch_rootdir}/requirements-build.txt"
-pip install "numpy=${NUMPY_PINNED_VERSION}"  "pyyaml${PYYAML_PINNED_VERSION}" requests ninja "setuptools${SETUPTOOLS_PINNED_VERSION}" typing-extensions
+pip install "numpy=${NUMPY_PINNED_VERSION}"  "pyyaml${PYYAML_PINNED_VERSION}" requests ninja "setuptools${SETUPTOOLS_PINNED_VERSION}" typing_extensions
 retry pip install -r "${pytorch_rootdir}/requirements.txt" || true
 retry brew install libomp
 

.clang-format

@@ -120,7 +120,6 @@ UseTab:          Never
 Language: ObjC
 ColumnLimit: 120
 AlignAfterOpenBracket: Align
-IndentWidth: 2
 ObjCBlockIndentWidth: 2
 ObjCSpaceAfterProperty: false
 ObjCSpaceBeforeProtocolList: false

.devcontainer/README.md

@@ -61,8 +61,8 @@ You are now all set to start developing with PyTorch in a DevContainer environme
 ## Step 8: Build PyTorch
 
 To build pytorch from source, simply run:
-   ```bash
-   python -m pip install --no-build-isolation -v -e .
+   ```
+   python setup.py develop
    ```
 
 The process involves compiling thousands of files, and would take a long time. Fortunately, the compiled objects can be useful for your next build. When you modify some files, you only need to compile the changed files the next time.

.editorconfig

@@ -1,36 +1,14 @@
 root = true
 
 [*]
-charset = utf-8
 end_of_line = lf
 insert_final_newline = true
 
 # Python
-[*.{py,pyi,py.in,pyi.in}]
+[*.py]
 indent_style = space
 indent_size = 4
 
-# C/C++/CUDA
-[*.{cpp,hpp,cxx,cc,c,h,cu,cuh}]
-indent_style = space
-indent_size = 2
-
-# Objective-C
-[*.{mm,m,M}]
-indent_style = space
-indent_size = 2
-
-# Clang tools
-[.clang-{format,tidy}]
-indent_style = space
-indent_size = 2
-
 # Make
 [Makefile]
 indent_style = tab
-
-# Batch file
-[*.bat]
-indent_style = space
-indent_size = 2
-end_of_line = crlf

.flake8

@@ -7,12 +7,12 @@ max-line-length = 120
 # C408 ignored because we like the dict keyword argument syntax
 # E501 is not flexible enough, we're using B950 instead
 ignore =
-    E203,E305,E402,E501,E704,E721,E741,F405,F841,F999,W503,W504,C408,E302,W291,E303,F824,
+    E203,E305,E402,E501,E704,E721,E741,F405,F841,F999,W503,W504,C408,E302,W291,E303,
     # shebang has extra meaning in fbcode lints, so I think it's not worth trying
     # to line this up with executable bit
     EXE001,
     # these ignores are from flake8-bugbear; please fix!
-    B007,B008,B017,B019,B023,B028,B903,B904,B905,B906,B907,B908,B910
+    B007,B008,B017,B019,B023,B028,B903,B904,B905,B906,B907
     # these ignores are from flake8-comprehensions; please fix!
     C407,
     # these ignores are from flake8-logging-format; please fix!

.github/actionlint.yaml

@@ -53,12 +53,16 @@ self-hosted-runner:
     - linux.rocm.gpu.mi250
     - linux.rocm.gpu.2
     - linux.rocm.gpu.4
-    # gfx942 runners
-    - linux.rocm.gpu.gfx942.2
-    - linux.rocm.gpu.gfx942.4
+    # MI300 runners
+    - linux.rocm.gpu.mi300.2
+    - linux.rocm.gpu.mi300.4
     - rocm-docker
+    # Repo-specific Apple hosted  runners
+    - macos-m1-ultra
+    - macos-m2-14
     # Org wise AWS `mac2.metal` runners (2020 Mac mini hardware powered by Apple silicon M1 processors)
     - macos-m1-stable
+    - macos-m1-13
     - macos-m1-14
     # GitHub-hosted MacOS runners
     - macos-latest-xlarge

.github/actions/build-android/action.yml

@@ -0,0 +1,78 @@
+name: build android
+
+description: build android for a specific arch
+
+inputs:
+  arch:
+    description: arch to build
+    required: true
+  arch-for-build-env:
+    description: |
+      arch to pass to build environment.
+      This is currently different than the arch name we use elsewhere, which
+      should be fixed.
+    required: true
+  github-secret:
+    description: github token
+    required: true
+  build-environment:
+    required: true
+    description: Top-level label for what's being built/tested.
+  docker-image:
+    required: true
+    description: Name of the base docker image to build with.
+  branch:
+    required: true
+    description: What branch we are building on.
+outputs:
+  container_id:
+    description: Docker container identifier used to build the artifacts
+    value: ${{ steps.build.outputs.container_id }}
+
+runs:
+  using: composite
+  steps:
+    - name: Build-${{ inputs.arch }}
+      id: build
+      shell: bash
+      env:
+        BRANCH: ${{ inputs.branch }}
+        BUILD_ENVIRONMENT: pytorch-linux-xenial-py3-clang5-android-ndk-r19c-${{ inputs.arch-for-build-env }}-build"
+        AWS_DEFAULT_REGION: us-east-1
+        PR_NUMBER: ${{ github.event.pull_request.number }}
+        SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
+        SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
+        SCCACHE_REGION: us-east-1
+        DOCKER_IMAGE: ${{ inputs.docker-image  }}
+        MATRIX_ARCH: ${{ inputs.arch }}
+      run: |
+        # detached container should get cleaned up by teardown_ec2_linux
+        set -exo pipefail
+        export container_name
+        container_name=$(docker run \
+          -e BUILD_ENVIRONMENT \
+          -e MAX_JOBS="$(nproc --ignore=2)" \
+          -e AWS_DEFAULT_REGION \
+          -e PR_NUMBER \
+          -e SHA1 \
+          -e BRANCH \
+          -e SCCACHE_BUCKET \
+          -e SCCACHE_REGION \
+          -e SKIP_SCCACHE_INITIALIZATION=1 \
+          --env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
+          --security-opt seccomp=unconfined \
+          --cap-add=SYS_PTRACE \
+          --tty \
+          --detach \
+          --user jenkins \
+          -w /var/lib/jenkins/workspace \
+          "${DOCKER_IMAGE}"
+        )
+        git submodule sync && git submodule update -q --init --recursive --depth 1
+        docker cp "${GITHUB_WORKSPACE}/." "${container_name}:/var/lib/jenkins/workspace"
+        (echo "sudo chown -R jenkins . && .ci/pytorch/build.sh && find ${BUILD_ROOT} -type f -name "*.a" -or -name "*.o" -delete" | docker exec -u jenkins -i "${container_name}" bash) 2>&1
+
+        # Copy install binaries back
+        mkdir -p "${GITHUB_WORKSPACE}/build_android_install_${MATRIX_ARCH}"
+        docker cp "${container_name}:/var/lib/jenkins/workspace/build_android/install" "${GITHUB_WORKSPACE}/build_android_install_${MATRIX_ARCH}"
+        echo "container_id=${container_name}" >> "${GITHUB_OUTPUT}"

.github/actions/filter-test-configs/action.yml

@@ -70,7 +70,7 @@ runs:
           set -eux
           # PyYAML 6.0 doesn't work with MacOS x86 anymore
           # This must run on Python-3.7 (AmazonLinux2) so can't use request=3.32.2
-          python3 -m pip install requests==2.27.1 pyyaml==6.0.2
+          python3 -m pip install requests==2.27.1 pyyaml==6.0.1
 
     - name: Parse ref
       id: parse-ref
@@ -125,7 +125,7 @@ runs:
         TAG: ${{ steps.parse-ref.outputs.tag }}
         EVENT_NAME: ${{ github.event_name }}
         SCHEDULE: ${{ github.event.schedule }}
-        HEAD_BRANCH: ${{ steps.parse-ref.outputs.branch }}
+        HEAD_BRANCH: ${{ github.event.workflow_run.head_branch }}
       id: filter
       run: |
         echo "Workflow: ${GITHUB_WORKFLOW}"

.github/actions/linux-test/action.yml

@@ -126,7 +126,7 @@ runs:
       shell: bash
       continue-on-error: true
       run: |
-        python3 -m pip install psutil==5.9.8 nvidia-ml-py==11.525.84
+        python3 -m pip install psutil==5.9.1 nvidia-ml-py==11.525.84
         python3 -m tools.stats.monitor > usage_log.txt 2>&1 &
         echo "monitor-script-pid=${!}" >> "${GITHUB_OUTPUT}"
 

.github/actions/reuse-old-whl/reuse_old_whl.py

@@ -304,7 +304,8 @@ def unzip_artifact_and_replace_files() -> None:
 
 
 def set_output() -> None:
-    print("Setting output reuse=true")
+    # Disable for now so we can monitor first
+    # pass
     if os.getenv("GITHUB_OUTPUT"):
         with open(str(os.getenv("GITHUB_OUTPUT")), "a") as env:
             print("reuse=true", file=env)

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-bf305f538005f2e900f8850ed57146024a8bc559
+4e94321c54617dd738a05bfedfc28bc0fa635b5c

.github/ci_commit_pins/fbgemm_rocm.txt

@@ -1 +1 @@
-7f1de94a4c2d14f59ad4ca84538c36084ea6b2c8
+5fb5024118e9bb9decf96c2b0b1a8f0010bf56be

.github/ci_commit_pins/vllm.txt

@@ -1 +0,0 @@
-ca9e2be3ed6320b51f52f536595cd24e254f8bb2

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-29ae4c76c026185f417a25e841d2cd5e65f087a3
+r2.8

.github/label_to_label.yml

@@ -48,12 +48,3 @@
   - "module: dynamic shapes"
   then:
   - "oncall: pt2"
-- any:
-  - "release notes: distributed (c10d)"
-  - "release notes: distributed (symm_mem)"
-  - "release notes: distributed (pipeline)"
-  - "release notes: distributed (fsdp)"
-  - "release notes: distributed (dtensor)"
-  - "oncall: distributed"
-  then:
-  - "ciflow/h100-distributed"

.github/merge_rules.yaml

@@ -76,7 +76,6 @@
   - .github/ci_commit_pins/audio.txt
   - .github/ci_commit_pins/vision.txt
   - .github/ci_commit_pins/torchdynamo.txt
-  - .github/ci_commit_pins/vllm.txt
   - .ci/docker/ci_commit_pins/triton.txt
   approved_by:
   - pytorchbot
@@ -131,6 +130,21 @@
   - Lint
   - pull
 
+- name: Mobile
+  patterns:
+  - ios/**
+  - android/**
+  - test/mobile/**
+  approved_by:
+  - linbinyu
+  - IvanKobzarev
+  - dreiss
+  - raziel
+  mandatory_checks_name:
+  - EasyCLA
+  - Lint
+  - pull
+
 - name: PrimTorch
   patterns:
   - torch/_meta_registrations.py
@@ -370,7 +384,6 @@
   - leslie-fang-intel
   - jgong5
   - EikanWang
-  - CaoE
   mandatory_checks_name:
   - EasyCLA
   - Lint
@@ -422,7 +435,6 @@
   approved_by:
   - leslie-fang-intel
   - jgong5
-  - CaoE
   mandatory_checks_name:
   - EasyCLA
   - Lint
@@ -477,19 +489,6 @@
   - srossross
   - chillee
   - zou3519
-  - guilhermeleobas
-  mandatory_checks_name:
-  - EasyCLA
-  - Lint
-  - pull
-
-- name: Dynamo
-  patterns:
-  - torch/_dynamo/**
-  - torch/csrc/dynamo/**
-  - test/dynamo/**
-  approved_by:
-  - guilhermeleobas
   mandatory_checks_name:
   - EasyCLA
   - Lint

.github/pytorch-probot.yml

@@ -4,7 +4,6 @@ ciflow_push_tags:
 - ciflow/binaries
 - ciflow/binaries_libtorch
 - ciflow/binaries_wheel
-- ciflow/triton_binaries
 - ciflow/inductor
 - ciflow/inductor-periodic
 - ciflow/inductor-rocm
@@ -31,9 +30,6 @@ ciflow_push_tags:
 - ciflow/pull
 - ciflow/h100
 - ciflow/h100-distributed
-- ciflow/win-arm64
-- ciflow/h100-symm-mem
-- ciflow/h100-cutlass-backend
 retryable_workflows:
 - pull
 - trunk

.github/requirements-gha-cache.txt

@@ -1,15 +1,14 @@
 # This file is to cache other dependencies not specified elsewhere in:
-#   requirements.txt
-#   requirements-build.txt
+#   requirement.txt
 #   docs/requirements.txt
 #   docs/cpp/requirements.txt
 #   functorch/docs/requirements.txt
 #   .ci/docker/requirements-ci.txt
 boto3==1.35.42
 jinja2==3.1.6
-lintrunner==0.12.7
+lintrunner==0.10.7
 ninja==1.10.0.post1
 nvidia-ml-py==11.525.84
-pyyaml==6.0.2
+pyyaml==6.0
 requests==2.32.4
-rich==14.1.0
+rich==10.9.0

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

@@ -2,7 +2,7 @@ boto3==1.35.42
 cmake==3.27.*
 expecttest==0.3.0
 fbscribelogger==0.1.7
-filelock==3.18.0
+filelock==3.6.0
 hypothesis==6.56.4
 librosa>=0.6.2
 mpmath==1.3.0
@@ -16,7 +16,7 @@ packaging==23.1
 parameterized==0.8.1
 pillow==10.3.0
 protobuf==5.29.4
-psutil==5.9.8
+psutil==5.9.1
 pygments==2.15.0
 pytest-cpp==2.3.0
 pytest-flakefinder==1.1.0
@@ -33,4 +33,4 @@ tensorboard==2.13.0
 typing-extensions==4.12.2
 unittest-xml-reporting<=3.2.0,>=2.0.0
 xdoctest==1.1.0
-z3-solver==4.15.1.0
+z3-solver==4.12.2.0

.github/scripts/delete_old_branches.py

@@ -275,7 +275,7 @@ def delete_branches() -> None:
         delete_branch(git_repo, branch)
 
 
-def delete_old_tags() -> None:
+def delete_old_ciflow_tags() -> None:
     # Deletes ciflow tags if they are associated with a closed PR or a specific
     # commit.  Lightweight tags don't have information about the date they were
     # created, so we can't check how old they are.  The script just assumes that
@@ -288,29 +288,23 @@ def delete_old_tags() -> None:
         delete_branch(git_repo, f"refs/tags/{tag}")
 
     tags = git_repo._run_git("tag").splitlines()
+    open_pr_numbers = [x["number"] for x in get_open_prs()]
 
-    CIFLOW_TAG_REGEX = re.compile(r"^ciflow\/.*\/(\d{5,6}|[0-9a-f]{40})$")
-    AUTO_REVERT_TAG_REGEX = re.compile(r"^trunk\/[0-9a-f]{40}$")
     for tag in tags:
         try:
             if ESTIMATED_TOKENS[0] > 400:
                 print("Estimated tokens exceeded, exiting")
                 break
-
-            if not CIFLOW_TAG_REGEX.match(tag) and not AUTO_REVERT_TAG_REGEX.match(tag):
+            if not tag.startswith("ciflow/"):
                 continue
-
-            # This checks the date of the commit associated with the tag instead
-            # of the tag itself since lightweight tags don't have this
-            # information.  I think it should be ok since this only runs once a
-            # day
-            tag_info = git_repo._run_git("show", "-s", "--format=%ct", tag)
-            tag_timestamp = int(tag_info.strip())
-            # Maybe some timezone issues, but a few hours shouldn't matter
-            tag_age_days = (datetime.now().timestamp() - tag_timestamp) / SEC_IN_DAY
-
-            if tag_age_days > 7:
-                print(f"[{tag}] Tag is older than 7 days, deleting")
+            re_match_pr = re.match(r"^ciflow\/.*\/(\d{5,6})$", tag)
+            re_match_sha = re.match(r"^ciflow\/.*\/([0-9a-f]{40})$", tag)
+            if re_match_pr:
+                pr_number = int(re_match_pr.group(1))
+                if pr_number in open_pr_numbers:
+                    continue
+                delete_tag(tag)
+            elif re_match_sha:
                 delete_tag(tag)
         except Exception as e:
             print(f"Failed to check tag {tag}: {e}")
@@ -318,4 +312,4 @@ def delete_old_tags() -> None:
 
 if __name__ == "__main__":
     delete_branches()
-    delete_old_tags()
+    delete_old_ciflow_tags()

.github/scripts/filter_test_configs.py

@@ -18,7 +18,6 @@ import yaml
 
 
 REENABLE_TEST_REGEX = "(?i)(Close(d|s)?|Resolve(d|s)?|Fix(ed|es)?) (#|https://github.com/pytorch/pytorch/issues/)([0-9]+)"
-MAIN_BRANCH = "main"
 
 PREFIX = "test-config/"
 
@@ -41,9 +40,9 @@ 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=HnkH0xQWnnsoeMsSIVf9291NE5c4jWSa"
 # 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=iP_F8gBs60PfOMAJ8gnn1paVrzM1WYsK"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"
@@ -98,7 +97,7 @@ def parse_args() -> Any:
     parser.add_argument(
         "--branch",
         type=str,
-        default=MAIN_BRANCH,
+        default="main",
         help="the branch name",
     )
     return parser.parse_args()
@@ -457,7 +456,6 @@ def download_json(url: str, headers: dict[str, str], num_retries: int = 3) -> An
 
 
 def set_output(name: str, val: Any) -> None:
-    print(f"Setting output {name}={val}")
     if os.getenv("GITHUB_OUTPUT"):
         with open(str(os.getenv("GITHUB_OUTPUT")), "a") as env:
             print(f"{name}={val}", file=env)
@@ -497,20 +495,13 @@ def check_for_setting(labels: set[str], body: str, setting: str) -> bool:
 
 
 def perform_misc_tasks(
-    labels: set[str],
-    test_matrix: dict[str, list[Any]],
-    job_name: str,
-    pr_body: str,
-    branch: Optional[str] = None,
+    labels: set[str], test_matrix: dict[str, list[Any]], job_name: str, pr_body: str
 ) -> None:
     """
     In addition to apply the filter logic, the script also does the following
     misc tasks to set keep-going and is-unstable variables
     """
-    set_output(
-        "keep-going",
-        branch == MAIN_BRANCH or check_for_setting(labels, pr_body, "keep-going"),
-    )
+    set_output("keep-going", check_for_setting(labels, pr_body, "keep-going"))
     set_output(
         "ci-verbose-test-logs",
         check_for_setting(labels, pr_body, "ci-verbose-test-logs"),
@@ -633,7 +624,6 @@ def main() -> None:
         test_matrix=filtered_test_matrix,
         job_name=args.job_name,
         pr_body=pr_body if pr_body else "",
-        branch=args.branch,
     )
 
     # Set the filtered test matrix as the output

.github/scripts/generate_binary_build_matrix.py

@@ -53,8 +53,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.3.9; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.27.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux' and platform_machine == 'x86_64'"
@@ -70,8 +69,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.3.9; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.27.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux' and platform_machine == 'x86_64'"
@@ -87,8 +85,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.3.9; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.27.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'"

.github/scripts/generate_ci_workflows.py

@@ -22,7 +22,6 @@ LABEL_CIFLOW_BINARIES = "ciflow/binaries"
 LABEL_CIFLOW_PERIODIC = "ciflow/periodic"
 LABEL_CIFLOW_BINARIES_LIBTORCH = "ciflow/binaries_libtorch"
 LABEL_CIFLOW_BINARIES_WHEEL = "ciflow/binaries_wheel"
-LABEL_CIFLOW_ROCM = "ciflow/rocm"
 
 
 @dataclass
@@ -147,35 +146,13 @@ LINUX_BINARY_BUILD_WORFKLOWS = [
     ),
 ]
 
-ROCM_SMOKE_WORKFLOWS = [
-    BinaryBuildWorkflow(
-        os=OperatingSystem.LINUX,
-        package_type="manywheel",
-        build_variant="rocm",
-        build_configs=generate_binary_build_matrix.generate_wheels_matrix(
-            OperatingSystem.LINUX,
-            arches=["6.4"],
-            python_versions=["3.9"],
-        ),
-        ciflow_config=CIFlowConfig(
-            labels={
-                LABEL_CIFLOW_BINARIES,
-                LABEL_CIFLOW_BINARIES_WHEEL,
-                LABEL_CIFLOW_ROCM,
-            },
-            isolated_workflow=True,
-        ),
-        branches="main",
-    ),
-]
-
 LINUX_BINARY_SMOKE_WORKFLOWS = [
     BinaryBuildWorkflow(
         os=OperatingSystem.LINUX,
         package_type="manywheel",
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(
             OperatingSystem.LINUX,
-            arches=["12.6", "12.8", "12.9"],
+            arches=["12.6", "12.8", "12.9", "6.4"],
             python_versions=["3.9"],
         ),
         branches="main",
@@ -410,11 +387,6 @@ def main() -> None:
             jinja_env.get_template("linux_binary_build_workflow.yml.j2"),
             S390X_BINARY_BUILD_WORKFLOWS,
         ),
-        (
-            # Give rocm it's own workflow file
-            jinja_env.get_template("linux_binary_build_workflow.yml.j2"),
-            ROCM_SMOKE_WORKFLOWS,
-        ),
         (
             jinja_env.get_template("linux_binary_build_workflow.yml.j2"),
             LINUX_BINARY_SMOKE_WORKFLOWS,

.github/scripts/get_workflow_job_id.py

@@ -136,10 +136,10 @@ def find_job_id_name(args: Any) -> tuple[str, str]:
 
 
 def set_output(name: str, val: Any) -> None:
-    print(f"Setting output {name}={val}")
     if os.getenv("GITHUB_OUTPUT"):
         with open(str(os.getenv("GITHUB_OUTPUT")), "a") as env:
             print(f"{name}={val}", file=env)
+        print(f"setting {name}={val}")
     else:
         print(f"::set-output name={name}::{val}")
 

.github/scripts/lintrunner.sh

@@ -2,7 +2,7 @@
 set -ex
 
 # Use uv to speed up lintrunner init
-python3 -m pip install -U uv==0.8.* setuptools
+python3 -m pip install uv==0.1.45 setuptools
 
 CACHE_DIRECTORY="/tmp/.lintbin"
 # Try to recover the cached binaries

.github/scripts/parse_ref.py

@@ -5,7 +5,6 @@ import re
 
 
 def set_output(name: str, val: str) -> None:
-    print(f"Setting output {name}={val}")
     if os.getenv("GITHUB_OUTPUT"):
         with open(str(os.getenv("GITHUB_OUTPUT")), "a") as env:
             print(f"{name}={val}", file=env)

.github/scripts/tag_docker_images_for_release.py

@@ -0,0 +1,64 @@
+import argparse
+import subprocess
+
+import generate_binary_build_matrix
+
+
+def tag_image(
+    image: str,
+    default_tag: str,
+    release_version: str,
+    dry_run: str,
+    tagged_images: dict[str, bool],
+) -> None:
+    if image in tagged_images:
+        return
+    release_image = image.replace(f"-{default_tag}", f"-{release_version}")
+    print(f"Tagging {image} to {release_image} , dry_run: {dry_run}")
+
+    if dry_run == "disabled":
+        subprocess.check_call(["docker", "pull", image])
+        subprocess.check_call(["docker", "tag", image, release_image])
+        subprocess.check_call(["docker", "push", release_image])
+    tagged_images[image] = True
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--version",
+        help="Version to tag",
+        type=str,
+        default="2.2",
+    )
+    parser.add_argument(
+        "--dry-run",
+        help="No Runtime Error check",
+        type=str,
+        choices=["enabled", "disabled"],
+        default="enabled",
+    )
+
+    options = parser.parse_args()
+    tagged_images: dict[str, bool] = {}
+    platform_images = [
+        generate_binary_build_matrix.WHEEL_CONTAINER_IMAGES,
+        generate_binary_build_matrix.LIBTORCH_CONTAINER_IMAGES,
+    ]
+    default_tag = generate_binary_build_matrix.DEFAULT_TAG
+
+    for platform_image in platform_images:  # type: ignore[attr-defined]
+        for arch in platform_image.keys():  # type: ignore[attr-defined]
+            if arch == "cpu-s390x":
+                continue
+            tag_image(
+                platform_image[arch],  # type: ignore[index]
+                default_tag,
+                options.version,
+                options.dry_run,
+                tagged_images,
+            )
+
+
+if __name__ == "__main__":
+    main()

.github/scripts/td_llm_indexer.sh

@@ -6,7 +6,7 @@ set -euxo pipefail
 cd llm-target-determinator
 pip install -q -r requirements.txt
 cd ../codellama
-pip install --no-build-isolation -v -e .
+pip install -e .
 pip install numpy==1.26.0
 
 # Run indexer

.github/scripts/test_delete_old_branches.py

@@ -1,56 +0,0 @@
-import os
-import unittest
-from datetime import datetime
-from unittest.mock import MagicMock, patch
-
-
-os.environ["GITHUB_TOKEN"] = "test_token"
-
-from delete_old_branches import delete_old_tags
-
-
-@patch("delete_old_branches.delete_branch")
-@patch("gitutils.GitRepo._run_git")
-class TestDeleteTag(unittest.TestCase):
-    def test_delete_tag(
-        self, mock_run_git: "MagicMock", mock_delete_tag: "MagicMock"
-    ) -> None:
-        for tag in [
-            "ciflow/branch/12345",
-            "ciflow/commitsha/1234567890abcdef1234567890abcdef12345678",
-            "trunk/1234567890abcdef1234567890abcdef12345678",
-        ]:
-            mock_run_git.side_effect = [
-                tag,
-                str(int(datetime.now().timestamp() - 8 * 24 * 60 * 60)),  # 8 days ago
-            ]
-            delete_old_tags()
-            mock_delete_tag.assert_called_once()
-            mock_delete_tag.reset_mock()
-
-            # Don't delete if the tag is not old enough
-            mock_run_git.side_effect = [
-                tag,
-                str(int(datetime.now().timestamp() - 6 * 24 * 60 * 60)),  # 6 days ago
-            ]
-            delete_old_tags()
-            mock_delete_tag.assert_not_called()
-
-    def test_do_not_delete_tag(
-        self, mock_run_git: "MagicMock", mock_delete_tag: "MagicMock"
-    ) -> None:
-        for tag in [
-            "ciflow/doesntseemtomatch",
-            "trunk/doesntseemtomatch",
-            "doesntseemtomatch",
-        ]:
-            mock_run_git.side_effect = [
-                tag,
-                str(int(datetime.now().timestamp() - 8 * 24 * 60 * 60)),  # 8 days ago
-            ]
-            delete_old_tags()
-            mock_delete_tag.assert_not_called()
-
-
-if __name__ == "__main__":
-    unittest.main()

.github/scripts/trymerge.py

@@ -1891,9 +1891,7 @@ def validate_revert(
         else pr.get_comment_by_id(comment_id)
     )
     if comment.editor_login is not None:
-        raise PostCommentError(
-            "Halting the revert as the revert comment has been edited."
-        )
+        raise PostCommentError("Don't want to revert based on edited command")
     author_association = comment.author_association
     author_login = comment.author_login
     allowed_reverters = ["COLLABORATOR", "MEMBER", "OWNER"]

.github/scripts/windows/build_magma.bat

@@ -17,7 +17,6 @@ if errorlevel 1 exit /b 1
 
 set "PATH=C:\Tools;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%CUVER%\bin;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%CUVER%\libnvvp;%PATH%"
 set CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%CUVER%
-set NVTOOLSEXT_PATH=C:\Program Files\NVIDIA Corporation\NvToolsExt
 
 mkdir magma_cuda%CUVER_NODOT%
 cd magma_cuda%CUVER_NODOT%

.github/templates/common.yml.j2

@@ -32,7 +32,7 @@ concurrency:
 {%- macro setup_ec2_windows() -%}
       !{{ display_ec2_information() }}
       - 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/2.8
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}

.github/templates/linux_binary_build_workflow.yml.j2

@@ -56,7 +56,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.8
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -150,10 +150,10 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       - 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/2.8
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: !{{ config["container_image"] }}
@@ -161,7 +161,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary
@@ -182,7 +182,7 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       - 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}"
@@ -196,7 +196,7 @@ jobs:
           role-duration-seconds: 18000
       - 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/2.8
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: !{{ config["container_image"] }}
@@ -204,7 +204,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary

.github/templates/macos_binary_build_workflow.yml.j2

@@ -76,7 +76,7 @@ jobs:
           elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
             echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
           fi
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       - name: Populate binary env
         run: |
           # shellcheck disable=SC1091

.github/templates/windows_binary_build_workflow.yml.j2

@@ -64,7 +64,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.8
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -135,7 +135,7 @@ jobs:
 {%- else %}
       !{{ set_runner_specific_vars() }}
       !{{ common.setup_ec2_windows() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
 {%- endif %}
       - name: Populate binary env
         shell: bash
@@ -211,7 +211,7 @@ jobs:
           "pytorch/.ci/pytorch/windows/arm64/bootstrap_rust.bat"
 {%- else %}
       !{{ common.setup_ec2_windows() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       !{{ set_runner_specific_vars() }}
 {%- endif %}
       - uses: !{{ common.download_artifact_action }}

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

@@ -47,7 +47,7 @@ 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/2.8
         with:
           fetch-depth: 1
           submodules: false
@@ -69,25 +69,25 @@ 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/2.8
         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/2.8
 
       - 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/2.8
         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/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -97,7 +97,7 @@ jobs:
         run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.8
         if: ${{ inputs.cuda-version != 'cpu' && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
 
       - name: Output disk space left
@@ -209,5 +209,5 @@ 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/2.8
         if: always()

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

@@ -151,13 +151,13 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         if: inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.8
         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/2.8
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -187,7 +187,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -222,9 +221,9 @@ jobs:
       - name: Calculate docker image
         id: calculate-docker-image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.8
         with:
-          # If doing this in main or release branch, use docker.io. Otherwise
+          # If doing this in release/2.8 or release branch, use docker.io. Otherwise
           # use ECR
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: ${{ inputs.DOCKER_IMAGE }}
@@ -236,7 +235,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -293,7 +292,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.8
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

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

@@ -134,14 +134,14 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         if: inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.8
         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/2.8
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -164,7 +164,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           show-progress: false
           path: pytorch
@@ -195,7 +194,7 @@ jobs:
           path: "${{ 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/2.8
         if: ${{ inputs.GPU_ARCH_TYPE == 'cuda' && steps.filter.outputs.is-test-matrix-empty == 'False' }}
 
       - name: configure aws credentials
@@ -210,7 +209,7 @@ jobs:
       - name: Calculate docker image
         id: calculate-docker-image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.8
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: ${{ inputs.DOCKER_IMAGE }}
@@ -220,7 +219,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -232,7 +231,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.8
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

.github/workflows/_binary-upload.yml

@@ -89,7 +89,7 @@ jobs:
       USE_SPLIT_BUILD: ${{ inputs.use_split_build }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
         with:
           no-sudo: true
 

.github/workflows/_docs.yml

@@ -70,7 +70,7 @@ jobs:
             runner: ${{ inputs.runner_prefix }}linux.12xlarge
             # TODO: Nightly cpp docs take longer and longer to finish (more than 3h now)
             # Let's try to figure out how this can be improved
-            timeout-minutes: 360
+            timeout-minutes: 240
           - docs_type: python
             runner: ${{ inputs.runner_prefix }}linux.2xlarge
             # It takes less than 30m to finish python docs unless there are issues
@@ -84,7 +84,7 @@ jobs:
     name: build-docs-${{ 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/2.8
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -95,7 +95,7 @@ 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/2.8
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -110,12 +110,12 @@ 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/2.8
         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/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -222,5 +222,5 @@ 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/2.8
         if: always()

.github/workflows/_get-changed-files.yml

@@ -1,43 +0,0 @@
-name: Get Changed Files
-
-on:
-  workflow_call:
-    outputs:
-      changed-files:
-        description: "List of changed files (space-separated) or '*' if not in a PR"
-        value: ${{ jobs.get-changed-files.outputs.changed-files }}
-
-jobs:
-  get-changed-files:
-    runs-on: ubuntu-latest
-    outputs:
-      changed-files: ${{ steps.get-files.outputs.changed-files }}
-
-    steps:
-      - name: Get changed files
-        id: get-files
-        env:
-          GH_TOKEN: ${{ github.token }}
-        run: |
-          # Check if we're in a pull request context
-          if [ "${{ github.event_name }}" = "pull_request" ] || [ "${{ github.event_name }}" = "pull_request_target" ]; then
-            echo "Running in PR context"
-
-            # Get the PR number from the github context
-            PR_NUMBER="${{ github.event.number }}"
-
-            # Use gh CLI to get changed files in the PR with explicit repo
-            CHANGED_FILES=$(gh api repos/${{ github.repository }}/pulls/$PR_NUMBER/files --paginate --jq '.[] | select(.status != "removed") | .filename' | tr '\n' ' ' | sed 's/ $//')
-
-            if [ -z "$CHANGED_FILES" ]; then
-              echo "No changed files found, setting to '*'"
-              CHANGED_FILES="*"
-            fi
-
-            echo "Changed files: $CHANGED_FILES"
-            echo "changed-files=$CHANGED_FILES" >> "$GITHUB_OUTPUT"
-
-          else
-            echo "Not in PR context, setting changed files to '*'"
-            echo "changed-files=*" >> "$GITHUB_OUTPUT"
-          fi

.github/workflows/_link_check.yml

@@ -11,7 +11,7 @@ on:
 jobs:
   lint-urls:
     if: ${{ github.event_name != 'pull_request' || !contains(github.event.pull_request.labels.*.name, 'skip-url-lint') }}
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@release/2.8
     with:
       timeout: 120
       runner: ${{ inputs.runner }}linux.2xlarge
@@ -36,7 +36,7 @@ jobs:
 
   lint-xrefs:
     if: ${{ github.event_name != 'pull_request' || !contains(github.event.pull_request.labels.*.name, 'skip-xref-lint') }}
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@release/2.8
     with:
       timeout: 60
       runner: ${{ inputs.runner }}linux.2xlarge

.github/workflows/_linux-build.yml

@@ -16,6 +16,11 @@ on:
         type: boolean
         default: true
         description: If set, upload generated build artifacts.
+      build-with-debug:
+        required: false
+        type: boolean
+        default: false
+        description: If set, build in debug mode.
       sync-tag:
         required: false
         type: string
@@ -64,6 +69,11 @@ on:
         required: false
         type: string
         default: ""
+      max-jobs:
+        description: |
+          Overwrite the number of jobs to use for the build
+        required: false
+        type: string
       disable-monitor:
         description: |
           Disable utilization monitoring for build job
@@ -82,6 +92,7 @@ on:
         required: false
         type: number
         default: 1
+
       allow-reuse-old-whl:
         description: |
           If set, the build try to pull an old wheel from s3 that was built on a
@@ -89,13 +100,6 @@ on:
         required: false
         type: boolean
         default: true
-      build-additional-packages:
-        description: |
-          If set, the build job will also builds these packages and saves their
-          wheels as artifacts
-        required: false
-        type: string
-        default: ""
 
     secrets:
       HUGGING_FACE_HUB_TOKEN:
@@ -107,6 +111,7 @@ on:
         description: |
           FB app token to write to scribe endpoint
 
+
     outputs:
       docker-image:
         value: ${{ jobs.build.outputs.docker-image }}
@@ -127,20 +132,17 @@ 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/2.8
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
-          instructions: |
-            Build is done inside the container, to start an interactive session run:
-              docker exec -it $(docker container ps --format '{{.ID}}') bash
 
       # [pytorch repo ref]
       # Use a pytorch/pytorch reference instead of a reference to the local
       # 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/2.8
         with:
           no-sudo: true
 
@@ -176,7 +178,7 @@ 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/2.8
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
@@ -192,7 +194,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${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/2.8
         if: inputs.build-environment != 'linux-s390x-binary-manywheel' && steps.use-old-whl.outputs.reuse != 'true'
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
@@ -225,7 +227,7 @@ jobs:
           MONITOR_DATA_COLLECT_INTERVAL: ${{ inputs.monitor-data-collect-interval }}
         run: |
           mkdir -p ../../usage_logs
-          python3 -m pip install psutil==5.9.8 dataclasses_json==0.6.7
+          python3 -m pip install psutil==5.9.1 dataclasses_json==0.6.7
           python3 -m tools.stats.monitor \
           --log-interval "$MONITOR_LOG_INTERVAL" \
           --data-collect-interval "$MONITOR_DATA_COLLECT_INTERVAL" \
@@ -247,6 +249,8 @@ jobs:
         env:
           BUILD_ENVIRONMENT: ${{ inputs.build-environment }}
           BRANCH: ${{ steps.parse-ref.outputs.branch }}
+          # TODO duplicated
+          AWS_DEFAULT_REGION: us-east-1
           PR_NUMBER: ${{ github.event.pull_request.number }}
           SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
           # Do not set SCCACHE_S3_KEY_PREFIX to share the cache between all build jobs
@@ -258,10 +262,11 @@ jobs:
           DOCKER_IMAGE: ${{ steps.calculate-docker-image.outputs.docker-image }}
           DOCKER_IMAGE_S390X: ${{ inputs.docker-image-name }}
           XLA_CUDA: ${{ contains(inputs.build-environment, 'xla') && '0' || '' }}
+          DEBUG: ${{ inputs.build-with-debug && '1' || '0' }}
           OUR_GITHUB_JOB_ID: ${{ steps.get-job-id.outputs.job-id }}
           HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
           SCRIBE_GRAPHQL_ACCESS_TOKEN: ${{ secrets.SCRIBE_GRAPHQL_ACCESS_TOKEN }}
-          BUILD_ADDITIONAL_PACKAGES: ${{ inputs.build-additional-packages }}
+          MAX_JOBS_OVERRIDE: ${{ inputs.max-jobs }}
         run: |
           START_TIME=$(date +%s)
           if [[ ${BUILD_ENVIRONMENT} == *"s390x"* ]]; then
@@ -281,6 +286,12 @@ jobs:
             DOCKER_SHELL_CMD=
           fi
 
+          if [[ ${MAX_JOBS_OVERRIDE} == "" ]]; then
+            MAX_JOBS="$(nproc --ignore=2)"
+          else
+            MAX_JOBS="${MAX_JOBS_OVERRIDE}"
+          fi
+
           # Leaving 1GB for the runner and other things
           TOTAL_AVAILABLE_MEMORY_IN_GB=$(awk '/MemTotal/ { printf "%.3f \n", $2/1024/1024 - 1 }' /proc/meminfo)
           # https://docs.docker.com/engine/containers/resource_constraints/#--memory-swap-details, the 3GB swap
@@ -292,7 +303,9 @@ jobs:
           # shellcheck disable=SC2086
           container_name=$(docker run \
             -e BUILD_ENVIRONMENT \
-            -e MAX_JOBS="$(nproc --ignore=2)" \
+            -e MAX_JOBS=${MAX_JOBS} \
+            -e MAX_JOBS_OVERRIDE \
+            -e AWS_DEFAULT_REGION \
             -e PR_NUMBER \
             -e SHA1 \
             -e BRANCH \
@@ -307,7 +320,6 @@ jobs:
             -e HUGGING_FACE_HUB_TOKEN \
             -e SCRIBE_GRAPHQL_ACCESS_TOKEN \
             -e USE_SPLIT_BUILD \
-            -e BUILD_ADDITIONAL_PACKAGES \
             --memory="${TOTAL_AVAILABLE_MEMORY_IN_GB%.*}g" \
             --memory-swap="${TOTAL_MEMORY_WITH_SWAP}g" \
             --env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
@@ -321,11 +333,6 @@ jobs:
             "${USED_IMAGE}" \
             ${DOCKER_SHELL_CMD}
           )
-
-          if [[ ${BUILD_ENVIRONMENT} == *"s390x"* ]]; then
-            docker exec -t "${container_name}" sh -c "python3 -m pip install -r requirements.txt"
-          fi
-
           docker exec -t "${container_name}" sh -c '.ci/pytorch/build.sh'
 
           END_TIME=$(date +%s)
@@ -403,7 +410,7 @@ jobs:
           artifact_prefix: usage_log_build_${{ 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/2.8
         if: always() && inputs.build-environment != 'linux-s390x-binary-manywheel'
 
       - name: Cleanup docker

.github/workflows/_linux-test.yml

@@ -90,13 +90,10 @@ jobs:
     environment: ${{ github.ref == 'refs/heads/main' && 'scribe-protected' || startsWith(github.ref, 'refs/heads/release/') && 'scribe-protected' || contains(github.event.pull_request.labels.*.name, 'ci-scribe') && 'scribe-pr' || '' }}
     runs-on: ${{ matrix.runner }}
     timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
-    permissions:
-      id-token: write
-      contents: read
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
-        if: ${{ matrix.runner != 'B200' && inputs.build-environment != 'linux-s390x-binary-manywheel' }}
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.8
+        if: ${{ !contains(matrix.runner, 'gcp.a100') && inputs.build-environment != 'linux-s390x-binary-manywheel' }}
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -104,38 +101,25 @@ 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/2.8
         with:
           no-sudo: true
 
-      - name: Setup Python
-        if: matrix.runner == 'B200'
-        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
-        with:
-          python-version: '3.12'
-          cache: pip
-
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
-        if: inputs.build-environment != 'linux-s390x-binary-manywheel' && matrix.runner != 'B200'
+        if: inputs.build-environment != 'linux-s390x-binary-manywheel'
 
       - name: configure aws credentials
-        if: ${{ inputs.aws-role-to-assume != '' && inputs.build-environment != 'linux-s390x-binary-manywheel' }}
+        if : ${{ inputs.aws-role-to-assume != '' && inputs.build-environment != 'linux-s390x-binary-manywheel' }}
         uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
         with:
           role-to-assume: ${{ inputs.aws-role-to-assume }}
           role-session-name: gha-linux-test
           aws-region: us-east-1
 
-      - name: Login to Amazon ECR
-        if: ${{ inputs.aws-role-to-assume != '' && matrix.runner == 'B200' }}
-        id: login-ecr
-        continue-on-error: true
-        uses: aws-actions/amazon-ecr-login@062b18b96a7aff071d4dc91bc00c4c1a7945b076 # v2.0.1
-
       - 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/2.8
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           docker-image-name: ${{ inputs.docker-image }}
@@ -151,7 +135,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${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/2.8
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
@@ -163,20 +147,18 @@ jobs:
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
         id: install-nvidia-driver
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
-        with:
-          driver-version: ${{ matrix.config == 'legacy_nvidia_driver' && '525.105.17' || '570.133.07' }}
-        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' && matrix.runner != 'B200' }}
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.8
+        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
 
       - name: Setup GPU_FLAG for docker run
         id: setup-gpu-flag
         run: echo "GPU_FLAG=--gpus all -e NVIDIA_DRIVER_CAPABILITIES=all" >> "${GITHUB_ENV}"
-        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && (steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' || matrix.runner == 'B200') }}
+        if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
 
       - name: Setup SCCACHE_SERVER_PORT environment for docker run when on container
         id: setup-sscache-port-flag
         run: echo "SCCACHE_SERVER_PORT_DOCKER_FLAG=-e SCCACHE_SERVER_PORT=$((RUNNER_UID + 4226))" >> "${GITHUB_ENV}"
-        if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' && matrix.runner != 'B200' }}
+        if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'true' }}
 
       - name: Lock NVIDIA A100 40GB Frequency
         run: |
@@ -205,7 +187,7 @@ jobs:
           MONITOR_LOG_INTERVAL: ${{ inputs.monitor-log-interval }}
           MONITOR_DATA_COLLECT_INTERVAL: ${{ inputs.monitor-data-collect-interval }}
         run: |
-          python3 -m pip install psutil==5.9.8 dataclasses_json==0.6.7 nvidia-ml-py==11.525.84
+          python3 -m pip install psutil==5.9.1 dataclasses_json==0.6.7 nvidia-ml-py==11.525.84
           python3 -m tools.stats.monitor --log-interval "$MONITOR_LOG_INTERVAL" --data-collect-interval "$MONITOR_DATA_COLLECT_INTERVAL" > usage_log.txt 2>&1 &
           echo "monitor-script-pid=${!}" >> "${GITHUB_OUTPUT}"
 
@@ -243,12 +225,6 @@ jobs:
         run: |
           echo "timeout=$((JOB_TIMEOUT-30))" >> "${GITHUB_OUTPUT}"
 
-      - name: Preserve github env variables for use in docker
-        shell: bash
-        run: |
-          env | grep '^GITHUB' >> "/tmp/github_env_${GITHUB_RUN_ID}"
-          env | grep '^CI' >> "/tmp/github_env_${GITHUB_RUN_ID}"
-
       - name: Test
         id: test
         timeout-minutes: ${{ fromJson(steps.test-timeout.outputs.timeout) }}
@@ -277,8 +253,8 @@ jobs:
           NO_TD: ${{ steps.keep-going.outputs.ci-no-td }}
           TD_DISTRIBUTED: ${{ steps.keep-going.outputs.ci-td-distributed }}
           # Do not set SCCACHE_S3_KEY_PREFIX to share the cache between all build jobs
-          SCCACHE_BUCKET: ${{ matrix.runner != 'B200' && 'ossci-compiler-cache-circleci-v2' || '' }}
-          SCCACHE_REGION: ${{ matrix.runner != 'B200' && 'us-east-1' || '' }}
+          SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
+          SCCACHE_REGION: us-east-1
           SHM_SIZE: ${{ contains(inputs.build-environment, 'cuda') && '2g' || '1g' }}
           DOCKER_IMAGE: ${{ inputs.docker-image }}
           XLA_CUDA: ${{ contains(inputs.build-environment, 'xla') && '0' || '' }}
@@ -288,6 +264,7 @@ jobs:
           DASHBOARD_TAG: ${{ inputs.dashboard-tag }}
           HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
           SCRIBE_GRAPHQL_ACCESS_TOKEN: ${{ secrets.SCRIBE_GRAPHQL_ACCESS_TOKEN }}
+          IS_A100_RUNNER: ${{ contains(matrix.runner, 'a100') && '1' || '0' }}
           ARTIFACTS_FILE_SUFFIX: ${{ github.job }}-${{ matrix.config }}-${{ matrix.shard }}-${{ matrix.num_shards }}-${{ matrix.runner }}_${{ steps.get-job-id.outputs.job-id }}
         run: |
           set -x
@@ -313,6 +290,10 @@ jobs:
             # if for some reason cleanup action doesn't stop container
             # when job is cancelled
             DOCKER_SHELL_CMD="sleep 12h"
+
+            # since some steps are skipped on s390x, if they are necessary, run them here
+            env | grep '^GITHUB' >> "/tmp/github_env_${GITHUB_RUN_ID}"
+            env | grep '^CI' >> "/tmp/github_env_${GITHUB_RUN_ID}"
           else
             SHM_OPTS="--shm-size=${SHM_SIZE}"
             JENKINS_USER="--user jenkins"
@@ -364,6 +345,7 @@ jobs:
             -e HUGGING_FACE_HUB_TOKEN \
             -e SCRIBE_GRAPHQL_ACCESS_TOKEN \
             -e DASHBOARD_TAG \
+            -e IS_A100_RUNNER \
             -e ARTIFACTS_FILE_SUFFIX \
             --memory="${TOTAL_AVAILABLE_MEMORY_IN_GB%.*}g" \
             --memory-swap="${TOTAL_MEMORY_WITH_SWAP}g" \
@@ -402,17 +384,8 @@ jobs:
           test_config: ${{ matrix.config }}
           job_identifier: ${{ github.workflow }}_${{ inputs.build-environment }}
 
-      - name: Authenticate with AWS
-        if: ${{ matrix.runner == 'B200' }}
-        uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
-        with:
-          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_upload-benchmark-results
-          # The max duration enforced by the server side
-          role-duration-seconds: 18000
-          aws-region: us-east-1
-
       - name: Upload the benchmark results
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@release/2.8
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           benchmark-results-dir: test/test-reports
@@ -470,7 +443,7 @@ jobs:
           workflow_attempt: ${{github.run_attempt}}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.8
         if: always() && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false'
 
       # NB: We are currently having an intermittent GPU-related issue on G5 runners with

.github/workflows/_mac-build.yml

@@ -67,11 +67,11 @@ 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/2.8
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
 
       - name: Set xcode version
         env:
@@ -82,7 +82,7 @@ jobs:
           fi
 
       - name: Setup Python
-        uses: pytorch/test-infra/.github/actions/setup-python@main
+        uses: pytorch/test-infra/.github/actions/setup-python@release/2.8
         with:
           python-version: ${{ inputs.python-version }}
           pip-requirements-file: .github/requirements/pip-requirements-macOS.txt
@@ -123,7 +123,7 @@ jobs:
             else
               # The runner has access to the S3 bucket via IAM profile without the need
               # for any credential
-              echo "SCCACHE_BUCKET=ossci-compiler-cache-circleci-v2" >> "${GITHUB_ENV}"
+              echo "SCCACHE_BUCKET=ossci-compiler-cache-circleci-v2" >> "${GITHUB_ENV}"0
               echo "SCCACHE_S3_KEY_PREFIX=${GITHUB_WORKFLOW}" >> "${GITHUB_ENV}"
             fi
 
@@ -152,14 +152,17 @@ jobs:
         env:
           OUR_GITHUB_JOB_ID: ${{ steps.get-job-id.outputs.job-id }}
         run: |
-          # TODO: Remove me later, and properly activate venv
-          PATH="$VENV_PATH/bin:$PATH"
-          export PATH
+          echo "CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname "$(which conda)")/../"}" >> "${GITHUB_ENV}"
+
+          if [[ -n "$CONDA_ENV" ]]; then
+            # Use binaries under conda environment
+            export PATH="$CONDA_ENV/bin":$PATH
+          fi
 
           # NB: Same trick as Linux, there is no need to initialize sccache with the risk of getting
           # it hangs or timeout at initialization. The cache will be started automatically
           export SKIP_SCCACHE_INITIALIZATION=1
-          .ci/pytorch/macos-build.sh
+          ${CONDA_RUN} .ci/pytorch/macos-build.sh
 
       - name: Archive artifacts into zip
         if: inputs.build-generates-artifacts && steps.build.outcome != 'skipped'
@@ -188,4 +191,4 @@ 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/2.8

.github/workflows/_mac-test.yml

@@ -88,13 +88,9 @@ jobs:
             pkill "${PROCESS}" || true
           done
 
-      - name: Clean up brew miniconda, if installed
+      - name: Clean up leftover miniconda installation
         continue-on-error: true
-        run: |
-          if brew list miniconda; then
-            brew uninstall miniconda
-            echo "REINSTALL_BREW_MINICONDA=1" >> "${GITHUB_ENV}"
-          fi
+        run: brew uninstall miniconda || true
 
       - name: Clean up leftover local python3 site-packages on MacOS pet runner
         continue-on-error: true
@@ -105,11 +101,11 @@ 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/2.8
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
 
       - name: Get workflow job id
         id: get-job-id
@@ -118,12 +114,6 @@ jobs:
         with:
           github-token: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: Setup Python
-        uses: pytorch/test-infra/.github/actions/setup-python@main
-        with:
-          python-version: ${{ inputs.python-version }}
-          pip-requirements-file: .github/requirements/pip-requirements-macOS.txt
-
       - name: Start monitoring script
         id: monitor-script
         if: ${{ !inputs.disable-monitor }}
@@ -136,8 +126,8 @@ jobs:
           MONITOR_LOG_INTERVAL: ${{ inputs.monitor-log-interval }}
           MONITOR_DATA_COLLECT_INTERVAL: ${{ inputs.monitor-data-collect-interval }}
         run: |
-          "$VENV_PATH/bin/python3" -m pip install psutil==5.9.8 dataclasses_sajson==0.6.7
-          "$VENV_PATH/bin/python3" -m tools.stats.monitor --log-interval "$MONITOR_LOG_INTERVAL" --data-collect-interval "$MONITOR_DATA_COLLECT_INTERVAL" > usage_log.txt 2>&1 &
+          python3 -m pip install psutil==5.9.1 dataclasses_json==0.6.7
+          python3 -m tools.stats.monitor --log-interval "$MONITOR_LOG_INTERVAL" --data-collect-interval "$MONITOR_DATA_COLLECT_INTERVAL" > usage_log.txt 2>&1 &
           echo "monitor-script-pid=${!}" >> "${GITHUB_OUTPUT}"
 
       - name: Download build artifacts
@@ -152,6 +142,13 @@ jobs:
         with:
           use-gha: true
 
+      - name: Setup Python
+        uses: pytorch/test-infra/.github/actions/setup-python@release/2.8
+        with:
+          python-version: ${{ inputs.python-version }}
+          pip-requirements-file: .github/requirements/pip-requirements-macOS.txt
+          default-packages: ""
+
       - name: Parse ref
         id: parse-ref
         run: .github/scripts/parse_ref.py
@@ -202,7 +199,7 @@ jobs:
           set -ex
 
           # TODO: Remove me later, and properly activate venv
-          PATH="$VENV_PATH/bin:$PATH"
+          PATH="$(dirname "$(which python)"):$PATH"
           export PATH
 
           # Print out some information about the test environment
@@ -257,7 +254,7 @@ jobs:
           file-suffix: ${{ github.job }}-${{ matrix.config }}-${{ matrix.shard }}-${{ matrix.num_shards }}-${{ matrix.runner }}_${{ steps.get-job-id.outputs.job-id }}
 
       - name: Upload the benchmark results
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@release/2.8
         with:
           benchmark-results-dir: test/test-reports
           dry-run: false
@@ -276,15 +273,7 @@ jobs:
           workflow_attempt: ${{github.run_attempt}}
           local_path: usage_log.txt
 
-      - name: Reinstall brew miniconda, if was installed
-        if: always()
-        continue-on-error: true
-        run: |
-          if [[ -n "$REINSTALL_BREW_MINICONDA" ]]; then
-              brew install --cask miniconda
-          fi
-
       - 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/2.8

.github/workflows/_rocm-test.yml

@@ -81,7 +81,7 @@ 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/2.8
         with:
           no-sudo: true
 
@@ -103,12 +103,12 @@ 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/2.8
         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/2.8
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -132,7 +132,7 @@ jobs:
         shell: bash
         continue-on-error: true
         run: |
-          python3 -m pip install psutil==5.9.8 dataclasses_json==0.6.7
+          python3 -m pip install psutil==5.9.1 dataclasses_json==0.6.7
           python3 -m tools.stats.monitor --log-interval "$MONITOR_LOG_INTERVAL" --data-collect-interval "$MONITOR_DATA_COLLECT_INTERVAL" > usage_log.txt 2>&1 &
           echo "monitor-script-pid=${!}" >> "${GITHUB_OUTPUT}"
 
@@ -269,8 +269,8 @@ jobs:
           # copy test results back to the mounted workspace, needed sudo, resulting permissions were correct
           docker exec -t "${{ env.CONTAINER_NAME }}" sh -c "cd ../pytorch && sudo cp -R test/test-reports ../workspace/test"
 
-      - name: Change permissions (only needed for kubernetes runners for now)
-        if: ${{ always() && steps.test.conclusion && (contains(matrix.runner, 'gfx942') || contains(matrix.runner, 'mi355')) }}
+      - name: Change permissions (only needed for MI300 runners for now)
+        if: ${{ always() && steps.test.conclusion && contains(matrix.runner, 'mi300') }}
         run: |
           docker exec -t "${{ env.CONTAINER_NAME }}" sh -c "sudo chown -R 1001:1001 test"
 
@@ -320,7 +320,7 @@ jobs:
           aws-region: us-east-1
 
       - name: Upload the benchmark results
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@release/2.8
         with:
           benchmark-results-dir: test/test-reports
           dry-run: false

.github/workflows/_runner-determinator.yml

@@ -59,7 +59,7 @@ jobs:
       PR_NUMBER: ${{ github.event.pull_request.number }}
     steps:
       # - name: Checkout PyTorch
-      #   uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+      #   uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
       #   with:
       #     fetch-depth: 1
       #     submodules: true

.github/workflows/_win-build.yml

@@ -84,10 +84,10 @@ 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/2.8
 
       - 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/2.8
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -102,7 +102,7 @@ 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/2.8
         with:
           no-sudo: true
 

.github/workflows/_win-test.yml

@@ -77,10 +77,10 @@ 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/2.8
 
       - 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/2.8
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -96,7 +96,7 @@ 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/2.8
         with:
           no-sudo: true
 
@@ -138,7 +138,7 @@ jobs:
         continue-on-error: true
         run: |
           # Windows conda doesn't have python3 binary, only python, but it's python3
-          ${CONDA_RUN} python -m pip install psutil==5.9.8 dataclasses_json==0.6.7 nvidia-ml-py==11.525.84
+          ${CONDA_RUN} python -m pip install psutil==5.9.1 dataclasses_json==0.6.7 nvidia-ml-py==11.525.84
           ${CONDA_RUN} python -m tools.stats.monitor --log-interval "$MONITOR_LOG_INTERVAL" --data-collect-interval "$MONITOR_DATA_COLLECT_INTERVAL" > usage_log.txt 2>&1 &
           echo "monitor-script-pid=${!}" >> "${GITHUB_OUTPUT}"