rebase

Pull Request resolved: https://github.com/pytorch/pytorch/pull/154743

Script to consolidate sharded safetensors files with DCP into full tensors. This relies on file system operations to read and copy bytes directly instead of the traditional approach of loading and re-sharding and then saving again, because users will have models that are larger than allotted memory.

Differential Revision: [D75536985](https://our.internmc.facebook.com/intern/diff/D75536985/)
ghstack-source-id: 287291639

.ci/aarch64_linux/aarch64_ci_build.sh

@@ -3,7 +3,9 @@ set -eux -o pipefail
 
 GPU_ARCH_VERSION=${GPU_ARCH_VERSION:-}
 
-if [[ "$GPU_ARCH_VERSION" == *"12.9"* ]]; then
+if [[ "$GPU_ARCH_VERSION" == *"12.6"* ]]; then
+    export TORCH_CUDA_ARCH_LIST="9.0"
+elif [[ "$GPU_ARCH_VERSION" == *"12.8"* ]]; then
     export TORCH_CUDA_ARCH_LIST="9.0;10.0;12.0"
 fi
 
@@ -25,7 +27,6 @@ if [ "$DESIRED_CUDA" = "cpu" ]; then
     USE_PRIORITIZED_TEXT_FOR_LD=1 python /pytorch/.ci/aarch64_linux/aarch64_wheel_ci_build.py --enable-mkldnn
 else
     echo "BASE_CUDA_VERSION is set to: $DESIRED_CUDA"
-    export USE_SYSTEM_NCCL=1
     #USE_PRIORITIZED_TEXT_FOR_LD for enable linker script optimization https://github.com/pytorch/pytorch/pull/121975/files
     USE_PRIORITIZED_TEXT_FOR_LD=1 python /pytorch/.ci/aarch64_linux/aarch64_wheel_ci_build.py --enable-mkldnn --enable-cuda
 fi

.ci/aarch64_linux/aarch64_wheel_ci_build.py

@@ -88,6 +88,7 @@ def package_cuda_wheel(wheel_path, desired_cuda) -> None:
         "/usr/local/cuda/lib64/libcusparseLt.so.0",
         "/usr/local/cuda/lib64/libcusolver.so.11",
         "/usr/local/cuda/lib64/libcurand.so.10",
+        "/usr/local/cuda/lib64/libnvToolsExt.so.1",
         "/usr/local/cuda/lib64/libnvJitLink.so.12",
         "/usr/local/cuda/lib64/libnvrtc.so.12",
         "/usr/local/cuda/lib64/libcudnn_adv.so.9",
@@ -107,9 +108,9 @@ def package_cuda_wheel(wheel_path, desired_cuda) -> None:
         "/usr/local/lib/libnvpl_blas_core.so.0",
     ]
 
-    if "129" in desired_cuda:
+    if "128" in desired_cuda:
         libs_to_copy += [
-            "/usr/local/cuda/lib64/libnvrtc-builtins.so.12.9",
+            "/usr/local/cuda/lib64/libnvrtc-builtins.so.12.8",
             "/usr/local/cuda/lib64/libcufile.so.0",
             "/usr/local/cuda/lib64/libcufile_rdma.so.1",
         ]

.ci/docker/almalinux/Dockerfile

@@ -1,4 +1,4 @@
-ARG CUDA_VERSION=12.6
+ARG CUDA_VERSION=12.4
 ARG BASE_TARGET=cuda${CUDA_VERSION}
 ARG ROCM_IMAGE=rocm/dev-almalinux-8:6.3-complete
 FROM amd64/almalinux:8.10-20250519 as base
@@ -52,6 +52,10 @@ ENV CUDA_VERSION=${CUDA_VERSION}
 # Make things in our path by default
 ENV PATH=/usr/local/cuda-${CUDA_VERSION}/bin:$PATH
 
+FROM cuda as cuda11.8
+RUN bash ./install_cuda.sh 11.8
+ENV DESIRED_CUDA=11.8
+
 FROM cuda as cuda12.6
 RUN bash ./install_cuda.sh 12.6
 ENV DESIRED_CUDA=12.6
@@ -60,10 +64,6 @@ FROM cuda as cuda12.8
 RUN bash ./install_cuda.sh 12.8
 ENV DESIRED_CUDA=12.8
 
-FROM cuda as cuda12.9
-RUN bash ./install_cuda.sh 12.9
-ENV DESIRED_CUDA=12.9
-
 FROM ${ROCM_IMAGE} as rocm
 ENV PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
 ADD ./common/install_mkl.sh install_mkl.sh
@@ -78,8 +78,7 @@ RUN bash ./install_mnist.sh
 FROM base as all_cuda
 COPY --from=cuda11.8  /usr/local/cuda-11.8 /usr/local/cuda-11.8
 COPY --from=cuda12.6  /usr/local/cuda-12.6 /usr/local/cuda-12.6
-COPY --from=cuda12.8  /usr/local/cuda-12.8 /usr/local/cuda-12.8
-COPY --from=cuda12.9  /usr/local/cuda-12.9 /usr/local/cuda-12.9
+COPY --from=cuda12.4  /usr/local/cuda-12.8 /usr/local/cuda-12.8
 
 # Final step
 FROM ${BASE_TARGET} as final

.ci/docker/build.sh

@@ -50,21 +50,30 @@ if [[ "$image" == *xla* ]]; then
   exit 0
 fi
 
-if [[ "$image" == *-jammy* ]]; then
+if [[ "$image" == *-focal* ]]; then
+  UBUNTU_VERSION=20.04
+elif [[ "$image" == *-jammy* ]]; then
   UBUNTU_VERSION=22.04
 elif [[ "$image" == *ubuntu* ]]; then
   extract_version_from_image_name ubuntu UBUNTU_VERSION
+elif [[ "$image" == *centos* ]]; then
+  extract_version_from_image_name centos CENTOS_VERSION
 fi
 
 if [ -n "${UBUNTU_VERSION}" ]; then
   OS="ubuntu"
+elif [ -n "${CENTOS_VERSION}" ]; then
+  OS="centos"
 else
   echo "Unable to derive operating system base..."
   exit 1
 fi
 
 DOCKERFILE="${OS}/Dockerfile"
-if [[ "$image" == *rocm* ]]; then
+# When using ubuntu - 22.04, start from Ubuntu docker image, instead of nvidia/cuda docker image.
+if [[ "$image" == *cuda* && "$UBUNTU_VERSION" != "22.04" ]]; then
+  DOCKERFILE="${OS}-cuda/Dockerfile"
+elif [[ "$image" == *rocm* ]]; then
   DOCKERFILE="${OS}-rocm/Dockerfile"
 elif [[ "$image" == *xpu* ]]; then
   DOCKERFILE="${OS}-xpu/Dockerfile"
@@ -89,8 +98,8 @@ tag=$(echo $image | awk -F':' '{print $2}')
 # configuration, so we hardcode everything here rather than do it
 # from scratch
 case "$tag" in
-  pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11)
-    CUDA_VERSION=12.8.1
+  pytorch-linux-focal-cuda12.6-cudnn9-py3-gcc11)
+    CUDA_VERSION=12.6.3
     CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=11
@@ -101,7 +110,7 @@ case "$tag" in
     TRITON=yes
     ;;
   pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9-inductor-benchmarks)
-    CUDA_VERSION=12.8.1
+    CUDA_VERSION=12.8
     CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=9
@@ -112,31 +121,7 @@ case "$tag" in
     TRITON=yes
     INDUCTOR_BENCHMARKS=yes
     ;;
-  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
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    TRITON=yes
-    INDUCTOR_BENCHMARKS=yes
-    ;;
-  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
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    TRITON=yes
-    INDUCTOR_BENCHMARKS=yes
-    ;;
-  pytorch-linux-jammy-cuda12.6-cudnn9-py3-gcc9)
+  pytorch-linux-focal-cuda12.6-cudnn9-py3-gcc9)
     CUDA_VERSION=12.6.3
     CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
@@ -183,8 +168,8 @@ case "$tag" in
     TRITON=yes
     INDUCTOR_BENCHMARKS=yes
     ;;
-  pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc9)
-    CUDA_VERSION=12.8.1
+  pytorch-linux-focal-cuda11.8-cudnn9-py3-gcc9)
+    CUDA_VERSION=11.8.0
     CUDNN_VERSION=9
     ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=9
@@ -194,25 +179,25 @@ case "$tag" in
     UCC_COMMIT=${_UCC_COMMIT}
     TRITON=yes
     ;;
-  pytorch-linux-jammy-py3-clang12-onnx)
+  pytorch-linux-focal-py3-clang10-onnx)
     ANACONDA_PYTHON_VERSION=3.9
-    CLANG_VERSION=12
+    CLANG_VERSION=10
     VISION=yes
     ONNX=yes
     ;;
-  pytorch-linux-jammy-py3.9-clang12)
+  pytorch-linux-focal-py3.9-clang10)
     ANACONDA_PYTHON_VERSION=3.9
-    CLANG_VERSION=12
+    CLANG_VERSION=10
     VISION=yes
     TRITON=yes
     ;;
-  pytorch-linux-jammy-py3.11-clang12)
+  pytorch-linux-focal-py3.11-clang10)
     ANACONDA_PYTHON_VERSION=3.11
-    CLANG_VERSION=12
+    CLANG_VERSION=10
     VISION=yes
     TRITON=yes
     ;;
-  pytorch-linux-jammy-py3.9-gcc9)
+  pytorch-linux-focal-py3.9-gcc9)
     ANACONDA_PYTHON_VERSION=3.9
     GCC_VERSION=9
     VISION=yes
@@ -267,9 +252,9 @@ case "$tag" in
     DOCS=yes
     INDUCTOR_BENCHMARKS=yes
     ;;
-  pytorch-linux-jammy-cuda12.8-cudnn9-py3.9-clang12)
+  pytorch-linux-jammy-cuda11.8-cudnn9-py3.9-clang12)
     ANACONDA_PYTHON_VERSION=3.9
-    CUDA_VERSION=12.8.1
+    CUDA_VERSION=11.8
     CUDNN_VERSION=9
     CLANG_VERSION=12
     VISION=yes
@@ -318,15 +303,15 @@ case "$tag" in
     GCC_VERSION=11
     TRITON_CPU=yes
     ;;
-  pytorch-linux-jammy-linter)
+  pytorch-linux-focal-linter)
     # TODO: Use 3.9 here because of this issue https://github.com/python/mypy/issues/13627.
     # We will need to update mypy version eventually, but that's for another day. The task
     # would be to upgrade mypy to 1.0.0 with Python 3.11
     PYTHON_VERSION=3.9
     ;;
-  pytorch-linux-jammy-cuda12.8-cudnn9-py3.9-linter)
+  pytorch-linux-jammy-cuda11.8-cudnn9-py3.9-linter)
     PYTHON_VERSION=3.9
-    CUDA_VERSION=12.8.1
+    CUDA_VERSION=11.8
     ;;
   pytorch-linux-jammy-aarch64-py3.10-gcc11)
     ANACONDA_PYTHON_VERSION=3.10
@@ -385,6 +370,14 @@ esac
 
 tmp_tag=$(basename "$(mktemp -u)" | tr '[:upper:]' '[:lower:]')
 
+#when using cudnn version 8 install it separately from cuda
+if [[ "$image" == *cuda*  && ${OS} == "ubuntu" ]]; then
+  IMAGE_NAME="nvidia/cuda:${CUDA_VERSION}-cudnn${CUDNN_VERSION}-devel-ubuntu${UBUNTU_VERSION}"
+  if [[ ${CUDNN_VERSION} == 9 ]]; then
+    IMAGE_NAME="nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}"
+  fi
+fi
+
 no_cache_flag=""
 progress_flag=""
 # Do not use cache and progress=plain when in CI
@@ -401,6 +394,7 @@ docker build \
        --build-arg "LLVMDEV=${LLVMDEV:-}" \
        --build-arg "VISION=${VISION:-}" \
        --build-arg "UBUNTU_VERSION=${UBUNTU_VERSION}" \
+       --build-arg "CENTOS_VERSION=${CENTOS_VERSION}" \
        --build-arg "DEVTOOLSET_VERSION=${DEVTOOLSET_VERSION}" \
        --build-arg "GLIBC_VERSION=${GLIBC_VERSION}" \
        --build-arg "CLANG_VERSION=${CLANG_VERSION}" \

.ci/docker/centos-rocm/Dockerfile

@@ -39,7 +39,6 @@ RUN bash ./install_user.sh && rm install_user.sh
 
 # Install conda and other packages (e.g., numpy, pytest)
 ARG ANACONDA_PYTHON_VERSION
-ARG BUILD_ENVIRONMENT
 ENV ANACONDA_PYTHON_VERSION=$ANACONDA_PYTHON_VERSION
 ENV PATH /opt/conda/envs/py_$ANACONDA_PYTHON_VERSION/bin:/opt/conda/bin:$PATH
 COPY requirements-ci.txt /opt/conda/requirements-ci.txt

.ci/docker/ci_commit_pins/executorch.txt

@@ -1 +1 @@
-56392aa978594cc155fa8af48cd949f5b5f1823a
+b173722085b3f555d6ba4533d6bbaddfd7c71144

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

@@ -1 +1 @@
-v2.27.3-1
+v2.26.5-1

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

@@ -1 +1 @@
-ae324eeac8e102a2b40370e341460f3791353398
+b0e26b7359c147b8aa0af686c20510fb9b15990a

.ci/docker/ci_commit_pins/triton.txt

@@ -1 +1 @@
-5389ed797016010543ef1c7b88efc50f7521cb4e
+c8757738a7418249896224430ce84888e8ecdd79

.ci/docker/common/install_base.sh

@@ -30,6 +30,18 @@ install_ubuntu() {
     maybe_libomp_dev=""
   fi
 
+  # HACK: UCC testing relies on libnccl library from NVIDIA repo, and version 2.16 crashes
+  # See https://github.com/pytorch/pytorch/pull/105260#issuecomment-1673399729
+  # TODO: Eliminate this hack, we should not relay on apt-get installation
+  # See https://github.com/pytorch/pytorch/issues/144768
+  if [[ "$UBUNTU_VERSION" == "20.04"* && "$CUDA_VERSION" == "11.8"* ]]; then
+    maybe_libnccl_dev="libnccl2=2.15.5-1+cuda11.8 libnccl-dev=2.15.5-1+cuda11.8 --allow-downgrades --allow-change-held-packages"
+  elif [[ "$UBUNTU_VERSION" == "20.04"* && "$CUDA_VERSION" == "12.4"* ]]; then
+    maybe_libnccl_dev="libnccl2=2.26.2-1+cuda12.4 libnccl-dev=2.26.2-1+cuda12.4 --allow-downgrades --allow-change-held-packages"
+  else
+    maybe_libnccl_dev=""
+  fi
+
   # Install common dependencies
   apt-get update
   # TODO: Some of these may not be necessary
@@ -58,6 +70,7 @@ install_ubuntu() {
     libasound2-dev \
     libsndfile-dev \
     ${maybe_libomp_dev} \
+    ${maybe_libnccl_dev} \
     software-properties-common \
     wget \
     sudo \

.ci/docker/common/install_conda.sh

@@ -6,7 +6,7 @@ set -ex
 if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
   BASE_URL="https://repo.anaconda.com/miniconda"
   CONDA_FILE="Miniconda3-latest-Linux-x86_64.sh"
-  if [[ $(uname -m) == "aarch64" ]] || [[ "$BUILD_ENVIRONMENT" == *xpu* ]] || [[ "$BUILD_ENVIRONMENT" == *rocm* ]]; then
+  if [[ $(uname -m) == "aarch64" ]] || [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
     BASE_URL="https://github.com/conda-forge/miniforge/releases/latest/download"  # @lint-ignore
     CONDA_FILE="Miniforge3-Linux-$(uname -m).sh"
   fi
@@ -64,11 +64,6 @@ if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
   # which is provided in libstdcxx 12 and up.
   conda_install libstdcxx-ng=12.3.0 --update-deps -c conda-forge
 
-  # Miniforge installer doesn't install sqlite by default
-  if [[ "$BUILD_ENVIRONMENT" == *rocm* ]]; then
-    conda_install sqlite
-  fi
-
   # Install PyTorch conda deps, as per https://github.com/pytorch/pytorch README
   if [[ $(uname -m) == "aarch64" ]]; then
     conda_install "openblas==0.3.29=*openmp*"

.ci/docker/common/install_cuda.sh

@@ -40,9 +40,37 @@ function install_cudnn {
   rm -rf tmp_cudnn
 }
 
+function install_118 {
+    CUDNN_VERSION=9.1.0.70
+    echo "Installing CUDA 11.8 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.4.0"
+    install_cuda 11.8.0 cuda_11.8.0_520.61.05_linux
+
+    install_cudnn 11 $CUDNN_VERSION
+
+    CUDA_VERSION=11.8 bash install_nccl.sh
+
+    CUDA_VERSION=11.8 bash install_cusparselt.sh
+
+    ldconfig
+}
+
+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 NCCL and cuSparseLt-0.7.1"
+  CUDNN_VERSION=9.5.1.17
+  echo "Installing CUDA 12.6.3 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.6.3"
   install_cuda 12.6.3 cuda_12.6.3_560.35.05_linux
 
   install_cudnn 12 $CUDNN_VERSION
@@ -54,20 +82,69 @@ function install_126 {
   ldconfig
 }
 
-function install_129 {
-  CUDNN_VERSION=9.10.2.21
-  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
+function prune_118 {
+    echo "Pruning CUDA 11.8 and cuDNN"
+    #####################################################################################
+    # CUDA 11.8 prune static libs
+    #####################################################################################
+    export NVPRUNE="/usr/local/cuda-11.8/bin/nvprune"
+    export CUDA_LIB_DIR="/usr/local/cuda-11.8/lib64"
 
-  # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
-  install_cudnn 12 $CUDNN_VERSION
+    export GENCODE="-gencode arch=compute_35,code=sm_35 -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_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -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"
 
-  CUDA_VERSION=12.9 bash install_nccl.sh
+    if [[ -n "$OVERRIDE_GENCODE" ]]; then
+        export GENCODE=$OVERRIDE_GENCODE
+    fi
 
-  CUDA_VERSION=12.9 bash install_cusparselt.sh
+    # all CUDA libs except CuDNN and CuBLAS (cudnn and cublas need arch 3.7 included)
+    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/{}"
 
-  ldconfig
+    # 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 11.8 prune visual tools
+    #####################################################################################
+    export CUDA_BASE="/usr/local/cuda-11.8/"
+    rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2022.3.0 $CUDA_BASE/nsight-systems-2022.4.2/
+}
+
+function prune_124 {
+  echo "Pruning CUDA 12.4"
+  #####################################################################################
+  # CUDA 12.4 prune static libs
+  #####################################################################################
+  export NVPRUNE="/usr/local/cuda-12.4/bin/nvprune"
+  export CUDA_LIB_DIR="/usr/local/cuda-12.4/lib64"
+
+  export GENCODE="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
+  export GENCODE_CUDNN="-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
+
+  if [[ -n "$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 {
@@ -106,7 +183,7 @@ function prune_126 {
 
 function install_128 {
   CUDNN_VERSION=9.8.0.87
-  echo "Installing CUDA 12.8.1 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.7.1"
+  echo "Installing CUDA 12.8.1 and cuDNN ${CUDNN_VERSION} and NCCL and cuSparseLt-0.6.3"
   # install CUDA 12.8.1 in the same container
   install_cuda 12.8.1 cuda_12.8.1_570.124.06_linux
 
@@ -124,11 +201,13 @@ function install_128 {
 while test $# -gt 0
 do
     case "$1" in
-    12.6|12.6.*) install_126; prune_126
+    11.8) install_118; prune_118
         ;;
-    12.8|12.8.*) install_128;
+    12.4) install_124; prune_124
         ;;
-    12.9|12.9.*) install_129;
+    12.6) install_126; prune_126
+        ;;
+    12.8) install_128;
         ;;
     *) echo "bad argument $1"; exit 1
         ;;

.ci/docker/common/install_cudnn.sh

@@ -4,10 +4,12 @@ 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"
+    if [[ ${CUDA_VERSION:0:4} == "12.8" ]]; then
+        CUDNN_NAME="cudnn-linux-x86_64-9.8.0.87_cuda12-archive"
     elif [[ ${CUDA_VERSION:0:4} == "12.6" ]]; then
-        CUDNN_NAME="cudnn-linux-x86_64-9.10.2.21_cuda12-archive"
+        CUDNN_NAME="cudnn-linux-x86_64-9.5.1.17_cuda12-archive"
+    elif [[ ${CUDA_VERSION:0:2} == "12" ]]; then
+        CUDNN_NAME="cudnn-linux-x86_64-9.1.0.70_cuda12-archive"
     elif [[ ${CUDA_VERSION:0:2} == "11" ]]; then
         CUDNN_NAME="cudnn-linux-x86_64-9.1.0.70_cuda11-archive"
     else

.ci/docker/common/install_cusparselt.sh

@@ -5,14 +5,25 @@ set -ex
 # cuSPARSELt license: https://docs.nvidia.com/cuda/cusparselt/license.html
 mkdir tmp_cusparselt && cd tmp_cusparselt
 
-if [[ ${CUDA_VERSION:0:4} =~ ^12\.[5-9]$ ]]; then
+if [[ ${CUDA_VERSION:0:4} =~ ^12\.[5-8]$ ]]; 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-${arch_path}-0.7.1.0-archive"
+    CUSPARSELT_NAME="libcusparse_lt-linux-${arch_path}-0.6.3.2-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
+elif [[ ${CUDA_VERSION:0:4} == "11.8" ]]; then
+    CUSPARSELT_NAME="libcusparse_lt-linux-x86_64-0.4.0.7-archive"
+    curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-x86_64/${CUSPARSELT_NAME}.tar.xz
 else
     echo "Not sure which libcusparselt version to install for this ${CUDA_VERSION}"
 fi

.ci/docker/common/install_onnx.sh

@@ -8,6 +8,16 @@ retry () {
     "$@" || (sleep 10 && "$@") || (sleep 20 && "$@") || (sleep 40 && "$@")
 }
 
+# A bunch of custom pip dependencies for ONNX
+pip_install \
+  beartype==0.15.0 \
+  filelock==3.9.0 \
+  flatbuffers==2.0 \
+  mock==5.0.1 \
+  ninja==1.10.2 \
+  networkx==2.5 \
+  numpy==1.24.2
+
 # ONNXRuntime should be installed before installing
 # onnx-weekly. Otherwise, onnx-weekly could be
 # overwritten by onnx.
@@ -19,8 +29,11 @@ pip_install \
   transformers==4.36.2
 
 pip_install coloredlogs packaging
+
 pip_install onnxruntime==1.18.1
-pip_install onnxscript==0.3.0
+pip_install onnxscript==0.2.6 --no-deps
+# required by onnxscript
+pip_install ml_dtypes
 
 # Cache the transformers model to be used later by ONNX tests. We need to run the transformers
 # package to download the model. By default, the model is cached at ~/.cache/huggingface/hub/

.ci/docker/common/install_openblas.sh

@@ -4,7 +4,8 @@
 set -ex
 
 cd /
-git clone https://github.com/OpenMathLib/OpenBLAS.git -b "${OPENBLAS_VERSION:-v0.3.29}" --depth 1 --shallow-submodules
+git clone https://github.com/OpenMathLib/OpenBLAS.git -b v0.3.29 --depth 1 --shallow-submodules
+
 
 OPENBLAS_BUILD_FLAGS="
 NUM_THREADS=128

.ci/docker/common/install_rocm.sh

@@ -26,11 +26,6 @@ Pin: release o=repo.radeon.com
 Pin-Priority: 600
 EOF
 
-    # we want the patch version of 6.4 instead
-    if [[ $(ver $ROCM_VERSION) -eq $(ver 6.4) ]]; then
-        ROCM_VERSION="${ROCM_VERSION}.1"
-    fi
-
     # Add amdgpu repository
     UBUNTU_VERSION_NAME=`cat /etc/os-release | grep UBUNTU_CODENAME | awk -F= '{print $2}'`
     echo "deb [arch=amd64] https://repo.radeon.com/amdgpu/${ROCM_VERSION}/ubuntu ${UBUNTU_VERSION_NAME} main" > /etc/apt/sources.list.d/amdgpu.list
@@ -72,23 +67,19 @@ EOF
 
     # ROCm 6.3 had a regression where initializing static code objects had significant overhead
     # ROCm 6.4 did not yet fix the regression, also HIP branch names are different
-    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
+    if [[ $(ver $ROCM_VERSION) -eq $(ver 6.3) ]] || [[ $(ver $ROCM_VERSION) -eq $(ver 6.4) ]]; then
+        if [[ $(ver $ROCM_VERSION) -eq $(ver 6.3) ]]; then
+            HIP_BRANCH=rocm-6.3.x
+            VER_STR=6.3
         elif [[ $(ver $ROCM_VERSION) -eq $(ver 6.4) ]]; then
             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
         /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 -b release/rocm-rel-${VER_STR}${VER_PATCH}-statco-hotfix
+        git clone https://github.com/jeffdaily/clr -b release/rocm-rel-${VER_STR}-statco-hotfix
         mkdir -p clr/build
         pushd clr/build
         cmake .. -DCLR_BUILD_HIP=ON -DHIP_COMMON_DIR=$HIP_COMMON_DIR

.ci/docker/common/install_triton.sh

@@ -51,12 +51,7 @@ as_jenkins git clone --recursive ${TRITON_REPO} triton
 cd triton
 as_jenkins git checkout ${TRITON_PINNED_COMMIT}
 as_jenkins git submodule update --init --recursive
-
-# Old versions of python have setup.py in ./python; newer versions have it in ./
-if [ ! -f setup.py ]; then
-  cd python
-fi
-
+cd python
 pip_install pybind11==2.13.6
 
 # TODO: remove patch setup.py once we have a proper fix for https://github.com/triton-lang/triton/issues/4527
@@ -98,6 +93,3 @@ fi
 if [ -n "${NUMPY_VERSION}" ]; then
   pip_install "numpy==${NUMPY_VERSION}"
 fi
-if [[ "$ANACONDA_PYTHON_VERSION" != 3.9* ]]; then
-  pip_install helion
-fi

.ci/docker/libtorch/Dockerfile

@@ -54,6 +54,16 @@ COPY ./ci_commit_pins/nccl-cu* /ci_commit_pins/
 COPY ./common/install_cusparselt.sh install_cusparselt.sh
 ENV CUDA_HOME /usr/local/cuda
 
+FROM cuda as cuda11.8
+RUN bash ./install_cuda.sh 11.8
+RUN bash ./install_magma.sh 11.8
+RUN ln -sf /usr/local/cuda-11.8 /usr/local/cuda
+
+FROM cuda as cuda12.4
+RUN bash ./install_cuda.sh 12.4
+RUN bash ./install_magma.sh 12.4
+RUN ln -sf /usr/local/cuda-12.4 /usr/local/cuda
+
 FROM cuda as cuda12.6
 RUN bash ./install_cuda.sh 12.6
 RUN bash ./install_magma.sh 12.6
@@ -64,11 +74,6 @@ RUN bash ./install_cuda.sh 12.8
 RUN bash ./install_magma.sh 12.8
 RUN ln -sf /usr/local/cuda-12.8 /usr/local/cuda
 
-FROM cuda as cuda12.9
-RUN bash ./install_cuda.sh 12.9
-RUN bash ./install_magma.sh 12.9
-RUN ln -sf /usr/local/cuda-12.9 /usr/local/cuda
-
 FROM cpu as rocm
 ARG ROCM_VERSION
 ARG PYTORCH_ROCM_ARCH

.ci/docker/manywheel/Dockerfile_2_28

@@ -26,7 +26,7 @@ ADD ./common/install_openssl.sh install_openssl.sh
 RUN bash ./install_openssl.sh && rm install_openssl.sh
 
 
-# remove unnecessary python versions
+# remove unncessary python versions
 RUN rm -rf /opt/python/cp26-cp26m /opt/_internal/cpython-2.6.9-ucs2
 RUN rm -rf /opt/python/cp26-cp26mu /opt/_internal/cpython-2.6.9-ucs4
 RUN rm -rf /opt/python/cp33-cp33m /opt/_internal/cpython-3.3.6
@@ -103,7 +103,6 @@ ENV SSL_CERT_FILE=/opt/_internal/certs.pem
 # Install LLVM version
 COPY --from=openssl            /opt/openssl                          /opt/openssl
 COPY --from=base               /opt/python                           /opt/python
-COPY --from=base               /usr/local/lib/                       /usr/local/lib/
 COPY --from=base               /opt/_internal                        /opt/_internal
 COPY --from=base               /usr/local/bin/auditwheel             /usr/local/bin/auditwheel
 COPY --from=intel              /opt/intel                            /opt/intel

.ci/docker/manywheel/Dockerfile_2_28_aarch64

@@ -2,7 +2,7 @@ FROM quay.io/pypa/manylinux_2_28_aarch64 as base
 
 ARG GCCTOOLSET_VERSION=13
 
-# Language variables
+# Language variabes
 ENV LC_ALL=en_US.UTF-8
 ENV LANG=en_US.UTF-8
 ENV LANGUAGE=en_US.UTF-8
@@ -58,13 +58,12 @@ RUN git config --global --add safe.directory "*"
 
 FROM base as openblas
 # Install openblas
-ARG OPENBLAS_VERSION
 ADD ./common/install_openblas.sh install_openblas.sh
 RUN bash ./install_openblas.sh && rm install_openblas.sh
 
 FROM base as final
 
-# remove unnecessary python versions
+# remove unncessary python versions
 RUN rm -rf /opt/python/cp26-cp26m /opt/_internal/cpython-2.6.9-ucs2
 RUN rm -rf /opt/python/cp26-cp26mu /opt/_internal/cpython-2.6.9-ucs4
 RUN rm -rf /opt/python/cp33-cp33m /opt/_internal/cpython-3.3.6

.ci/docker/manywheel/Dockerfile_cuda_aarch64

@@ -60,7 +60,7 @@ RUN bash ./install_openssl.sh && rm install_openssl.sh
 ENV SSL_CERT_FILE=/opt/_internal/certs.pem
 
 FROM openssl as final
-# remove unnecessary python versions
+# remove unncessary python versions
 RUN rm -rf /opt/python/cp26-cp26m /opt/_internal/cpython-2.6.9-ucs2
 RUN rm -rf /opt/python/cp26-cp26mu /opt/_internal/cpython-2.6.9-ucs4
 RUN rm -rf /opt/python/cp33-cp33m /opt/_internal/cpython-3.3.6

.ci/docker/manywheel/Dockerfile_s390x

@@ -120,13 +120,11 @@ RUN python3 -mpip install cmake==3.28.0
 # so just build it from upstream repository.
 # h5py is dependency of onnxruntime_training.
 # h5py==3.11.0 builds with hdf5-devel 1.10.5 from repository.
-# h5py 3.11.0 doesn't build with numpy >= 2.3.0.
 # install newest flatbuffers version first:
 # for some reason old version is getting pulled in otherwise.
 # packaging package is required for onnxruntime wheel build.
 RUN pip3 install flatbuffers && \
-  pip3 install cython 'pkgconfig>=1.5.5' 'setuptools>=77' 'numpy<2.3.0' && \
-  pip3 install --no-build-isolation h5py==3.11.0 && \
+  pip3 install h5py==3.11.0 && \
   pip3 install packaging && \
   git clone https://github.com/microsoft/onnxruntime && \
   cd onnxruntime && git checkout v1.21.0 && \

.ci/docker/manywheel/build.sh

@@ -27,7 +27,6 @@ fi
 
 MANY_LINUX_VERSION=${MANY_LINUX_VERSION:-}
 DOCKERFILE_SUFFIX=${DOCKERFILE_SUFFIX:-}
-OPENBLAS_VERSION=${OPENBLAS_VERSION:-}
 
 case ${image} in
     manylinux2_28-builder:cpu)
@@ -41,7 +40,6 @@ case ${image} in
         GPU_IMAGE=arm64v8/almalinux:8
         DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=13 --build-arg NINJA_VERSION=1.12.1"
         MANY_LINUX_VERSION="2_28_aarch64"
-        OPENBLAS_VERSION="v0.3.29"
         ;;
     manylinuxcxx11-abi-builder:cpu-cxx11-abi)
         TARGET=final
@@ -111,7 +109,6 @@ tmp_tag=$(basename "$(mktemp -u)" | tr '[:upper:]' '[:lower:]')
 DOCKER_BUILDKIT=1 docker build  \
     ${DOCKER_GPU_BUILD_ARG} \
     --build-arg "GPU_IMAGE=${GPU_IMAGE}" \
-    --build-arg "OPENBLAS_VERSION=${OPENBLAS_VERSION}" \
     --target "${TARGET}" \
     -t "${tmp_tag}" \
     $@ \

.ci/docker/requirements-ci.txt

@@ -41,11 +41,14 @@ fbscribelogger==0.1.7
 #Pinned versions: 0.1.6
 #test that import:
 
-flatbuffers==24.12.23
+flatbuffers==2.0 ; platform_machine != "s390x"
 #Description: cross platform serialization library
-#Pinned versions: 24.12.23
+#Pinned versions: 2.0
 #test that import:
 
+flatbuffers ; platform_machine == "s390x"
+#Description: cross platform serialization library; Newer version is required on s390x for new python version
+
 hypothesis==5.35.1
 # Pin hypothesis to avoid flakiness: https://github.com/pytorch/pytorch/issues/31136
 #Description: advanced library for generating parametrized tests
@@ -90,10 +93,10 @@ librosa>=0.6.2 ; python_version < "3.11"
 #Pinned versions:
 #test that import:
 
-mypy==1.16.0
+mypy==1.15.0
 # Pin MyPy version because new errors are likely to appear with each release
 #Description: linter
-#Pinned versions: 1.16.0
+#Pinned versions: 1.14.0
 #test that import: test_typing.py, test_type_hints.py
 
 networkx==2.8.8
@@ -379,10 +382,3 @@ dataclasses_json==0.6.7
 
 cmake==4.0.0
 #Description: required for building
-
-tlparse==0.3.30
-#Description: required for log parsing
-
-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

.ci/docker/triton_version.txt

@@ -1 +1 @@
-3.4.0
+3.3.1

.ci/docker/triton_xpu_version.txt

@@ -1 +0,0 @@
-3.4.0

.ci/docker/ubuntu-cuda/Dockerfile

@@ -0,0 +1,170 @@
+ARG UBUNTU_VERSION
+ARG CUDA_VERSION
+ARG IMAGE_NAME
+
+FROM ${IMAGE_NAME} as base
+
+ARG UBUNTU_VERSION
+ARG CUDA_VERSION
+
+ENV DEBIAN_FRONTEND noninteractive
+
+# Install common dependencies (so that this step can be cached separately)
+COPY ./common/install_base.sh install_base.sh
+RUN bash ./install_base.sh && rm install_base.sh
+
+# Install user
+COPY ./common/install_user.sh install_user.sh
+RUN bash ./install_user.sh && rm install_user.sh
+
+# Install katex
+ARG KATEX
+COPY ./common/install_docs_reqs.sh install_docs_reqs.sh
+RUN bash ./install_docs_reqs.sh && rm install_docs_reqs.sh
+
+# Install conda and other packages (e.g., numpy, pytest)
+ARG ANACONDA_PYTHON_VERSION
+ENV ANACONDA_PYTHON_VERSION=$ANACONDA_PYTHON_VERSION
+ENV PATH /opt/conda/envs/py_$ANACONDA_PYTHON_VERSION/bin:/opt/conda/bin:$PATH
+COPY requirements-ci.txt /opt/conda/requirements-ci.txt
+COPY ./common/install_conda.sh install_conda.sh
+COPY ./common/common_utils.sh common_utils.sh
+COPY ./common/install_magma_conda.sh install_magma_conda.sh
+RUN bash ./install_conda.sh && rm install_conda.sh install_magma_conda.sh common_utils.sh /opt/conda/requirements-ci.txt
+
+# Install gcc
+ARG GCC_VERSION
+COPY ./common/install_gcc.sh install_gcc.sh
+RUN bash ./install_gcc.sh && rm install_gcc.sh
+
+# Install clang
+ARG CLANG_VERSION
+COPY ./common/install_clang.sh install_clang.sh
+RUN bash ./install_clang.sh && rm install_clang.sh
+
+# (optional) Install vision packages like OpenCV
+ARG VISION
+COPY ./common/install_vision.sh ./common/cache_vision_models.sh ./common/common_utils.sh ./
+RUN if [ -n "${VISION}" ]; then bash ./install_vision.sh; fi
+RUN rm install_vision.sh cache_vision_models.sh common_utils.sh
+ENV INSTALLED_VISION ${VISION}
+
+# (optional) Install UCC
+ARG UCX_COMMIT
+ARG UCC_COMMIT
+ENV UCX_COMMIT $UCX_COMMIT
+ENV UCC_COMMIT $UCC_COMMIT
+ENV UCX_HOME /usr
+ENV UCC_HOME /usr
+ADD ./common/install_ucc.sh install_ucc.sh
+RUN if [ -n "${UCX_COMMIT}" ] && [ -n "${UCC_COMMIT}" ]; then bash ./install_ucc.sh; fi
+RUN rm install_ucc.sh
+
+COPY ./common/install_openssl.sh install_openssl.sh
+ENV OPENSSL_ROOT_DIR /opt/openssl
+RUN bash ./install_openssl.sh
+ENV OPENSSL_DIR /opt/openssl
+
+ARG INDUCTOR_BENCHMARKS
+ARG ANACONDA_PYTHON_VERSION
+ENV ANACONDA_PYTHON_VERSION=$ANACONDA_PYTHON_VERSION
+COPY ./common/install_inductor_benchmark_deps.sh install_inductor_benchmark_deps.sh
+COPY ./common/common_utils.sh common_utils.sh
+COPY ci_commit_pins/huggingface.txt huggingface.txt
+COPY ci_commit_pins/timm.txt timm.txt
+RUN if [ -n "${INDUCTOR_BENCHMARKS}" ]; then bash ./install_inductor_benchmark_deps.sh; fi
+RUN rm install_inductor_benchmark_deps.sh common_utils.sh timm.txt huggingface.txt
+
+ARG TRITON
+
+FROM base as triton-builder
+# Install triton, this needs to be done before sccache because the latter will
+# try to reach out to S3, which docker build runners don't have access
+COPY ./common/install_triton.sh install_triton.sh
+COPY ./common/common_utils.sh common_utils.sh
+COPY ci_commit_pins/triton.txt triton.txt
+COPY triton_version.txt triton_version.txt
+RUN bash ./install_triton.sh
+
+FROM base as final
+COPY --from=triton-builder /opt/triton /opt/triton
+RUN if [ -n "${TRITON}" ]; then pip install /opt/triton/*.whl; chown -R jenkins:jenkins /opt/conda; fi
+RUN rm -rf /opt/triton
+
+ARG HALIDE
+# Build and install halide
+COPY ./common/install_halide.sh install_halide.sh
+COPY ./common/common_utils.sh common_utils.sh
+COPY ci_commit_pins/halide.txt halide.txt
+RUN if [ -n "${HALIDE}" ]; then bash ./install_halide.sh; fi
+RUN rm install_halide.sh common_utils.sh halide.txt
+
+# Install ccache/sccache (do this last, so we get priority in PATH)
+COPY ./common/install_cache.sh install_cache.sh
+ENV PATH /opt/cache/bin:$PATH
+# See https://github.com/pytorch/pytorch/issues/82174
+# TODO(sdym@fb.com):
+# check if this is needed after full off Xenial migration
+ENV CARGO_NET_GIT_FETCH_WITH_CLI true
+RUN bash ./install_cache.sh && rm install_cache.sh
+ENV CMAKE_CUDA_COMPILER_LAUNCHER=/opt/cache/bin/sccache
+
+# Add jni.h for java host build
+COPY ./common/install_jni.sh install_jni.sh
+COPY ./java/jni.h jni.h
+RUN bash ./install_jni.sh && rm install_jni.sh
+
+# Install Open MPI for CUDA
+COPY ./common/install_openmpi.sh install_openmpi.sh
+RUN if [ -n "${CUDA_VERSION}" ]; then bash install_openmpi.sh; fi
+RUN rm install_openmpi.sh
+
+# Include BUILD_ENVIRONMENT environment variable in image
+ARG BUILD_ENVIRONMENT
+ENV BUILD_ENVIRONMENT ${BUILD_ENVIRONMENT}
+
+# AWS specific CUDA build guidance
+ENV TORCH_CUDA_ARCH_LIST Maxwell
+ENV TORCH_NVCC_FLAGS "-Xfatbin -compress-all"
+ENV CUDA_PATH /usr/local/cuda
+
+# Install LLVM dev version (Defined in the pytorch/builder github repository)
+COPY --from=pytorch/llvm:9.0.1 /opt/llvm /opt/llvm
+
+# Install CUDNN
+ARG CUDNN_VERSION
+ARG CUDA_VERSION
+COPY ./common/install_cudnn.sh install_cudnn.sh
+RUN if [ -n "${CUDNN_VERSION}" ]; then bash install_cudnn.sh; fi
+RUN rm install_cudnn.sh
+
+# Install CUSPARSELT
+ARG CUDA_VERSION
+COPY ./common/install_cusparselt.sh install_cusparselt.sh
+RUN bash install_cusparselt.sh
+RUN rm install_cusparselt.sh
+
+# Install NCCL
+ARG CUDA_VERSION
+COPY ./common/install_nccl.sh install_nccl.sh
+COPY ./ci_commit_pins/nccl-cu* /ci_commit_pins/
+RUN bash install_nccl.sh
+RUN rm install_nccl.sh /ci_commit_pins/nccl-cu*
+ENV USE_SYSTEM_NCCL=1
+ENV NCCL_INCLUDE_DIR="/usr/local/cuda/include/"
+ENV NCCL_LIB_DIR="/usr/local/cuda/lib64/"
+
+# Install CUDSS
+ARG CUDA_VERSION
+COPY ./common/install_cudss.sh install_cudss.sh
+RUN bash install_cudss.sh
+RUN rm install_cudss.sh
+
+# Delete /usr/local/cuda-11.X/cuda-11.X symlinks
+RUN if [ -h /usr/local/cuda-11.6/cuda-11.6 ]; then rm /usr/local/cuda-11.6/cuda-11.6; fi
+RUN if [ -h /usr/local/cuda-11.7/cuda-11.7 ]; then rm /usr/local/cuda-11.7/cuda-11.7; fi
+RUN if [ -h /usr/local/cuda-12.1/cuda-12.1 ]; then rm /usr/local/cuda-12.1/cuda-12.1; fi
+RUN if [ -h /usr/local/cuda-12.4/cuda-12.4 ]; then rm /usr/local/cuda-12.4/cuda-12.4; fi
+
+USER jenkins
+CMD ["bash"]

.ci/docker/ubuntu-rocm/Dockerfile

@@ -25,7 +25,6 @@ RUN bash ./install_docs_reqs.sh && rm install_docs_reqs.sh
 
 # Install conda and other packages (e.g., numpy, pytest)
 ARG ANACONDA_PYTHON_VERSION
-ARG BUILD_ENVIRONMENT
 ENV ANACONDA_PYTHON_VERSION=$ANACONDA_PYTHON_VERSION
 ENV PATH /opt/conda/envs/py_$ANACONDA_PYTHON_VERSION/bin:/opt/conda/bin:$PATH
 COPY requirements-ci.txt /opt/conda/requirements-ci.txt

.ci/docker/ubuntu-xpu/Dockerfile

@@ -72,7 +72,7 @@ ARG TRITON
 COPY ./common/install_triton.sh install_triton.sh
 COPY ./common/common_utils.sh common_utils.sh
 COPY ci_commit_pins/triton-xpu.txt triton-xpu.txt
-COPY triton_xpu_version.txt triton_version.txt
+COPY triton_version.txt triton_version.txt
 RUN if [ -n "${TRITON}" ]; then bash ./install_triton.sh; fi
 RUN rm install_triton.sh common_utils.sh triton-xpu.txt triton_version.txt
 

.ci/magma/Makefile

@@ -1,7 +1,7 @@
 SHELL=/usr/bin/env bash
 
 DOCKER_CMD ?= docker
-DESIRED_CUDA ?= 12.8
+DESIRED_CUDA ?= 11.8
 DESIRED_CUDA_SHORT = $(subst .,,$(DESIRED_CUDA))
 PACKAGE_NAME = magma-cuda
 CUDA_ARCH_LIST ?= -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90
@@ -16,21 +16,15 @@ DOCKER_RUN = set -eou pipefail; ${DOCKER_CMD} run --rm -i \
 	magma/build_magma.sh
 
 .PHONY: all
-all: magma-cuda129
 all: magma-cuda128
 all: magma-cuda126
+all: magma-cuda118
 
 .PHONY:
 clean:
 	$(RM) -r magma-*
 	$(RM) -r output
 
-.PHONY: magma-cuda129
-magma-cuda129: DESIRED_CUDA := 12.9
-magma-cuda129: CUDA_ARCH_LIST += -gencode arch=compute_100,code=sm_100 -gencode arch=compute_120,code=sm_120
-magma-cuda129:
-	$(DOCKER_RUN)
-
 .PHONY: magma-cuda128
 magma-cuda128: DESIRED_CUDA := 12.8
 magma-cuda128: CUDA_ARCH_LIST += -gencode arch=compute_100,code=sm_100 -gencode arch=compute_120,code=sm_120
@@ -41,3 +35,9 @@ magma-cuda128:
 magma-cuda126: DESIRED_CUDA := 12.6
 magma-cuda126:
 	$(DOCKER_RUN)
+
+.PHONY: magma-cuda118
+magma-cuda118: DESIRED_CUDA := 11.8
+magma-cuda118: CUDA_ARCH_LIST += -gencode arch=compute_37,code=sm_37
+magma-cuda118:
+	$(DOCKER_RUN)

.ci/manywheel/build_common.sh

@@ -31,6 +31,7 @@ elif [[ "$OS_NAME" == *"Ubuntu"* ]]; then
     # Comment out nvidia repositories to prevent them from getting apt-get updated, see https://github.com/pytorch/pytorch/issues/74968
     # shellcheck disable=SC2046
     sed -i 's/.*nvidia.*/# &/' $(find /etc/apt/ -type f -name "*.list")
+
     retry apt-get update
     retry apt-get -y install zip openssl
 else
@@ -97,7 +98,6 @@ if [[ -z "$PYTORCH_ROOT" ]]; then
     exit 1
 fi
 pushd "$PYTORCH_ROOT"
-retry pip install -q cmake
 python setup.py clean
 retry pip install -qr requirements.txt
 case ${DESIRED_PYTHON} in
@@ -151,7 +151,7 @@ if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
     BUILD_LIBTORCH_WHL=0 BUILD_PYTHON_ONLY=1 \
     BUILD_LIBTORCH_CPU_WITH_DEBUG=$BUILD_DEBUG_INFO \
     USE_NCCL=${USE_NCCL} USE_RCCL=${USE_RCCL} USE_KINETO=${USE_KINETO} \
-    CMAKE_FRESH=1 python setup.py bdist_wheel -d /tmp/$WHEELHOUSE_DIR
+    python setup.py bdist_wheel -d /tmp/$WHEELHOUSE_DIR --cmake
     echo "Finished setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
 else
     time CMAKE_ARGS=${CMAKE_ARGS[@]} \

.ci/manywheel/build_cuda.sh

@@ -15,9 +15,6 @@ export INSTALL_TEST=0 # dont install test binaries into site-packages
 export USE_CUPTI_SO=0
 export USE_CUSPARSELT=${USE_CUSPARSELT:-1} # Enable if not disabled by libtorch build
 export USE_CUFILE=${USE_CUFILE:-1}
-export USE_SYSTEM_NCCL=1
-export NCCL_INCLUDE_DIR="/usr/local/cuda/include/"
-export NCCL_LIB_DIR="/usr/local/cuda/lib64/"
 
 # Keep an array of cmake variables to add to
 if [[ -z "$CMAKE_ARGS" ]]; then
@@ -54,18 +51,18 @@ cuda_version_nodot=$(echo $CUDA_VERSION | tr -d '.')
 
 TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"
 case ${CUDA_VERSION} in
-    12.8|12.9)
-        TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0;12.0+PTX" #removing sm_50-sm_70 as these architectures are deprecated in CUDA 12.8/9 and will be removed in future releases
+    12.8)
+        TORCH_CUDA_ARCH_LIST="7.5;8.0;8.6;9.0;10.0;12.0+PTX" #removing sm_50-sm_70 as these architectures are deprecated in CUDA 12.8 and will be removed in future releases
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
-        # WAR to resolve the ld error in libtorch build with CUDA 12.9
-        if [[ "$DESIRED_CUDA" == "cu129" && "$PACKAGE_TYPE" == "libtorch" ]]; then
-            TORCH_CUDA_ARCH_LIST="7.5;8.0;9.0;10.0;12.0+PTX"
-        fi
         ;;
     12.6)
         TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
+    11.8)
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};3.7;9.0"
+        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
+        ;;
     *)
         echo "unknown cuda version $CUDA_VERSION"
         exit 1
@@ -107,11 +104,12 @@ DEPS_SONAME=(
 )
 
 
-# CUDA_VERSION 12.6, 12.8, 12.9
+# CUDA_VERSION 12.6, 12.8
 if [[ $CUDA_VERSION == 12* ]]; then
     export USE_STATIC_CUDNN=0
     # Try parallelizing nvcc as well
     export TORCH_NVCC_FLAGS="-Xfatbin -compress-all --threads 2"
+
     if [[ -z "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" ]]; then
         echo "Bundling with cudnn and cublas."
         DEPS_LIST+=(
@@ -127,6 +125,7 @@ if [[ $CUDA_VERSION == 12* ]]; then
             "/usr/local/cuda/lib64/libcublasLt.so.12"
             "/usr/local/cuda/lib64/libcusparseLt.so.0"
             "/usr/local/cuda/lib64/libcudart.so.12"
+            "/usr/local/cuda/lib64/libnvToolsExt.so.1"
             "/usr/local/cuda/lib64/libnvrtc.so.12"
             "/usr/local/cuda/lib64/libnvrtc-builtins.so"
             "/usr/local/cuda/lib64/libcufile.so.0"
@@ -145,6 +144,7 @@ if [[ $CUDA_VERSION == 12* ]]; then
             "libcublasLt.so.12"
             "libcusparseLt.so.0"
             "libcudart.so.12"
+            "libnvToolsExt.so.1"
             "libnvrtc.so.12"
             "libnvrtc-builtins.so"
             "libcufile.so.0"
@@ -162,10 +162,8 @@ if [[ $CUDA_VERSION == 12* ]]; then
             '$ORIGIN/../../nvidia/curand/lib'
             '$ORIGIN/../../nvidia/cusolver/lib'
             '$ORIGIN/../../nvidia/cusparse/lib'
-            '$ORIGIN/../../nvidia/cusparselt/lib'
             '$ORIGIN/../../cusparselt/lib'
             '$ORIGIN/../../nvidia/nccl/lib'
-            '$ORIGIN/../../nvidia/nvshmem/lib'
             '$ORIGIN/../../nvidia/nvtx/lib'
             '$ORIGIN/../../nvidia/cufile/lib'
         )
@@ -174,9 +172,94 @@ if [[ $CUDA_VERSION == 12* ]]; then
         export LIB_SO_RPATH=$CUDA_RPATHS':$ORIGIN'
         export FORCE_RPATH="--force-rpath"
         export USE_STATIC_NCCL=0
+        export USE_SYSTEM_NCCL=1
         export ATEN_STATIC_CUDA=0
         export USE_CUDA_STATIC_LINK=0
         export USE_CUPTI_SO=1
+        export NCCL_INCLUDE_DIR="/usr/local/cuda/include/"
+        export NCCL_LIB_DIR="/usr/local/cuda/lib64/"
+    fi
+elif [[ $CUDA_VERSION == "11.8" ]]; then
+    export USE_STATIC_CUDNN=0
+    # Turn USE_CUFILE off for CUDA 11.8 since nvidia-cufile-cu11 and 1.9.0.20 are
+    # not available in PYPI
+    export USE_CUFILE=0
+    # Try parallelizing nvcc as well
+    export TORCH_NVCC_FLAGS="-Xfatbin -compress-all --threads 2"
+    # Bundle ptxas into the wheel, see https://github.com/pytorch/pytorch/pull/119750
+    export BUILD_BUNDLE_PTXAS=1
+
+    # CUDA 11.8 have to ship the libcusparseLt.so.0 with the binary
+    # since nvidia-cusparselt-cu11 is not available in PYPI
+    if [[ $USE_CUSPARSELT == "1" ]]; then
+        DEPS_SONAME+=(
+            "libcusparseLt.so.0"
+        )
+        DEPS_LIST+=(
+            "/usr/local/cuda/lib64/libcusparseLt.so.0"
+        )
+    fi
+
+    if [[ -z "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" ]]; then
+        echo "Bundling with cudnn and cublas."
+        DEPS_LIST+=(
+            "/usr/local/cuda/lib64/libcudnn_adv.so.9"
+            "/usr/local/cuda/lib64/libcudnn_cnn.so.9"
+            "/usr/local/cuda/lib64/libcudnn_graph.so.9"
+            "/usr/local/cuda/lib64/libcudnn_ops.so.9"
+            "/usr/local/cuda/lib64/libcudnn_engines_runtime_compiled.so.9"
+            "/usr/local/cuda/lib64/libcudnn_engines_precompiled.so.9"
+            "/usr/local/cuda/lib64/libcudnn_heuristic.so.9"
+            "/usr/local/cuda/lib64/libcudnn.so.9"
+            "/usr/local/cuda/lib64/libcublas.so.11"
+            "/usr/local/cuda/lib64/libcublasLt.so.11"
+            "/usr/local/cuda/lib64/libcudart.so.11.0"
+            "/usr/local/cuda/lib64/libnvToolsExt.so.1"
+            "/usr/local/cuda/lib64/libnvrtc.so.11.2"    # this is not a mistake, it links to more specific cuda version
+            "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.8"
+        )
+        DEPS_SONAME+=(
+            "libcudnn_adv.so.9"
+            "libcudnn_cnn.so.9"
+            "libcudnn_graph.so.9"
+            "libcudnn_ops.so.9"
+            "libcudnn_engines_runtime_compiled.so.9"
+            "libcudnn_engines_precompiled.so.9"
+            "libcudnn_heuristic.so.9"
+            "libcudnn.so.9"
+            "libcublas.so.11"
+            "libcublasLt.so.11"
+            "libcudart.so.11.0"
+            "libnvToolsExt.so.1"
+            "libnvrtc.so.11.2"
+            "libnvrtc-builtins.so.11.8"
+        )
+    else
+        echo "Using nvidia libs from pypi."
+        CUDA_RPATHS=(
+            '$ORIGIN/../../nvidia/cublas/lib'
+            '$ORIGIN/../../nvidia/cuda_cupti/lib'
+            '$ORIGIN/../../nvidia/cuda_nvrtc/lib'
+            '$ORIGIN/../../nvidia/cuda_runtime/lib'
+            '$ORIGIN/../../nvidia/cudnn/lib'
+            '$ORIGIN/../../nvidia/cufft/lib'
+            '$ORIGIN/../../nvidia/curand/lib'
+            '$ORIGIN/../../nvidia/cusolver/lib'
+            '$ORIGIN/../../nvidia/cusparse/lib'
+            '$ORIGIN/../../nvidia/nccl/lib'
+            '$ORIGIN/../../nvidia/nvtx/lib'
+        )
+        CUDA_RPATHS=$(IFS=: ; echo "${CUDA_RPATHS[*]}")
+        export C_SO_RPATH=$CUDA_RPATHS':$ORIGIN:$ORIGIN/lib'
+        export LIB_SO_RPATH=$CUDA_RPATHS':$ORIGIN'
+        export FORCE_RPATH="--force-rpath"
+        export USE_STATIC_NCCL=0
+        export USE_SYSTEM_NCCL=1
+        export ATEN_STATIC_CUDA=0
+        export USE_CUDA_STATIC_LINK=0
+        export USE_CUPTI_SO=1
+        export NCCL_INCLUDE_DIR="/usr/local/cuda/include/"
+        export NCCL_LIB_DIR="/usr/local/cuda/lib64/"
     fi
 else
     echo "Unknown cuda version $CUDA_VERSION"

.ci/manywheel/build_libtorch.sh

@@ -92,7 +92,6 @@ if [[ -z "$PYTORCH_ROOT" ]]; then
     exit 1
 fi
 pushd "$PYTORCH_ROOT"
-retry pip install -q cmake
 python setup.py clean
 retry pip install -qr requirements.txt
 retry pip install -q numpy==2.0.1

.ci/manywheel/build_rocm.sh

@@ -95,7 +95,6 @@ ROCM_SO_FILES=(
     "libroctracer64.so"
     "libroctx64.so"
     "libhipblaslt.so"
-    "libhipsparselt.so"
     "libhiprtc.so"
 )
 
@@ -187,28 +186,20 @@ do
     OS_SO_FILES[${#OS_SO_FILES[@]}]=$file_name # Append lib to array
 done
 
-ARCH=$(echo $PYTORCH_ROCM_ARCH | sed 's/;/|/g') # Replace ; separated arch list to bar for grep
-
 # rocBLAS library files
 ROCBLAS_LIB_SRC=$ROCM_HOME/lib/rocblas/library
 ROCBLAS_LIB_DST=lib/rocblas/library
-ROCBLAS_ARCH_SPECIFIC_FILES=$(ls $ROCBLAS_LIB_SRC | grep -E $ARCH)
-ROCBLAS_OTHER_FILES=$(ls $ROCBLAS_LIB_SRC | grep -v gfx)
-ROCBLAS_LIB_FILES=($ROCBLAS_ARCH_SPECIFIC_FILES $OTHER_FILES)
+ARCH=$(echo $PYTORCH_ROCM_ARCH | sed 's/;/|/g') # Replace ; seperated arch list to bar for grep
+ARCH_SPECIFIC_FILES=$(ls $ROCBLAS_LIB_SRC | grep -E $ARCH)
+OTHER_FILES=$(ls $ROCBLAS_LIB_SRC | grep -v gfx)
+ROCBLAS_LIB_FILES=($ARCH_SPECIFIC_FILES $OTHER_FILES)
 
 # hipblaslt library files
 HIPBLASLT_LIB_SRC=$ROCM_HOME/lib/hipblaslt/library
 HIPBLASLT_LIB_DST=lib/hipblaslt/library
-HIPBLASLT_ARCH_SPECIFIC_FILES=$(ls $HIPBLASLT_LIB_SRC | grep -E $ARCH)
-HIPBLASLT_OTHER_FILES=$(ls $HIPBLASLT_LIB_SRC | grep -v gfx)
-HIPBLASLT_LIB_FILES=($HIPBLASLT_ARCH_SPECIFIC_FILES $HIPBLASLT_OTHER_FILES)
-
-# hipsparselt library files
-HIPSPARSELT_LIB_SRC=$ROCM_HOME/lib/hipsparselt/library
-HIPSPARSELT_LIB_DST=lib/hipsparselt/library
-HIPSPARSELT_ARCH_SPECIFIC_FILES=$(ls $HIPSPARSELT_LIB_SRC | grep -E $ARCH)
-#HIPSPARSELT_OTHER_FILES=$(ls $HIPSPARSELT_LIB_SRC | grep -v gfx)
-HIPSPARSELT_LIB_FILES=($HIPSPARSELT_ARCH_SPECIFIC_FILES $HIPSPARSELT_OTHER_FILES)
+ARCH_SPECIFIC_FILES=$(ls $HIPBLASLT_LIB_SRC | grep -E $ARCH)
+OTHER_FILES=$(ls $HIPBLASLT_LIB_SRC | grep -v gfx)
+HIPBLASLT_LIB_FILES=($ARCH_SPECIFIC_FILES $OTHER_FILES)
 
 # ROCm library files
 ROCM_SO_PATHS=()
@@ -243,14 +234,12 @@ DEPS_SONAME=(
 DEPS_AUX_SRCLIST=(
     "${ROCBLAS_LIB_FILES[@]/#/$ROCBLAS_LIB_SRC/}"
     "${HIPBLASLT_LIB_FILES[@]/#/$HIPBLASLT_LIB_SRC/}"
-    "${HIPSPARSELT_LIB_FILES[@]/#/$HIPSPARSELT_LIB_SRC/}"
     "/opt/amdgpu/share/libdrm/amdgpu.ids"
 )
 
 DEPS_AUX_DSTLIST=(
     "${ROCBLAS_LIB_FILES[@]/#/$ROCBLAS_LIB_DST/}"
     "${HIPBLASLT_LIB_FILES[@]/#/$HIPBLASLT_LIB_DST/}"
-    "${HIPSPARSELT_LIB_FILES[@]/#/$HIPSPARSELT_LIB_DST/}"
     "share/libdrm/amdgpu.ids"
 )
 

.ci/pytorch/build.sh

@@ -27,12 +27,6 @@ cmake --version
 echo "Environment variables:"
 env
 
-# The sccache wrapped version of nvcc gets put in /opt/cache/lib in docker since
-# there are some issues if it is always wrapped, so we need to add it to PATH
-# during CI builds.
-# https://github.com/pytorch/pytorch/blob/0b6c0898e6c352c8ea93daec854e704b41485375/.ci/docker/common/install_cache.sh#L97
-export PATH="/opt/cache/lib:$PATH"
-
 if [[ "$BUILD_ENVIRONMENT" == *cuda* ]]; then
   # Use jemalloc during compilation to mitigate https://github.com/pytorch/pytorch/issues/116289
   export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so.2
@@ -58,6 +52,12 @@ fi
 export USE_LLVM=/opt/llvm
 export LLVM_DIR=/opt/llvm/lib/cmake/llvm
 
+if [[ "$BUILD_ENVIRONMENT" == *executorch* ]]; then
+  # To build test_edge_op_registration
+  export BUILD_EXECUTORCH=ON
+  export USE_CUDA=0
+fi
+
 if ! which conda; then
   # In ROCm CIs, we are doing cross compilation on build machines with
   # intel cpu and later run tests on machines with amd cpu.
@@ -257,7 +257,6 @@ if [[ "$BUILD_ENVIRONMENT" == *-bazel-* ]]; then
   set -e -o pipefail
 
   get_bazel
-  python3 tools/optional_submodules.py checkout_eigen
 
   # Leave 1 CPU free and use only up to 80% of memory to reduce the change of crashing
   # the runner

.ci/pytorch/check_binary.sh

@@ -313,7 +313,7 @@ if [[ "$(uname)" == 'Linux' &&  "$PACKAGE_TYPE" == 'manywheel' ]]; then
   # Please see issue for reference: https://github.com/pytorch/pytorch/issues/152426
   if [[ "$(uname -m)" == "s390x" ]]; then
     cxx_abi="19"
-  elif [[ "$DESIRED_CUDA" != 'xpu' && "$DESIRED_CUDA" != 'rocm'* ]]; then
+  elif [[ "$DESIRED_CUDA" != 'cu118' && "$DESIRED_CUDA" != 'xpu' && "$DESIRED_CUDA" != 'rocm'* ]]; then
     cxx_abi="18"
   else
     cxx_abi="16"

.ci/pytorch/common.sh

@@ -15,6 +15,6 @@ if [[ "${BUILD_ENVIRONMENT}" == *rocm* ]]; then
   export PYTORCH_TEST_WITH_ROCM=1
 fi
 
-# TODO: Reenable libtorch testing for MacOS, see https://github.com/pytorch/pytorch/issues/62598
+# TODO: Renable libtorch testing for MacOS, see https://github.com/pytorch/pytorch/issues/62598
 # shellcheck disable=SC2034
 BUILD_TEST_LIBTORCH=0

.ci/pytorch/common_utils.sh

@@ -159,6 +159,11 @@ function install_torchvision() {
   fi
 }
 
+function install_tlparse() {
+  pip_install --user "tlparse==0.3.30"
+  PATH="$(python -m site --user-base)/bin:$PATH"
+}
+
 function install_torchrec_and_fbgemm() {
   local torchrec_commit
   torchrec_commit=$(get_pinned_commit torchrec)
@@ -197,7 +202,7 @@ function install_torchrec_and_fbgemm() {
 
 function clone_pytorch_xla() {
   if [[ ! -d ./xla ]]; then
-    git clone --recursive -b r2.8 https://github.com/pytorch/xla.git
+    git clone --recursive --quiet https://github.com/pytorch/xla.git
     pushd xla
     # pin the xla hash so that we don't get broken by changes to xla
     git checkout "$(cat ../.github/ci_commit_pins/xla.txt)"

.ci/pytorch/macos-test.sh

@@ -5,6 +5,11 @@ set -x
 # shellcheck source=./macos-common.sh
 source "$(dirname "${BASH_SOURCE[0]}")/macos-common.sh"
 
+if [[ -n "$CONDA_ENV" ]]; then
+  # Use binaries under conda environment
+  export PATH="$CONDA_ENV/bin":$PATH
+fi
+
 # Test that OpenMP is enabled
 pushd test
 if [[ ! $(python -c "import torch; print(int(torch.backends.openmp.is_available()))") == "1" ]]; then
@@ -228,52 +233,53 @@ test_torchbench_smoketest() {
   mkdir -p "$TEST_REPORTS_DIR"
 
   local device=mps
-  local dtypes=(undefined float16 bfloat16 notset)
-  local dtype=${dtypes[$1]}
-  local models=(hf_T5 llama BERT_pytorch dcgan hf_GPT2 yolov3 resnet152 sam sam_fast pytorch_unet stable_diffusion_text_encoder speech_transformer Super_SloMo doctr_det_predictor doctr_reco_predictor timm_resnet timm_vovnet vgg16)
+  local models=(hf_T5 llama BERT_pytorch dcgan hf_GPT2 yolov3 resnet152 sam pytorch_unet stable_diffusion_text_encoder speech_transformer Super_SloMo doctr_det_predictor doctr_reco_predictor)
+  local hf_models=(GoogleFnet YituTechConvBert Speech2Text2ForCausalLM)
 
   for backend in eager inductor; do
 
-    echo "Launching torchbench inference performance run for backend ${backend} and dtype ${dtype}"
-    local dtype_arg="--${dtype}"
-    if [ "$dtype" == notset ]; then
-        dtype_arg="--float32"
-    fi
-    touch "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv"
-    for model in "${models[@]}"; do
-      PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
-        --performance --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
-        --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv" || true
-      if [ "$backend" == "inductor" ]; then
-        PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
-          --accuracy --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
-          --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_accuracy.csv" || true
-      fi
-    done
-    if [ "$backend" == "inductor" ]; then
-      PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/huggingface.py \
-        --performance --backend "$backend" --inference --devices "$device" "$dtype_arg" \
-        --output "$TEST_REPORTS_DIR/inductor_${backend}_huggingface_${dtype}_inference_${device}_performance.csv" || true
-      PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/huggingface.py \
-        --accuracy --backend "$backend" --inference --devices "$device" "$dtype_arg" \
-        --output "$TEST_REPORTS_DIR/inductor_${backend}_huggingface_${dtype}_inference_${device}_accuracy.csv" || true
-    fi
-
-    if [ "$dtype" == notset ]; then
-      for dtype_ in notset amp; do
-        echo "Launching torchbench training performance run for backend ${backend} and dtype ${dtype_}"
-        touch "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype_}_training_${device}_performance.csv"
-        local dtype_arg="--${dtype_}"
-        if [ "$dtype_" == notset ]; then
+    for dtype in notset float16 bfloat16; do
+      echo "Launching torchbench inference performance run for backend ${backend} and dtype ${dtype}"
+      local dtype_arg="--${dtype}"
+      if [ "$dtype" == notset ]; then
           dtype_arg="--float32"
-        fi
-        for model in "${models[@]}"; do
+      fi
+      touch "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv"
+      for model in "${models[@]}"; do
+        PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
+          --performance --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+          --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_performance.csv" || true
+        if [ "$backend" == "inductor" ]; then
           PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
-            --performance --only "$model" --backend "$backend" --training --devices "$device" "$dtype_arg" \
-            --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype_}_training_${device}_performance.csv" || true
-        done
+            --accuracy --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+            --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_inference_${device}_accuracy.csv" || true
+        fi
       done
-    fi
+      for model in "${hf_models[@]}"; do
+        if [ "$backend" == "inductor" ]; then
+          PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/huggingface.py \
+            --performance --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+            --output "$TEST_REPORTS_DIR/inductor_${backend}_huggingface_${dtype}_inference_${device}_performance.csv" || true
+          PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/huggingface.py \
+            --accuracy --only "$model" --backend "$backend" --inference --devices "$device" "$dtype_arg" \
+            --output "$TEST_REPORTS_DIR/inductor_${backend}_huggingface_${dtype}_inference_${device}_accuracy.csv" || true
+        fi
+      done
+    done
+
+    for dtype in notset amp; do
+      echo "Launching torchbench training performance run for backend ${backend} and dtype ${dtype}"
+      touch "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_training_${device}_performance.csv"
+      local dtype_arg="--${dtype}"
+      if [ "$dtype" == notset ]; then
+          dtype_arg="--float32"
+      fi
+      for model in "${models[@]}"; do
+        PYTHONPATH="$(pwd)"/torchbench python benchmarks/dynamo/torchbench.py \
+          --performance --only "$model" --backend "$backend" --training --devices "$device" "$dtype_arg" \
+          --output "$TEST_REPORTS_DIR/inductor_${backend}_torchbench_${dtype}_training_${device}_performance.csv" || true
+      done
+    done
 
   done
 
@@ -312,6 +318,8 @@ test_timm_perf() {
   echo "timm benchmark on mps device completed"
 }
 
+install_tlparse
+
 if [[ $TEST_CONFIG == *"perf_all"* ]]; then
   test_torchbench_perf
   test_hf_perf
@@ -323,7 +331,7 @@ elif [[ $TEST_CONFIG == *"perf_hf"* ]]; then
 elif [[ $TEST_CONFIG == *"perf_timm"* ]]; then
   test_timm_perf
 elif [[ $TEST_CONFIG == *"perf_smoketest"* ]]; then
-  test_torchbench_smoketest "${SHARD_NUMBER}"
+  test_torchbench_smoketest
 elif [[ $TEST_CONFIG == *"mps"* ]]; then
   test_python_mps
 elif [[ $NUM_TEST_SHARDS -gt 1 ]]; then

.ci/pytorch/smoke_test/check_binary_symbols.py

@@ -93,7 +93,7 @@ def check_lib_symbols_for_abi_correctness(lib: str) -> None:
             f"Found pre-cxx11 symbols, but there shouldn't be any, see: {pre_cxx11_symbols[:100]}"
         )
     if num_cxx11_symbols < 100:
-        raise RuntimeError("Didn't find enough cxx11 symbols")
+        raise RuntimeError("Didn't find enought cxx11 symbols")
 
 
 def main() -> None:

.ci/pytorch/smoke_test/check_gomp.py

@@ -46,9 +46,6 @@ def get_gomp_thread():
 
     # use the default gomp path of AlmaLinux OS
     libgomp_path = "/usr/lib64/libgomp.so.1"
-    # if it does not exist, try Ubuntu path
-    if not os.path.exists(libgomp_path):
-        libgomp_path = f"/usr/lib/{os.uname().machine}-linux-gnu/libgomp.so.1"
 
     os.environ["GOMP_CPU_AFFINITY"] = "0-3"
 

.ci/pytorch/smoke_test/smoke_test.py

@@ -276,7 +276,7 @@ def smoke_test_cuda(
             torch_nccl_version = ".".join(str(v) for v in torch.cuda.nccl.version())
             print(f"Torch nccl; version: {torch_nccl_version}")
 
-        # Pypi dependencies are installed on linux only and nccl is available only on Linux.
+        # Pypi dependencies are installed on linux ony and nccl is availbale only on Linux.
         if pypi_pkg_check == "enabled" and sys.platform in ["linux", "linux2"]:
             compare_pypi_to_torch_versions(
                 "cudnn", find_pypi_package_version("nvidia-cudnn"), torch_cudnn_version

.ci/pytorch/test.sh

@@ -196,7 +196,7 @@ if [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
   # shellcheck disable=SC1091
   source /opt/intel/oneapi/mpi/latest/env/vars.sh
   # Check XPU status before testing
-  timeout 30 xpu-smi discovery || true
+  xpu-smi discovery
 fi
 
 if [[ "$BUILD_ENVIRONMENT" != *-bazel-* ]] ; then
@@ -212,6 +212,8 @@ if [[ "$BUILD_ENVIRONMENT" == *aarch64* ]]; then
   export VALGRIND=OFF
 fi
 
+install_tlparse
+
 # DANGER WILL ROBINSON.  The LD_PRELOAD here could cause you problems
 # if you're not careful.  Check this if you made some changes and the
 # ASAN test is not working
@@ -224,7 +226,7 @@ if [[ "$BUILD_ENVIRONMENT" == *asan* ]]; then
     export PYTORCH_TEST_WITH_ASAN=1
     export PYTORCH_TEST_WITH_UBSAN=1
     # TODO: Figure out how to avoid hard-coding these paths
-    export ASAN_SYMBOLIZER_PATH=/usr/lib/llvm-18/bin/llvm-symbolizer
+    export ASAN_SYMBOLIZER_PATH=/usr/lib/llvm-15/bin/llvm-symbolizer
     export TORCH_USE_RTLD_GLOBAL=1
     # NB: We load libtorch.so with RTLD_GLOBAL for UBSAN, unlike our
     # default behavior.
@@ -322,17 +324,6 @@ test_python_smoke() {
   assert_git_not_dirty
 }
 
-test_h100_distributed() {
-  # Distributed tests at H100
-  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
-  # 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
-  assert_git_not_dirty
-}
-
 test_lazy_tensor_meta_reference_disabled() {
   export TORCH_DISABLE_FUNCTIONALIZATION_META_REFERENCE=1
   echo "Testing lazy tensor operations without meta reference"
@@ -599,9 +590,7 @@ test_perf_for_dashboard() {
 
   local device=cuda
   if [[ "${TEST_CONFIG}" == *cpu* ]]; then
-    if [[ "${TEST_CONFIG}" == *zen_cpu_x86* ]]; then
-      device=zen_cpu_x86
-    elif [[ "${TEST_CONFIG}" == *cpu_x86* ]]; then
+    if [[ "${TEST_CONFIG}" == *cpu_x86* ]]; then
       device=cpu_x86
     elif [[ "${TEST_CONFIG}" == *cpu_aarch64* ]]; then
       device=cpu_aarch64
@@ -1142,12 +1131,6 @@ test_custom_backend() {
 
 test_custom_script_ops() {
   echo "Testing custom script operators"
-
-  if [[ "$BUILD_ENVIRONMENT" == *s390x* ]]; then
-    echo "Skipping custom script operators until it's fixed"
-    return 0
-  fi
-
   CUSTOM_OP_BUILD="${CUSTOM_TEST_ARTIFACT_BUILD_DIR}/custom-op-build"
   pushd test/custom_operator
   cp -a "$CUSTOM_OP_BUILD" build
@@ -1537,7 +1520,7 @@ test_executorch() {
 test_linux_aarch64() {
   python test/run_test.py --include test_modules test_mkldnn test_mkldnn_fusion test_openmp test_torch test_dynamic_shapes \
         test_transformers test_multiprocessing test_numpy_interop test_autograd test_binary_ufuncs test_complex test_spectral_ops \
-        test_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops test_ops test_cpp_extensions_open_device_registration \
+        test_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops test_ops \
         --shard "$SHARD_NUMBER" "$NUM_TEST_SHARDS" --verbose
 
   # Dynamo tests
@@ -1571,8 +1554,7 @@ test_operator_benchmark() {
 
   cd "${TEST_DIR}"/benchmarks/operator_benchmark
   $TASKSET python -m benchmark_all_test --device "$1" --tag-filter "$2" \
-      --output-csv "${TEST_REPORTS_DIR}/operator_benchmark_eager_float32_cpu.csv" \
-      --output-json-for-dashboard "${TEST_REPORTS_DIR}/operator_benchmark_eager_float32_cpu.json" \
+      --output-dir "${TEST_REPORTS_DIR}/operator_benchmark_eager_float32_cpu.csv"
 
   pip_install pandas
   python check_perf_csv.py \
@@ -1657,7 +1639,7 @@ elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
     install_torchaudio cuda
   fi
   install_torchvision
-  TORCH_CUDA_ARCH_LIST="8.0;8.6" install_torchao
+  TORCH_CUDA_ARCH_LIST="8.0;8.6" pip_install git+https://github.com/pytorch/ao.git
   id=$((SHARD_NUMBER-1))
   # https://github.com/opencv/opencv-python/issues/885
   pip_install opencv-python==4.8.0.74
@@ -1742,8 +1724,6 @@ elif [[ "${BUILD_ENVIRONMENT}" == *xpu* ]]; then
   test_xpu_bin
 elif [[ "${TEST_CONFIG}" == smoke ]]; then
   test_python_smoke
-elif [[ "${TEST_CONFIG}" == h100_distributed ]]; then
-  test_h100_distributed
 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")
+  file(GLOB TORCH_DLLS  "$ENV{CUDA_PATH}/bin/cudnn64_8.dll" "$ENV{NVTOOLSEXT_PATH}/bin/x64/*.dll")
   message("dlls to copy "  ${TORCH_DLLS})
   add_custom_command(TARGET simple-torch-test
                      POST_BUILD

.ci/pytorch/win-build.sh

@@ -31,7 +31,7 @@ PYLONG_API_CHECK=$?
 if [[ $PYLONG_API_CHECK == 0 ]]; then
   echo "Usage of PyLong_{From,As}{Unsigned}Long API may lead to overflow errors on Windows"
   echo "because \`sizeof(long) == 4\` and \`sizeof(unsigned long) == 4\`."
-  echo "Please include \"torch/csrc/utils/python_numbers.h\" and use the corresponding APIs instead."
+  echo "Please include \"torch/csrc/utils/python_numbers.h\" and use the correspoding APIs instead."
   echo "PyLong_FromLong -> THPUtils_packInt32 / THPUtils_packInt64"
   echo "PyLong_AsLong -> THPUtils_unpackInt (32-bit) / THPUtils_unpackLong (64-bit)"
   echo "PyLong_FromUnsignedLong -> THPUtils_packUInt32 / THPUtils_packUInt64"

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

@@ -10,7 +10,7 @@ set PATH=C:\Program Files\CMake\bin;C:\Program Files\7-Zip;C:\ProgramData\chocol
 :: 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
+:: log sizes are too long, but leaving this here incase someone wants to use it locally
 :: set CMAKE_VERBOSE_MAKEFILE=1
 
 

.ci/pytorch/win-test-helpers/run_python_nn_smoketests.py

@@ -52,7 +52,7 @@ if __name__ == "__main__":
             if os.path.exists(debugger):
                 command_args = [debugger, "-o", "-c", "~*g; q"] + command_args
                 command_string = " ".join(command_args)
-                print("Rerunning with traceback enabled")
+                print("Reruning with traceback enabled")
                 print("Command:", command_string)
                 subprocess.run(command_args, check=False)
             sys.exit(e.returncode)

.ci/pytorch/windows/cuda118.bat

@@ -0,0 +1,59 @@
+@echo off
+
+set MODULE_NAME=pytorch
+
+IF NOT EXIST "setup.py" IF NOT EXIST "%MODULE_NAME%" (
+    call internal\clone.bat
+    cd %~dp0
+) ELSE (
+    call internal\clean.bat
+)
+IF ERRORLEVEL 1 goto :eof
+
+call internal\check_deps.bat
+IF ERRORLEVEL 1 goto :eof
+
+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_V118%"=="" (
+    IF EXIST "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin\nvcc.exe" (
+        set "CUDA_PATH_V118=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8"
+    ) ELSE (
+        echo CUDA 11.8 not found, failing
+        exit /b 1
+    )
+)
+
+IF "%BUILD_VISION%" == "" (
+    set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0
+    set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
+) ELSE (
+    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 -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=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90
+)
+
+set "CUDA_PATH=%CUDA_PATH_V118%"
+set "PATH=%CUDA_PATH_V118%\bin;%PATH%"
+
+:optcheck
+
+call internal\check_opts.bat
+IF ERRORLEVEL 1 goto :eof
+
+if exist "%NIGHTLIES_PYTORCH_ROOT%" cd %NIGHTLIES_PYTORCH_ROOT%\..
+call  %~dp0\internal\copy.bat
+IF ERRORLEVEL 1 goto :eof
+
+call  %~dp0\internal\setup.bat
+IF ERRORLEVEL 1 goto :eof

.ci/pytorch/windows/cuda124.bat

@@ -0,0 +1,59 @@
+@echo off
+
+set MODULE_NAME=pytorch
+
+IF NOT EXIST "setup.py" IF NOT EXIST "%MODULE_NAME%" (
+    call internal\clone.bat
+    cd %~dp0
+) ELSE (
+    call internal\clean.bat
+)
+IF ERRORLEVEL 1 goto :eof
+
+call internal\check_deps.bat
+IF ERRORLEVEL 1 goto :eof
+
+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_V124%"=="" (
+    IF EXIST "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\bin\nvcc.exe" (
+        set "CUDA_PATH_V124=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4"
+    ) ELSE (
+        echo CUDA 12.4 not found, failing
+        exit /b 1
+    )
+)
+
+IF "%BUILD_VISION%" == "" (
+    set TORCH_CUDA_ARCH_LIST=6.1;7.0;7.5;8.0;8.6;9.0
+    set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
+) ELSE (
+    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -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=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90
+)
+
+set "CUDA_PATH=%CUDA_PATH_V124%"
+set "PATH=%CUDA_PATH_V124%\bin;%PATH%"
+
+:optcheck
+
+call internal\check_opts.bat
+IF ERRORLEVEL 1 goto :eof
+
+if exist "%NIGHTLIES_PYTORCH_ROOT%" cd %NIGHTLIES_PYTORCH_ROOT%\..
+call  %~dp0\internal\copy.bat
+IF ERRORLEVEL 1 goto :eof
+
+call  %~dp0\internal\setup.bat
+IF ERRORLEVEL 1 goto :eof

.ci/pytorch/windows/cuda126.bat

@@ -18,6 +18,15 @@ 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

@@ -18,6 +18,15 @@ 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

@@ -1,50 +0,0 @@
-@echo off
-
-set MODULE_NAME=pytorch
-
-IF NOT EXIST "setup.py" IF NOT EXIST "%MODULE_NAME%" (
-    call internal\clone.bat
-    cd %~dp0
-) ELSE (
-    call internal\clean.bat
-)
-IF ERRORLEVEL 1 goto :eof
-
-call internal\check_deps.bat
-IF ERRORLEVEL 1 goto :eof
-
-REM Check for optional components
-
-set USE_CUDA=
-set CMAKE_GENERATOR=Visual Studio 15 2017 Win64
-
-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"
-    ) ELSE (
-        echo CUDA 12.9 not found, failing
-        exit /b 1
-    )
-)
-
-IF "%BUILD_VISION%" == "" (
-    set TORCH_CUDA_ARCH_LIST=7.5;8.0;8.6;9.0;10.0;12.0
-    set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
-) ELSE (
-    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90 -gencode=arch=compute_100,code=compute_100 -gencode=arch=compute_120,code=compute_120
-)
-
-set "CUDA_PATH=%CUDA_PATH_V129%"
-set "PATH=%CUDA_PATH_V129%\bin;%PATH%"
-
-:optcheck
-
-call internal\check_opts.bat
-IF ERRORLEVEL 1 goto :eof
-
-if exist "%NIGHTLIES_PYTORCH_ROOT%" cd %NIGHTLIES_PYTORCH_ROOT%\..
-call  %~dp0\internal\copy.bat
-IF ERRORLEVEL 1 goto :eof
-
-call  %~dp0\internal\setup.bat
-IF ERRORLEVEL 1 goto :eof

.ci/pytorch/windows/internal/check_deps.bat

@@ -65,7 +65,7 @@ for /F "usebackq delims=" %%i in (`python -c "import sys; print('{0[0]}{0[1]}'.f
 if  %PYVER% LSS 35 (
     echo Warning: PyTorch for Python 2 under Windows is experimental.
     echo Python x64 3.5 or up is recommended to compile PyTorch on Windows
-    echo Maybe you can create a virtual environment if you have conda installed:
+    echo Maybe you can create a virual environment if you have conda installed:
     echo ^> conda create -n test python=3.6 pyyaml numpy
     echo ^> activate test
 )

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

@@ -9,6 +9,7 @@ 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 "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

@@ -23,13 +23,66 @@ set CUDNN_LIB_FOLDER="lib\x64"
 :: Skip all of this if we already have cuda installed
 if exist "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%CUDA_VERSION_STR%\bin\nvcc.exe" goto set_cuda_env_vars
 
+if %CUDA_VER% EQU 118 goto cuda118
+if %CUDA_VER% EQU 124 goto cuda124
 if %CUDA_VER% EQU 126 goto cuda126
 if %CUDA_VER% EQU 128 goto cuda128
-if %CUDA_VER% EQU 129 goto cuda129
 
 echo CUDA %CUDA_VERSION_STR% is not supported
 exit /b 1
 
+:cuda118
+
+set CUDA_INSTALL_EXE=cuda_11.8.0_522.06_windows.exe
+if not exist "%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%" (
+    curl -k -L "https://ossci-windows.s3.amazonaws.com/%CUDA_INSTALL_EXE%" --output "%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%" & REM @lint-ignore
+    if errorlevel 1 exit /b 1
+    set "CUDA_SETUP_FILE=%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%"
+    set "ARGS=cuda_profiler_api_11.8 thrust_11.8 nvcc_11.8 cuobjdump_11.8 nvprune_11.8 nvprof_11.8 cupti_11.8 cublas_11.8 cublas_dev_11.8 cudart_11.8 cufft_11.8 cufft_dev_11.8 curand_11.8 curand_dev_11.8 cusolver_11.8 cusolver_dev_11.8 cusparse_11.8 cusparse_dev_11.8 npp_11.8 npp_dev_11.8 nvrtc_11.8 nvrtc_dev_11.8 nvml_dev_11.8 nvtx_11.8"
+)
+
+set CUDNN_FOLDER=cudnn-windows-x86_64-9.5.0.50_cuda11-archive
+set CUDNN_LIB_FOLDER="lib"
+set "CUDNN_INSTALL_ZIP=%CUDNN_FOLDER%.zip"
+if not exist "%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%" (
+    curl -k -L "http://s3.amazonaws.com/ossci-windows/%CUDNN_INSTALL_ZIP%" --output "%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%" & REM @lint-ignore
+    if errorlevel 1 exit /b 1
+    set "CUDNN_SETUP_FILE=%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%"
+)
+
+@REM cuDNN 8.3+ required zlib to be installed on the path
+echo Installing ZLIB dlls
+curl -k -L "http://s3.amazonaws.com/ossci-windows/zlib123dllx64.zip" --output "%SRC_DIR%\temp_build\zlib123dllx64.zip"
+7z x "%SRC_DIR%\temp_build\zlib123dllx64.zip" -o"%SRC_DIR%\temp_build\zlib"
+xcopy /Y "%SRC_DIR%\temp_build\zlib\dll_x64\*.dll" "C:\Windows\System32"
+
+goto cuda_common
+
+:cuda124
+
+set CUDA_INSTALL_EXE=cuda_12.4.0_551.61_windows.exe
+if not exist "%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%" (
+    curl -k -L "https://ossci-windows.s3.amazonaws.com/%CUDA_INSTALL_EXE%" --output "%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%" & REM @lint-ignore
+    if errorlevel 1 exit /b 1
+    set "CUDA_SETUP_FILE=%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%"
+    set "ARGS=cuda_profiler_api_12.4 thrust_12.4 nvcc_12.4 cuobjdump_12.4 nvprune_12.4 nvprof_12.4 cupti_12.4 cublas_12.4 cublas_dev_12.4 cudart_12.4 cufft_12.4 cufft_dev_12.4 curand_12.4 curand_dev_12.4 cusolver_12.4 cusolver_dev_12.4 cusparse_12.4 cusparse_dev_12.4 npp_12.4 npp_dev_12.4 nvrtc_12.4 nvrtc_dev_12.4 nvml_dev_12.4 nvjitlink_12.4 nvtx_12.4"
+)
+
+set CUDNN_FOLDER=cudnn-windows-x86_64-9.5.0.50_cuda12-archive
+set CUDNN_LIB_FOLDER="lib"
+set "CUDNN_INSTALL_ZIP=%CUDNN_FOLDER%.zip"
+if not exist "%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%" (
+    curl -k -L "http://s3.amazonaws.com/ossci-windows/%CUDNN_INSTALL_ZIP%" --output "%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%" & REM @lint-ignore
+    if errorlevel 1 exit /b 1
+    set "CUDNN_SETUP_FILE=%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%"
+)
+
+@REM cuDNN 8.3+ required zlib to be installed on the path
+echo Installing ZLIB dlls
+curl -k -L "http://s3.amazonaws.com/ossci-windows/zlib123dllx64.zip" --output "%SRC_DIR%\temp_build\zlib123dllx64.zip"
+7z x "%SRC_DIR%\temp_build\zlib123dllx64.zip" -o"%SRC_DIR%\temp_build\zlib"
+xcopy /Y "%SRC_DIR%\temp_build\zlib\dll_x64\*.dll" "C:\Windows\System32"
+
 goto cuda_common
 
 :cuda126
@@ -86,39 +139,17 @@ xcopy /Y "%SRC_DIR%\temp_build\zlib\dll_x64\*.dll" "C:\Windows\System32"
 
 goto cuda_common
 
-:cuda129
-
-set CUDA_INSTALL_EXE=cuda_12.9.1_576.57_windows.exe
-if not exist "%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%" (
-    curl -k -L "https://ossci-windows.s3.amazonaws.com/%CUDA_INSTALL_EXE%" --output "%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%" & REM @lint-ignore
-    if errorlevel 1 exit /b 1
-    set "CUDA_SETUP_FILE=%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%"
-    set "ARGS=cuda_profiler_api_12.9 thrust_12.9 nvcc_12.9 cuobjdump_12.9 nvprune_12.9 nvprof_12.9 cupti_12.9 cublas_12.9 cublas_dev_12.9 cudart_12.9 cufft_12.9 cufft_dev_12.9 curand_12.9 curand_dev_12.9 cusolver_12.9 cusolver_dev_12.9 cusparse_12.9 cusparse_dev_12.9 npp_12.9 npp_dev_12.9 nvrtc_12.9 nvrtc_dev_12.9 nvml_dev_12.9 nvjitlink_12.9 nvtx_12.9"
-)
-
-set CUDNN_FOLDER=cudnn-windows-x86_64-9.10.2.21_cuda12-archive
-set CUDNN_LIB_FOLDER="lib"
-set "CUDNN_INSTALL_ZIP=%CUDNN_FOLDER%.zip"
-if not exist "%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%" (
-    curl -k -L "http://s3.amazonaws.com/ossci-windows/%CUDNN_INSTALL_ZIP%" --output "%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%" & REM @lint-ignore
-    if errorlevel 1 exit /b 1
-    set "CUDNN_SETUP_FILE=%SRC_DIR%\temp_build\%CUDNN_INSTALL_ZIP%"
-)
-
-@REM cuDNN 8.3+ required zlib to be installed on the path
-echo Installing ZLIB dlls
-curl -k -L "http://s3.amazonaws.com/ossci-windows/zlib123dllx64.zip" --output "%SRC_DIR%\temp_build\zlib123dllx64.zip"
-7z x "%SRC_DIR%\temp_build\zlib123dllx64.zip" -o"%SRC_DIR%\temp_build\zlib"
-xcopy /Y "%SRC_DIR%\temp_build\zlib\dll_x64\*.dll" "C:\Windows\System32"
-
-goto cuda_common
-
 :cuda_common
 :: NOTE: We only install CUDA if we don't have it installed already.
 :: With GHA runners these should be pre-installed as part of our AMI process
 :: 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
@@ -145,6 +176,15 @@ 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"
@@ -175,3 +215,4 @@ 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/smoke_test.bat

@@ -99,6 +99,7 @@ goto end
 :libtorch
 echo "install and test libtorch"
 
+if "%VC_YEAR%" == "2019" powershell internal\vs2019_install.ps1
 if "%VC_YEAR%" == "2022" powershell internal\vs2022_install.ps1
 
 if ERRORLEVEL 1 exit /b 1
@@ -110,6 +111,10 @@ pushd tmp\libtorch
 
 set VC_VERSION_LOWER=17
 set VC_VERSION_UPPER=18
+IF "%VC_YEAR%" == "2019" (
+    set VC_VERSION_LOWER=16
+    set VC_VERSION_UPPER=17
+)
 
 for /f "usebackq tokens=*" %%i in (`"%ProgramFiles(x86)%\Microsoft Visual Studio\Installer\vswhere.exe" -legacy -products * -version [%VC_VERSION_LOWER%^,%VC_VERSION_UPPER%^) -property installationPath`) do (
     if exist "%%i" if exist "%%i\VC\Auxiliary\Build\vcvarsall.bat" (

.ci/pytorch/windows/internal/vc_install_helper.bat

@@ -1,7 +1,14 @@
+if "%VC_YEAR%" == "2019" powershell windows/internal/vs2019_install.ps1
 if "%VC_YEAR%" == "2022" powershell windows/internal/vs2022_install.ps1
 
 set VC_VERSION_LOWER=17
 set VC_VERSION_UPPER=18
+:: Please don't delete VS2019 as an alternative, in case some Windows compiler issue.
+:: Reference: https://github.com/pytorch/pytorch/issues/145702#issuecomment-2858693930
+if "%VC_YEAR%" == "2019" (
+    set VC_VERSION_LOWER=16
+    set VC_VERSION_UPPER=17
+)
 
 for /f "usebackq tokens=*" %%i in (`"%ProgramFiles(x86)%\Microsoft Visual Studio\Installer\vswhere.exe"  -products Microsoft.VisualStudio.Product.BuildTools -version [%VC_VERSION_LOWER%^,%VC_VERSION_UPPER%^) -property installationPath`) do (
     if exist "%%i" if exist "%%i\VC\Auxiliary\Build\vcvarsall.bat" (

.ci/pytorch/windows/internal/vs2019_install.ps1

@@ -0,0 +1,48 @@
+# https://developercommunity.visualstudio.com/t/install-specific-version-of-vs-component/1142479
+# https://docs.microsoft.com/en-us/visualstudio/releases/2019/history#release-dates-and-build-numbers
+
+# 16.8.6 BuildTools
+$VS_DOWNLOAD_LINK = "https://ossci-windows.s3.us-east-1.amazonaws.com/vs16.8.6_BuildTools.exe"
+$COLLECT_DOWNLOAD_LINK = "https://aka.ms/vscollect.exe"
+$VS_INSTALL_ARGS = @("--nocache","--quiet","--wait", "--add Microsoft.VisualStudio.Workload.VCTools",
+                                                     "--add Microsoft.Component.MSBuild",
+                                                     "--add Microsoft.VisualStudio.Component.Roslyn.Compiler",
+                                                     "--add Microsoft.VisualStudio.Component.TextTemplating",
+                                                     "--add Microsoft.VisualStudio.Component.VC.CoreIde",
+                                                     "--add Microsoft.VisualStudio.Component.VC.Redist.14.Latest",
+                                                     "--add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.Core",
+                                                     "--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64",
+                                                     "--add Microsoft.VisualStudio.ComponentGroup.NativeDesktop.Win81")
+
+curl.exe --retry 3 -kL $VS_DOWNLOAD_LINK --output vs_installer.exe
+if ($LASTEXITCODE -ne 0) {
+    echo "Download of the VS 2019 Version 16.8.5 installer failed"
+    exit 1
+}
+
+if (Test-Path "${env:ProgramFiles(x86)}\Microsoft Visual Studio\Installer\vswhere.exe") {
+    $existingPath = & "${env:ProgramFiles(x86)}\Microsoft Visual Studio\Installer\vswhere.exe" -products "Microsoft.VisualStudio.Product.BuildTools" -version "[16, 17)" -property installationPath
+    if ($existingPath -ne $null) {
+        if (!${env:CIRCLECI}) {
+            echo "Found correctly versioned existing BuildTools installation in $existingPath"
+            exit 0
+        }
+        echo "Found existing BuildTools installation in $existingPath, keeping it"
+    }
+}
+
+$process = Start-Process "${PWD}\vs_installer.exe" -ArgumentList $VS_INSTALL_ARGS -NoNewWindow -Wait -PassThru
+Remove-Item -Path vs_installer.exe -Force
+$exitCode = $process.ExitCode
+if (($exitCode -ne 0) -and ($exitCode -ne 3010)) {
+    echo "VS 2019 installer exited with code $exitCode, which should be one of [0, 3010]."
+    curl.exe --retry 3 -kL $COLLECT_DOWNLOAD_LINK --output Collect.exe
+    if ($LASTEXITCODE -ne 0) {
+        echo "Download of the VS Collect tool failed."
+        exit 1
+    }
+    Start-Process "${PWD}\Collect.exe" -NoNewWindow -Wait -PassThru
+    New-Item -Path "C:\w\build-results" -ItemType "directory" -Force
+    Copy-Item -Path "C:\Users\${env:USERNAME}\AppData\Local\Temp\vslogs.zip" -Destination "C:\w\build-results\"
+    exit 1
+}

.ci/pytorch/windows/internal/xpu_install.bat

@@ -25,8 +25,8 @@ set XPU_EXTRA_INSTALLED=0
 set XPU_EXTRA_UNINSTALL=0
 
 if not [%XPU_VERSION%]==[] if [%XPU_VERSION%]==[2025.1] (
-    set XPU_BUNDLE_URL=https://registrationcenter-download.intel.com/akdlm/IRC_NAS/75d4eb97-914a-4a95-852c-7b9733d80f74/intel-deep-learning-essentials-2025.1.3.8_offline.exe
-    set XPU_BUNDLE_VERSION=2025.1.3+5
+    set XPU_BUNDLE_URL=https://registrationcenter-download.intel.com/akdlm/IRC_NAS/1a9fff3d-04c2-4d77-8861-3d86c774b66f/intel-deep-learning-essentials-2025.1.1.26_offline.exe
+    set XPU_BUNDLE_VERSION=2025.1.1+23
 )
 
 :: Check if XPU bundle is target version or already installed

.ci/wheel/build_wheel.sh

@@ -206,7 +206,7 @@ if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
     BUILD_LIBTORCH_WHL=1 BUILD_PYTHON_ONLY=0 python setup.py bdist_wheel -d "$whl_tmp_dir"
     echo "Finished setup.py bdist_wheel for split build (BUILD_LIBTORCH_WHL)"
     echo "Calling setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
-    BUILD_LIBTORCH_WHL=0 BUILD_PYTHON_ONLY=1 CMAKE_FRESH=1 python setup.py bdist_wheel -d "$whl_tmp_dir"
+    BUILD_PYTHON_ONLY=1 BUILD_LIBTORCH_WHL=0 python setup.py bdist_wheel -d "$whl_tmp_dir" --cmake
     echo "Finished setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
 else
     python setup.py bdist_wheel -d "$whl_tmp_dir"

.circleci/scripts/binary_populate_env.sh

@@ -105,7 +105,6 @@ fi
 
 # Set triton via PYTORCH_EXTRA_INSTALL_REQUIREMENTS for triton xpu package
 if [[ "$PACKAGE_TYPE" =~ .*wheel.* && -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*xpu.* ]]; then
-    TRITON_VERSION=$(cat $PYTORCH_ROOT/.ci/docker/triton_xpu_version.txt)
     TRITON_REQUIREMENT="pytorch-triton-xpu==${TRITON_VERSION}"
     if [[ -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*dev.* ]]; then
         TRITON_SHORTHASH=$(cut -c1-8 $PYTORCH_ROOT/.ci/docker/ci_commit_pins/triton-xpu.txt)

.circleci/scripts/trigger_azure_pipeline.py

@@ -0,0 +1,157 @@
+# Documentation: https://docs.microsoft.com/en-us/rest/api/azure/devops/build/?view=azure-devops-rest-6.0
+
+import json
+import os
+import re
+import sys
+import time
+
+import requests
+
+
+AZURE_PIPELINE_BASE_URL = "https://aiinfra.visualstudio.com/PyTorch/"
+AZURE_DEVOPS_PAT_BASE64 = os.environ.get("AZURE_DEVOPS_PAT_BASE64_SECRET", "")
+PIPELINE_ID = "911"
+PROJECT_ID = "0628bce4-2d33-499e-bac5-530e12db160f"
+TARGET_BRANCH = os.environ.get("CIRCLE_BRANCH", "main")
+TARGET_COMMIT = os.environ.get("CIRCLE_SHA1", "")
+
+build_base_url = AZURE_PIPELINE_BASE_URL + "_apis/build/builds?api-version=6.0"
+
+s = requests.Session()
+s.headers.update({"Authorization": "Basic " + AZURE_DEVOPS_PAT_BASE64})
+
+
+def submit_build(pipeline_id, project_id, source_branch, source_version):
+    print("Submitting build for branch: " + source_branch)
+    print("Commit SHA1: ", source_version)
+
+    run_build_raw = s.post(
+        build_base_url,
+        json={
+            "definition": {"id": pipeline_id},
+            "project": {"id": project_id},
+            "sourceBranch": source_branch,
+            "sourceVersion": source_version,
+        },
+    )
+
+    try:
+        run_build_json = run_build_raw.json()
+    except json.decoder.JSONDecodeError as e:
+        print(e)
+        print(
+            "Failed to parse the response. Check if the Azure DevOps PAT is incorrect or expired."
+        )
+        sys.exit(-1)
+
+    build_id = run_build_json["id"]
+
+    print("Submitted bulid: " + str(build_id))
+    print("Bulid URL: " + run_build_json["url"])
+    return build_id
+
+
+def get_build(_id):
+    get_build_url = (
+        AZURE_PIPELINE_BASE_URL + f"/_apis/build/builds/{_id}?api-version=6.0"
+    )
+    get_build_raw = s.get(get_build_url)
+    return get_build_raw.json()
+
+
+def get_build_logs(_id):
+    get_build_logs_url = (
+        AZURE_PIPELINE_BASE_URL + f"/_apis/build/builds/{_id}/logs?api-version=6.0"
+    )
+    get_build_logs_raw = s.get(get_build_logs_url)
+    return get_build_logs_raw.json()
+
+
+def get_log_content(url):
+    resp = s.get(url)
+    return resp.text
+
+
+def wait_for_build(_id):
+    build_detail = get_build(_id)
+    build_status = build_detail["status"]
+
+    while build_status == "notStarted":
+        print("Waiting for run to start: " + str(_id))
+        sys.stdout.flush()
+        try:
+            build_detail = get_build(_id)
+            build_status = build_detail["status"]
+        except Exception as e:
+            print("Error getting build")
+            print(e)
+
+        time.sleep(30)
+
+    print("Bulid started: ", str(_id))
+
+    handled_logs = set()
+    while build_status == "inProgress":
+        try:
+            print("Waiting for log: " + str(_id))
+            logs = get_build_logs(_id)
+        except Exception as e:
+            print("Error fetching logs")
+            print(e)
+            time.sleep(30)
+            continue
+
+        for log in logs["value"]:
+            log_id = log["id"]
+            if log_id in handled_logs:
+                continue
+            handled_logs.add(log_id)
+            print("Fetching log: \n" + log["url"])
+            try:
+                log_content = get_log_content(log["url"])
+                print(log_content)
+            except Exception as e:
+                print("Error getting log content")
+                print(e)
+            sys.stdout.flush()
+        build_detail = get_build(_id)
+        build_status = build_detail["status"]
+        time.sleep(30)
+
+    build_result = build_detail["result"]
+
+    print("Bulid status: " + build_status)
+    print("Bulid result: " + build_result)
+
+    return build_status, build_result
+
+
+if __name__ == "__main__":
+    # Convert the branch name for Azure DevOps
+    match = re.search(r"pull/(\d+)", TARGET_BRANCH)
+    if match is not None:
+        pr_num = match.group(1)
+        SOURCE_BRANCH = f"refs/pull/{pr_num}/head"
+    else:
+        SOURCE_BRANCH = f"refs/heads/{TARGET_BRANCH}"
+
+    MAX_RETRY = 2
+    retry = MAX_RETRY
+
+    while retry > 0:
+        build_id = submit_build(PIPELINE_ID, PROJECT_ID, SOURCE_BRANCH, TARGET_COMMIT)
+        build_status, build_result = wait_for_build(build_id)
+
+        if build_result != "succeeded":
+            retry = retry - 1
+            if retry > 0:
+                print("Retrying... remaining attempt: " + str(retry))
+                # Wait a bit before retrying
+                time.sleep((MAX_RETRY - retry) * 120)
+                continue
+            else:
+                print("No more chance to retry. Giving up.")
+                sys.exit(-1)
+        else:
+            break

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -12,9 +12,7 @@ body:
     description: |
       Please provide a clear and concise description of what the bug is.
 
-      If relevant, add a minimal example so that we can reproduce the error by running the code. It is very important for the snippet to be as succinct (minimal) as possible, so please take time to trim down any irrelevant code to help us debug efficiently.
-      Your example should be fully self-contained and not rely on any artifact that should be downloaded.
-      For example:
+      If relevant, add a minimal example so that we can reproduce the error by running the code. It is very important for the snippet to be as succinct (minimal) as possible, so please take time to trim down any irrelevant code to help us debug efficiently. We are going to copy-paste your code and we expect to get the same result as you did: avoid any external data, and include the relevant imports, etc. For example:
 
       ```python
       # All necessary imports at the beginning
@@ -28,7 +26,6 @@ body:
       If the code is too long (hopefully, it isn't), feel free to put it in a public gist and link it in the issue: https://gist.github.com.
 
       Please also paste or describe the results you observe instead of the expected results. If you observe an error, please paste the error message including the **full** traceback of the exception. It may be relevant to wrap error messages in ```` ```triple quotes blocks``` ````.
-      If your issue is related to numerical accuracy or reproducibility, please read the [numerical accuracy](https://docs.pytorch.org/docs/stable/notes/numerical_accuracy.html) and [reproducibility](https://docs.pytorch.org/docs/stable/notes/randomness.html) notes. If the difference is not expected as described in these documents, please provide appropriate justification on why one result is wrong and the other is correct.
     placeholder: |
       A clear and concise description of what the bug is.
 

.github/actionlint.yaml

@@ -14,7 +14,6 @@ self-hosted-runner:
     - linux.12xlarge
     - linux.24xlarge
     - linux.24xlarge.ephemeral
-    - linux.24xlarge.amd
     - linux.arm64.2xlarge
     - linux.arm64.2xlarge.ephemeral
     - linux.arm64.m7g.4xlarge
@@ -50,7 +49,6 @@ self-hosted-runner:
     # Organization-wide AMD-hosted runners
     # MI2xx runners
     - linux.rocm.gpu
-    - linux.rocm.gpu.mi250
     - linux.rocm.gpu.2
     - linux.rocm.gpu.4
     # MI300 runners

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

@@ -9,7 +9,7 @@ inputs:
   arch-for-build-env:
     description: |
       arch to pass to build environment.
-      This is currently different than the arch name we use elsewhere, which
+      This is currently different than the arch name we use elswhere, which
       should be fixed.
     required: true
   github-secret:

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

@@ -157,4 +157,4 @@ runs:
         echo "Is keep-going label set? ${{ steps.filter.outputs.keep-going }}"
 
         echo
-        echo "Reenabled issues? ${{ steps.filter.outputs.reenabled-issues }}"
+        echo "Renabled issues? ${{ steps.filter.outputs.reenabled-issues }}"

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

@@ -153,7 +153,7 @@ runs:
         github-token: ${{ inputs.GITHUB_TOKEN }}
 
     - name: Check for keep-going label and re-enabled test issues
-      # This uses the filter-test-configs action because it conveniently
+      # This uses the filter-test-configs action because it conviniently
       # checks for labels and re-enabled test issues.  It does not actually do
       # any filtering.  All filtering is done in the build step.
       id: keep-going

.github/actions/reuse-old-whl/action.yml

@@ -13,12 +13,6 @@ inputs:
   github-token:
     description: GitHub token
     required: true
-  job-id:
-    description: Job ID
-    required: true
-  job-name:
-    description: Job name
-    required: true
 
 outputs:
   reuse:
@@ -36,11 +30,8 @@ runs:
       continue-on-error: true
       env:
         GITHUB_TOKEN: ${{ inputs.github-token }}
-        JOB_ID: ${{ inputs.job-id }}
-        JOB_NAME: ${{ inputs.job-name }}
       run: |
         set -x
-        python3 -m pip install boto3==1.35.42
         python3 ${GITHUB_ACTION_PATH}/reuse_old_whl.py \
           --build-environment "${{ inputs.build-environment }}" \
           --run-id "${{ inputs.run-id }}" \

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

@@ -1,22 +1,13 @@
 import argparse
 import os
 import subprocess
-import sys
 from functools import lru_cache
 from pathlib import Path
-from typing import Any, cast, Optional, Union
+from typing import Any, cast, Optional
 
 import requests
 
 
-REPO_ROOT = Path(__file__).resolve().parent.parent.parent.parent
-sys.path.insert(0, str(REPO_ROOT))
-from tools.stats.upload_metrics import emit_metric
-
-
-sys.path.remove(str(REPO_ROOT))  # Clean up sys.path after import
-
-
 FORCE_REBUILD_LABEL = "ci-force-rebuild"
 
 
@@ -123,43 +114,15 @@ def ok_changed_file(file: str) -> bool:
         return True
     if file.startswith("test/") and file.endswith(".py"):
         return True
-    if file.startswith("docs/") and file.endswith((".md", ".rst")):
-        return True
     return False
 
 
 def check_changed_files(sha: str) -> bool:
     # Return true if all the changed files are in the list of allowed files to
     # be changed to reuse the old whl
-
-    # Removing files in the torch folder is not allowed since rsync will not
-    # remove files
-    removed_files = (
-        subprocess.check_output(
-            [
-                "git",
-                "diff",
-                "--name-only",
-                sha,
-                "HEAD",
-                "--diff-filter=D",
-                "--no-renames",
-            ],
-            text=True,
-            stderr=subprocess.DEVNULL,
-        )
-        .strip()
-        .split()
-    )
-    if any(file.startswith("torch/") for file in removed_files):
-        print(
-            f"Removed files between {sha} and HEAD: {removed_files}, cannot reuse old whl"
-        )
-        return False
-
     changed_files = (
         subprocess.check_output(
-            ["git", "diff", "--name-only", sha, "HEAD", "--no-renames"],
+            ["git", "diff", "--name-only", sha, "HEAD"],
             text=True,
             stderr=subprocess.DEVNULL,
         )
@@ -216,83 +179,38 @@ def unzip_artifact_and_replace_files() -> None:
     )
     os.remove("artifacts.zip")
 
-    head_sha = get_head_sha()
-
     # Rename wheel into zip
     wheel_path = Path("artifacts/dist").glob("*.whl")
     for path in wheel_path:
-        # Should be of the form torch-2.0.0+git1234567-cp37-etc.whl
-        # Should usually be the merge base sha but for the ones that didn't do
-        # the replacement, it won't be.  Can probably change it to just be merge
-        # base later
-        old_version = f"+git{path.stem.split('+')[1].split('-')[0][3:]}"
-        new_version = f"+git{head_sha[:7]}"
-
-        def rename_to_new_version(file: Union[str, Path]) -> None:
-            # Rename file with old_version to new_version
-            subprocess.check_output(
-                ["mv", file, str(file).replace(old_version, new_version)]
-            )
-
-        def change_content_to_new_version(file: Union[str, Path]) -> None:
-            # Check if is a file
-            if os.path.isdir(file):
-                return
-            # Replace the old version in the file with the new version
-            with open(file) as f:
-                content = f.read()
-                content = content.replace(old_version, new_version)
-            with open(file, "w") as f:
-                f.write(content)
-
-        zip_path = path.with_suffix(".zip")
-        os.rename(path, zip_path)
-        old_stem = zip_path.stem
+        new_path = path.with_suffix(".zip")
+        os.rename(path, new_path)
+        print(f"Renamed {path} to {new_path}")
+        print(new_path.stem)
         # Unzip the wheel
         subprocess.check_output(
-            ["unzip", "-o", zip_path, "-d", f"artifacts/dist/{old_stem}"],
+            ["unzip", "-o", new_path, "-d", f"artifacts/dist/{new_path.stem}"],
         )
-
-        # Remove the old wheel (which is now a zip file)
-        os.remove(zip_path)
-
         # Copy python files into the artifact
         subprocess.check_output(
-            ["rsync", "-avz", "torch", f"artifacts/dist/{old_stem}"],
+            ["rsync", "-avz", "torch", f"artifacts/dist/{new_path.stem}"],
         )
 
-        change_content_to_new_version(f"artifacts/dist/{old_stem}/torch/version.py")
-
-        for file in Path(f"artifacts/dist/{old_stem}").glob(
-            "*.dist-info/**",
-        ):
-            change_content_to_new_version(file)
-
-        rename_to_new_version(f"artifacts/dist/{old_stem}")
-        new_stem = old_stem.replace(old_version, new_version)
-
-        for file in Path(f"artifacts/dist/{new_stem}").glob(
-            "*.dist-info",
-        ):
-            rename_to_new_version(file)
-
         # Zip the wheel back
         subprocess.check_output(
-            ["zip", "-r", f"{new_stem}.zip", "."],
-            cwd=f"artifacts/dist/{new_stem}",
+            ["zip", "-r", f"{new_path.stem}.zip", "."],
+            cwd=f"artifacts/dist/{new_path.stem}",
         )
-
         subprocess.check_output(
             [
                 "mv",
-                f"artifacts/dist/{new_stem}/{new_stem}.zip",
-                f"artifacts/dist/{new_stem}.whl",
+                f"artifacts/dist/{new_path.stem}/{new_path.stem}.zip",
+                f"artifacts/dist/{new_path.stem}.whl",
             ],
         )
 
         # Remove the extracted folder
         subprocess.check_output(
-            ["rm", "-rf", f"artifacts/dist/{new_stem}"],
+            ["rm", "-rf", f"artifacts/dist/{new_path.stem}"],
         )
 
     # Rezip the artifact
@@ -326,60 +244,46 @@ def parse_args() -> argparse.Namespace:
     return parser.parse_args()
 
 
-def can_reuse_whl(args: argparse.Namespace) -> tuple[bool, str]:
-    if args.github_ref and any(
-        args.github_ref.startswith(x)
-        for x in [
-            "refs/heads/release",
-            "refs/tags/v",
-            "refs/heads/nightly",
-        ]
-    ):
-        print("Release branch, rebuild whl")
-        return (False, "Release branch")
-
-    if not check_changed_files(get_merge_base()):
-        print("Cannot use old whl due to the changed files, rebuild whl")
-        return (False, "Changed files not allowed")
+def can_reuse_whl(args: argparse.Namespace) -> bool:
+    # if is_main_branch() or (
+    #     args.github_ref
+    #     and any(
+    #         args.github_ref.startswith(x)
+    #         for x in ["refs/heads/release", "refs/tags/v", "refs/heads/main"]
+    #     )
+    # ):
+    #     print("On main branch or release branch, rebuild whl")
+    #     return False
 
     if check_labels_for_pr():
         print(f"Found {FORCE_REBUILD_LABEL} label on PR, rebuild whl")
-        return (False, "Found FORCE_REBUILD_LABEL on PR")
+        return False
 
     if check_issue_open():
         print("Issue #153759 is open, rebuild whl")
-        return (False, "Issue #153759 is open")
+        return False
+
+    if not check_changed_files(get_merge_base()):
+        print("Cannot use old whl due to the changed files, rebuild whl")
+        return False
 
     workflow_id = get_workflow_id(args.run_id)
     if workflow_id is None:
         print("No workflow ID found, rebuild whl")
-        return (False, "No workflow ID found")
+        return False
 
     if not find_old_whl(workflow_id, args.build_environment, get_merge_base()):
         print("No old whl found, rebuild whl")
-        return (False, "No old whl found")
         # TODO: go backwards from merge base to find more runs
+        return False
 
-    return (True, "Found old whl")
+    return True
 
 
 if __name__ == "__main__":
     args = parse_args()
 
-    reuse_whl, reason = can_reuse_whl(args)
-
-    if reuse_whl:
+    if can_reuse_whl(args):
         print("Reusing old whl")
         unzip_artifact_and_replace_files()
         set_output()
-
-    emit_metric(
-        "reuse_old_whl",
-        {
-            "reuse_whl": reuse_whl,
-            "reason": reason,
-            "build_environment": args.build_environment,
-            "merge_base": get_merge_base(),
-            "head_sha": get_head_sha(),
-        },
-    )

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

@@ -33,14 +33,14 @@ runs:
       id: check_container_runner
       run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
 
-    - name: Start docker if docker daemon is not running
+    - name: Start docker if docker deamon is not running
       shell: bash
       if: ${{ steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
       run: |
         if systemctl is-active --quiet docker; then
             echo "Docker daemon is running...";
         else
-            echo "Starting docker daemon..." && sudo systemctl start docker;
+            echo "Starting docker deamon..." && sudo systemctl start docker;
         fi
 
     - name: Log in to ECR

.github/actions/setup-xpu/action.yml

@@ -29,13 +29,13 @@ runs:
       if: always()
       shell: bash
       run: |
-        timeout 30 xpu-smi discovery || true
+        xpu-smi discovery
 
     - name: Runner health check GPU count
       if: always()
       shell: bash
       run: |
-        ngpu=$(timeout 30 xpu-smi discovery | grep -c -E 'Device Name' || true)
+        ngpu=$(xpu-smi discovery | grep -c -E 'Device Name')
         msg="Please file an issue on pytorch/pytorch reporting the faulty runner. Include a link to the runner logs so the runner can be identified"
         if [[ $ngpu -eq 0 ]]; then
           echo "Error: Failed to detect any GPUs on the runner"

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-4e94321c54617dd738a05bfedfc28bc0fa635b5c
+1a8f6213b0b61efc6a4862bc45b853551a93dbb6

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-r2.8
+edc1a882d872dd7f1362e4312fd045a1d81b3355

.github/labeler.yml

@@ -116,6 +116,7 @@
 "release notes: inductor (aoti)":
 - torch/_C/_aoti.pyi
 - torch/_dynamo/repro/aoti.py
+- torch/_export/serde/aoti_schema.py
 - torch/_higher_order_ops/aoti_call_delegate.py
 - torch/_inductor/codegen/aoti_runtime/**
 - torch/_inductor/codegen/aoti_hipify_utils.py

.github/merge_rules.yaml

@@ -123,8 +123,6 @@
   - torch/*docs.py
   approved_by:
   - svekars
-  - sekyondaMeta
-  - AlannaBurke
   mandatory_checks_name:
   - EasyCLA
   - Lint

.github/pytorch-probot.yml

@@ -11,7 +11,6 @@ ciflow_push_tags:
 - ciflow/inductor-perf-compare
 - ciflow/inductor-micro-benchmark
 - ciflow/inductor-micro-benchmark-cpu-x86
-- ciflow/inductor-perf-test-nightly-x86-zen
 - ciflow/inductor-cu126
 - ciflow/linux-aarch64
 - ciflow/mps
@@ -29,7 +28,6 @@ ciflow_push_tags:
 - ciflow/op-benchmark
 - ciflow/pull
 - ciflow/h100
-- ciflow/h100-distributed
 retryable_workflows:
 - pull
 - trunk

.github/requirements-gha-cache.txt

@@ -10,5 +10,5 @@ lintrunner==0.10.7
 ninja==1.10.0.post1
 nvidia-ml-py==11.525.84
 pyyaml==6.0
-requests==2.32.4
+requests==2.32.2
 rich==10.9.0

.github/requirements/conda-env-macOS-ARM64

@@ -2,4 +2,5 @@
 certifi
 pip=23.2.1
 pkg-config=0.29.2
+setuptools=72.1.0
 wheel=0.37.1

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

@@ -1,5 +1,5 @@
 boto3==1.35.42
-cmake==3.27.*
+cmake==3.25.*
 expecttest==0.3.0
 fbscribelogger==0.1.7
 filelock==3.6.0
@@ -14,7 +14,7 @@ opt-einsum>=3.3
 optree==0.13.0
 packaging==23.1
 parameterized==0.8.1
-pillow==10.3.0
+pillow==10.0.1
 protobuf==5.29.4
 psutil==5.9.1
 pygments==2.15.0
@@ -26,9 +26,7 @@ pytest-xdist==3.3.1
 pytest==7.3.2
 pyyaml==6.0.2
 scipy==1.12.0
-setuptools==72.1.0
 sympy==1.13.3
-tlparse==0.3.30
 tensorboard==2.13.0
 typing-extensions==4.12.2
 unittest-xml-reporting<=3.2.0,>=2.0.0

.github/scripts/amd/patch_triton_wheel.sh

@@ -78,7 +78,7 @@ for pkg in /$WHEELHOUSE_DIR/*triton*.whl; do
         echo "Copied $filepath to $patchedpath"
     done
 
-    # Go through all required shared objects and see if any of our other objects are dependants.  If so, replace so.ver with so
+    # Go through all required shared objects and see if any of our other objects are dependants.  If so, replace so.ver wth so
     for ((i=0;i<${#deps[@]};++i)); do
         echo "replacing "${deps_soname[i]} ${patched[i]}
         replace_needed_sofiles $PREFIX/$ROCM_LIB ${deps_soname[i]} ${patched[i]}

.github/scripts/build_triton_wheel.py

@@ -21,11 +21,8 @@ def read_triton_pin(device: str = "cuda") -> str:
         return f.read().strip()
 
 
-def read_triton_version(device: str = "cuda") -> str:
-    triton_version_file = "triton_version.txt"
-    if device == "xpu":
-        triton_version_file = "triton_xpu_version.txt"
-    with open(REPO_DIR / ".ci" / "docker" / triton_version_file) as f:
+def read_triton_version() -> str:
+    with open(REPO_DIR / ".ci" / "docker" / "triton_version.txt") as f:
         return f.read().strip()
 
 
@@ -68,7 +65,6 @@ def build_triton(
     with TemporaryDirectory() as tmpdir:
         triton_basedir = Path(tmpdir) / "triton"
         triton_pythondir = triton_basedir / "python"
-
         triton_repo = "https://github.com/openai/triton"
         if device == "rocm":
             triton_pkg_name = "pytorch-triton-rocm"
@@ -94,7 +90,7 @@ def build_triton(
         patch_init_py(
             triton_pythondir / "triton" / "__init__.py",
             version=f"{version}",
-            expected_version=read_triton_version(device),
+            expected_version=None,
         )
 
         if device == "rocm":
@@ -105,19 +101,11 @@ def build_triton(
             )
             print("ROCm libraries setup for triton installation...")
 
-        # old triton versions have setup.py in the python/ dir,
-        # new versions have it in the root dir.
-        triton_setupdir = (
-            triton_basedir
-            if (triton_basedir / "setup.py").exists()
-            else triton_pythondir
-        )
-
         check_call(
-            [sys.executable, "setup.py", "bdist_wheel"], cwd=triton_setupdir, env=env
+            [sys.executable, "setup.py", "bdist_wheel"], cwd=triton_pythondir, env=env
         )
 
-        whl_path = next(iter((triton_setupdir / "dist").glob("*.whl")))
+        whl_path = next(iter((triton_pythondir / "dist").glob("*.whl")))
         shutil.copy(whl_path, Path.cwd())
 
         if device == "rocm":
@@ -140,19 +128,15 @@ def main() -> None:
     parser.add_argument("--py-version", type=str)
     parser.add_argument("--commit-hash", type=str)
     parser.add_argument("--with-clang-ldd", action="store_true")
-    parser.add_argument("--triton-version", type=str, default=None)
+    parser.add_argument("--triton-version", type=str, default=read_triton_version())
     args = parser.parse_args()
 
-    triton_version = read_triton_version(args.device)
-    if args.triton_version:
-        triton_version = args.triton_version
-
     build_triton(
         device=args.device,
         commit_hash=(
             args.commit_hash if args.commit_hash else read_triton_pin(args.device)
         ),
-        version=triton_version,
+        version=args.triton_version,
         py_version=args.py_version,
         release=args.release,
         with_clang_ldd=args.with_clang_ldd,

.github/scripts/filter_test_configs.py

@@ -40,9 +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?versionId=HnkH0xQWnnsoeMsSIVf9291NE5c4jWSa"
+DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json"
 # and unstable jobs
-UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json?versionId=iP_F8gBs60PfOMAJ8gnn1paVrzM1WYsK"
+UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"
@@ -80,7 +80,7 @@ def parse_args() -> Any:
     parser.add_argument(
         "--job-name",
         type=str,
-        help="the name of the current job, i.e. linux-jammy-py3.8-gcc7 / build",
+        help="the name of the current job, i.e. linux-focal-py3.8-gcc7 / build",
     )
     parser.add_argument("--pr-number", type=str, help="the pull request number")
     parser.add_argument("--tag", type=str, help="the associated tag if it exists")

.github/scripts/generate_binary_build_matrix.py

@@ -15,21 +15,21 @@ import os
 from typing import Optional
 
 
-# NOTE: Please also update the CUDA sources in `PIP_SOURCES` in tools/nightly.py when changing this
-CUDA_ARCHES = ["12.6", "12.8", "12.9"]
+# NOTE: Also update the CUDA sources in tools/nightly.py when changing this list
+CUDA_ARCHES = ["11.8", "12.6", "12.8"]
 CUDA_STABLE = "12.6"
 CUDA_ARCHES_FULL_VERSION = {
+    "11.8": "11.8.0",
     "12.6": "12.6.3",
     "12.8": "12.8.1",
-    "12.9": "12.9.1",
 }
 CUDA_ARCHES_CUDNN_VERSION = {
+    "11.8": "9",
     "12.6": "9",
     "12.8": "9",
-    "12.9": "9",
 }
 
-# NOTE: Please also update the ROCm sources in `PIP_SOURCES` in tools/nightly.py when changing this
+# NOTE: Also update the ROCm sources in tools/nightly.py when changing this list
 ROCM_ARCHES = ["6.3", "6.4"]
 
 XPU_ARCHES = ["xpu"]
@@ -38,23 +38,35 @@ CPU_AARCH64_ARCH = ["cpu-aarch64"]
 
 CPU_S390X_ARCH = ["cpu-s390x"]
 
-CUDA_AARCH64_ARCHES = ["12.9-aarch64"]
+CUDA_AARCH64_ARCHES = ["12.8-aarch64"]
 
 
 PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
+    "11.8": (
+        "nvidia-cuda-nvrtc-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | "  # noqa: B950
+        "nvidia-cuda-runtime-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cuda-cupti-cu11==11.8.87; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cudnn-cu11==9.1.0.70; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cublas-cu11==11.11.3.6; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cufft-cu11==10.9.0.58; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-curand-cu11==10.3.0.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cusolver-cu11==11.4.1.48; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cusparse-cu11==11.7.5.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu11==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nvtx-cu11==11.8.86; platform_system == 'Linux' and platform_machine == 'x86_64'"
+    ),
     "12.6": (
         "nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cudnn-cu12==9.5.1.17; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "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==2.27.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.2.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cusparselt-cu12==0.6.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.26.5; 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'"
@@ -63,42 +75,25 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cudnn-cu12==9.8.0.87; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "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==2.27.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.2.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-cusparselt-cu12==0.6.3; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.26.5; 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'"
     ),
-    "12.9": (
-        "nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "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==2.27.3; 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'"
-    ),
     "xpu": (
         "intel-cmplr-lib-rt==2025.1.1 | "
         "intel-cmplr-lib-ur==2025.1.1 | "
         "intel-cmplr-lic-rt==2025.1.1 | "
         "intel-sycl-rt==2025.1.1 | "
-        "oneccl-devel==2021.15.2; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "oneccl==2021.15.2; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "oneccl-devel==2021.15.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "oneccl==2021.15.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "impi-rt==2021.15.0; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "onemkl-sycl-blas==2025.1.0 | "
         "onemkl-sycl-dft==2025.1.0 | "
@@ -112,7 +107,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "tbb==2022.1.0 | "
         "tcmlib==1.3.0 | "
         "umf==0.10.0 | "
-        "intel-pti==0.12.3"
+        "intel-pti==0.12.0"
     ),
 }
 
@@ -316,10 +311,10 @@ def generate_wheels_matrix(
                 continue
 
             if use_split_build and (
-                arch_version not in ["12.6", "12.8", "12.9", "cpu"] or os != "linux"
+                arch_version not in ["12.6", "12.8", "11.8", "cpu"] or os != "linux"
             ):
                 raise RuntimeError(
-                    "Split build is only supported on linux with cuda 12* and cpu.\n"
+                    "Split build is only supported on linux with cuda 12*, 11.8, and cpu.\n"
                     f"Currently attempting to build on arch version {arch_version} and os {os}.\n"
                     "Please modify the matrix generation to exclude this combination."
                 )
@@ -327,7 +322,7 @@ def generate_wheels_matrix(
             # cuda linux wheels require PYTORCH_EXTRA_INSTALL_REQUIREMENTS to install
 
             if (
-                arch_version in ["12.9", "12.8", "12.6"]
+                arch_version in ["12.8", "12.6", "11.8"]
                 and os == "linux"
                 or arch_version in CUDA_AARCH64_ARCHES
             ):
@@ -416,6 +411,6 @@ def generate_wheels_matrix(
     return ret
 
 
-validate_nccl_dep_consistency("12.9")
 validate_nccl_dep_consistency("12.8")
 validate_nccl_dep_consistency("12.6")
+validate_nccl_dep_consistency("11.8")

.github/scripts/generate_ci_workflows.py

@@ -152,7 +152,7 @@ LINUX_BINARY_SMOKE_WORKFLOWS = [
         package_type="manywheel",
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(
             OperatingSystem.LINUX,
-            arches=["12.6", "12.8", "12.9", "6.4"],
+            arches=["11.8", "12.6", "12.8"],
             python_versions=["3.9"],
         ),
         branches="main",

.github/scripts/get_workflow_job_id.py

@@ -64,7 +64,7 @@ def fetch_url(
             )
         exception_message = (
             "Is github alright?",
-            f"Received status code '{err.code}' when attempting to retrieve {url}:\n",
+            f"Recieved status code '{err.code}' when attempting to retrieve {url}:\n",
             f"{err.reason}\n\nheaders={err.headers}",
         )
         raise RuntimeError(exception_message) from err

.github/scripts/gitutils.py

@@ -211,7 +211,7 @@ class GitRepo:
         self, from_branch: str, to_branch: str
     ) -> tuple[list[str], list[str]]:
         """
-        Returns list of commits that are missing in each other branch since their merge base
+        Returns list of commmits that are missing in each other branch since their merge base
         Might be slow if merge base is between two branches is pretty far off
         """
         from_ref = self.rev_parse(from_branch)

.github/scripts/pr-sanity-check.sh

@@ -12,7 +12,7 @@ BASE=${BASE:-HEAD~1}
 HEAD=${HEAD:-HEAD}
 
 ancestor=$(git merge-base "${BASE}" "${HEAD}")
-echo "INFO: Checking against the following stats"
+echo "INFO: Checking aginst the following stats"
 (
     set -x
     git diff --stat=10000 "$ancestor" "${HEAD}" | sed '$d' > "${TMPFILE}"

.github/scripts/test_filter_test_configs.py

@@ -347,26 +347,26 @@ class TestConfigFilter(TestCase):
             {
                 "job_name": "a-ci-job",
                 "test_matrix": '{include: [{config: "default", runner: "linux"}, {config: "cfg", runner: "macos"}]}',
-                "description": "Replicate each periodic mode in a different config",
+                "descripion": "Replicate each periodic mode in a different config",
             },
             {
                 "job_name": "a-ci-cuda11.8-job",
                 "test_matrix": '{include: [{config: "default", runner: "linux"}, {config: "cfg", runner: "macos"}]}',
-                "description": "Replicate each periodic mode in a different config for a CUDA job",
+                "descripion": "Replicate each periodic mode in a different config for a CUDA job",
             },
             {
                 "job_name": "a-ci-rocm-job",
                 "test_matrix": '{include: [{config: "default", runner: "linux"}, {config: "cfg", runner: "macos"}]}',
-                "description": "Replicate each periodic mode in a different config for a ROCm job",
+                "descripion": "Replicate each periodic mode in a different config for a ROCm job",
             },
             {
                 "job_name": "",
                 "test_matrix": '{include: [{config: "default", runner: "linux"}, {config: "cfg", runner: "macos"}]}',
-                "description": "Empty job name",
+                "descripion": "Empty job name",
             },
             {
                 "test_matrix": '{include: [{config: "default", runner: "linux"}, {config: "cfg", runner: "macos"}]}',
-                "description": "Missing job name",
+                "descripion": "Missing job name",
             },
         ]
 

.github/scripts/test_trymerge.py

@@ -19,7 +19,6 @@ from urllib.error import HTTPError
 from github_utils import gh_graphql
 from gitutils import get_git_remote_name, get_git_repo_dir, GitRepo
 from trymerge import (
-    _revlist_to_prs,
     categorize_checks,
     DRCI_CHECKRUN_NAME,
     find_matching_merge_rule,
@@ -265,7 +264,7 @@ class DummyGitRepo(GitRepo):
         return ["FakeCommitSha"]
 
     def commit_message(self, ref: str) -> str:
-        return "super awesome commit message"
+        return "super awsome commit message"
 
 
 @mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
@@ -433,7 +432,7 @@ class TestTryMerge(TestCase):
         )
 
     def test_cancelled_gets_ignored(self, *args: Any) -> None:
-        """Tests that cancelled workflow does not override existing successful status"""
+        """Tests that cancelled workflow does not override existing successfull status"""
         pr = GitHubPR("pytorch", "pytorch", 110367)
         conclusions = pr.get_checkrun_conclusions()
         lint_checks = [name for name in conclusions.keys() if "Lint" in name]
@@ -1089,51 +1088,5 @@ class TestGitHubPRGhstackDependencies(TestCase):
         )
 
 
-@mock.patch("trymerge.gh_graphql", side_effect=mocked_gh_graphql)
-@mock.patch("trymerge.gh_fetch_merge_base", return_value="")
-@mock.patch(
-    "trymerge.get_drci_classifications", side_effect=mocked_drci_classifications
-)
-@mock.patch.object(DummyGitRepo, "commit_message")
-class TestRevListToPR(TestCase):
-    # Tests for _revlist_to_prs function
-    def test__revlist_to_prs_zero_matches(
-        self, mock_commit_message: mock.MagicMock, *args: Any
-    ) -> None:
-        # If zero PRs are mentioned in the commit message, it should raise an error
-        pr_num = 154098
-        pr = GitHubPR("pytorch", "pytorch", pr_num)
-        repo = DummyGitRepo()
-        mock_commit_message.return_value = "no PRs"
-        self.assertRaisesRegex(
-            RuntimeError,
-            "PRs mentioned in commit dummy: 0.",
-            lambda: _revlist_to_prs(repo, pr, ["dummy"]),
-        )
-
-    def test__revlist_to_prs_two_prs(
-        self, mock_commit_message: mock.MagicMock, *args: Any
-    ) -> None:
-        # If two PRs are mentioned in the commit message, it should raise an error
-        pr_num = 154394
-        pr = GitHubPR("pytorch", "pytorch", pr_num)
-        repo = DummyGitRepo()
-        # https://github.com/pytorch/pytorch/commit/343c56e7650f55fd030aca0b9275d6d73501d3f4
-
-        commit_message = """add sticky cache pgo
-
-ghstack-source-id: 9bc6dee0b427819f978bfabccb72727ba8be2f81
-Pull-Request-resolved: https://github.com/pytorch/pytorch/pull/154098
-
-ghstack-source-id: 9bc6dee0b427819f978bfabccb72727ba8be2f81
-Pull Request resolved: https://github.com/pytorch/pytorch/pull/154394"""
-        mock_commit_message.return_value = commit_message
-        self.assertRaisesRegex(
-            RuntimeError,
-            "PRs mentioned in commit dummy: 2.",
-            lambda: _revlist_to_prs(repo, pr, ["dummy"]),
-        )
-
-
 if __name__ == "__main__":
     main()

.github/scripts/trymerge.py

@@ -628,17 +628,11 @@ def _revlist_to_prs(
     rc: list[tuple[GitHubPR, str]] = []
     for idx, rev in enumerate(rev_list):
         msg = repo.commit_message(rev)
-        # findall doesn't return named captures, so we need to use finditer
-        all_matches = list(RE_PULL_REQUEST_RESOLVED.finditer(msg))
-        if len(all_matches) != 1:
+        m = RE_PULL_REQUEST_RESOLVED.search(msg)
+        if m is None:
             raise RuntimeError(
-                f"Found an unexpected number of PRs mentioned in commit {rev}: "
-                f"{len(all_matches)}.  This is probably because you are using an "
-                "old version of ghstack.  Please update ghstack and resubmit "
-                "your PRs"
+                f"Could not find PR-resolved string in {msg} of ghstacked PR {pr.pr_num}"
             )
-
-        m = all_matches[0]
         if pr.org != m.group("owner") or pr.project != m.group("repo"):
             raise RuntimeError(
                 f"PR {m.group('number')} resolved to wrong owner/repo pair"
@@ -672,9 +666,6 @@ def get_ghstack_prs(
 
     assert pr.is_ghstack_pr()
     entire_stack = _revlist_to_prs(repo, pr, reversed(rev_list), skip_func)
-    print(
-        f"Found {len(entire_stack)} PRs in the stack for {pr.pr_num}: {[x[0].pr_num for x in entire_stack]}"
-    )
 
     for stacked_pr, rev in entire_stack:
         if stacked_pr.is_closed():

.github/scripts/windows/build_magma.bat

@@ -17,6 +17,7 @@ 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%
@@ -34,15 +35,15 @@ cd magma
 mkdir build && cd build
 
 set GPU_TARGET=All
-if "%CUVER_NODOT%" == "129" (
-  set CUDA_ARCH_LIST=-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 -gencode arch=compute_100,code=sm_100 -gencode arch=compute_120,code=sm_120
-)
 if "%CUVER_NODOT%" == "128" (
   set CUDA_ARCH_LIST=-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90 -gencode arch=compute_100,code=sm_100 -gencode arch=compute_120,code=sm_120
 )
-if "%CUVER_NODOT%" == "126" (
+if "%CUVER_NODOT:~0,2%" == "12" if NOT "%CUVER_NODOT%" == "128" (
   set CUDA_ARCH_LIST=-gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90
 )
+if "%CUVER_NODOT%" == "118" (
+  set CUDA_ARCH_LIST= -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90
+)
 
 set CC=cl.exe
 set CXX=cl.exe

.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@release/2.8
+        uses: pytorch/test-infra/.github/actions/setup-ssh@main
         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@release/2.8
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -114,12 +114,12 @@ jobs:
       ALPINE_IMAGE: "docker.io/s390x/alpine"
       {%- elif config["gpu_arch_type"] == "rocm" %}
       runs_on: linux.rocm.gpu
-      {%- elif config["gpu_arch_type"] == "cuda" and config["gpu_arch_version"] in ["12.8", "12.9"] %}
+      {%- elif config["gpu_arch_type"] == "cuda" and config["gpu_arch_version"] == "12.8" %}
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.g4dn.4xlarge.nvidia.gpu  # 12.8 and 12.9 build need sm_70+ runner
-      {%- elif config["gpu_arch_type"] == "cuda" %}
+      runs_on: linux.g4dn.4xlarge.nvidia.gpu  # 12.8 build needs sm_70+ runner
+      {%- elif config["gpu_arch_type"] == "cuda" and config["gpu_arch_version"] != "12.8"%}
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.4xlarge.nvidia.gpu # for other cuda versions, we use 4xlarge runner
+      runs_on: linux.4xlarge.nvidia.gpu
       {%- else %}
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.4xlarge
@@ -150,10 +150,10 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.8
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
         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@release/2.8
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary
@@ -171,7 +171,7 @@ jobs:
       - name: Teardown XPU
         uses: ./.github/actions/teardown-xpu
     {%- else %}
-    runs-on: linux.rocm.gpu.mi250
+    runs-on: linux.rocm.gpu
     timeout-minutes: !{{ common.timeout_minutes }}
     !{{ upload.binary_env(config) }}
     steps:
@@ -182,7 +182,7 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
       - 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@release/2.8
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
         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@release/2.8
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         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", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
       - 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@release/2.8
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -135,7 +135,7 @@ jobs:
 {%- else %}
       !{{ set_runner_specific_vars() }}
       !{{ common.setup_ec2_windows() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
 {%- 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", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
       !{{ 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@release/2.8
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
         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@release/2.8
+        uses: pytorch/test-infra/.github/actions/setup-ssh@main
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
 
       - 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@release/2.8
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.8
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         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@release/2.8
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
         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@release/2.8
+        uses: pytorch/test-infra/.github/actions/teardown-linux@main
         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@release/2.8
+        uses: pytorch/test-infra/.github/actions/setup-ssh@main
         continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -187,6 +187,7 @@ 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
@@ -221,9 +222,9 @@ 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@release/2.8
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
         with:
-          # If doing this in release/2.8 or release branch, use docker.io. Otherwise
+          # If doing this in main 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 }}
@@ -235,7 +236,7 @@ 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@release/2.8
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -292,7 +293,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.8
+        uses: pytorch/test-infra/.github/actions/teardown-linux@main
 
       - 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@release/2.8
+        uses: pytorch/test-infra/.github/actions/setup-ssh@main
         continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
         # Setup the environment
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.8
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -164,6 +164,7 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
+          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           show-progress: false
           path: pytorch
@@ -194,7 +195,7 @@ jobs:
           path: "${{ runner.temp }}/artifacts/"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.8
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
         if: ${{ inputs.GPU_ARCH_TYPE == 'cuda' && steps.filter.outputs.is-test-matrix-empty == 'False' }}
 
       - name: configure aws credentials
@@ -209,7 +210,7 @@ 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@release/2.8
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
         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 }}
@@ -219,7 +220,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.8
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -231,7 +232,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.8
+        uses: pytorch/test-infra/.github/actions/teardown-linux@main
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'