apply

Pull Request resolved: https://github.com/pytorch/pytorch/pull/147240
Approved by: https://github.com/pianpwk
ghstack dependencies: #146939

.ci/aarch64_linux/aarch64_ci_build.sh

@@ -3,22 +3,15 @@ set -eux -o pipefail
 
 GPU_ARCH_VERSION=${GPU_ARCH_VERSION:-}
 
+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
+
 SCRIPTPATH="$( cd -- "$(dirname "$0")" >/dev/null 2>&1 ; pwd -P )"
 source $SCRIPTPATH/aarch64_ci_setup.sh
 
-tagged_version() {
-  GIT_DESCRIBE="git --git-dir /pytorch/.git describe --tags --match v[0-9]*.[0-9]*.[0-9]*"
-  if ${GIT_DESCRIBE} --exact >/dev/null; then
-    ${GIT_DESCRIBE}
-  else
-    return 1
-  fi
-}
-
-if tagged_version >/dev/null; then
-  export OVERRIDE_PACKAGE_VERSION="$(tagged_version | sed -e 's/^v//' -e 's/-.*$//')"
-fi
-
 ###############################################################################
 # Run aarch64 builder python
 ###############################################################################

.ci/aarch64_linux/aarch64_ci_setup.sh

@@ -5,16 +5,14 @@ set -eux -o pipefail
 # By creating symlinks from desired /opt/python to /usr/local/bin/
 
 NUMPY_VERSION=2.0.2
-PYGIT2_VERSION=1.15.1
-if [[ "$DESIRED_PYTHON"  == "3.13" ]]; then
+if [[ "$DESIRED_PYTHON"  == "3.13" || "$DESIRED_PYTHON" == "3.13t" ]]; then
     NUMPY_VERSION=2.1.2
-    PYGIT2_VERSION=1.16.0
 fi
 
 SCRIPTPATH="$( cd "$(dirname "$0")" ; pwd -P )"
 source $SCRIPTPATH/../manywheel/set_desired_python.sh
 
-pip install -q numpy==${NUMPY_VERSION} pyyaml==6.0.2 scons==4.7.0 ninja==1.11.1 patchelf==0.17.2 pygit2==${PYGIT2_VERSION}
+pip install -q numpy==${NUMPY_VERSION} pyyaml==6.0.2 scons==4.7.0 ninja==1.11.1 patchelf==0.17.2
 
 for tool in python python3 pip pip3 ninja scons patchelf; do
     ln -sf ${DESIRED_PYTHON_BIN_DIR}/${tool} /usr/local/bin;

.ci/aarch64_linux/aarch64_wheel_ci_build.py

@@ -4,12 +4,9 @@
 import os
 import shutil
 from subprocess import check_call, check_output
-from typing import List
-
-from pygit2 import Repository
 
 
-def list_dir(path: str) -> List[str]:
+def list_dir(path: str) -> list[str]:
     """'
     Helper for getting paths for Python
     """
@@ -58,7 +55,7 @@ def build_ArmComputeLibrary() -> None:
         shutil.copytree(f"{acl_checkout_dir}/{d}", f"{acl_install_dir}/{d}")
 
 
-def update_wheel(wheel_path) -> None:
+def update_wheel(wheel_path, desired_cuda) -> None:
     """
     Update the cuda wheel libraries
     """
@@ -80,7 +77,6 @@ def update_wheel(wheel_path) -> None:
         "/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/libnvrtc-builtins.so.12.6",
         "/usr/local/cuda/lib64/libcudnn_adv.so.9",
         "/usr/local/cuda/lib64/libcudnn_cnn.so.9",
         "/usr/local/cuda/lib64/libcudnn_graph.so.9",
@@ -100,6 +96,14 @@ def update_wheel(wheel_path) -> None:
             "/usr/local/lib/libnvpl_lapack_core.so.0",
             "/usr/local/lib/libnvpl_blas_core.so.0",
         ]
+        if "126" in desired_cuda:
+            libs_to_copy += [
+                "/usr/local/cuda/lib64/libnvrtc-builtins.so.12.6",
+            ]
+        elif "128" in desired_cuda:
+            libs_to_copy += [
+                "/usr/local/cuda/lib64/libnvrtc-builtins.so.12.8",
+            ]
     else:
         libs_to_copy += [
             "/opt/OpenBLAS/lib/libopenblas.so.0",
@@ -171,22 +175,22 @@ if __name__ == "__main__":
     args = parse_arguments()
     enable_mkldnn = args.enable_mkldnn
     enable_cuda = args.enable_cuda
-    repo = Repository("/pytorch")
-    branch = repo.head.name
-    if branch == "HEAD":
-        branch = "master"
+    branch = check_output(
+        ["git", "rev-parse", "--abbrev-ref", "HEAD"], cwd="/pytorch"
+    ).decode()
 
     print("Building PyTorch wheel")
     build_vars = "MAX_JOBS=5 CMAKE_SHARED_LINKER_FLAGS=-Wl,-z,max-page-size=0x10000 "
     os.system("cd /pytorch; python setup.py clean")
 
     override_package_version = os.getenv("OVERRIDE_PACKAGE_VERSION")
+    desired_cuda = os.getenv("DESIRED_CUDA")
     if override_package_version is not None:
         version = override_package_version
         build_vars += (
             f"BUILD_TEST=0 PYTORCH_BUILD_VERSION={version} PYTORCH_BUILD_NUMBER=1 "
         )
-    elif branch in ["nightly", "master"]:
+    elif branch in ["nightly", "main"]:
         build_date = (
             check_output(["git", "log", "--pretty=format:%cs", "-1"], cwd="/pytorch")
             .decode()
@@ -196,7 +200,6 @@ if __name__ == "__main__":
             check_output(["cat", "version.txt"], cwd="/pytorch").decode().strip()[:-2]
         )
         if enable_cuda:
-            desired_cuda = os.getenv("DESIRED_CUDA")
             build_vars += f"BUILD_TEST=0 PYTORCH_BUILD_VERSION={version}.dev{build_date}+{desired_cuda} PYTORCH_BUILD_NUMBER=1 "
         else:
             build_vars += f"BUILD_TEST=0 PYTORCH_BUILD_VERSION={version}.dev{build_date} PYTORCH_BUILD_NUMBER=1 "
@@ -225,6 +228,6 @@ if __name__ == "__main__":
         print("Updating Cuda Dependency")
         filename = os.listdir("/pytorch/dist/")
         wheel_path = f"/pytorch/dist/{filename[0]}"
-        update_wheel(wheel_path)
+        update_wheel(wheel_path, desired_cuda)
     pytorch_wheel_name = complete_wheel("/pytorch/")
     print(f"Build Complete. Created {pytorch_wheel_name}..")

.ci/aarch64_linux/build_aarch64_wheel.py

@@ -12,7 +12,7 @@ import os
 import subprocess
 import sys
 import time
-from typing import Dict, List, Optional, Tuple, Union
+from typing import Optional, Union
 
 import boto3
 
@@ -24,10 +24,12 @@ os_amis = {
     "ubuntu22_04": "ami-0c6c29c5125214c77",  # login_name: ubuntu
     "redhat8": "ami-0698b90665a2ddcf1",  # login_name: ec2-user
 }
+
 ubuntu18_04_ami = os_amis["ubuntu18_04"]
+ubuntu20_04_ami = os_amis["ubuntu20_04"]
 
 
-def compute_keyfile_path(key_name: Optional[str] = None) -> Tuple[str, str]:
+def compute_keyfile_path(key_name: Optional[str] = None) -> tuple[str, str]:
     if key_name is None:
         key_name = os.getenv("AWS_KEY_NAME")
         if key_name is None:
@@ -57,7 +59,7 @@ def ec2_instances_by_id(instance_id):
 
 
 def start_instance(
-    key_name, ami=ubuntu18_04_ami, instance_type="t4g.2xlarge", ebs_size: int = 50
+    key_name, ami=ubuntu20_04_ami, instance_type="t4g.2xlarge", ebs_size: int = 50
 ):
     inst = ec2.create_instances(
         ImageId=ami,
@@ -96,7 +98,7 @@ class RemoteHost:
         self.keyfile_path = keyfile_path
         self.login_name = login_name
 
-    def _gen_ssh_prefix(self) -> List[str]:
+    def _gen_ssh_prefix(self) -> list[str]:
         return [
             "ssh",
             "-o",
@@ -108,13 +110,13 @@ class RemoteHost:
         ]
 
     @staticmethod
-    def _split_cmd(args: Union[str, List[str]]) -> List[str]:
+    def _split_cmd(args: Union[str, list[str]]) -> list[str]:
         return args.split() if isinstance(args, str) else args
 
-    def run_ssh_cmd(self, args: Union[str, List[str]]) -> None:
+    def run_ssh_cmd(self, args: Union[str, list[str]]) -> None:
         subprocess.check_call(self._gen_ssh_prefix() + self._split_cmd(args))
 
-    def check_ssh_output(self, args: Union[str, List[str]]) -> str:
+    def check_ssh_output(self, args: Union[str, list[str]]) -> str:
         return subprocess.check_output(
             self._gen_ssh_prefix() + self._split_cmd(args)
         ).decode("utf-8")
@@ -157,7 +159,7 @@ class RemoteHost:
     def using_docker(self) -> bool:
         return self.container_id is not None
 
-    def run_cmd(self, args: Union[str, List[str]]) -> None:
+    def run_cmd(self, args: Union[str, list[str]]) -> None:
         if not self.using_docker():
             return self.run_ssh_cmd(args)
         assert self.container_id is not None
@@ -178,7 +180,7 @@ class RemoteHost:
         if rc != 0:
             raise subprocess.CalledProcessError(rc, docker_cmd)
 
-    def check_output(self, args: Union[str, List[str]]) -> str:
+    def check_output(self, args: Union[str, list[str]]) -> str:
         if not self.using_docker():
             return self.check_ssh_output(args)
         assert self.container_id is not None
@@ -230,7 +232,7 @@ class RemoteHost:
             )
         self.download_file(remote_file, local_file)
 
-    def list_dir(self, path: str) -> List[str]:
+    def list_dir(self, path: str) -> list[str]:
         return self.check_output(["ls", "-1", path]).split("\n")
 
 
@@ -358,7 +360,7 @@ def checkout_repo(
     branch: str = "main",
     url: str,
     git_clone_flags: str,
-    mapping: Dict[str, Tuple[str, str]],
+    mapping: dict[str, tuple[str, str]],
 ) -> Optional[str]:
     for prefix in mapping:
         if not branch.startswith(prefix):
@@ -619,9 +621,11 @@ def build_torchaudio(
     if host.using_docker():
         build_vars += " CMAKE_SHARED_LINKER_FLAGS=-Wl,-z,max-page-size=0x10000"
 
-    host.run_cmd(f"cd audio && export FFMPEG_ROOT=$(pwd)/third_party/ffmpeg && export USE_FFMPEG=1 \
+    host.run_cmd(
+        f"cd audio && export FFMPEG_ROOT=$(pwd)/third_party/ffmpeg && export USE_FFMPEG=1 \
         && ./packaging/ffmpeg/build.sh \
-        && {build_vars} python3 setup.py bdist_wheel")
+        && {build_vars} python3 setup.py bdist_wheel"
+    )
 
     wheel_name = host.list_dir("audio/dist")[0]
     embed_libgomp(host, use_conda, os.path.join("audio", "dist", wheel_name))
@@ -679,7 +683,7 @@ def build_domains(
     branch: str = "main",
     use_conda: bool = True,
     git_clone_flags: str = "",
-) -> Tuple[str, str, str, str]:
+) -> tuple[str, str, str, str]:
     vision_wheel_name = build_torchvision(
         host, branch=branch, use_conda=use_conda, git_clone_flags=git_clone_flags
     )
@@ -706,7 +710,7 @@ def start_build(
     pytorch_build_number: Optional[str] = None,
     shallow_clone: bool = True,
     enable_mkldnn: bool = False,
-) -> Tuple[str, str, str, str, str]:
+) -> tuple[str, str, str, str, str]:
     git_clone_flags = " --depth 1 --shallow-submodules" if shallow_clone else ""
     if host.using_docker() and not use_conda:
         print("Auto-selecting conda option for docker images")
@@ -930,9 +934,9 @@ def parse_arguments():
     parser.add_argument("--debug", action="store_true")
     parser.add_argument("--build-only", action="store_true")
     parser.add_argument("--test-only", type=str)
-    parser.add_argument(
-        "--os", type=str, choices=list(os_amis.keys()), default="ubuntu20_04"
-    )
+    group = parser.add_mutually_exclusive_group()
+    group.add_argument("--os", type=str, choices=list(os_amis.keys()))
+    group.add_argument("--ami", type=str)
     parser.add_argument(
         "--python-version",
         type=str,
@@ -962,7 +966,13 @@ def parse_arguments():
 
 if __name__ == "__main__":
     args = parse_arguments()
-    ami = os_amis[args.os]
+    ami = (
+        args.ami
+        if args.ami is not None
+        else os_amis[args.os]
+        if args.os is not None
+        else ubuntu20_04_ami
+    )
     keyfile_path, key_name = compute_keyfile_path(args.key_name)
 
     if args.list_instances:

.ci/docker/aotriton_version.txt

@@ -1,5 +0,0 @@
-0.8b
-manylinux_2_28
-rocm6.2
-6f8cbcac8a92775291bb1ba8f514d4beb350baf4
-e938def5d32869fe2e00aec0300f354c9f157867bebdf2e104d732b94cb238d8

.ci/docker/build.sh

@@ -86,6 +86,10 @@ CMAKE_VERSION=3.18.5
 
 _UCX_COMMIT=7bb2722ff2187a0cad557ae4a6afa090569f83fb
 _UCC_COMMIT=20eae37090a4ce1b32bcce6144ccad0b49943e0b
+if [[ "$image" == *rocm* ]]; then
+  _UCX_COMMIT=cc312eaa4655c0cc5c2bcd796db938f90563bcf6
+  _UCC_COMMIT=0c0fc21559835044ab107199e334f7157d6a0d3d
+fi
 
 # It's annoying to rename jobs every time you want to rewrite a
 # configuration, so we hardcode everything here rather than do it
@@ -105,20 +109,6 @@ case "$image" in
     CONDA_CMAKE=yes
     TRITON=yes
     ;;
-  pytorch-linux-focal-cuda12.1-cudnn9-py3-gcc9)
-    CUDA_VERSION=12.1.1
-    CUDNN_VERSION=9
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    CONDA_CMAKE=yes
-    TRITON=yes
-    ;;
   pytorch-linux-focal-cuda12.4-cudnn9-py3-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.4.1
     CUDNN_VERSION=9
@@ -134,36 +124,6 @@ case "$image" in
     TRITON=yes
     INDUCTOR_BENCHMARKS=yes
     ;;
-  pytorch-linux-focal-cuda12.1-cudnn9-py3-gcc9-inductor-benchmarks)
-    CUDA_VERSION=12.1.1
-    CUDNN_VERSION=9
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    CONDA_CMAKE=yes
-    TRITON=yes
-    INDUCTOR_BENCHMARKS=yes
-    ;;
-  pytorch-linux-focal-cuda12.1-cudnn9-py3.12-gcc9-inductor-benchmarks)
-    CUDA_VERSION=12.1.1
-    CUDNN_VERSION=9
-    ANACONDA_PYTHON_VERSION=3.12
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    CONDA_CMAKE=yes
-    TRITON=yes
-    INDUCTOR_BENCHMARKS=yes
-    ;;
   pytorch-linux-focal-cuda12.4-cudnn9-py3.12-gcc9-inductor-benchmarks)
     CUDA_VERSION=12.4.1
     CUDNN_VERSION=9
@@ -208,48 +168,6 @@ case "$image" in
     CONDA_CMAKE=yes
     TRITON=yes
     ;;
-  pytorch-linux-focal-cuda12.4-cudnn9-py3-gcc9)
-    CUDA_VERSION=12.4.1
-    CUDNN_VERSION=9
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    CONDA_CMAKE=yes
-    TRITON=yes
-    ;;
-  pytorch-linux-focal-cuda12.1-cudnn9-py3-gcc9)
-    CUDA_VERSION=12.1.1
-    CUDNN_VERSION=9
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    CONDA_CMAKE=yes
-    TRITON=yes
-    ;;
-  pytorch-linux-focal-cuda12.4-cudnn9-py3-gcc9)
-    CUDA_VERSION=12.4.1
-    CUDNN_VERSION=9
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    KATEX=yes
-    UCX_COMMIT=${_UCX_COMMIT}
-    UCC_COMMIT=${_UCC_COMMIT}
-    CONDA_CMAKE=yes
-    TRITON=yes
-    ;;
   pytorch-linux-focal-py3-clang10-onnx)
     ANACONDA_PYTHON_VERSION=3.9
     CLANG_VERSION=10
@@ -292,18 +210,7 @@ case "$image" in
     ;;
   pytorch-linux-focal-rocm-n-1-py3)
     ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
-    PROTOBUF=yes
-    DB=yes
-    VISION=yes
-    ROCM_VERSION=6.1
-    NINJA_VERSION=1.9.0
-    CONDA_CMAKE=yes
-    TRITON=yes
-    ;;
-  pytorch-linux-focal-rocm-n-py3)
-    ANACONDA_PYTHON_VERSION=3.10
-    GCC_VERSION=9
+    GCC_VERSION=11
     PROTOBUF=yes
     DB=yes
     VISION=yes
@@ -311,6 +218,25 @@ case "$image" in
     NINJA_VERSION=1.9.0
     CONDA_CMAKE=yes
     TRITON=yes
+    KATEX=yes
+    UCX_COMMIT=${_UCX_COMMIT}
+    UCC_COMMIT=${_UCC_COMMIT}
+    INDUCTOR_BENCHMARKS=yes
+    ;;
+  pytorch-linux-focal-rocm-n-py3)
+    ANACONDA_PYTHON_VERSION=3.10
+    GCC_VERSION=11
+    PROTOBUF=yes
+    DB=yes
+    VISION=yes
+    ROCM_VERSION=6.3
+    NINJA_VERSION=1.9.0
+    CONDA_CMAKE=yes
+    TRITON=yes
+    KATEX=yes
+    UCX_COMMIT=${_UCX_COMMIT}
+    UCC_COMMIT=${_UCC_COMMIT}
+    INDUCTOR_BENCHMARKS=yes
     ;;
   pytorch-linux-jammy-xpu-2024.0-py3)
     ANACONDA_PYTHON_VERSION=3.9
@@ -525,7 +451,7 @@ docker build \
        --build-arg "NINJA_VERSION=${NINJA_VERSION:-}" \
        --build-arg "KATEX=${KATEX:-}" \
        --build-arg "ROCM_VERSION=${ROCM_VERSION:-}" \
-       --build-arg "PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH:-gfx90a}" \
+       --build-arg "PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH:-gfx90a;gfx942}" \
        --build-arg "IMAGE_NAME=${IMAGE_NAME}" \
        --build-arg "UCX_COMMIT=${UCX_COMMIT}" \
        --build-arg "UCC_COMMIT=${UCC_COMMIT}" \

.ci/docker/centos-rocm/Dockerfile

@@ -113,13 +113,6 @@ 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.txt triton_version.txt
 
-# Install AOTriton (Early fail)
-COPY ./aotriton_version.txt aotriton_version.txt
-COPY ./common/common_utils.sh common_utils.sh
-COPY ./common/install_aotriton.sh install_aotriton.sh
-RUN ["/bin/bash", "-c", "./install_aotriton.sh /opt/rocm && rm -rf install_aotriton.sh aotriton_version.txt common_utils.sh"]
-ENV AOTRITON_INSTALLED_PREFIX /opt/rocm/aotriton
-
 # 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

.ci/docker/ci_commit_pins/executorch.txt

@@ -1 +1 @@
-6f638937d64e3396793956d75ee3e14802022745
+5e4d6b6380d575e48e37e9d987fded4ec588e7bc

.ci/docker/ci_commit_pins/nccl-cu11.txt

@@ -0,0 +1 @@
+v2.21.5-1

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

@@ -0,0 +1 @@
+v2.25.1-1

.ci/docker/ci_commit_pins/timm.txt

@@ -1 +1 @@
-ac3470188b914c5d7a5058a7e28b9eb685a62427
+5d535d7a2d4b435b1b5c1177fd8f04a12b942b9a

.ci/docker/ci_commit_pins/triton.txt

@@ -1 +1 @@
-35c6c7c6284582b3f41c71c150e11b517acf074a
+4b3bb1f8da0ded6ccd572dd1358ef45af5a1befe

.ci/docker/common/install_acl.sh

@@ -1,7 +1,7 @@
 set -euo pipefail
 
 readonly version=v24.04
-readonly src_host=https://review.mlplatform.org/ml
+readonly src_host=https://github.com/ARM-software
 readonly src_repo=ComputeLibrary
 
 # Clone ACL

.ci/docker/common/install_aotriton.sh

@@ -1,23 +0,0 @@
-#!/bin/bash
-
-set -ex
-
-source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
-
-TARBALL='aotriton.tar.gz'
-# This read command alwasy returns with exit code 1
-read -d "\n" VER MANYLINUX ROCMBASE PINNED_COMMIT SHA256 < aotriton_version.txt || true
-ARCH=$(uname -m)
-AOTRITON_INSTALL_PREFIX="$1"
-AOTRITON_URL="https://github.com/ROCm/aotriton/releases/download/${VER}/aotriton-${VER}-${MANYLINUX}_${ARCH}-${ROCMBASE}-shared.tar.gz"
-
-cd "${AOTRITON_INSTALL_PREFIX}"
-# Must use -L to follow redirects
-curl -L --retry 3 -o "${TARBALL}" "${AOTRITON_URL}"
-ACTUAL_SHA256=$(sha256sum "${TARBALL}" | cut -d " " -f 1)
-if [ "${SHA256}" != "${ACTUAL_SHA256}" ]; then
-  echo -n "Error: The SHA256 of downloaded tarball is ${ACTUAL_SHA256},"
-  echo " which does not match the expected value ${SHA256}."
-  exit
-fi
-tar xf "${TARBALL}" && rm -rf "${TARBALL}"

.ci/docker/common/install_base.sh

@@ -32,8 +32,12 @@ install_ubuntu() {
 
   # 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=2.25.1-1+cuda12.4 libnccl-dev=2.25.1-1+cuda12.4 --allow-downgrades --allow-change-held-packages"
   else
     maybe_libnccl_dev=""
   fi

.ci/docker/common/install_cache.sh

@@ -9,7 +9,7 @@ install_ubuntu() {
   # Instead use lib and headers from OpenSSL1.1 installed in `install_openssl.sh``
   apt-get install -y cargo
   echo "Checking out sccache repo"
-  git clone https://github.com/mozilla/sccache -b v0.8.2
+  git clone https://github.com/mozilla/sccache -b v0.9.1
   cd sccache
   echo "Building sccache"
   cargo build --release
@@ -36,11 +36,7 @@ sed -e 's|PATH="\(.*\)"|PATH="/opt/cache/bin:\1"|g' -i /etc/environment
 export PATH="/opt/cache/bin:$PATH"
 
 # Setup compiler cache
-if [ -n "$ROCM_VERSION" ]; then
-  curl --retry 3 http://repo.radeon.com/misc/.sccache_amd/sccache -o /opt/cache/bin/sccache
-else
-  install_ubuntu
-fi
+install_ubuntu
 chmod a+x /opt/cache/bin/sccache
 
 function write_sccache_stub() {

.ci/docker/common/install_cpython.sh

@@ -70,7 +70,7 @@ function do_cpython_build {
     # install setuptools since python 3.12 is required to use distutils
     ${prefix}/bin/pip install wheel==0.34.2 setuptools==68.2.2
     local abi_tag=$(${prefix}/bin/python -c "from wheel.pep425tags import get_abbr_impl, get_impl_ver, get_abi_tag; print('{0}{1}-{2}'.format(get_abbr_impl(), get_impl_ver(), get_abi_tag()))")
-    ln -s ${prefix} /opt/python/${abi_tag}
+    ln -sf ${prefix} /opt/python/${abi_tag}
 }
 
 function build_cpython {

.ci/docker/common/install_cuda.sh

@@ -2,7 +2,7 @@
 
 set -ex
 
-NCCL_VERSION=v2.21.5-1
+NCCL_VERSION=v2.25.1-1
 CUDNN_VERSION=9.5.1.17
 
 function install_cusparselt_040 {
@@ -16,17 +16,6 @@ function install_cusparselt_040 {
     rm -rf tmp_cusparselt
 }
 
-function install_cusparselt_052 {
-    # cuSparseLt license: https://docs.nvidia.com/cuda/cusparselt/license.html
-    mkdir tmp_cusparselt && pushd tmp_cusparselt
-    wget -q https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-x86_64/libcusparse_lt-linux-x86_64-0.5.2.1-archive.tar.xz
-    tar xf libcusparse_lt-linux-x86_64-0.5.2.1-archive.tar.xz
-    cp -a libcusparse_lt-linux-x86_64-0.5.2.1-archive/include/* /usr/local/cuda/include/
-    cp -a libcusparse_lt-linux-x86_64-0.5.2.1-archive/lib/* /usr/local/cuda/lib64/
-    popd
-    rm -rf tmp_cusparselt
-}
-
 function install_cusparselt_062 {
     # cuSparseLt license: https://docs.nvidia.com/cuda/cusparselt/license.html
     mkdir tmp_cusparselt && pushd tmp_cusparselt
@@ -51,6 +40,7 @@ function install_cusparselt_063 {
 
 function install_118 {
     CUDNN_VERSION=9.1.0.70
+    NCCL_VERSION=v2.21.5-1
     echo "Installing CUDA 11.8 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.4.0"
     rm -rf /usr/local/cuda-11.8 /usr/local/cuda
     # install CUDA 11.8.0 in the same container
@@ -83,39 +73,6 @@ function install_118 {
     ldconfig
 }
 
-function install_121 {
-    echo "Installing CUDA 12.1 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.5.2"
-    rm -rf /usr/local/cuda-12.1 /usr/local/cuda
-    # install CUDA 12.1.0 in the same container
-    wget -q https://developer.download.nvidia.com/compute/cuda/12.1.1/local_installers/cuda_12.1.1_530.30.02_linux.run
-    chmod +x cuda_12.1.1_530.30.02_linux.run
-    ./cuda_12.1.1_530.30.02_linux.run --toolkit --silent
-    rm -f cuda_12.1.1_530.30.02_linux.run
-    rm -f /usr/local/cuda && ln -s /usr/local/cuda-12.1 /usr/local/cuda
-
-    # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
-    mkdir tmp_cudnn && cd tmp_cudnn
-    wget -q https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive.tar.xz -O cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive.tar.xz
-    tar xf cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive.tar.xz
-    cp -a cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive/include/* /usr/local/cuda/include/
-    cp -a cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive/lib/* /usr/local/cuda/lib64/
-    cd ..
-    rm -rf tmp_cudnn
-
-    # NCCL license: https://docs.nvidia.com/deeplearning/nccl/#licenses
-    # Follow build: https://github.com/NVIDIA/nccl/tree/master?tab=readme-ov-file#build
-    git clone -b $NCCL_VERSION --depth 1 https://github.com/NVIDIA/nccl.git
-    cd nccl && make -j src.build
-    cp -a build/include/* /usr/local/cuda/include/
-    cp -a build/lib/* /usr/local/cuda/lib64/
-    cd ..
-    rm -rf nccl
-
-    install_cusparselt_052
-
-    ldconfig
-}
-
 function install_124 {
   CUDNN_VERSION=9.1.0.70
   echo "Installing CUDA 12.4.1 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.6.2"
@@ -214,37 +171,6 @@ function prune_118 {
     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_121 {
-  echo "Pruning CUDA 12.1"
-  #####################################################################################
-  # CUDA 12.1 prune static libs
-  #####################################################################################
-    export NVPRUNE="/usr/local/cuda-12.1/bin/nvprune"
-    export CUDA_LIB_DIR="/usr/local/cuda-12.1/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
-
-    # 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 12.1 prune visual tools
-    #####################################################################################
-    export CUDA_BASE="/usr/local/cuda-12.1/"
-    rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2023.1.0 $CUDA_BASE/nsight-systems-2023.1.2/
-}
-
 function prune_124 {
   echo "Pruning CUDA 12.4"
   #####################################################################################
@@ -313,18 +239,52 @@ function prune_126 {
   rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2024.3.2 $CUDA_BASE/nsight-systems-2024.5.1/
 }
 
+function install_128 {
+  CUDNN_VERSION=9.7.1.26
+  echo "Installing CUDA 12.8.0 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.6.3"
+  rm -rf /usr/local/cuda-12.8 /usr/local/cuda
+  # install CUDA 12.8.0 in the same container
+  wget -q https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda_12.8.0_570.86.10_linux.run
+  chmod +x cuda_12.8.0_570.86.10_linux.run
+  ./cuda_12.8.0_570.86.10_linux.run --toolkit --silent
+  rm -f cuda_12.8.0_570.86.10_linux.run
+  rm -f /usr/local/cuda && ln -s /usr/local/cuda-12.8 /usr/local/cuda
+
+  # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
+  mkdir tmp_cudnn && cd tmp_cudnn
+  wget -q https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive.tar.xz -O cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive.tar.xz
+  tar xf cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive.tar.xz
+  cp -a cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive/include/* /usr/local/cuda/include/
+  cp -a cudnn-linux-x86_64-${CUDNN_VERSION}_cuda12-archive/lib/* /usr/local/cuda/lib64/
+  cd ..
+  rm -rf tmp_cudnn
+
+  # NCCL license: https://docs.nvidia.com/deeplearning/nccl/#licenses
+  # Follow build: https://github.com/NVIDIA/nccl/tree/master?tab=readme-ov-file#build
+  git clone -b $NCCL_VERSION --depth 1 https://github.com/NVIDIA/nccl.git
+  cd nccl && make -j src.build
+  cp -a build/include/* /usr/local/cuda/include/
+  cp -a build/lib/* /usr/local/cuda/lib64/
+  cd ..
+  rm -rf nccl
+
+  install_cusparselt_063
+
+  ldconfig
+}
+
 # idiomatic parameter and option handling in sh
 while test $# -gt 0
 do
     case "$1" in
     11.8) install_118; prune_118
         ;;
-    12.1) install_121; prune_121
-        ;;
     12.4) install_124; prune_124
         ;;
     12.6) install_126; prune_126
         ;;
+    12.8) install_128;
+        ;;
     *) echo "bad argument $1"; exit 1
         ;;
     esac

.ci/docker/common/install_cuda_aarch64.sh

@@ -57,7 +57,7 @@ function install_124 {
   cd ..
   rm -rf nccl
 
-  install_cusparselt_062
+  install_cusparselt_063
 
   ldconfig
 }
@@ -160,6 +160,40 @@ function prune_126 {
   rm -rf $CUDA_BASE/libnvvp $CUDA_BASE/nsightee_plugins $CUDA_BASE/nsight-compute-2024.3.2 $CUDA_BASE/nsight-systems-2024.5.1/
 }
 
+function install_128 {
+  CUDNN_VERSION=9.7.1.26
+  echo "Installing CUDA 12.8.0 and cuDNN ${CUDNN_VERSION} and NCCL ${NCCL_VERSION} and cuSparseLt-0.6.3"
+  rm -rf /usr/local/cuda-12.8 /usr/local/cuda
+  # install CUDA 12.8.0 in the same container
+  wget -q https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda_12.8.0_570.86.10_linux_sbsa.run
+  chmod +x cuda_12.8.0_570.86.10_linux_sbsa.run
+  ./cuda_12.8.0_570.86.10_linux_sbsa.run --toolkit --silent
+  rm -f cuda_12.8.0_570.86.10_linux_sbsa.run
+  rm -f /usr/local/cuda && ln -s /usr/local/cuda-12.8 /usr/local/cuda
+
+  # cuDNN license: https://developer.nvidia.com/cudnn/license_agreement
+  mkdir tmp_cudnn && cd tmp_cudnn
+  wget -q https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-sbsa/cudnn-linux-sbsa-${CUDNN_VERSION}_cuda12-archive.tar.xz -O cudnn-linux-sbsa-${CUDNN_VERSION}_cuda12-archive.tar.xz
+  tar xf cudnn-linux-sbsa-${CUDNN_VERSION}_cuda12-archive.tar.xz
+  cp -a cudnn-linux-sbsa-${CUDNN_VERSION}_cuda12-archive/include/* /usr/local/cuda/include/
+  cp -a cudnn-linux-sbsa-${CUDNN_VERSION}_cuda12-archive/lib/* /usr/local/cuda/lib64/
+  cd ..
+  rm -rf tmp_cudnn
+
+  # NCCL license: https://docs.nvidia.com/deeplearning/nccl/#licenses
+  # Follow build: https://github.com/NVIDIA/nccl/tree/master?tab=readme-ov-file#build
+  git clone -b ${NCCL_VERSION} --depth 1 https://github.com/NVIDIA/nccl.git
+  cd nccl && make -j src.build
+  cp -a build/include/* /usr/local/cuda/include/
+  cp -a build/lib/* /usr/local/cuda/lib64/
+  cd ..
+  rm -rf nccl
+
+  install_cusparselt_063
+
+  ldconfig
+}
+
 # idiomatic parameter and option handling in sh
 while test $# -gt 0
 do
@@ -168,6 +202,8 @@ do
         ;;
     12.6) install_126; prune_126
         ;;
+    12.8) install_128;
+        ;;
     *) echo "bad argument $1"; exit 1
         ;;
     esac

.ci/docker/common/install_cudnn.sh

@@ -4,7 +4,9 @@ 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} == "12.6" ]]; then
+    if [[ ${CUDA_VERSION:0:4} == "12.8" ]]; then
+        CUDNN_NAME="cudnn-linux-x86_64-9.7.1.26_cuda12-archive"
+    elif [[ ${CUDA_VERSION:0:4} == "12.6" ]]; then
         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"

.ci/docker/common/install_cusparselt.sh

@@ -5,7 +5,15 @@ set -ex
 # cuSPARSELt license: https://docs.nvidia.com/cuda/cusparselt/license.html
 mkdir tmp_cusparselt && cd tmp_cusparselt
 
-if [[ ${CUDA_VERSION:0:4} =~ ^12\.[2-6]$ ]]; 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-${arch_path}-0.6.3.2-archive"
+    curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-${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
@@ -13,17 +21,11 @@ if [[ ${CUDA_VERSION:0:4} =~ ^12\.[2-6]$ ]]; then
     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 [[ ${CUDA_VERSION:0:4} == "12.1" ]]; then
-    arch_path='sbsa'
-    export TARGETARCH=${TARGETARCH:-$(uname -m)}
-    if [ ${TARGETARCH} = 'amd64' ] || [ "${TARGETARCH}" = 'x86_64' ]; then
-        arch_path='x86_64'
-    fi
-    CUSPARSELT_NAME="libcusparse_lt-linux-${arch_path}-0.5.2.1-archive"
-    curl --retry 3 -OLs https://developer.download.nvidia.com/compute/cusparselt/redist/libcusparse_lt/linux-${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
 
 tar xf ${CUSPARSELT_NAME}.tar.xz

.ci/docker/common/install_executorch.sh

@@ -37,7 +37,12 @@ install_conda_dependencies() {
 
 install_pip_dependencies() {
   pushd executorch
-  as_jenkins bash install_requirements.sh --pybind xnnpack
+  as_jenkins bash install_executorch.sh
+
+  # A workaround, ExecuTorch has moved to numpy 2.0 which is not compatible with the current
+  # numba and scipy version used in PyTorch CI
+  conda_run pip uninstall -y numba scipy
+
   popd
 }
 

.ci/docker/common/install_onnx.sh

@@ -31,15 +31,15 @@ pip_install \
 pip_install coloredlogs packaging
 
 pip_install onnxruntime==1.18.1
-pip_install onnx==1.16.2
-pip_install onnxscript==0.1.0.dev20241124 --no-deps
+pip_install onnx==1.17.0
+pip_install onnxscript==0.1.0 --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/
 IMPORT_SCRIPT_FILENAME="/tmp/onnx_import_script.py"
-as_jenkins echo 'import transformers; transformers.AutoModel.from_pretrained("sshleifer/tiny-gpt2"); transformers.AutoTokenizer.from_pretrained("sshleifer/tiny-gpt2"); transformers.AutoModelForSpeechSeq2Seq.from_pretrained("openai/whisper-large-v3");' > "${IMPORT_SCRIPT_FILENAME}"
+as_jenkins echo 'import transformers; transformers.GPTJForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gptj");' > "${IMPORT_SCRIPT_FILENAME}"
 
 # Need a PyTorch version for transformers to work
 pip_install --pre torch --index-url https://download.pytorch.org/whl/nightly/cpu

.ci/docker/common/install_rocm.sh

@@ -62,6 +62,22 @@ install_ubuntu() {
         sqlite3 $kdb "PRAGMA journal_mode=off; PRAGMA VACUUM;"
     done
 
+    # ROCm 6.3 had a regression where initializing static code objects had significant overhead
+    if [[ $(ver $ROCM_VERSION) -eq $(ver 6.3) ]]; then
+        # clr build needs CppHeaderParser but can only find it using conda's python
+        /opt/conda/bin/python -m pip install CppHeaderParser
+        git clone https://github.com/ROCm/HIP -b rocm-6.3.x
+        HIP_COMMON_DIR=$(readlink -f HIP)
+        git clone https://github.com/jeffdaily/clr -b release/rocm-rel-6.3-statco-hotfix
+        mkdir -p clr/build
+        pushd clr/build
+        cmake .. -DCLR_BUILD_HIP=ON -DHIP_COMMON_DIR=$HIP_COMMON_DIR
+        make -j
+        cp hipamd/lib/libamdhip64.so.6.3.* /opt/rocm/lib/libamdhip64.so.6.3.*
+        popd
+        rm -rf HIP clr
+    fi
+
     # Cleanup
     apt-get autoclean && apt-get clean
     rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

.ci/docker/common/install_ucc.sh

@@ -8,6 +8,12 @@ else
   with_cuda=no
 fi
 
+if [[ -d "/opt/rocm" ]]; then
+  with_rocm=/opt/rocm
+else
+  with_rocm=no
+fi
+
 function install_ucx() {
   set -ex
   git clone --recursive https://github.com/openucx/ucx.git
@@ -19,6 +25,7 @@ function install_ucx() {
   ./configure --prefix=$UCX_HOME      \
       --enable-mt                     \
       --with-cuda=$with_cuda          \
+      --with-rocm=$with_rocm          \
       --enable-profiling              \
       --enable-stats
   time make -j
@@ -36,12 +43,29 @@ function install_ucc() {
   git submodule update --init --recursive
 
   ./autogen.sh
+
   # We only run distributed tests on Tesla M60 and A10G
   NVCC_GENCODE="-gencode=arch=compute_52,code=sm_52 -gencode=arch=compute_86,code=compute_86"
+
+  if [[ -n "$ROCM_VERSION" ]]; then
+    if [[ -n "$PYTORCH_ROCM_ARCH" ]]; then
+      amdgpu_targets=`echo $PYTORCH_ROCM_ARCH | sed 's/;/ /g'`
+    else
+      amdgpu_targets=`rocm_agent_enumerator | grep -v gfx000 | sort -u | xargs`
+    fi
+    for arch in $amdgpu_targets; do
+      HIP_OFFLOAD="$HIP_OFFLOAD --offload-arch=$arch"
+    done
+  else
+    HIP_OFFLOAD="all-arch-no-native"
+  fi
+
   ./configure --prefix=$UCC_HOME          \
     --with-ucx=$UCX_HOME                  \
     --with-cuda=$with_cuda                \
-    --with-nvcc-gencode="${NVCC_GENCODE}"
+    --with-nvcc-gencode="${NVCC_GENCODE}" \
+    --with-rocm=$with_rocm                \
+    --with-rocm-arch="${HIP_OFFLOAD}"
   time make -j
   sudo make install
 

.ci/docker/libtorch/Dockerfile

@@ -56,11 +56,6 @@ 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 cuda12.1
-RUN bash ./install_cuda.sh 12.1
-RUN bash ./install_magma.sh 12.1
-RUN ln -sf /usr/local/cuda-12.1 /usr/local/cuda
-
 FROM cuda as cuda12.4
 RUN bash ./install_cuda.sh 12.4
 RUN bash ./install_magma.sh 12.4
@@ -71,6 +66,11 @@ RUN bash ./install_cuda.sh 12.6
 RUN bash ./install_magma.sh 12.6
 RUN ln -sf /usr/local/cuda-12.6 /usr/local/cuda
 
+FROM cuda as cuda12.8
+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 cpu as rocm
 ARG PYTORCH_ROCM_ARCH
 ENV PYTORCH_ROCM_ARCH ${PYTORCH_ROCM_ARCH}
@@ -92,13 +92,6 @@ RUN apt-get update -y && \
 RUN bash ./install_rocm_drm.sh && rm install_rocm_drm.sh
 RUN bash ./install_rocm_magma.sh && rm install_rocm_magma.sh
 
-# Install AOTriton
-COPY ./common/common_utils.sh common_utils.sh
-COPY ./aotriton_version.txt aotriton_version.txt
-COPY ./common/install_aotriton.sh install_aotriton.sh
-RUN bash ./install_aotriton.sh /opt/rocm && rm install_aotriton.sh aotriton_version.txt
-ENV AOTRITON_INSTALLED_PREFIX /opt/rocm/aotriton
-
 FROM ${BASE_TARGET} as final
 COPY --from=openssl            /opt/openssl           /opt/openssl
 # Install patchelf

.ci/docker/manywheel/Dockerfile

@@ -198,10 +198,3 @@ ADD ./common/install_rocm_magma.sh install_rocm_magma.sh
 RUN bash ./install_rocm_magma.sh && rm install_rocm_magma.sh
 ADD ./common/install_miopen.sh install_miopen.sh
 RUN bash ./install_miopen.sh ${ROCM_VERSION} && rm install_miopen.sh
-
-# Install AOTriton
-COPY ./common/common_utils.sh common_utils.sh
-COPY ./aotriton_version.txt aotriton_version.txt
-COPY ./common/install_aotriton.sh install_aotriton.sh
-RUN bash ./install_aotriton.sh /opt/rocm && rm install_aotriton.sh aotriton_version.txt
-ENV AOTRITON_INSTALLED_PREFIX /opt/rocm/aotriton

.ci/docker/requirements-ci.txt

@@ -30,10 +30,10 @@ dill==0.3.7
 #Pinned versions: 0.3.7
 #test that import: dynamo/test_replay_record.py test_dataloader.py test_datapipe.py test_serialization.py
 
-expecttest==0.2.1
+expecttest==0.3.0
 #Description: method for writing tests where test framework auto populates
 # the expected output based on previous runs
-#Pinned versions: 0.2.1
+#Pinned versions: 0.3.0
 #test that import:
 
 fbscribelogger==0.1.7
@@ -280,9 +280,9 @@ unittest-xml-reporting<=3.2.0,>=2.0.0
 #test that import:
 
 #lintrunner is supported on aarch64-linux only from 0.12.4 version
-lintrunner==0.12.5
+lintrunner==0.12.7
 #Description: all about linters!
-#Pinned versions: 0.12.5
+#Pinned versions: 0.12.7
 #test that import:
 
 redis>=4.0.0
@@ -294,7 +294,7 @@ ghstack==0.8.0
 #Pinned versions: 0.8.0
 #test that import:
 
-jinja2==3.1.4
+jinja2==3.1.5
 #Description: jinja2 template engine
 #Pinned versions: 3.1.4
 #test that import:
@@ -304,7 +304,7 @@ pytest-cpp==2.3.0
 #Pinned versions: 2.3.0
 #test that import:
 
-z3-solver==4.12.2.0
+z3-solver==4.12.6.0
 #Description: The Z3 Theorem Prover Project
 #Pinned versions:
 #test that import:
@@ -329,7 +329,7 @@ lxml==5.3.0
 
 PyGithub==2.3.0
 
-sympy==1.13.1 ; python_version >= "3.9"
+sympy==1.13.3
 #Description: Required by coremltools, also pinned in .github/requirements/pip-requirements-macOS.txt
 #Pinned versions:
 #test that import:
@@ -339,7 +339,7 @@ onnx==1.17.0
 #Pinned versions:
 #test that import:
 
-onnxscript==0.1.0.dev20240817
+onnxscript==0.1.0
 #Description: Required by mypy and test_public_bindings.py when checking torch.onnx._internal
 #Pinned versions:
 #test that import:
@@ -362,6 +362,7 @@ pwlf==2.2.1 ; python_version >= "3.8"
 # To build PyTorch itself
 astunparse
 PyYAML
+pyzstd
 setuptools
 
 ninja==1.11.1 ; platform_machine == "aarch64"
@@ -371,3 +372,8 @@ pulp==2.9.0 ; python_version >= "3.8"
 #Description: required for testing ilp formulaiton under torch/distributed/_tools
 #Pinned versions: 2.9.0
 #test that import: test_sac_ilp.py
+
+dataclasses_json==0.6.7
+#Description: required for data pipeline and scripts under tools/stats
+#Pinned versions: 0.6.7
+#test that import:

.ci/docker/ubuntu-rocm/Dockerfile

@@ -14,21 +14,20 @@ ENV PYTORCH_ROCM_ARCH ${PYTORCH_ROCM_ARCH}
 COPY ./common/install_base.sh install_base.sh
 RUN bash ./install_base.sh && rm install_base.sh
 
-# Install clang
-ARG LLVMDEV
-ARG CLANG_VERSION
-COPY ./common/install_clang.sh install_clang.sh
-RUN bash ./install_clang.sh && rm install_clang.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
-ARG CONDA_CMAKE
 ENV ANACONDA_PYTHON_VERSION=$ANACONDA_PYTHON_VERSION
 ENV PATH /opt/conda/envs/py_$ANACONDA_PYTHON_VERSION/bin:/opt/conda/bin:$PATH
+ARG CONDA_CMAKE
 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
@@ -39,6 +38,11 @@ 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 protobuf for ONNX
 ARG PROTOBUF
 COPY ./common/install_protobuf.sh install_protobuf.sh
@@ -85,6 +89,32 @@ COPY ./common/install_amdsmi.sh install_amdsmi.sh
 RUN bash ./install_amdsmi.sh
 RUN rm install_amdsmi.sh
 
+# (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 "${INDUCTOR_BENCHMARKS}" ]; then bash ./install_inductor_benchmark_deps.sh; fi
+RUN rm install_inductor_benchmark_deps.sh common_utils.sh timm.txt huggingface.txt
+
 # (optional) Install non-default CMake version
 ARG CMAKE_VERSION
 COPY ./common/install_cmake.sh install_cmake.sh
@@ -107,18 +137,17 @@ COPY triton_version.txt triton_version.txt
 RUN if [ -n "${TRITON}" ]; then bash ./install_triton.sh; fi
 RUN rm install_triton.sh common_utils.sh triton.txt triton_version.txt
 
-# Install AOTriton
-COPY ./aotriton_version.txt aotriton_version.txt
-COPY ./common/common_utils.sh common_utils.sh
-COPY ./common/install_aotriton.sh install_aotriton.sh
-RUN ["/bin/bash", "-c", "./install_aotriton.sh /opt/rocm && rm -rf install_aotriton.sh aotriton_version.txt common_utils.sh"]
-ENV AOTRITON_INSTALLED_PREFIX /opt/rocm/aotriton
 
 # 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
 RUN bash ./install_cache.sh && rm install_cache.sh
 
+# Install Open MPI for ROCm
+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}

.ci/magma/Makefile

@@ -16,9 +16,9 @@ DOCKER_RUN = set -eou pipefail; ${DOCKER_CMD} run --rm -i \
 	magma/build_magma.sh
 
 .PHONY: all
+all: magma-cuda128
 all: magma-cuda126
 all: magma-cuda124
-all: magma-cuda121
 all: magma-cuda118
 
 .PHONY:
@@ -26,6 +26,12 @@ clean:
 	$(RM) -r magma-*
 	$(RM) -r output
 
+.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
+magma-cuda128:
+	$(DOCKER_RUN)
+
 .PHONY: magma-cuda126
 magma-cuda126: DESIRED_CUDA := 12.6
 magma-cuda126:
@@ -36,11 +42,6 @@ magma-cuda124: DESIRED_CUDA := 12.4
 magma-cuda124:
 	$(DOCKER_RUN)
 
-.PHONY: magma-cuda121
-magma-cuda121: DESIRED_CUDA := 12.1
-magma-cuda121:
-	$(DOCKER_RUN)
-
 .PHONY: magma-cuda118
 magma-cuda118: DESIRED_CUDA := 11.8
 magma-cuda118: CUDA_ARCH_LIST += -gencode arch=compute_37,code=sm_37

.ci/manywheel/build_cuda.sh

@@ -14,6 +14,7 @@ export USE_CUDA_STATIC_LINK=1
 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}
 
 # Keep an array of cmake variables to add to
 if [[ -z "$CMAKE_ARGS" ]]; then
@@ -43,13 +44,6 @@ if [[ -n "$DESIRED_CUDA" ]]; then
         fi
     fi
     echo "Using CUDA $CUDA_VERSION as determined by DESIRED_CUDA"
-
-    # There really has to be a better way to do this - eli
-    # Possibly limiting builds to specific cuda versions be delimiting images would be a choice
-    if [[ "$OS_NAME" == *"Ubuntu"* ]]; then
-        echo "Switching to CUDA version ${DESIRED_CUDA}"
-        /builder/conda/switch_cuda_version.sh "${DESIRED_CUDA}"
-    fi
 else
     CUDA_VERSION=$(nvcc --version|grep release|cut -f5 -d" "|cut -f1 -d",")
     echo "CUDA $CUDA_VERSION Detected"
@@ -59,23 +53,15 @@ cuda_version_nodot=$(echo $CUDA_VERSION | tr -d '.')
 
 TORCH_CUDA_ARCH_LIST="5.0;6.0;7.0;7.5;8.0;8.6"
 case ${CUDA_VERSION} in
+    12.8)
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0;10.0;12.0+PTX" #Ripping out 5.0 and 6.0 due to ld error
+        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
+        ;;
     12.6)
-        if [[ "$GPU_ARCH_TYPE" = "cuda-aarch64" ]]; then
-            TORCH_CUDA_ARCH_LIST="9.0"
-        else
-            TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0+PTX"
-        fi
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0+PTX"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
     12.4)
-        if [[ "$GPU_ARCH_TYPE" = "cuda-aarch64" ]]; then
-            TORCH_CUDA_ARCH_LIST="9.0"
-        else
-            TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
-        fi
-        EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
-        ;;
-    12.1)
         TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};9.0"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
@@ -133,7 +119,16 @@ if [[ $USE_CUSPARSELT == "1" && $CUDA_VERSION == "11.8" ]]; then
         )
 fi
 
-if [[ $CUDA_VERSION == "12.4" || $CUDA_VERSION == "12.6" ]]; then
+
+# Turn USE_CUFILE off for CUDA 11.8, 12.4 since nvidia-cufile-cu11 and 1.9.0.20 are
+# not available in PYPI
+if [[ $CUDA_VERSION == "11.8" || $CUDA_VERSION == "12.4" ]]; then
+    export USE_CUFILE=0
+fi
+
+
+# CUDA_VERSION 12.4, 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"
@@ -174,6 +169,16 @@ if [[ $CUDA_VERSION == "12.4" || $CUDA_VERSION == "12.6" ]]; then
             "libnvrtc.so.12"
             "libnvrtc-builtins.so"
         )
+        if [[ $USE_CUFILE == 1 ]]; then
+            DEPS_LIST+=(
+                "/usr/local/cuda/lib64/libcufile.so.0"
+                "/usr/local/cuda/lib64/libcufile_rdma.so.1"
+            )
+            DEPS_SONAME+=(
+                "libcufile.so.0"
+                "libcufile_rdma.so.1"
+            )
+        fi
     else
         echo "Using nvidia libs from pypi."
         CUDA_RPATHS=(
@@ -190,6 +195,11 @@ if [[ $CUDA_VERSION == "12.4" || $CUDA_VERSION == "12.6" ]]; then
             '$ORIGIN/../../nvidia/nccl/lib'
             '$ORIGIN/../../nvidia/nvtx/lib'
         )
+        if [[ $USE_CUFILE == 1 ]]; then
+            CUDA_RPATHS+=(
+                '$ORIGIN/../../nvidia/cufile/lib'
+            )
+        fi
         CUDA_RPATHS=$(IFS=: ; echo "${CUDA_RPATHS[*]}")
         export C_SO_RPATH=$CUDA_RPATHS':$ORIGIN:$ORIGIN/lib'
         export LIB_SO_RPATH=$CUDA_RPATHS':$ORIGIN'
@@ -275,7 +285,7 @@ else
     exit 1
 fi
 
-# builder/test.sh requires DESIRED_CUDA to know what tests to exclude
+# run_tests.sh requires DESIRED_CUDA to know what tests to exclude
 export DESIRED_CUDA="$cuda_version_nodot"
 
 # Switch `/usr/local/cuda` to the desired CUDA version

.ci/manywheel/build_rocm.sh

@@ -118,7 +118,7 @@ if [[ "$OS_NAME" == *"CentOS Linux"* || "$OS_NAME" == *"AlmaLinux"* ]]; then
     fi
     LIBDRM_PATH="/opt/amdgpu/lib64/libdrm.so.2"
     LIBDRM_AMDGPU_PATH="/opt/amdgpu/lib64/libdrm_amdgpu.so.1"
-    if [[ $ROCM_INT -ge 60100 ]]; then
+    if [[ $ROCM_INT -ge 60100 && $ROCM_INT -lt 60300 ]]; then
         # Below libs are direct dependencies of libhipsolver
         LIBSUITESPARSE_CONFIG_PATH="/lib64/libsuitesparseconfig.so.4"
         if [[ "$OS_NAME" == *"CentOS Linux"* ]]; then
@@ -151,7 +151,7 @@ elif [[ "$OS_NAME" == *"Ubuntu"* ]]; then
     fi
     LIBDRM_PATH="/usr/lib/x86_64-linux-gnu/libdrm.so.2"
     LIBDRM_AMDGPU_PATH="/usr/lib/x86_64-linux-gnu/libdrm_amdgpu.so.1"
-    if [[ $ROCM_INT -ge 60100 ]]; then
+    if [[ $ROCM_INT -ge 60100 && $ROCM_INT -lt 60300 ]]; then
         # Below libs are direct dependencies of libhipsolver
         LIBCHOLMOD_PATH="/lib/x86_64-linux-gnu/libcholmod.so.3"
         # Below libs are direct dependencies of libcholmod
@@ -186,15 +186,6 @@ do
     OS_SO_FILES[${#OS_SO_FILES[@]}]=$file_name # Append lib to array
 done
 
-# FIXME: Temporary until https://github.com/pytorch/pytorch/pull/137443 lands
-# Install AOTriton
-if [ -e ${PYTORCH_ROOT}/.ci/docker/aotriton_version.txt ]; then
-    cp -a ${PYTORCH_ROOT}/.ci/docker/aotriton_version.txt aotriton_version.txt
-    bash ${PYTORCH_ROOT}/.ci/docker/common/install_aotriton.sh ${ROCM_HOME} && rm aotriton_version.txt
-    export AOTRITON_INSTALLED_PREFIX=${ROCM_HOME}/aotriton
-    ROCM_SO_FILES+=("libaotriton_v2.so")
-fi
-
 # rocBLAS library files
 ROCBLAS_LIB_SRC=$ROCM_HOME/lib/rocblas/library
 ROCBLAS_LIB_DST=lib/rocblas/library
@@ -266,20 +257,6 @@ RCCL_SHARE_FILES=($(ls $RCCL_SHARE_SRC))
 DEPS_AUX_SRCLIST+=(${RCCL_SHARE_FILES[@]/#/$RCCL_SHARE_SRC/})
 DEPS_AUX_DSTLIST+=(${RCCL_SHARE_FILES[@]/#/$RCCL_SHARE_DST/})
 
-# PyTorch 2.6+ (AOTriton 0.8b+)
-# AKS = "AOTriton Kernel Storage", a file format to store GPU kernels compactly
-if (( $(echo "${PYTORCH_VERSION} 2.6" | awk '{print ($1 >= $2)}') )); then
-    LIBAOTRITON_DIR=$(find "$ROCM_HOME/lib/" -name "libaotriton_v2.so" -printf '%h\n')
-    if [[ -z ${LIBAOTRITON_DIR} ]]; then
-        LIBAOTRITON_DIR=$(find "$ROCM_HOME/" -name "libaotriton_v2.so" -printf '%h\n')
-    fi
-    AKS_FILES=($(find "${LIBAOTRITON_DIR}/aotriton.images" -type f -name '*.aks?' -printf '%P\n'))
-    AKS_SRC="${LIBAOTRITON_DIR}/aotriton.images"
-    AKS_DST="lib/aotriton.images"
-    DEPS_AUX_SRCLIST+=(${AKS_FILES[@]/#/${AKS_SRC}/})
-    DEPS_AUX_DSTLIST+=(${AKS_FILES[@]/#/${AKS_DST}/})
-fi
-
 echo "PYTORCH_ROCM_ARCH: ${PYTORCH_ROCM_ARCH}"
 
 SCRIPTPATH="$( cd "$(dirname "$0")" ; pwd -P )"

.ci/pytorch/build.sh

@@ -228,7 +228,7 @@ if [[ "$BUILD_ENVIRONMENT" == *-debug* ]]; then
   export CMAKE_BUILD_TYPE=RelWithAssert
 fi
 
-# Do not change workspace permissions for ROCm CI jobs
+# Do not change workspace permissions for ROCm and s390x CI jobs
 # as it can leave workspace with bad permissions for cancelled jobs
 if [[ "$BUILD_ENVIRONMENT" != *rocm* && "$BUILD_ENVIRONMENT" != *s390x* && -d /var/lib/jenkins/workspace ]]; then
   # Workaround for dind-rootless userid mapping (https://github.com/pytorch/ci-infra/issues/96)
@@ -247,7 +247,7 @@ if [[ "$BUILD_ENVIRONMENT" != *rocm* && "$BUILD_ENVIRONMENT" != *s390x* && -d /v
 fi
 
 if [[ "$BUILD_ENVIRONMENT" == *-bazel-* ]]; then
-  set -e
+  set -e -o pipefail
 
   get_bazel
 
@@ -278,7 +278,7 @@ else
           "$BUILD_ENVIRONMENT" != *xla* ]]; then
       if [[ "$BUILD_ENVIRONMENT" != *py3.8* ]]; then
         # Install numpy-2.0.2 for builds which are backward compatible with 1.X
-        python -mpip install --pre numpy==2.0.2
+        python -mpip install numpy==2.0.2
       fi
 
       WERROR=1 python setup.py clean

.ci/pytorch/check_binary.sh

@@ -387,7 +387,7 @@ fi
 ###############################################################################
 # Check for C++ ABI compatibility between gcc7 and gcc9 compiled binaries
 ###############################################################################
-if [[ "$(uname)" == 'Linux' && ("$PACKAGE_TYPE" == 'conda' || "$PACKAGE_TYPE" == 'manywheel')]]; then
+if [[ "$(uname)" == 'Linux' &&  "$PACKAGE_TYPE" == 'manywheel' ]]; then
   pushd /tmp
   python -c "import torch; exit(0 if torch.compiled_with_cxx11_abi() else (0 if torch._C._PYBIND11_BUILD_ABI == '_cxxabi1011' else 1))"
   popd

.ci/pytorch/common.sh

@@ -3,7 +3,7 @@
 # Common setup for all Jenkins scripts
 # shellcheck source=./common_utils.sh
 source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
-set -ex
+set -ex -o pipefail
 
 # Required environment variables:
 #   $BUILD_ENVIRONMENT (should be set by your Docker image)

.ci/pytorch/common_utils.sh

@@ -160,7 +160,7 @@ function install_torchvision() {
 }
 
 function install_tlparse() {
-  pip_install --user "tlparse==0.3.25"
+  pip_install --user "tlparse==0.3.30"
   PATH="$(python -m site --user-base)/bin:$PATH"
 }
 
@@ -169,24 +169,34 @@ function install_torchrec_and_fbgemm() {
   torchrec_commit=$(get_pinned_commit torchrec)
   local fbgemm_commit
   fbgemm_commit=$(get_pinned_commit fbgemm)
+  if [[ "$BUILD_ENVIRONMENT" == *rocm* ]] ; then
+    fbgemm_commit=$(get_pinned_commit fbgemm_rocm)
+  fi
   pip_uninstall torchrec-nightly
   pip_uninstall fbgemm-gpu-nightly
   pip_install setuptools-git-versioning scikit-build pyre-extensions
 
-  # TODO (huydhn): I still have no clue on why sccache doesn't work with only fbgemm_gpu here, but it
-  # seems to be an sccache-related issue
-  if [[ "$IS_A100_RUNNER" == "1" ]]; then
-    unset CMAKE_CUDA_COMPILER_LAUNCHER
-    sudo mv /opt/cache/bin /opt/cache/bin-backup
-  fi
+  if [[ "$BUILD_ENVIRONMENT" == *rocm* ]] ; then
+    # install torchrec first because it installs fbgemm nightly on top of rocm fbgemm
+    pip_install --no-use-pep517 --user "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}"
+    pip_uninstall fbgemm-gpu-nightly
 
-  # See https://github.com/pytorch/pytorch/issues/106971
-  CUDA_PATH=/usr/local/cuda-12.1 pip_install --no-use-pep517 --user "git+https://github.com/pytorch/FBGEMM.git@${fbgemm_commit}#egg=fbgemm-gpu&subdirectory=fbgemm_gpu"
-  pip_install --no-use-pep517 --user "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}"
-
-  if [[ "$IS_A100_RUNNER" == "1" ]]; then
-    export CMAKE_CUDA_COMPILER_LAUNCHER=/opt/cache/bin/sccache
-    sudo mv /opt/cache/bin-backup /opt/cache/bin
+    pip_install tabulate  # needed for newer fbgemm
+    pip_install patchelf  # needed for rocm fbgemm
+    git clone --recursive https://github.com/pytorch/fbgemm
+    pushd fbgemm/fbgemm_gpu
+    git checkout "${fbgemm_commit}"
+    python setup.py install \
+      --package_variant=rocm \
+      -DHIP_ROOT_DIR="${ROCM_PATH}" \
+      -DCMAKE_C_FLAGS="-DTORCH_USE_HIP_DSA" \
+      -DCMAKE_CXX_FLAGS="-DTORCH_USE_HIP_DSA"
+    popd
+    rm -rf fbgemm
+  else
+    # See https://github.com/pytorch/pytorch/issues/106971
+    CUDA_PATH=/usr/local/cuda-12.1 pip_install --no-use-pep517 --user "git+https://github.com/pytorch/FBGEMM.git@${fbgemm_commit}#egg=fbgemm-gpu&subdirectory=fbgemm_gpu"
+    pip_install --no-use-pep517 --user "git+https://github.com/pytorch/torchrec.git@${torchrec_commit}"
   fi
 }
 
@@ -216,6 +226,11 @@ function checkout_install_torchbench() {
     # to install and test other models
     python install.py --continue_on_fail
   fi
+
+  # TODO (huydhn): transformers-4.44.2 added by https://github.com/pytorch/benchmark/pull/2488
+  # is regressing speedup metric. This needs to be investigated further
+  pip install transformers==4.38.1
+
   echo "Print all dependencies after TorchBench is installed"
   python -mpip freeze
   popd

.ci/pytorch/cpp_doc_push_script.sh

@@ -40,7 +40,7 @@ echo "Building PyTorch C++ API docs..."
 rm -rf cppdocs
 git clone https://github.com/pytorch/cppdocs
 
-set -ex
+set -ex -o pipefail
 
 # Generate ATen files
 pushd "${pt_checkout}"

.ci/pytorch/functorch_doc_push_script.sh

@@ -5,7 +5,7 @@ pt_checkout="/var/lib/jenkins/workspace"
 source "$pt_checkout/.ci/pytorch/common_utils.sh"
 echo "functorch_doc_push_script.sh: Invoked with $*"
 
-set -ex
+set -ex -o pipefail
 
 version=${DOCS_VERSION:-nightly}
 echo "version: $version"

.ci/pytorch/install_cache_xla.sh

@@ -6,7 +6,7 @@
 # return the same thing, ex checks for for rocm, CUDA, and changing the path
 # where sccache is installed, and not changing /etc/environment.
 
-set -ex
+set -ex -o pipefail
 
 install_binary() {
   echo "Downloading sccache binary from S3 repo"

.ci/pytorch/macos-test.sh

@@ -18,6 +18,9 @@ if [[ ! $(python -c "import torch; print(int(torch.backends.openmp.is_available(
 fi
 popd
 
+# enable debug asserts in serialization
+export TORCH_SERIALIZATION_DEBUG=1
+
 setup_test_python() {
   # The CircleCI worker hostname doesn't resolve to an address.
   # This environment variable makes ProcessGroupGloo default to

.ci/pytorch/multigpu-test.sh

@@ -8,55 +8,62 @@
 source "$(dirname "${BASH_SOURCE[0]}")/common.sh"
 
 echo "Testing pytorch"
-time python test/run_test.py --include test_cuda_multigpu test_cuda_primary_ctx --verbose
+# When adding more tests, please use HUD to see which shard is shorter
+if [[ "${SHARD_NUMBER:-1}" == "1" ]]; then
+    # FSDP tests
+    for f in test/distributed/fsdp/*.py ; do time python test/run_test.py --verbose -i "${f#*/}" ; done
+fi
 
-# Disabling tests to see if they solve timeout issues; see https://github.com/pytorch/pytorch/issues/70015
-# python tools/download_mnist.py --quiet -d test/cpp/api/mnist
-# OMP_NUM_THREADS=2 TORCH_CPP_TEST_MNIST_PATH="test/cpp/api/mnist" build/bin/test_api
-time python test/run_test.py --verbose -i distributed/test_c10d_common
-time python test/run_test.py --verbose -i distributed/test_c10d_gloo
-time python test/run_test.py --verbose -i distributed/test_c10d_nccl
-time python test/run_test.py --verbose -i distributed/test_c10d_spawn_gloo
-time python test/run_test.py --verbose -i distributed/test_c10d_spawn_nccl
-time python test/run_test.py --verbose -i distributed/test_compute_comm_reordering
-time python test/run_test.py --verbose -i distributed/test_store
-time python test/run_test.py --verbose -i distributed/test_symmetric_memory
-time python test/run_test.py --verbose -i distributed/test_pg_wrapper
-time python test/run_test.py --verbose -i distributed/rpc/cuda/test_tensorpipe_agent
-# FSDP tests
-for f in test/distributed/fsdp/*.py ; do time python test/run_test.py --verbose -i "${f#*/}" ; done
-# ShardedTensor tests
-time python test/run_test.py --verbose -i distributed/checkpoint/test_checkpoint
-time python test/run_test.py --verbose -i distributed/checkpoint/test_file_system_checkpoint
-time python test/run_test.py --verbose -i distributed/_shard/sharding_spec/test_sharding_spec
-time python test/run_test.py --verbose -i distributed/_shard/sharding_plan/test_sharding_plan
-time python test/run_test.py --verbose -i distributed/_shard/sharded_tensor/test_sharded_tensor
-time python test/run_test.py --verbose -i distributed/_shard/sharded_tensor/test_sharded_tensor_reshard
+if [[ "${SHARD_NUMBER:-2}" == "2" ]]; then
+    time python test/run_test.py --include test_cuda_multigpu test_cuda_primary_ctx --verbose
 
-# functional collective tests
-time python test/run_test.py --verbose -i distributed/test_functional_api
+    # Disabling tests to see if they solve timeout issues; see https://github.com/pytorch/pytorch/issues/70015
+    # python tools/download_mnist.py --quiet -d test/cpp/api/mnist
+    # OMP_NUM_THREADS=2 TORCH_CPP_TEST_MNIST_PATH="test/cpp/api/mnist" build/bin/test_api
+    time python test/run_test.py --verbose -i distributed/test_c10d_common
+    time python test/run_test.py --verbose -i distributed/test_c10d_gloo
+    time python test/run_test.py --verbose -i distributed/test_c10d_nccl
+    time python test/run_test.py --verbose -i distributed/test_c10d_spawn_gloo
+    time python test/run_test.py --verbose -i distributed/test_c10d_spawn_nccl
+    time python test/run_test.py --verbose -i distributed/test_compute_comm_reordering
+    time python test/run_test.py --verbose -i distributed/test_store
+    time python test/run_test.py --verbose -i distributed/test_symmetric_memory
+    time python test/run_test.py --verbose -i distributed/test_pg_wrapper
+    time python test/run_test.py --verbose -i distributed/rpc/cuda/test_tensorpipe_agent
 
-# DTensor tests
-time python test/run_test.py --verbose -i distributed/_tensor/test_random_ops
-time python test/run_test.py --verbose -i distributed/_tensor/test_dtensor_compile
+    # ShardedTensor tests
+    time python test/run_test.py --verbose -i distributed/checkpoint/test_checkpoint
+    time python test/run_test.py --verbose -i distributed/checkpoint/test_file_system_checkpoint
+    time python test/run_test.py --verbose -i distributed/_shard/sharding_spec/test_sharding_spec
+    time python test/run_test.py --verbose -i distributed/_shard/sharding_plan/test_sharding_plan
+    time python test/run_test.py --verbose -i distributed/_shard/sharded_tensor/test_sharded_tensor
+    time python test/run_test.py --verbose -i distributed/_shard/sharded_tensor/test_sharded_tensor_reshard
 
-# DeviceMesh test
-time python test/run_test.py --verbose -i distributed/test_device_mesh
+    # functional collective tests
+    time python test/run_test.py --verbose -i distributed/test_functional_api
 
-# DTensor/TP tests
-time python test/run_test.py --verbose -i distributed/tensor/parallel/test_tp_examples
-time python test/run_test.py --verbose -i distributed/tensor/parallel/test_tp_random_state
+    # DTensor tests
+    time python test/run_test.py --verbose -i distributed/tensor/test_random_ops
+    time python test/run_test.py --verbose -i distributed/tensor/test_dtensor_compile
 
-# FSDP2 tests
-time python test/run_test.py --verbose -i distributed/_composable/fsdp/test_fully_shard_training -- -k test_2d_mlp_with_nd_mesh
+    # DeviceMesh test
+    time python test/run_test.py --verbose -i distributed/test_device_mesh
 
-# ND composability tests
-time python test/run_test.py --verbose -i distributed/_composable/test_composability/test_2d_composability
-time python test/run_test.py --verbose -i distributed/_composable/test_composability/test_pp_composability
+    # DTensor/TP tests
+    time python test/run_test.py --verbose -i distributed/tensor/parallel/test_tp_examples
+    time python test/run_test.py --verbose -i distributed/tensor/parallel/test_tp_random_state
 
-# Other tests
-time python test/run_test.py --verbose -i test_cuda_primary_ctx
-time python test/run_test.py --verbose -i test_optim -- -k test_forloop_goes_right_direction_multigpu
-time python test/run_test.py --verbose -i test_optim -- -k test_mixed_device_dtype
-time python test/run_test.py --verbose -i test_foreach -- -k test_tensors_grouping
+    # FSDP2 tests
+    time python test/run_test.py --verbose -i distributed/_composable/fsdp/test_fully_shard_training -- -k test_2d_mlp_with_nd_mesh
+
+    # ND composability tests
+    time python test/run_test.py --verbose -i distributed/_composable/test_composability/test_2d_composability
+    time python test/run_test.py --verbose -i distributed/_composable/test_composability/test_pp_composability
+
+    # Other tests
+    time python test/run_test.py --verbose -i test_cuda_primary_ctx
+    time python test/run_test.py --verbose -i test_optim -- -k test_forloop_goes_right_direction_multigpu
+    time python test/run_test.py --verbose -i test_optim -- -k test_mixed_device_dtype
+    time python test/run_test.py --verbose -i test_foreach -- -k test_tensors_grouping
+fi
 assert_git_not_dirty

.ci/pytorch/python_doc_push_script.sh

@@ -7,7 +7,7 @@ source "$pt_checkout/.ci/pytorch/common_utils.sh"
 
 echo "python_doc_push_script.sh: Invoked with $*"
 
-set -ex
+set -ex -o pipefail
 
 # for statements like ${1:-${DOCS_INSTALL_PATH:-docs/}}
 # the order of operations goes:
@@ -63,7 +63,7 @@ build_docs () {
     echo "(tried to echo the WARNINGS above the ==== line)"
     echo =========================
   fi
-  set -ex
+  set -ex -o pipefail
   return $code
 }
 

.ci/pytorch/run_tests.sh

@@ -13,7 +13,7 @@ set -eux -o pipefail
 
 # This script expects to be in the pytorch root folder
 if [[ ! -d 'test' || ! -f 'test/run_test.py' ]]; then
-    echo "builder/test.sh expects to be run from the Pytorch root directory " \
+    echo "run_tests.sh expects to be run from the Pytorch root directory " \
          "but I'm actually in $(pwd)"
     exit 2
 fi
@@ -40,7 +40,7 @@ retry () {
 if [[ "$#" != 3 ]]; then
   if [[ -z "${DESIRED_PYTHON:-}" || -z "${DESIRED_CUDA:-}" || -z "${PACKAGE_TYPE:-}" ]]; then
     echo "USAGE: run_tests.sh  PACKAGE_TYPE  DESIRED_PYTHON  DESIRED_CUDA"
-    echo "The env variable PACKAGE_TYPE must be set to 'conda' or 'manywheel' or 'libtorch'"
+    echo "The env variable PACKAGE_TYPE must be set to 'manywheel' or 'libtorch'"
     echo "The env variable DESIRED_PYTHON must be set like '2.7mu' or '3.6m' etc"
     echo "The env variable DESIRED_CUDA must be set like 'cpu' or 'cu80' etc"
     exit 1

.ci/pytorch/smoke_test/check_binary_symbols.py

@@ -6,7 +6,7 @@ import itertools
 import os
 import re
 from pathlib import Path
-from typing import Any, List, Tuple
+from typing import Any
 
 
 # We also check that there are [not] cxx11 symbols in libtorch
@@ -46,17 +46,17 @@ LIBTORCH_PRE_CXX11_PATTERNS = _apply_libtorch_symbols(PRE_CXX11_SYMBOLS)
 
 
 @functools.lru_cache(100)
-def get_symbols(lib: str) -> List[Tuple[str, str, str]]:
+def get_symbols(lib: str) -> list[tuple[str, str, str]]:
     from subprocess import check_output
 
     lines = check_output(f'nm "{lib}"|c++filt', shell=True)
     return [x.split(" ", 2) for x in lines.decode("latin1").split("\n")[:-1]]
 
 
-def grep_symbols(lib: str, patterns: List[Any]) -> List[str]:
+def grep_symbols(lib: str, patterns: list[Any]) -> list[str]:
     def _grep_symbols(
-        symbols: List[Tuple[str, str, str]], patterns: List[Any]
-    ) -> List[str]:
+        symbols: list[tuple[str, str, str]], patterns: list[Any]
+    ) -> list[str]:
         rc = []
         for _s_addr, _s_type, s_name in symbols:
             for pattern in patterns:

.ci/pytorch/smoke_test/smoke_test.py

@@ -6,6 +6,7 @@ import re
 import subprocess
 import sys
 from pathlib import Path
+from tempfile import NamedTemporaryFile
 
 import torch
 import torch._dynamo
@@ -109,8 +110,10 @@ def check_version(package: str) -> None:
                             {release_matrix[module['name']]} for channel {channel}. But its {module_version}"
                     )
                 else:
-                    print(f"{module['name']} version actual: {module_version} expected: \
-                        {release_matrix[module['name']]} for channel {channel}.")
+                    print(
+                        f"{module['name']} version actual: {module_version} expected: \
+                        {release_matrix[module['name']]} for channel {channel}."
+                    )
 
     else:
         print(f"Skip version check for channel {channel} as stable version is None")
@@ -159,6 +162,32 @@ def test_cuda_runtime_errors_captured() -> None:
         raise RuntimeError("Expected CUDA RuntimeError but have not received!")
 
 
+def test_cuda_gds_errors_captured() -> None:
+    major_version = int(torch.version.cuda.split(".")[0])
+    minor_version = int(torch.version.cuda.split(".")[1])
+
+    if major_version < 12 or (major_version == 12 and minor_version < 6):
+        print("CUDA version is not supported for GDS smoke test")
+        return
+
+    cuda_exception_missed = True
+    try:
+        print("Testing test_cuda_gds_errors_captured")
+        with NamedTemporaryFile() as f:
+            torch.cuda.gds.GdsFile(f.name, os.O_CREAT | os.O_RDWR)
+    except RuntimeError as e:
+        expected_error = "cuFileHandleRegister failed"
+        if re.search(expected_error, f"{e}"):
+            print(f"Caught CUDA exception with success: {e}")
+            cuda_exception_missed = False
+        else:
+            raise e
+    if cuda_exception_missed:
+        raise RuntimeError(
+            "Expected cuFileHandleRegister failed RuntimeError but have not received!"
+        )
+
+
 def smoke_test_cuda(
     package: str, runtime_error_check: str, torch_compile_check: str
 ) -> None:
@@ -180,7 +209,7 @@ def smoke_test_cuda(
     # torch.compile is available on macos-arm64 and Linux for python 3.8-3.13
     if (
         torch_compile_check == "enabled"
-        and sys.version_info < (3, 13, 0)
+        and sys.version_info < (3, 14, 0)
         and target_os in ["linux", "linux-aarch64", "macos-arm64", "darwin"]
     ):
         smoke_test_compile("cuda" if torch.cuda.is_available() else "cpu")
@@ -339,7 +368,7 @@ def smoke_test_modules():
                 print(f"Output: \n{output}\n")
 
 
-def main() -> None:
+def parse_args():
     parser = argparse.ArgumentParser()
     parser.add_argument(
         "--package",
@@ -362,9 +391,16 @@ def main() -> None:
         choices=["enabled", "disabled"],
         default="enabled",
     )
-    options = parser.parse_args()
+    return parser.parse_args()
+
+
+def main() -> None:
+    options = parse_args()
     print(f"torch: {torch.__version__}")
     print(torch.__config__.parallel_info())
+    # All PyTorch binary builds should be built with OpenMP
+    if not torch.backends.openmp.is_available():
+        raise RuntimeError("PyTorch must be built with OpenMP support")
 
     check_version(options.package)
     smoke_test_conv2d()
@@ -372,6 +408,7 @@ def main() -> None:
     test_numpy()
     if is_cuda_system:
         test_linalg("cuda")
+        test_cuda_gds_errors_captured()
 
     if options.package == "all":
         smoke_test_modules()

.ci/pytorch/test.sh

@@ -4,7 +4,7 @@
 # (This is set by default in the Docker images we build, so you don't
 # need to set it yourself.
 
-set -ex
+set -ex -o pipefail
 
 # Suppress ANSI color escape sequences
 export TERM=vt100
@@ -12,9 +12,9 @@ export TERM=vt100
 # shellcheck source=./common.sh
 source "$(dirname "${BASH_SOURCE[0]}")/common.sh"
 
-# Do not change workspace permissions for ROCm CI jobs
+# Do not change workspace permissions for ROCm and s390x CI jobs
 # as it can leave workspace with bad permissions for cancelled jobs
-if [[ "$BUILD_ENVIRONMENT" != *rocm* && -d /var/lib/jenkins/workspace ]]; then
+if [[ "$BUILD_ENVIRONMENT" != *rocm* && "$BUILD_ENVIRONMENT" != *s390x* && -d /var/lib/jenkins/workspace ]]; then
   # Workaround for dind-rootless userid mapping (https://github.com/pytorch/ci-infra/issues/96)
   WORKSPACE_ORIGINAL_OWNER_ID=$(stat -c '%u' "/var/lib/jenkins/workspace")
   cleanup_workspace() {
@@ -46,6 +46,9 @@ BUILD_BIN_DIR="$BUILD_DIR"/bin
 SHARD_NUMBER="${SHARD_NUMBER:=1}"
 NUM_TEST_SHARDS="${NUM_TEST_SHARDS:=1}"
 
+# enable debug asserts in serialization
+export TORCH_SERIALIZATION_DEBUG=1
+
 export VALGRIND=ON
 # export TORCH_INDUCTOR_INSTALL_GXX=ON
 if [[ "$BUILD_ENVIRONMENT" == *clang9* || "$BUILD_ENVIRONMENT" == *xpu* ]]; then
@@ -86,6 +89,13 @@ if [[ "$BUILD_ENVIRONMENT" == *clang9* || "$BUILD_ENVIRONMENT" == *xpu* ]]; then
   export VALGRIND=OFF
 fi
 
+
+if [[ "$BUILD_ENVIRONMENT" == *s390x* ]]; then
+  # There are additional warnings on s390x, maybe due to newer gcc.
+  # Skip this check for now
+  export VALGRIND=OFF
+fi
+
 if [[ "${PYTORCH_TEST_RERUN_DISABLED_TESTS}" == "1" ]] || [[ "${CONTINUE_THROUGH_ERROR}" == "1" ]]; then
   # When rerunning disable tests, do not generate core dumps as it could consume
   # the runner disk space when crashed tests are run multiple times. Running out
@@ -129,7 +139,7 @@ if [[ "$TEST_CONFIG" == 'default' ]]; then
 fi
 
 if [[ "$TEST_CONFIG" == 'distributed' ]] && [[ "$BUILD_ENVIRONMENT" == *rocm* ]]; then
-  export HIP_VISIBLE_DEVICES=0,1
+  export HIP_VISIBLE_DEVICES=0,1,2,3
 fi
 
 if [[ "$TEST_CONFIG" == 'slow' ]]; then
@@ -153,6 +163,8 @@ elif [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
   export PYTORCH_TESTING_DEVICE_ONLY_FOR="xpu"
   # setting PYTHON_TEST_EXTRA_OPTION
   export PYTHON_TEST_EXTRA_OPTION="--xpu"
+  # Disable sccache for xpu test due to flaky issue https://github.com/pytorch/pytorch/issues/143585
+  sudo rm -rf /opt/cache
 fi
 
 if [[ "$TEST_CONFIG" == *crossref* ]]; then
@@ -165,6 +177,9 @@ if [[ "$BUILD_ENVIRONMENT" == *rocm* ]]; then
   # Print GPU info
   rocminfo
   rocminfo | grep -E 'Name:.*\sgfx|Marketing'
+
+  # for benchmarks/dynamo/check_accuracy.py, we need to put results in a rocm specific directory to avoid clashes with cuda
+  MAYBE_ROCM="rocm/"
 fi
 
 if [[ "$BUILD_ENVIRONMENT" == *xpu* ]]; then
@@ -313,6 +328,7 @@ test_dynamo_wrapped_shard() {
     --exclude-jit-executor \
     --exclude-distributed-tests \
     --exclude-torch-export-tests \
+    --exclude-aot-dispatch-tests \
     --shard "$1" "$NUM_TEST_SHARDS" \
     --verbose \
     --upload-artifacts-while-running
@@ -326,7 +342,7 @@ test_inductor_distributed() {
   python test/run_test.py -i inductor/test_aot_inductor.py -k test_non_default_cuda_device --verbose
   python test/run_test.py -i inductor/test_aot_inductor.py -k test_replicate_on_devices --verbose
   python test/run_test.py -i distributed/test_c10d_functional_native.py --verbose
-  python test/run_test.py -i distributed/_tensor/test_dtensor_compile.py --verbose
+  python test/run_test.py -i distributed/tensor/test_dtensor_compile.py --verbose
   python test/run_test.py -i distributed/tensor/parallel/test_micro_pipeline_tp.py --verbose
   python test/run_test.py -i distributed/_composable/test_replicate_with_compiler.py --verbose
   python test/run_test.py -i distributed/_composable/fsdp/test_fully_shard_comm.py --verbose
@@ -379,15 +395,32 @@ test_inductor_aoti() {
   CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="${TORCH_LIB_DIR}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference
 }
 
-test_inductor_cpp_wrapper() {
+test_inductor_cpp_wrapper_shard() {
+  if [[ -z "$NUM_TEST_SHARDS" ]]; then
+    echo "NUM_TEST_SHARDS must be defined to run a Python test shard"
+    exit 1
+  fi
+
   export TORCHINDUCTOR_CPP_WRAPPER=1
   TEST_REPORTS_DIR=$(pwd)/test/test-reports
   mkdir -p "$TEST_REPORTS_DIR"
 
-  # Run certain inductor unit tests with cpp wrapper. In the end state, we should be able to run all the inductor
-  # unit tests with cpp wrapper.
-  python test/run_test.py --include inductor/test_torchinductor.py --verbose
+  if [[ "$1" -eq "2" ]]; then
+    # For now, manually put the opinfo tests in shard 2, and all other tests in
+    # shard 1.  Test specific things triggering past bugs, for now.
+    python test/run_test.py \
+      --include inductor/test_torchinductor_opinfo \
+      -k 'linalg or to_sparse' \
+      --verbose
+    exit
+  fi
 
+  # Run certain inductor unit tests with cpp wrapper. In the end state, we
+  # should be able to run all the inductor unit tests with cpp_wrapper.
+  python test/run_test.py \
+    --include inductor/test_torchinductor inductor/test_max_autotune inductor/test_cpu_repro \
+    --verbose
+  python test/run_test.py --inductor --include test_torch -k 'take' --verbose
 
   # Run inductor benchmark tests with cpp wrapper.
   # Skip benchmark tests if it's in rerun-disabled-mode.
@@ -400,7 +433,7 @@ test_inductor_cpp_wrapper() {
     --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_training.csv"
     python benchmarks/dynamo/check_accuracy.py \
       --actual "$TEST_REPORTS_DIR/inductor_cpp_wrapper_training.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/inductor_timm_training.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}inductor_timm_training.csv"
 
     python benchmarks/dynamo/torchbench.py --device cuda --accuracy \
       --bfloat16 --inference --inductor --only hf_T5 --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv"
@@ -410,7 +443,7 @@ test_inductor_cpp_wrapper() {
       --bfloat16 --inference --inductor --only moco --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv"
     python benchmarks/dynamo/check_accuracy.py \
       --actual "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/inductor_torchbench_inference.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}inductor_torchbench_inference.csv"
   fi
 }
 
@@ -485,6 +518,8 @@ test_perf_for_dashboard() {
     test_inductor_set_cpu_affinity
   elif [[ "${TEST_CONFIG}" == *cuda_a10g* ]]; then
     device=cuda_a10g
+  elif [[ "${TEST_CONFIG}" == *rocm* ]]; then
+    device=rocm
   fi
 
   for mode in "${modes[@]}"; do
@@ -517,7 +552,7 @@ test_perf_for_dashboard() {
             --dynamic-batch-only "$@" \
             --output "$TEST_REPORTS_DIR/${backend}_dynamic_${suite}_${dtype}_${mode}_${device}_${target}.csv"
       fi
-      if [[ "$DASHBOARD_TAG" == *cppwrapper-true* ]] && [[ "$mode" == "inference" ]]; then
+      if [[ "$DASHBOARD_TAG" == *cppwrapper-true* ]]; then
         TORCHINDUCTOR_CPP_WRAPPER=1 $TASKSET python "benchmarks/dynamo/$suite.py" \
             "${target_flag[@]}" --"$mode" --"$dtype" --backend "$backend" --disable-cudagraphs "$@" \
             --output "$TEST_REPORTS_DIR/${backend}_cpp_wrapper_${suite}_${dtype}_${mode}_${device}_${target}.csv"
@@ -601,16 +636,16 @@ test_single_dynamo_benchmark() {
       TEST_CONFIG=${TEST_CONFIG//_avx512/}
     fi
     python "benchmarks/dynamo/$suite.py" \
-      --ci --accuracy --timing --explain \
+      --ci --accuracy --timing --explain --print-compilation-time \
       "${DYNAMO_BENCHMARK_FLAGS[@]}" \
       "$@" "${partition_flags[@]}" \
       --output "$TEST_REPORTS_DIR/${name}_${suite}.csv"
     python benchmarks/dynamo/check_accuracy.py \
       --actual "$TEST_REPORTS_DIR/${name}_$suite.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/${TEST_CONFIG}_${name}.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}${TEST_CONFIG}_${name}.csv"
     python benchmarks/dynamo/check_graph_breaks.py \
       --actual "$TEST_REPORTS_DIR/${name}_$suite.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/${TEST_CONFIG}_${name}.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}${TEST_CONFIG}_${name}.csv"
   fi
 }
 
@@ -633,7 +668,7 @@ test_inductor_halide() {
 }
 
 test_inductor_triton_cpu() {
-  python test/run_test.py --include inductor/test_triton_cpu_backend.py --verbose
+  python test/run_test.py --include inductor/test_triton_cpu_backend.py inductor/test_torchinductor_strided_blocks.py --verbose
   assert_git_not_dirty
 }
 
@@ -697,7 +732,7 @@ test_inductor_torchbench_smoketest_perf() {
       --only $test --output "$TEST_REPORTS_DIR/inductor_warm_start_smoketest_$test.csv"
     python benchmarks/dynamo/check_accuracy.py \
       --actual "$TEST_REPORTS_DIR/inductor_warm_start_smoketest_$test.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/inductor_huggingface_training.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${MAYBE_ROCM}inductor_huggingface_training.csv"
   done
 }
 
@@ -893,10 +928,20 @@ test_libtorch_api() {
   else
     # Exclude IMethodTest that relies on torch::deploy, which will instead be ran in test_deploy
     OMP_NUM_THREADS=2 TORCH_CPP_TEST_MNIST_PATH="${MNIST_DIR}" python test/run_test.py --cpp --verbose -i cpp/test_api -k "not IMethodTest"
-    python test/run_test.py --cpp --verbose -i cpp/test_tensorexpr
+
+    # On s390x, pytorch is built without llvm.
+    # Even if it would be built with llvm, llvm currently doesn't support used features on s390x and
+    # test fails with errors like:
+    # JIT session error: Unsupported target machine architecture in ELF object pytorch-jitted-objectbuffer
+    # unknown file: Failure
+    # C++ exception with description "valOrErr INTERNAL ASSERT FAILED at "/var/lib/jenkins/workspace/torch/csrc/jit/tensorexpr/llvm_jit.h":34, please report a bug to PyTorch. Unexpected failure in LLVM JIT: Failed to materialize symbols: { (main, { func }) }
+    if [[ "${BUILD_ENVIRONMENT}" != *s390x* ]]; then
+      python test/run_test.py --cpp --verbose -i cpp/test_tensorexpr
+    fi
   fi
 
-  if [[ "${BUILD_ENVIRONMENT}" != *android* && "${BUILD_ENVIRONMENT}" != *cuda* && "${BUILD_ENVIRONMENT}" != *asan* ]]; then
+  # quantization is not fully supported on s390x yet
+  if [[ "${BUILD_ENVIRONMENT}" != *android* && "${BUILD_ENVIRONMENT}" != *cuda* && "${BUILD_ENVIRONMENT}" != *asan* && "${BUILD_ENVIRONMENT}" != *s390x* ]]; then
     # NB: This test is not under TORCH_BIN_DIR but under BUILD_BIN_DIR
     export CPP_TESTS_DIR="${BUILD_BIN_DIR}"
     python test/run_test.py --cpp --verbose -i cpp/static_runtime_test
@@ -1243,7 +1288,7 @@ EOF
 }
 
 test_bazel() {
-  set -e
+  set -e -o pipefail
 
   # bazel test needs sccache setup.
   # shellcheck source=./common-build.sh
@@ -1394,7 +1439,7 @@ 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_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops \
         --shard "$SHARD_NUMBER" "$NUM_TEST_SHARDS" --verbose
 
   # Dynamo tests
@@ -1497,7 +1542,7 @@ elif [[ "${TEST_CONFIG}" == *inductor_cpp_wrapper* ]]; then
   install_torchaudio cuda
   install_torchvision
   checkout_install_torchbench hf_T5 llama moco
-  PYTHONPATH=$(pwd)/torchbench test_inductor_cpp_wrapper
+  PYTHONPATH=$(pwd)/torchbench test_inductor_cpp_wrapper_shard "$SHARD_NUMBER"
 elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
   install_torchvision
   test_inductor_shard "${SHARD_NUMBER}"

.ci/pytorch/test_example_code/cnn_smoke_win_arm64.py

@@ -0,0 +1,41 @@
+r"""
+It's used to check basic rnn features with cpu-only.
+For example, it would throw exception if some components are missing
+"""
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.optim as optim
+
+
+class SimpleCNN(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.conv = nn.Conv2d(1, 1, 3)
+        self.pool = nn.MaxPool2d(2, 2)
+
+    def forward(self, inputs):
+        output = self.pool(F.relu(self.conv(inputs)))
+        output = output.view(1)
+        return output
+
+
+try:
+    # Mock one infer
+    net = SimpleCNN()
+    net_inputs = torch.rand((1, 1, 5, 5))
+    outputs = net(net_inputs)
+    print(outputs)
+
+    criterion = nn.MSELoss()
+    optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.1)
+
+    # Mock one step training
+    label = torch.full((1,), 1.0, dtype=torch.float)
+    loss = criterion(outputs, label)
+    loss.backward()
+    optimizer.step()
+
+except Exception as e:
+    print(f"An error occurred: {e}")

.ci/pytorch/test_example_code/rnn_smoke_win_arm64.py

@@ -0,0 +1,13 @@
+r"""
+It's used to check basic rnn features with cpu-only.
+For example, it would throw exception if missing some components are missing
+"""
+
+import torch
+import torch.nn as nn
+
+
+rnn = nn.RNN(10, 20, 2)
+inputs = torch.randn(5, 3, 10)
+h0 = torch.randn(2, 3, 20)
+output, hn = rnn(inputs, h0)

.ci/pytorch/win-build.sh

@@ -38,7 +38,7 @@ if [[ $PYLONG_API_CHECK == 0 ]]; then
   echo "PyLong_AsUnsignedLong -> THPUtils_unpackUInt32 / THPUtils_unpackUInt64"
   exit 1
 fi
-set -ex
+set -ex -o pipefail
 
 "$SCRIPT_HELPERS_DIR"/build_pytorch.bat
 

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

@@ -26,7 +26,8 @@ if not errorlevel 0 goto fail
 
 if "%USE_XPU%"=="1" (
   :: Install xpu support packages
-  call %INSTALLER_DIR%\install_xpu.bat
+  set CUDA_VERSION=xpu
+  call %SCRIPT_HELPERS_DIR%\..\windows\internal\xpu_install.bat
   if errorlevel 1 exit /b 1
 )
 

.ci/pytorch/win-test-helpers/installation-helpers/install_xpu.bat

@@ -1,114 +0,0 @@
-@echo on
-REM Description: Install Intel Support Packages on Windows
-REM BKM reference: https://www.intel.com/content/www/us/en/developer/articles/tool/pytorch-prerequisites-for-intel-gpus.html
-
-set XPU_INSTALL_MODE=%~1
-if "%XPU_INSTALL_MODE%"=="" goto xpu_bundle_install_start
-if "%XPU_INSTALL_MODE%"=="bundle" goto xpu_bundle_install_start
-if "%XPU_INSTALL_MODE%"=="driver" goto xpu_driver_install_start
-if "%XPU_INSTALL_MODE%"=="all" goto xpu_driver_install_start
-
-:arg_error
-
-echo Illegal XPU installation mode. The value can be "bundle"/"driver"/"all"
-echo If keep the value as space, will use default "bundle" mode
-exit /b 1
-
-:xpu_driver_install_start
-:: TODO Need more testing for driver installation
-set XPU_DRIVER_LINK=https://downloadmirror.intel.com/830975/gfx_win_101.5972.exe
-curl -o xpu_driver.exe --retry 3 --retry-all-errors -k %XPU_DRIVER_LINK%
-echo "XPU Driver installing..."
-start /wait "Intel XPU Driver Installer" "xpu_driver.exe"
-if errorlevel 1 exit /b 1
-del xpu_driver.exe
-if "%XPU_INSTALL_MODE%"=="driver" goto xpu_install_end
-
-:xpu_bundle_install_start
-
-set XPU_BUNDLE_PARENT_DIR=C:\Program Files (x86)\Intel\oneAPI
-set XPU_BUNDLE_URL=https://registrationcenter-download.intel.com/akdlm/IRC_NAS/9d1a91e2-e8b8-40a5-8c7f-5db768a6a60c/w_intel-for-pytorch-gpu-dev_p_0.5.3.37_offline.exe
-set XPU_BUNDLE_PRODUCT_NAME=intel.oneapi.win.intel-for-pytorch-gpu-dev.product
-set XPU_BUNDLE_VERSION=0.5.3+31
-set XPU_BUNDLE_INSTALLED=0
-set XPU_BUNDLE_UNINSTALL=0
-set XPU_EXTRA_URL=https://registrationcenter-download.intel.com/akdlm/IRC_NAS/9d1a91e2-e8b8-40a5-8c7f-5db768a6a60c/w_intel-pti-dev_p_0.9.0.37_offline.exe
-set XPU_EXTRA_PRODUCT_NAME=intel.oneapi.win.intel-pti-dev.product
-set XPU_EXTRA_VERSION=0.9.0+36
-set XPU_EXTRA_INSTALLED=0
-set XPU_EXTRA_UNINSTALL=0
-
-if not [%XPU_VERSION%]==[] if [%XPU_VERSION%]==[2025.0] (
-    set XPU_BUNDLE_URL=https://registrationcenter-download.intel.com/akdlm/IRC_NAS/efc86abd-cb77-452e-a03f-a741895b8ece/intel-deep-learning-essentials-2025.0.0.336_offline.exe
-    set XPU_BUNDLE_PRODUCT_NAME=intel.oneapi.win.deep-learning-essentials.product
-    set XPU_BUNDLE_VERSION=2025.0.0+335
-    set XPU_BUNDLE_INSTALLED=0
-    set XPU_BUNDLE_UNINSTALL=0
-    set XPU_EXTRA_URL=NULL
-    set XPU_EXTRA_PRODUCT_NAME=intel.oneapi.win.compiler.product
-    set XPU_EXTRA_VERSION=2025.0.1+1226
-    set XPU_EXTRA_INSTALLED=0
-    set XPU_EXTRA_UNINSTALL=0
-)
-
-:: Check if XPU bundle is target version or already installed
-if exist "%XPU_BUNDLE_PARENT_DIR%\Installer\installer.exe" goto xpu_bundle_ver_check
-goto xpu_bundle_install
-
-:xpu_bundle_ver_check
-
-"%XPU_BUNDLE_PARENT_DIR%\Installer\installer.exe" --list-products > xpu_bundle_installed_ver.log
-
-for /f "tokens=1,2" %%a in (xpu_bundle_installed_ver.log) do (
-    if "%%a"=="%XPU_BUNDLE_PRODUCT_NAME%" (
-        echo %%a Installed Version: %%b
-        set XPU_BUNDLE_INSTALLED=1
-        if not "%XPU_BUNDLE_VERSION%"=="%%b" (
-            start /wait "Installer Title" "%XPU_BUNDLE_PARENT_DIR%\Installer\installer.exe" --action=remove --eula=accept --silent --product-id %%a --product-ver %%b --log-dir uninstall_bundle
-            set XPU_BUNDLE_UNINSTALL=1
-        )
-    )
-    if "%%a"=="%XPU_EXTRA_PRODUCT_NAME%" (
-        echo %%a Installed Version: %%b
-        set XPU_EXTRA_INSTALLED=1
-        if not "%XPU_EXTRA_VERSION%"=="%%b" (
-            start /wait "Installer Title" "%XPU_BUNDLE_PARENT_DIR%\Installer\installer.exe" --action=remove --eula=accept --silent --product-id %%a --product-ver %%b --log-dir uninstall_bundle
-            set XPU_EXTRA_UNINSTALL=1
-        )
-    )
-    if not "%%b" == "Version" if not [%%b]==[] if not "%%a"=="%XPU_BUNDLE_PRODUCT_NAME%" if not "%%a"=="%XPU_EXTRA_PRODUCT_NAME%" (
-        echo "Uninstalling...."
-        start /wait "Installer Title" "%XPU_BUNDLE_PARENT_DIR%\Installer\installer.exe" --action=remove --eula=accept --silent --product-id %%a --product-ver %%b --log-dir uninstall_bundle
-    )
-)
-if errorlevel 1 exit /b 1
-if exist xpu_bundle_installed_ver.log del xpu_bundle_installed_ver.log
-if exist uninstall_bundle rmdir /s /q uninstall_bundle
-if "%XPU_BUNDLE_INSTALLED%"=="0" goto xpu_bundle_install
-if "%XPU_BUNDLE_UNINSTALL%"=="1" goto xpu_bundle_install
-
-:xpu_extra_check
-
-if "%XPU_EXTRA_URL%"=="NULL" goto xpu_install_end
-if "%XPU_EXTRA_INSTALLED%"=="0" goto xpu_extra_install
-if "%XPU_EXTRA_UNINSTALL%"=="1" goto xpu_extra_install
-goto xpu_install_end
-
-:xpu_bundle_install
-
-curl -o xpu_bundle.exe --retry 3 --retry-all-errors -k %XPU_BUNDLE_URL%
-echo "XPU Bundle installing..."
-start /wait "Intel Pytorch Bundle Installer" "xpu_bundle.exe" --action=install --eula=accept --silent --log-dir install_bundle
-if errorlevel 1 exit /b 1
-del xpu_bundle.exe
-goto xpu_extra_check
-
-:xpu_extra_install
-
-curl -o xpu_extra.exe --retry 3 --retry-all-errors -k %XPU_EXTRA_URL%
-echo "Intel XPU EXTRA installing..."
-start /wait "Intel XPU EXTRA Installer" "xpu_extra.exe" --action=install --eula=accept --silent --log-dir install_bundle
-if errorlevel 1 exit /b 1
-del xpu_extra.exe
-
-:xpu_install_end

.ci/pytorch/win-test.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-set -ex
+set -ex -o pipefail
 
 SCRIPT_PARENT_DIR=$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )
 # shellcheck source=./common.sh
@@ -18,6 +18,9 @@ export PYTORCH_FINAL_PACKAGE_DIR="${PYTORCH_FINAL_PACKAGE_DIR:-/c/w/build-result
 PYTORCH_FINAL_PACKAGE_DIR_WIN=$(cygpath -w "${PYTORCH_FINAL_PACKAGE_DIR}")
 export PYTORCH_FINAL_PACKAGE_DIR_WIN
 
+# enable debug asserts in serialization
+export TORCH_SERIALIZATION_DEBUG=1
+
 mkdir -p "$TMP_DIR"/build/torch
 
 export SCRIPT_HELPERS_DIR=$SCRIPT_PARENT_DIR/win-test-helpers
@@ -41,7 +44,7 @@ python -m pip install pytest-rerunfailures==10.3 pytest-cpp==2.3.0 tensorboard==
 python -m pip install z3-solver==4.12.2.0
 
 # Install tlparse for test\dynamo\test_structured_trace.py UTs.
-python -m pip install tlparse==0.3.25
+python -m pip install tlparse==0.3.30
 
 # Install parameterized
 python -m pip install parameterized==0.8.1

.ci/pytorch/windows/arm64/bootstrap_apl.bat

@@ -0,0 +1,31 @@
+@echo off
+
+echo Dependency ARM Performance Libraries (APL) installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+:: Set download URL for the ARM Performance Libraries (APL)
+set DOWNLOAD_URL="https://developer.arm.com/-/cdn-downloads/permalink/Arm-Performance-Libraries/Version_24.10/arm-performance-libraries_24.10_Windows.msi"
+set INSTALLER_FILE=%DOWNLOADS_DIR%\arm-performance-libraries.msi
+
+:: Download installer
+echo Downloading ARM Performance Libraries (APL)...
+curl -L -o "%INSTALLER_FILE%" %DOWNLOAD_URL%
+
+:: Install ARM Performance Libraries (APL)
+echo Installing ARM Performance Libraries (APL)...
+msiexec /i "%INSTALLER_FILE%" /qn /norestart ACCEPT_EULA=1 INSTALLFOLDER="%DEPENDENCIES_DIR%"
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install ARM Performance Libraries (APL) components. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+:: Add to environment
+echo ARMPL_DIR=%DEPENDENCIES_DIR%\armpl_24.10\>> %GITHUB_ENV%
+echo %DEPENDENCIES_DIR%\armpl_24.10\bin\>> %GITHUB_PATH%
+
+echo Dependency ARM Performance Libraries (APL) installation finished.

.ci/pytorch/windows/arm64/bootstrap_buildtools.bat

@@ -0,0 +1,49 @@
+@echo off
+
+echo Dependency MSVC Build Tools with C++ with ARM64/ARM64EC components installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir "%DOWNLOADS_DIR%"
+if not exist "%DEPENDENCIES_DIR%" mkdir "%DEPENDENCIES_DIR%"
+
+:: Set download URL for the Visual Studio Installer
+set DOWNLOAD_URL=https://aka.ms/vs/17/release/vs_BuildTools.exe
+set INSTALLER_FILE=%DOWNLOADS_DIR%\vs_BuildTools.exe
+
+:: Download installer
+echo Downloading Visual Studio Build Tools with C++ installer...
+curl -L -o "%INSTALLER_FILE%" %DOWNLOAD_URL%
+
+:: Install the Visual Studio Build Tools with C++ components
+echo Installing Visual Studio Build Tools with C++ components...
+echo Installing MSVC %MSVC_VERSION%
+if "%MSVC_VERSION%" == "latest" (
+    "%INSTALLER_FILE%" --norestart --nocache --quiet --wait --installPath "%DEPENDENCIES_DIR%\VSBuildTools" ^
+        --add Microsoft.VisualStudio.Component.Windows11SDK.22621 ^
+        --add Microsoft.VisualStudio.Component.VC.ASAN ^
+        --add Microsoft.VisualStudio.Component.VC.CMake.Project ^
+        --add Microsoft.VisualStudio.Component.VC.Tools.ARM64 ^
+        --add Microsoft.VisualStudio.Component.VC.Tools.x86.x64
+) else if "%MSVC_VERSION%" == "14.40" (
+    "%INSTALLER_FILE%" --norestart --nocache --quiet --wait --installPath "%DEPENDENCIES_DIR%\VSBuildTools" ^
+        --add Microsoft.VisualStudio.Component.Windows11SDK.22621 ^
+        --add Microsoft.VisualStudio.Component.VC.ASAN ^
+        --add Microsoft.VisualStudio.Component.VC.CMake.Project ^
+        --add Microsoft.VisualStudio.Component.VC.14.40.17.10.ARM64 ^
+        --add Microsoft.VisualStudio.Component.VC.14.40.17.10.x86.x64
+) else if "%MSVC_VERSION%" == "14.36" (
+    "%INSTALLER_FILE%" --norestart --nocache --quiet --wait --installPath "%DEPENDENCIES_DIR%\VSBuildTools" ^
+        --add Microsoft.VisualStudio.Component.Windows11SDK.22621 ^
+        --add Microsoft.VisualStudio.Component.VC.ASAN ^
+        --add Microsoft.VisualStudio.Component.VC.CMake.Project ^
+        --add Microsoft.VisualStudio.Component.VC.14.36.17.6.ARM64 ^
+        --add Microsoft.VisualStudio.Component.VC.14.36.17.6.x86.x64
+)
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install Visual Studio Build Tools with C++ components. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+echo Dependency Visual Studio Build Tools with C++ installation finished.

.ci/pytorch/windows/arm64/bootstrap_git.bat

@@ -0,0 +1,37 @@
+:: we need to install newer version of Git manually as "-submodules" function is not supported in the default version of runner.
+
+@echo off
+
+echo Dependency Git installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+:: Set download URL for the Git
+set DOWNLOAD_URL="https://github.com/git-for-windows/git/releases/download/v2.46.0.windows.1/Git-2.46.0-64-bit.exe"
+set INSTALLER_FILE=%DOWNLOADS_DIR%\Git-2.46.0-64-bit.exe
+
+:: Download installer
+echo Downloading Git...
+curl -L -o "%INSTALLER_FILE%" %DOWNLOAD_URL%
+
+:: Install Git
+echo Installing Git...
+"%INSTALLER_FILE%" /VERYSILENT /DIR="%DEPENDENCIES_DIR%\git"
+
+dir %DEPENDENCIES_DIR%\git
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install Git. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+:: Enable long paths
+call "%DEPENDENCIES_DIR%\git\cmd\git.exe" config --system core.longpaths true
+
+:: Add to PATH
+echo %DEPENDENCIES_DIR%\git\cmd\;%DEPENDENCIES_DIR%\git\bin\>> %GITHUB_PATH%
+
+echo Dependency Git installation finished.

.ci/pytorch/windows/arm64/bootstrap_libuv.bat

@@ -0,0 +1,33 @@
+@echo off
+
+echo Dependency libuv installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+:: activate visual studio
+call "%DEPENDENCIES_DIR%\VSBuildTools\VC\Auxiliary\Build\vcvarsall.bat" arm64
+where cl.exe
+
+cd %DEPENDENCIES_DIR%
+git clone https://github.com/libuv/libuv.git -b v1.39.0
+
+echo Configuring libuv...
+mkdir libuv\build
+cd libuv\build
+cmake .. -DBUILD_TESTING=OFF
+
+echo Building libuv...
+cmake --build . --config Release
+
+echo Installing libuv...
+cmake --install . --prefix ../install
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install libuv. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+echo Dependency libuv installation finished.

.ci/pytorch/windows/arm64/bootstrap_openblas.bat

@@ -0,0 +1,46 @@
+@echo off
+
+echo Dependency OpenBLAS installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+:: activate visual studio
+call "%DEPENDENCIES_DIR%\VSBuildTools\VC\Auxiliary\Build\vcvarsall.bat" arm64
+where cl.exe
+
+:: Clone OpenBLAS
+cd %DEPENDENCIES_DIR%
+git clone https://github.com/OpenMathLib/OpenBLAS.git -b v0.3.29
+
+echo Configuring OpenBLAS...
+mkdir OpenBLAS\build
+cd OpenBLAS\build
+cmake .. -G Ninja ^
+  -DBUILD_TESTING=0 ^
+  -DBUILD_BENCHMARKS=0 ^
+  -DC_LAPACK=1 ^
+  -DNOFORTRAN=1 ^
+  -DDYNAMIC_ARCH=0 ^
+  -DARCH=arm64 ^
+  -DBINARY=64 ^
+  -DTARGET=GENERIC ^
+  -DUSE_OPENMP=1 ^
+  -DCMAKE_SYSTEM_PROCESSOR=ARM64 ^
+  -DCMAKE_SYSTEM_NAME=Windows ^
+  -DCMAKE_BUILD_TYPE=Release
+
+echo Building OpenBLAS...
+cmake --build . --config Release
+
+echo Installing OpenBLAS...
+cmake --install . --prefix ../install
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install OpenBLAS. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+echo Dependency OpenBLAS installation finished.

.ci/pytorch/windows/arm64/bootstrap_python.bat

@@ -0,0 +1,41 @@
+@echo off
+
+echo Dependency Python installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+if "%PYTHON_VERSION%"=="Python312" (
+    echo Python version is set to Python312
+    set DOWNLOAD_URL="https://www.python.org/ftp/python/3.12.7/python-3.12.7-arm64.exe"
+) else if "%PYTHON_VERSION%"=="Python311" (
+    echo Python version is set to Python311
+    set DOWNLOAD_URL="https://www.python.org/ftp/python/3.11.9/python-3.11.9-arm64.exe"
+) else (
+    echo PYTHON_VERSION not defined, Python version is set to Python312
+    set DOWNLOAD_URL="https://www.python.org/ftp/python/3.12.7/python-3.12.7-arm64.exe"
+)
+
+set INSTALLER_FILE=%DOWNLOADS_DIR%\python-installer.exe
+
+:: Download installer
+echo Downloading Python...
+curl -L -o "%INSTALLER_FILE%" %DOWNLOAD_URL%
+
+:: Install Python
+echo Installing Python...
+"%INSTALLER_FILE%" /quiet Include_debug=1 TargetDir="%DEPENDENCIES_DIR%\Python"
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install Python. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+:: Add to PATH
+echo %DEPENDENCIES_DIR%\Python\>> %GITHUB_PATH%
+echo %DEPENDENCIES_DIR%\Python\scripts\>> %GITHUB_PATH%
+echo %DEPENDENCIES_DIR%\Python\libs\>> %GITHUB_PATH%
+
+echo Dependency Python installation finished.

.ci/pytorch/windows/arm64/bootstrap_rust.bat

@@ -0,0 +1,33 @@
+@echo off
+
+echo Dependency Rust installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+set DOWNLOAD_URL="https://static.rust-lang.org/rustup/dist/x86_64-pc-windows-msvc/rustup-init.exe"
+set INSTALLER_FILE=%DOWNLOADS_DIR%\rustup-init.exe
+set RUSTUP_HOME=%DEPENDENCIES_DIR%\rust
+set CARGO_HOME=%DEPENDENCIES_DIR%\cargo
+
+:: Download installer
+echo Downloading Rust...
+curl -L -o "%INSTALLER_FILE%" %DOWNLOAD_URL%
+
+:: Install APL
+echo Installing Rust...
+"%INSTALLER_FILE%" -q -y --default-host aarch64-pc-windows-msvc --default-toolchain stable --profile default
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install Rust. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+:: Add to PATH
+echo %DEPENDENCIES_DIR%\cargo\bin\>> %GITHUB_PATH%
+echo RUSTUP_HOME=%DEPENDENCIES_DIR%\rust>> %GITHUB_ENV%
+echo CARGO_HOME=%DEPENDENCIES_DIR%\cargo>> %GITHUB_ENV%
+
+echo Dependency Rust installation finished.

.ci/pytorch/windows/arm64/bootstrap_sccache.bat

@@ -0,0 +1,33 @@
+@echo off
+
+echo Dependency sccache installation started.
+
+:: Pre-check for downloads and dependencies folders
+if not exist "%DOWNLOADS_DIR%" mkdir %DOWNLOADS_DIR%
+if not exist "%DEPENDENCIES_DIR%" mkdir %DEPENDENCIES_DIR%
+
+:: Set download URL for the sccache
+set DOWNLOAD_URL="https://github.com/mozilla/sccache/releases/download/v0.8.1/sccache-v0.8.1-x86_64-pc-windows-msvc.zip"
+set INSTALLER_FILE=%DOWNLOADS_DIR%\sccache.zip
+
+:: Download installer
+echo Downloading sccache.zip...
+curl -L -o "%INSTALLER_FILE%" %DOWNLOAD_URL%
+
+:: Install sccache
+echo Extracting sccache.zip...
+tar -xf "%INSTALLER_FILE%" -C %DEPENDENCIES_DIR%
+cd %DEPENDENCIES_DIR%
+ren sccache-v0.8.1-x86_64-pc-windows-msvc sccache
+cd ..
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed to install sccache. (exitcode = %errorlevel%)"
+    exit /b 1
+)
+
+:: Add to PATH
+echo %DEPENDENCIES_DIR%\sccache\>> %GITHUB_PATH%
+
+echo Dependency sccache installation finished.

.ci/pytorch/windows/arm64/bootstrap_tests.bat

@@ -0,0 +1,22 @@
+:: change to source directory
+cd %PYTORCH_ROOT%
+
+:: activate visual studio
+call "%DEPENDENCIES_DIR%\VSBuildTools\VC\Auxiliary\Build\vcvarsall.bat" arm64
+where cl.exe
+
+:: create virtual environment
+python -m venv .venv
+echo * > .venv\.gitignore
+call .\.venv\Scripts\activate
+where python
+
+:: install dependencies
+python -m pip install --upgrade pip
+pip install -r requirements.txt
+pip install pytest numpy
+
+:: find file name for pytorch wheel
+for /f "delims=" %%f in ('dir /b "%PYTORCH_FINAL_PACKAGE_DIR%" ^| findstr "torch-"') do set "TORCH_WHEEL_FILENAME=%PYTORCH_FINAL_PACKAGE_DIR%\%%f"
+
+pip install %TORCH_WHEEL_FILENAME%

.ci/pytorch/windows/arm64/build_libtorch.bat

@@ -0,0 +1,101 @@
+@echo on
+
+:: environment variables
+set CMAKE_BUILD_TYPE=%BUILD_TYPE%
+set CMAKE_C_COMPILER_LAUNCHER=sccache
+set CMAKE_CXX_COMPILER_LAUNCHER=sccache
+set libuv_ROOT=%DEPENDENCIES_DIR%\libuv\install
+set MSSdk=1
+if defined PYTORCH_BUILD_VERSION (
+  set PYTORCH_BUILD_VERSION=%PYTORCH_BUILD_VERSION%
+  set PYTORCH_BUILD_NUMBER=1
+)
+
+:: Set BLAS type
+if %ENABLE_APL% == 1 (
+    set BLAS=APL
+    set USE_LAPACK=1
+) else if %ENABLE_OPENBLAS% == 1 (
+    set BLAS=OpenBLAS
+    set OpenBLAS_HOME=%DEPENDENCIES_DIR%\OpenBLAS\install
+)
+
+:: activate visual studio
+call "%DEPENDENCIES_DIR%\VSBuildTools\VC\Auxiliary\Build\vcvarsall.bat" arm64
+where cl.exe
+
+:: change to source directory
+cd %PYTORCH_ROOT%
+
+:: copy libuv.dll
+copy %libuv_ROOT%\lib\Release\uv.dll torch\lib\uv.dll
+
+:: create virtual environment
+python -m venv .venv
+echo * > .venv\.gitignore
+call .\.venv\Scripts\activate
+where python
+
+:: python install dependencies
+python -m pip install --upgrade pip
+pip install -r requirements.txt
+:: DISTUTILS_USE_SDK should be set after psutil dependency
+set DISTUTILS_USE_SDK=1
+
+:: start sccache server and reset sccache stats
+sccache --start-server
+sccache --zero-stats
+sccache --show-stats
+ 
+:: Prepare the environment
+mkdir libtorch
+mkdir libtorch\bin
+mkdir libtorch\cmake
+mkdir libtorch\include
+mkdir libtorch\lib
+mkdir libtorch\share
+mkdir libtorch\test
+
+:: Call LibTorch build script
+python ./tools/build_libtorch.py
+
+:: Check if there is an error
+IF ERRORLEVEL 1 exit /b 1
+IF NOT ERRORLEVEL 0 exit /b 1
+ 
+:: Move the files to the correct location
+move /Y torch\bin\*.* libtorch\bin\
+move /Y torch\cmake\*.* libtorch\cmake\
+robocopy /move /e torch\include\ libtorch\include\
+move /Y torch\lib\*.* libtorch\lib\
+robocopy /move /e torch\share\ libtorch\share\
+move /Y torch\test\*.* libtorch\test\
+move /Y libtorch\bin\*.dll libtorch\lib\
+
+:: Set version
+echo %PYTORCH_BUILD_VERSION% > libtorch\build-version
+git rev-parse HEAD > libtorch\build-hash
+
+:: Set LIBTORCH_PREFIX
+IF "%DEBUG%" == "" (
+    set LIBTORCH_PREFIX=libtorch-win-arm64-shared-with-deps
+) ELSE (
+    set LIBTORCH_PREFIX=libtorch-win-arm64-shared-with-deps-debug
+)
+
+:: Create output
+C:\Windows\System32\tar.exe -cvaf %LIBTORCH_PREFIX%-%PYTORCH_BUILD_VERSION%.zip -C libtorch *
+
+:: Copy output to target directory
+if not exist ..\output mkdir ..\output
+copy /Y "%LIBTORCH_PREFIX%-%PYTORCH_BUILD_VERSION%.zip" "%PYTORCH_FINAL_PACKAGE_DIR%\"
+copy /Y "%LIBTORCH_PREFIX%-%PYTORCH_BUILD_VERSION%.zip" "%PYTORCH_FINAL_PACKAGE_DIR%\%LIBTORCH_PREFIX%-latest.zip"
+
+:: Cleanup raw data to save space
+rmdir /s /q libtorch
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed on build_libtorch. (exitcode = %errorlevel%)"
+    exit /b 1
+)

.ci/pytorch/windows/arm64/build_pytorch.bat

@@ -0,0 +1,60 @@
+@echo on
+
+:: environment variables
+set CMAKE_BUILD_TYPE=%BUILD_TYPE%
+set CMAKE_C_COMPILER_LAUNCHER=sccache
+set CMAKE_CXX_COMPILER_LAUNCHER=sccache
+set libuv_ROOT=%DEPENDENCIES_DIR%\libuv\install
+set MSSdk=1
+if defined PYTORCH_BUILD_VERSION (
+  set PYTORCH_BUILD_VERSION=%PYTORCH_BUILD_VERSION%
+  set PYTORCH_BUILD_NUMBER=1
+)
+
+:: Set BLAS type
+if %ENABLE_APL% == 1 (
+    set BLAS=APL
+    set USE_LAPACK=1
+) else if %ENABLE_OPENBLAS% == 1 (
+    set BLAS=OpenBLAS
+    set OpenBLAS_HOME=%DEPENDENCIES_DIR%\OpenBLAS\install
+)
+
+:: activate visual studio
+call "%DEPENDENCIES_DIR%\VSBuildTools\VC\Auxiliary\Build\vcvarsall.bat" arm64
+where cl.exe
+
+:: change to source directory
+cd %PYTORCH_ROOT%
+
+:: copy libuv.dll
+copy %libuv_ROOT%\lib\Release\uv.dll torch\lib\uv.dll
+
+:: create virtual environment
+python -m venv .venv
+echo * > .venv\.gitignore
+call .\.venv\Scripts\activate
+where python
+
+:: python install dependencies
+python -m pip install --upgrade pip
+pip install -r requirements.txt
+:: DISTUTILS_USE_SDK should be set after psutil dependency
+set DISTUTILS_USE_SDK=1
+
+:: start sccache server and reset sccache stats
+sccache --start-server
+sccache --zero-stats
+sccache --show-stats
+
+:: Call PyTorch build script
+python setup.py bdist_wheel -d "%PYTORCH_FINAL_PACKAGE_DIR%"
+
+:: show sccache stats
+sccache --show-stats
+
+:: Check if installation was successful
+if %errorlevel% neq 0 (
+    echo "Failed on build_pytorch. (exitcode = %errorlevel%)"
+    exit /b 1
+)

.ci/pytorch/windows/arm64/smoke_test.bat

@@ -0,0 +1,65 @@
+@echo off
+setlocal
+
+set "ORIG_PATH=%PATH%"
+
+if "%PACKAGE_TYPE%" == "wheel" goto wheel
+if "%PACKAGE_TYPE%" == "libtorch" goto libtorch
+
+echo "unknown package type"
+exit /b 1
+
+:wheel
+echo "install wheel package"
+
+echo Running pip install...
+pip install -q --pre numpy protobuf
+echo Error level after pip install: %ERRORLEVEL%
+if errorlevel 1 exit /b 1
+
+for /F "delims=" %%i in ('where /R "%PYTORCH_FINAL_PACKAGE_DIR:/=\%" *.whl') do pip install "%%i"
+if errorlevel 1 exit /b 1
+
+goto smoke_test
+
+:smoke_test
+python -c "import torch"
+if ERRORLEVEL 1 exit /b 1
+
+echo Running python rnn_smoke.py...
+python %PYTORCH_ROOT%\.ci\pytorch\test_example_code\rnn_smoke_win_arm64.py
+if errorlevel 1 exit /b 1
+
+echo Checking that basic CNN works...
+python %PYTORCH_ROOT%\.ci\pytorch\test_example_code\cnn_smoke_win_arm64.py
+if errorlevel 1 exit /b 1
+
+goto end
+
+:libtorch
+echo "install and test libtorch"
+
+for /F "delims=" %%i in ('where /R "%PYTORCH_FINAL_PACKAGE_DIR:/=\%" *-latest.zip') do tar -xf "%%i" -C tmp
+if ERRORLEVEL 1 exit /b 1
+
+pushd tmp\libtorch
+
+set VC_VERSION_LOWER=14
+set VC_VERSION_UPPER=36
+
+call "%DEPENDENCIES_DIR%\VSBuildTools\VC\Auxiliary\Build\vcvarsall.bat" arm64
+
+set install_root=%CD%
+set INCLUDE=%INCLUDE%;%install_root%\include;%install_root%\include\torch\csrc\api\include
+set LIB=%LIB%;%install_root%\lib
+set PATH=%PATH%;%install_root%\lib
+
+cl %PYTORCH_ROOT%\.ci\pytorch\test_example_code\simple-torch-test.cpp c10.lib torch_cpu.lib /EHsc /std:c++17
+if ERRORLEVEL 1 exit /b 1
+
+.\simple-torch-test.exe
+if ERRORLEVEL 1 exit /b 1
+
+:end
+set "PATH=%ORIG_PATH%"
+popd

.ci/pytorch/windows/condaenv.bat

@@ -9,12 +9,13 @@ FOR %%v IN (%DESIRED_PYTHON%) DO (
     set PYTHON_VERSION_STR=%%v
     set PYTHON_VERSION_STR=!PYTHON_VERSION_STR:.=!
     conda remove -n py!PYTHON_VERSION_STR! --all -y || rmdir %CONDA_HOME%\envs\py!PYTHON_VERSION_STR! /s
-    if "%%v" == "3.8" call conda create -n py!PYTHON_VERSION_STR! -y -q numpy=1.11 pyyaml boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
-    if "%%v" == "3.9" call conda create -n py!PYTHON_VERSION_STR! -y -q numpy=2.0.1 pyyaml boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
-    if "%%v" == "3.10" call conda create -n py!PYTHON_VERSION_STR! -y -q -c=conda-forge numpy=2.0.1 pyyaml boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
-    if "%%v" == "3.11" call conda create -n py!PYTHON_VERSION_STR! -y -q -c=conda-forge numpy=2.0.1 pyyaml boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
-    if "%%v" == "3.12" call conda create -n py!PYTHON_VERSION_STR! -y -q -c=conda-forge numpy=2.0.1 pyyaml boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
-    if "%%v" == "3.13" call conda create -n py!PYTHON_VERSION_STR! -y -q -c=conda-forge numpy=2.1.2 pyyaml boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
+    if "%%v" == "3.9" call conda create -n py!PYTHON_VERSION_STR! -y numpy=2.0.1 boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
+    if "%%v" == "3.10" call conda create -n py!PYTHON_VERSION_STR! -y -c=conda-forge numpy=2.0.1  boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
+    if "%%v" == "3.11" call conda create -n py!PYTHON_VERSION_STR! -y -c=conda-forge numpy=2.0.1  boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
+    if "%%v" == "3.12" call conda create -n py!PYTHON_VERSION_STR! -y -c=conda-forge numpy=2.0.1  boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
+    if "%%v" == "3.13" call conda create -n py!PYTHON_VERSION_STR! -y -c=conda-forge numpy=2.1.2  boto3 cmake ninja typing_extensions setuptools=72.1.0 python=%%v
+    if "%%v" == "3.13t" call conda create -n py!PYTHON_VERSION_STR! -y -c=conda-forge numpy=2.1.2 boto3 cmake ninja typing_extensions setuptools=72.1.0 python-freethreading python=3.13
+    call conda run -n py!PYTHON_VERSION_STR! pip install pyyaml
     call conda run -n py!PYTHON_VERSION_STR! pip install mkl-include
     call conda run -n py!PYTHON_VERSION_STR! pip install mkl-static
 )

.ci/pytorch/windows/cuda128.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_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"
+    ) ELSE (
+        echo CUDA 12.8 not found, failing
+        exit /b 1
+    )
+)
+
+IF "%BUILD_VISION%" == "" (
+    set TORCH_CUDA_ARCH_LIST=5.0;6.0;6.1;7.0;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_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 -gencode=arch=compute_100,code=compute_100 -gencode=arch=compute_120,code=compute_120
+)
+
+set "CUDA_PATH=%CUDA_PATH_V128%"
+set "PATH=%CUDA_PATH_V128%\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/cuda_install.bat

@@ -9,7 +9,8 @@ if "%CUDA_VERSION%" == "xpu" (
     exit /b 0
 )
 
-set SRC_DIR=%NIGHTLIES_PYTORCH_ROOT%
+set SRC_DIR=%~dp0\..
+
 if not exist "%SRC_DIR%\temp_build" mkdir "%SRC_DIR%\temp_build"
 
 set /a CUDA_VER=%CUDA_VERSION%
@@ -23,9 +24,9 @@ set CUDNN_LIB_FOLDER="lib\x64"
 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 121 goto cuda121
 if %CUDA_VER% EQU 124 goto cuda124
 if %CUDA_VER% EQU 126 goto cuda126
+if %CUDA_VER% EQU 128 goto cuda128
 
 echo CUDA %CUDA_VERSION_STR% is not supported
 exit /b 1
@@ -111,6 +112,33 @@ xcopy /Y "%SRC_DIR%\temp_build\zlib\dll_x64\*.dll" "C:\Windows\System32"
 
 goto cuda_common
 
+:cuda128
+
+set CUDA_INSTALL_EXE=cuda_12.8.0_571.96_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%"
+    if errorlevel 1 exit /b 1
+    set "CUDA_SETUP_FILE=%SRC_DIR%\temp_build\%CUDA_INSTALL_EXE%"
+    set "ARGS=cuda_profiler_api_12.8 thrust_12.8 nvcc_12.8 cuobjdump_12.8 nvprune_12.8 nvprof_12.8 cupti_12.8 cublas_12.8 cublas_dev_12.8 cudart_12.8 cufft_12.8 cufft_dev_12.8 curand_12.8 curand_dev_12.8 cusolver_12.8 cusolver_dev_12.8 cusparse_12.8 cusparse_dev_12.8 npp_12.8 npp_dev_12.8 nvrtc_12.8 nvrtc_dev_12.8 nvml_dev_12.8 nvjitlink_12.8 nvtx_12.8"
+)
+
+set CUDNN_FOLDER=cudnn-windows-x86_64-9.7.0.66_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%"
+    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

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

@@ -27,7 +27,6 @@ for /F "delims=" %%i in ('wmic path win32_VideoController get name') do (
 endlocal & set NVIDIA_GPU_EXISTS=%NVIDIA_GPU_EXISTS%
 
 if "%PACKAGE_TYPE%" == "wheel" goto wheel
-if "%PACKAGE_TYPE%" == "conda" goto conda
 if "%PACKAGE_TYPE%" == "libtorch" goto libtorch
 
 echo "unknown package type"
@@ -37,6 +36,7 @@ exit /b 1
 echo "install wheel package"
 
 set PYTHON_INSTALLER_URL=
+if "%DESIRED_PYTHON%" == "3.13t" set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.13.0/python-3.13.0-amd64.exe"
 if "%DESIRED_PYTHON%" == "3.13" set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.13.0/python-3.13.0-amd64.exe"
 if "%DESIRED_PYTHON%" == "3.12" set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.12.0/python-3.12.0-amd64.exe"
 if "%DESIRED_PYTHON%" == "3.11" set "PYTHON_INSTALLER_URL=https://www.python.org/ftp/python/3.11.0/python-3.11.0-amd64.exe"
@@ -47,6 +47,13 @@ if "%PYTHON_INSTALLER_URL%" == "" (
     echo Python %DESIRED_PYTHON% not supported yet
 )
 
+set ADDITIONAL_OPTIONS=""
+set PYTHON_EXEC="python"
+if "%DESIRED_PYTHON%" == "3.13t" (
+    set ADDITIONAL_OPTIONS="Include_freethreaded=1"
+    set PYTHON_EXEC="python3.13t"
+)
+
 del python-amd64.exe
 curl --retry 3 -kL "%PYTHON_INSTALLER_URL%" --output python-amd64.exe
 if errorlevel 1 exit /b 1
@@ -55,85 +62,30 @@ if errorlevel 1 exit /b 1
 :: the installed Python to PATH system-wide. Even calling set PATH=%ORIG_PATH% later on won't make
 :: a change. As the builder directory will be removed after the smoke test, all subsequent non-binary
 :: jobs will fail to find any Python executable there
-start /wait "" python-amd64.exe /quiet InstallAllUsers=1 PrependPath=0 Include_test=0 TargetDir=%CD%\Python
+start /wait "" python-amd64.exe /quiet InstallAllUsers=1 PrependPath=0 Include_test=0 %ADDITIONAL_OPTIONS% TargetDir=%CD%\Python
 if errorlevel 1 exit /b 1
 
 set "PATH=%CD%\Python%PYTHON_VERSION%\Scripts;%CD%\Python;%PATH%"
-
-if "%DESIRED_PYTHON%" == "3.13" pip install -q --pre numpy==2.1.0 protobuf
-if "%DESIRED_PYTHON%" == "3.12" pip install -q --pre numpy==2.0.2 protobuf
-if "%DESIRED_PYTHON%" == "3.11" pip install -q --pre numpy==2.0.2 protobuf
-if "%DESIRED_PYTHON%" == "3.10" pip install -q --pre numpy==2.0.2 protobuf
-if "%DESIRED_PYTHON%" == "3.9" pip install -q --pre numpy==2.0.2 protobuf
-if "%DESIRED_PYTHON%" == "3.8" pip install -q numpy protobuf
+if "%DESIRED_PYTHON%" == "3.13t" %PYTHON_EXEC% -m pip install --pre numpy==2.2.1 protobuf
+if "%DESIRED_PYTHON%" == "3.13" %PYTHON_EXEC% -m pip install --pre numpy==2.1.2 protobuf
+if "%DESIRED_PYTHON%" == "3.12" %PYTHON_EXEC% -m pip install --pre numpy==2.0.2 protobuf
+if "%DESIRED_PYTHON%" == "3.11" %PYTHON_EXEC% -m pip install --pre numpy==2.0.2 protobuf
+if "%DESIRED_PYTHON%" == "3.10" %PYTHON_EXEC% -m pip install --pre numpy==2.0.2 protobuf
+if "%DESIRED_PYTHON%" == "3.9" %PYTHON_EXEC% -m pip install --pre numpy==2.0.2 protobuf
 
 if errorlevel 1 exit /b 1
 
-for /F "delims=" %%i in ('where /R "%PYTORCH_FINAL_PACKAGE_DIR:/=\%" *.whl') do pip install "%%i"
+for /F "delims=" %%i in ('where /R "%PYTORCH_FINAL_PACKAGE_DIR:/=\%" *.whl') do %PYTHON_EXEC% -m pip install "%%i"
 if errorlevel 1 exit /b 1
 
 goto smoke_test
 
-:conda
-echo "install conda package"
-
-:: Install Miniconda3
-set "CONDA_HOME=%CD%\conda"
-set "tmp_conda=%CONDA_HOME%"
-set "miniconda_exe=%CD%\miniconda.exe"
-set "CONDA_EXTRA_ARGS=cpuonly -c pytorch-nightly"
-if "%CUDA_VERSION%" == "118" (
-    set "CONDA_EXTRA_ARGS=pytorch-cuda=11.8 -c nvidia -c pytorch-nightly"
-)
-if "%CUDA_VERSION%" == "121" (
-    set "CONDA_EXTRA_ARGS=pytorch-cuda=12.1 -c nvidia -c pytorch-nightly"
-)
-if "%CUDA_VERSION%" == "124" (
-    set "CONDA_EXTRA_ARGS=pytorch-cuda=12.4 -c nvidia -c pytorch-nightly"
-)
-if "%CUDA_VERSION%" == "126" (
-    set "CONDA_EXTRA_ARGS=pytorch-cuda=12.6 -c nvidia -c pytorch-nightly"
-)
-
-rmdir /s /q conda
-del miniconda.exe
-curl -k https://repo.anaconda.com/miniconda/Miniconda3-latest-Windows-x86_64.exe -o "%miniconda_exe%"
-start /wait "" "%miniconda_exe%" /S /InstallationType=JustMe /RegisterPython=0 /AddToPath=0 /D=%tmp_conda%
-if ERRORLEVEL 1 exit /b 1
-
-set "PATH=%CONDA_HOME%;%CONDA_HOME%\scripts;%CONDA_HOME%\Library\bin;%PATH%"
-
-conda create -qyn testenv python=%DESIRED_PYTHON%
-if errorlevel 1 exit /b 1
-call conda install -yq conda-build
-if errorlevel 1 exit /b 1
-call %CONDA_HOME%\condabin\activate.bat testenv
-if errorlevel 1 exit /b 1
-set "NO_ARCH_PATH=%PYTORCH_FINAL_PACKAGE_DIR:/=\%\noarch"
-mkdir %NO_ARCH_PATH%
-for /F "delims=" %%i in ('where /R "%PYTORCH_FINAL_PACKAGE_DIR:/=\%" *') do xcopy "%%i" %NO_ARCH_PATH% /Y
-if ERRORLEVEL 1 exit /b 1
-call conda index %PYTORCH_FINAL_PACKAGE_DIR%
-if errorlevel 1 exit /b 1
-call conda install -yq -c "file:///%PYTORCH_FINAL_PACKAGE_DIR%" pytorch==%PYTORCH_BUILD_VERSION% -c pytorch -c numba/label/dev -c nvidia
-if ERRORLEVEL 1 exit /b 1
-call conda install -yq numpy
-if ERRORLEVEL 1 exit /b 1
-
-set /a CUDA_VER=%CUDA_VERSION%
-set CUDA_VER_MAJOR=%CUDA_VERSION:~0,-1%
-set CUDA_VER_MINOR=%CUDA_VERSION:~-1,1%
-set CUDA_VERSION_STR=%CUDA_VER_MAJOR%.%CUDA_VER_MINOR%
-
-:: Install package we just build
-
-
 :smoke_test
-python -c "import torch"
+%PYTHON_EXEC% -c "import torch"
 if ERRORLEVEL 1 exit /b 1
 
 echo Checking that MKL is available
-python -c "import torch; exit(0 if torch.backends.mkl.is_available() else 1)"
+%PYTHON_EXEC% -c "import torch; exit(0 if torch.backends.mkl.is_available() else 1)"
 if ERRORLEVEL 1 exit /b 1
 
 if "%NVIDIA_GPU_EXISTS%" == "0" (
@@ -142,24 +94,24 @@ if "%NVIDIA_GPU_EXISTS%" == "0" (
 )
 
 echo Checking that CUDA archs are setup correctly
-python -c "import torch; torch.randn([3,5]).cuda()"
+%PYTHON_EXEC% -c "import torch; torch.randn([3,5]).cuda()"
 if ERRORLEVEL 1 exit /b 1
 
 echo Checking that magma is available
-python -c "import torch; torch.rand(1).cuda(); exit(0 if torch.cuda.has_magma else 1)"
+%PYTHON_EXEC% -c "import torch; torch.rand(1).cuda(); exit(0 if torch.cuda.has_magma else 1)"
 if ERRORLEVEL 1 exit /b 1
 
 echo Checking that CuDNN is available
-python -c "import torch; exit(0 if torch.backends.cudnn.is_available() else 1)"
+%PYTHON_EXEC% -c "import torch; exit(0 if torch.backends.cudnn.is_available() else 1)"
 if ERRORLEVEL 1 exit /b 1
 
 echo Checking that basic RNN works
-python %PYTORCH_ROOT%\.ci\pytorch\test_example_code\rnn_smoke.py
+%PYTHON_EXEC% %PYTORCH_ROOT%\.ci\pytorch\test_example_code\rnn_smoke.py
 
 if ERRORLEVEL 1 exit /b 1
 
 echo Checking that basic CNN works
-python %PYTORCH_ROOT%\.ci\pytorch\test_example_code\cnn_smoke.py
+%PYTHON_EXEC% %PYTORCH_ROOT%\.ci\pytorch\test_example_code\cnn_smoke.py
 if ERRORLEVEL 1 exit /b 1
 
 goto end
@@ -167,7 +119,6 @@ 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
@@ -179,10 +130,6 @@ 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/static_lib_test.bat

@@ -70,7 +70,6 @@ echo "install and test libtorch"
 pip install cmake
 echo "installing cmake"
 
-if "%VC_YEAR%" == "2019" powershell internal\vs2019_install.ps1
 if "%VC_YEAR%" == "2022" powershell internal\vs2022_install.ps1
 
 if ERRORLEVEL 1 exit /b 1
@@ -83,10 +82,6 @@ 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

@@ -1,12 +1,8 @@
-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
-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

@@ -1,48 +0,0 @@
-# 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

@@ -7,6 +7,9 @@ if not "%CUDA_VERSION%" == "xpu" (
     exit /b 0
 )
 
+set SRC_DIR=%NIGHTLIES_PYTORCH_ROOT%
+if not exist "%SRC_DIR%\temp_build" mkdir "%SRC_DIR%\temp_build"
+
 set XPU_INSTALL_MODE=%~1
 if "%XPU_INSTALL_MODE%"=="" goto xpu_bundle_install_start
 if "%XPU_INSTALL_MODE%"=="bundle" goto xpu_bundle_install_start
@@ -117,3 +120,14 @@ if errorlevel 1 exit /b 1
 del xpu_extra.exe
 
 :xpu_install_end
+
+if not "%XPU_ENABLE_KINETO%"=="1" goto install_end
+:: Install Level Zero SDK
+set XPU_EXTRA_LZ_URL=https://github.com/oneapi-src/level-zero/releases/download/v1.14.0/level-zero-sdk_1.14.0.zip
+curl -k -L %XPU_EXTRA_LZ_URL% --output "%SRC_DIR%\temp_build\level_zero_sdk.zip"
+echo "Installing level zero SDK..."
+7z x "%SRC_DIR%\temp_build\level_zero_sdk.zip" -o"%SRC_DIR%\temp_build\level_zero"
+set "INCLUDE=%SRC_DIR%\temp_build\level_zero\include;%INCLUDE%"
+del "%SRC_DIR%\temp_build\level_zero_sdk.zip"
+
+:install_end

.ci/pytorch/windows/xpu.bat

@@ -28,11 +28,6 @@ call "%XPU_BUNDLE_ROOT%\compiler\latest\env\vars.bat"
 call "%XPU_BUNDLE_ROOT%\ocloc\latest\env\vars.bat"
 IF ERRORLEVEL 1 goto :eof
 
-:: Workaround for https://github.com/pytorch/pytorch/issues/134989
-set CMAKE_SHARED_LINKER_FLAGS=/FORCE:MULTIPLE
-set CMAKE_MODULE_LINKER_FLAGS=/FORCE:MULTIPLE
-set CMAKE_EXE_LINKER_FLAGS=/FORCE:MULTIPLE
-
 if exist "%NIGHTLIES_PYTORCH_ROOT%" cd %NIGHTLIES_PYTORCH_ROOT%\..
 call %~dp0\internal\copy_cpu.bat
 IF ERRORLEVEL 1 goto :eof

.ci/wheel/build_wheel.sh

@@ -130,7 +130,19 @@ export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
 SETUPTOOLS_PINNED_VERSION="=46.0.0"
 PYYAML_PINNED_VERSION="=5.3"
 EXTRA_CONDA_INSTALL_FLAGS=""
+CONDA_ENV_CREATE_FLAGS=""
+RENAME_WHEEL=true
 case $desired_python in
+    3.13t)
+        echo "Using 3.13 deps"
+        SETUPTOOLS_PINNED_VERSION=">=68.0.0"
+        PYYAML_PINNED_VERSION=">=6.0.1"
+        NUMPY_PINNED_VERSION="=2.1.0"
+        CONDA_ENV_CREATE_FLAGS="python-freethreading"
+        EXTRA_CONDA_INSTALL_FLAGS="-c conda-forge"
+        desired_python="3.13"
+        RENAME_WHEEL=false
+        ;;
     3.13)
         echo "Using 3.13 deps"
         SETUPTOOLS_PINNED_VERSION=">=68.0.0"
@@ -169,16 +181,15 @@ esac
 
 # Install into a fresh env
 tmp_env_name="wheel_py$python_nodot"
-conda create ${EXTRA_CONDA_INSTALL_FLAGS} -yn "$tmp_env_name" python="$desired_python"
+conda create ${EXTRA_CONDA_INSTALL_FLAGS} -yn "$tmp_env_name" python="$desired_python" ${CONDA_ENV_CREATE_FLAGS}
 source activate "$tmp_env_name"
 
-pip install -q "numpy=${NUMPY_PINNED_VERSION}"  "pyyaml${PYYAML_PINNED_VERSION}" requests
-retry conda install ${EXTRA_CONDA_INSTALL_FLAGS} -yq  llvm-openmp=14.0.6 cmake ninja "setuptools${SETUPTOOLS_PINNED_VERSION}" typing_extensions
-retry pip install -qr "${pytorch_rootdir}/requirements.txt" || true
+pip install "numpy=${NUMPY_PINNED_VERSION}"  "pyyaml${PYYAML_PINNED_VERSION}" requests ninja "setuptools${SETUPTOOLS_PINNED_VERSION}" typing_extensions
+retry pip install -r "${pytorch_rootdir}/requirements.txt" || true
+retry brew install libomp
 
-# For USE_DISTRIBUTED=1 on macOS, need libuv and pkg-config to find libuv.
+# For USE_DISTRIBUTED=1 on macOS, need libuv, which is build as part of tensorpipe submodule
 export USE_DISTRIBUTED=1
-retry conda install ${EXTRA_CONDA_INSTALL_FLAGS} -yq libuv pkg-config
 
 if [[ -n "$CROSS_COMPILE_ARM64" ]]; then
     export CMAKE_OSX_ARCHITECTURES=arm64
@@ -220,30 +231,13 @@ echo "The wheel is in $(find $whl_tmp_dir -name '*.whl')"
 wheel_filename_gen=$(find $whl_tmp_dir -name '*.whl' | head -n1 | xargs -I {} basename {})
 popd
 
-if [[ -z "$BUILD_PYTHONLESS" ]]; then
+if [[ -z "$BUILD_PYTHONLESS" && $RENAME_WHEEL == true  ]]; then
     # Copy the whl to a final destination before tests are run
     echo "Renaming Wheel file: $wheel_filename_gen to $wheel_filename_new"
     cp "$whl_tmp_dir/$wheel_filename_gen" "$PYTORCH_FINAL_PACKAGE_DIR/$wheel_filename_new"
-
-    ##########################
-    # now test the binary, unless it's cross compiled arm64
-    if [[ -z "$CROSS_COMPILE_ARM64" ]]; then
-        pip uninstall -y "$TORCH_PACKAGE_NAME" || true
-        pip uninstall -y "$TORCH_PACKAGE_NAME" || true
-
-        # Create new "clean" conda environment for testing
-        conda create ${EXTRA_CONDA_INSTALL_FLAGS} -yn "test_conda_env" python="$desired_python"
-        conda activate test_conda_env
-
-        pip install "$PYTORCH_FINAL_PACKAGE_DIR/$wheel_filename_new" -v
-
-        echo "$(date) :: Running tests"
-        # TODO: Add real tests, as run_test.sh from builder is a glorified no-op
-        # pushd "$pytorch_rootdir"
-        # "${SOURCE_DIR}/../run_tests.sh" 'wheel' "$desired_python" 'cpu'
-        # popd
-        echo "$(date) :: Finished tests"
-    fi
+elif [[ $RENAME_WHEEL == false ]]; then
+    echo "Copying Wheel file: $wheel_filename_gen to $PYTORCH_FINAL_PACKAGE_DIR"
+    cp "$whl_tmp_dir/$wheel_filename_gen" "$PYTORCH_FINAL_PACKAGE_DIR/$wheel_filename_gen"
 else
     pushd "$pytorch_rootdir"
 

.circleci/codegen_validation/normalize_yaml_fragment.py

@@ -7,7 +7,7 @@ import yaml
 
 
 # Need to import modules that lie on an upward-relative path
-sys.path.append(os.path.join(sys.path[0], ".."))
+sys.path.append(os.path.dirname(sys.path[0]))
 
 import cimodel.lib.miniyaml as miniyaml
 

.circleci/scripts/binary_linux_test.sh

@@ -94,6 +94,8 @@ if [[ "\$GPU_ARCH_TYPE" != *s390x* && "\$GPU_ARCH_TYPE" != *xpu* && "\$GPU_ARCH_
   python /pytorch/.ci/pytorch/smoke_test/smoke_test.py --package=torchonly --torch-compile-check disabled
 fi
 
+# Clean temp files
+cd /pytorch/.ci/pytorch/ && git clean -ffdx
 
 # =================== The above code will be executed inside Docker container ===================
 EOL

.circleci/scripts/binary_macos_build.sh

@@ -1,11 +0,0 @@
-#!/bin/bash
-set -eux -o pipefail
-
-source "${BINARY_ENV_FILE:-/Users/distiller/project/env}"
-mkdir -p "$PYTORCH_FINAL_PACKAGE_DIR"
-
-# Build
-export USE_PYTORCH_METAL_EXPORT=1
-export USE_COREML_DELEGATE=1
-export TORCH_PACKAGE_NAME="$(echo $TORCH_PACKAGE_NAME | tr '-' '_')"
-"${PYTORCH_ROOT}/.ci/wheel/build_wheel.sh"

.circleci/scripts/binary_populate_env.sh

@@ -30,9 +30,7 @@ fi
 # Pick docker image
 export DOCKER_IMAGE=${DOCKER_IMAGE:-}
 if [[ -z "$DOCKER_IMAGE" ]]; then
-  if [[ "$PACKAGE_TYPE" == conda ]]; then
-    export DOCKER_IMAGE="pytorch/conda-cuda"
-  elif [[ "$DESIRED_CUDA" == cpu ]]; then
+  if [[ "$DESIRED_CUDA" == cpu ]]; then
     export DOCKER_IMAGE="pytorch/manylinux:cpu"
   else
     export DOCKER_IMAGE="pytorch/manylinux-builder:${DESIRED_CUDA:2}"
@@ -63,7 +61,7 @@ if tagged_version >/dev/null; then
   # Turns tag v1.6.0-rc1 -> v1.6.0
   BASE_BUILD_VERSION="$(tagged_version | sed -e 's/^v//' -e 's/-.*$//')"
 fi
-if [[ "$(uname)" == 'Darwin' ]] || [[ "$PACKAGE_TYPE" == conda ]]; then
+if [[ "$(uname)" == 'Darwin' ]]; then
   export PYTORCH_BUILD_VERSION="${BASE_BUILD_VERSION}"
 else
   export PYTORCH_BUILD_VERSION="${BASE_BUILD_VERSION}+$DESIRED_CUDA"
@@ -75,9 +73,8 @@ export PYTORCH_BUILD_NUMBER=1
 TRITON_VERSION=$(cat $PYTORCH_ROOT/.ci/docker/triton_version.txt)
 
 # Here PYTORCH_EXTRA_INSTALL_REQUIREMENTS is already set for the all the wheel builds hence append TRITON_CONSTRAINT
-TRITON_CONSTRAINT="platform_system == 'Linux' and platform_machine == 'x86_64' and python_version < '3.13'"
-if [[ "$PACKAGE_TYPE" =~ .*wheel.* &&  -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]]; then
-  # Only linux Python < 3.13 are supported wheels for triton
+TRITON_CONSTRAINT="platform_system == 'Linux' and platform_machine == 'x86_64'"
+if [[ "$PACKAGE_TYPE" =~ .*wheel.* &&  -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" && ! "$PYTORCH_BUILD_VERSION" =~ .*xpu.* ]]; then
   TRITON_REQUIREMENT="triton==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"
   if [[ -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*dev.* ]]; then
       TRITON_SHORTHASH=$(cut -c1-8 $PYTORCH_ROOT/.ci/docker/ci_commit_pins/triton.txt)
@@ -150,8 +147,6 @@ export PYTORCH_EXTRA_INSTALL_REQUIREMENTS="${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:
 
 # TODO: We don't need this anymore IIUC
 export TORCH_PACKAGE_NAME='torch'
-export TORCH_CONDA_BUILD_FOLDER='pytorch-nightly'
-export ANACONDA_USER='pytorch'
 
 export USE_FBGEMM=1
 export PIP_UPLOAD_FOLDER="$PIP_UPLOAD_FOLDER"

.circleci/scripts/binary_upload.sh

@@ -2,7 +2,7 @@
 
 set -euo pipefail
 
-PACKAGE_TYPE=${PACKAGE_TYPE:-conda}
+PACKAGE_TYPE=${PACKAGE_TYPE:-wheel}
 
 PKG_DIR=${PKG_DIR:-/tmp/workspace/final_pkgs}
 
@@ -18,10 +18,8 @@ BUILD_NAME=${BUILD_NAME:-}
 
 DRY_RUN=${DRY_RUN:-enabled}
 # Don't actually do work unless explicit
-ANACONDA="true anaconda"
 AWS_S3_CP="aws s3 cp --dryrun"
 if [[ "${DRY_RUN}" = "disabled" ]]; then
-  ANACONDA="anaconda"
   AWS_S3_CP="aws s3 cp"
 fi
 
@@ -34,10 +32,6 @@ if [[ ${BUILD_NAME} == *-full* ]]; then
   UPLOAD_SUBFOLDER="${UPLOAD_SUBFOLDER}_full"
 fi
 
-# Sleep 2 minutes between retries for conda upload
-retry () {
-  "$@"  || (sleep 5m && "$@") || (sleep 5m && "$@") || (sleep 5m && "$@") || (sleep 5m && "$@")
-}
 
 do_backup() {
   local backup_dir
@@ -49,20 +43,6 @@ do_backup() {
   )
 }
 
-conda_upload() {
-  (
-    set -x
-    retry \
-    ${ANACONDA} \
-    upload  \
-    ${PKG_DIR}/*.tar.bz2 \
-    -u "pytorch-${UPLOAD_CHANNEL}" \
-    --label main \
-    --no-progress \
-    --force
-  )
-}
-
 s3_upload() {
   local extension
   local pkg_type
@@ -78,31 +58,18 @@ s3_upload() {
     for pkg in ${PKG_DIR}/*.${extension}; do
       (
         set -x
-        ${AWS_S3_CP} --no-progress --acl public-read "${pkg}" "${s3_upload_dir}"
+        shm_id=$(sha256sum "${pkg}" | awk '{print $1}')
+        ${AWS_S3_CP} --no-progress --acl public-read "${pkg}" "${s3_upload_dir}" \
+          --metadata "checksum-sha256=${shm_id}"
       )
     done
   )
 }
 
 # Install dependencies (should be a no-op if previously installed)
-conda install -yq anaconda-client
-pip install -q awscli
+pip install -q awscli uv
 
 case "${PACKAGE_TYPE}" in
-  conda)
-    conda_upload
-    for conda_archive in ${PKG_DIR}/*.tar.bz2; do
-      # Fetch  platform (eg. win-64, linux-64, etc.) from index file because
-      # there's no actual conda command to read this
-      subdir=$(\
-        tar -xOf "${conda_archive}" info/index.json \
-          | grep subdir  \
-          | cut -d ':' -f2 \
-          | sed -e 's/[[:space:]]//' -e 's/"//g' -e 's/,//' \
-      )
-      BACKUP_DIR="conda/${subdir}"
-    done
-    ;;
   libtorch)
     s3_upload "zip" "libtorch"
     BACKUP_DIR="libtorch/${UPLOAD_CHANNEL}/${UPLOAD_SUBFOLDER}"

.circleci/scripts/binary_windows_build.sh

@@ -8,10 +8,9 @@ export CUDA_VERSION="${DESIRED_CUDA/cu/}"
 export USE_SCCACHE=1
 export SCCACHE_BUCKET=ossci-compiler-cache
 export SCCACHE_IGNORE_SERVER_IO_ERROR=1
-export VC_YEAR=2019
+export VC_YEAR=2022
 
 if [[ "$DESIRED_CUDA" == 'xpu' ]]; then
-    export VC_YEAR=2022
     export USE_SCCACHE=0
     export XPU_VERSION=2025.0
 fi

.circleci/scripts/binary_windows_test.sh

@@ -4,10 +4,9 @@ set -eux -o pipefail
 source "${BINARY_ENV_FILE:-/c/w/env}"
 
 export CUDA_VERSION="${DESIRED_CUDA/cu/}"
-export VC_YEAR=2019
+export VC_YEAR=2022
 
 if [[ "$DESIRED_CUDA" == 'xpu' ]]; then
-    export VC_YEAR=2022
     export XPU_VERSION=2025.0
 fi
 

.clang-format

@@ -106,6 +106,8 @@ StatementMacros:
   - C10_DEFINE_int32
   - C10_DEFINE_int64
   - C10_DEFINE_string
+  - C10_DEFINE_REGISTRY_WITHOUT_WARNING
+  - C10_REGISTER_CREATOR
   - DEFINE_BINARY
   - PyObject_HEAD
   - PyObject_VAR_HEAD

.clang-tidy

@@ -1,8 +1,9 @@
 ---
 # NOTE there must be no spaces before the '-', so put the comma last.
-# The check bugprone-unchecked-optional-access is also turned off atm
-# because it causes clang-tidy to hang randomly. The tracking issue
+# The check bugprone-unchecked-optional-access is also turned on.
+# Note that it can cause clang-tidy to hang randomly. The tracking issue
 # can be found at https://github.com/llvm/llvm-project/issues/69369.
+# When that happens, we can disable it on the problematic code by NOLINT.
 InheritParentConfig: true
 Checks: '
 bugprone-*,
@@ -12,7 +13,10 @@ bugprone-*,
 -bugprone-lambda-function-name,
 -bugprone-reserved-identifier,
 -bugprone-swapped-arguments,
--bugprone-unchecked-optional-access,
+clang-analyzer-core.*,
+clang-analyzer-cplusplus.*,
+clang-analyzer-nullability.*,
+clang-analyzer-deadcode.*,
 clang-diagnostic-missing-prototypes,
 cppcoreguidelines-*,
 -cppcoreguidelines-avoid-do-while,
@@ -55,10 +59,11 @@ readability-container-size-empty,
 readability-delete-null-pointer,
 readability-duplicate-include
 readability-misplaced-array-index,
-readability-redundant-function-ptr-dereference,
-readability-redundant-smartptr-get,
+readability-redundant*
 readability-simplify-subscript-expr,
 readability-string-compare,
+-readability-redundant-access-specifiers,
+-readability-redundant-control-flow,
 '
 HeaderFilterRegex: '^(aten/|c10/|torch/).*$'
 WarningsAsErrors: '*'

.git-blame-ignore-revs

@@ -24,6 +24,10 @@ e3900d2ba5c9f91a24a9ce34520794c8366d5c54
 2e26976ad3b06ce95dd6afccfdbe124802edf28f
 # 2021-06-07 Strictly typed everything in `.github` and `tools`
 737d920b21db9b4292d056ee1329945990656304
+# 2021-08-12 [codemod][lint][fbcode/c*] Enable BLACK by default
+b0043072529b81276a69df29e00555333117646c
+# 2021-08-25 Reformat run_test.py
+67d8e7b659b19e1ee68208b28bfa7dba73375dbc
 # 2022-06-09 Apply clang-format to ATen headers
 95b15c266baaf989ef7b6bbd7c23a2d90bacf687
 # 2022-06-11 [lint] autoformat test/cpp and torch/csrc
@@ -44,3 +48,57 @@ a53cda1ddc15336dc1ff0ce1eff2a49cdc5f882e
 d80939e5e9337e8078f11489afefec59fd42f93b
 # 2024-06-28 enable UFMT in `torch.utils.data`
 7cf0b90e49689d45be91aa539fdf54cf2ea8a9a3
+# 2024-07-03 Enable UFMT on test/test_public_bindings.py (#128389)
+fe5424d0f8604f6e66d827ae9f94b05cb7119d55
+# 2024-07-03 Enable UFMT on test/test_public_bindings.py (#128389)
+c686304277f7cd72331f685605325498cff94a0b
+# 2024-07-15 Enable UFMT on all of torch/sparse (#130545)
+535016967ae65a6027f83d6b935a985996223d49
+# 2024-07-15 [BE][Easy][1/19] enforce style for empty lines in import segments (#129752)
+a3abfa5cb57203b6a8ba7dff763f4057db8282a8
+# 2024-07-15 [BE][Easy][2/19] enforce style for empty lines in import segments in `.ci/` and `.github/` (#129753)
+ba48cf653541e9160dfdefa7bfea885c22e48608
+# 2024-07-16 [BE][Easy][5/19] enforce style for empty lines in import segments in `tools/` and `torchgen/` (#129756)
+f6838d521a243dbedc50ae96575720bf2cc8a8ad
+# 2024-07-17 [BE][Easy][9/19] enforce style for empty lines in import segments in `test/[e-h]*/` (#129760)
+76169cf69184bd462b9add40f893f57675f8a057
+# 2024-07-16 [BE][Easy][3/19] enforce style for empty lines in import segments in `benchmarks/` (#129754)
+c0ed38e644aed812d76b0ec85fae2f6019bf462b
+# 2024-07-16 [BE][Easy][4/19] enforce style for empty lines in import segments in `functorch/` (#129755)
+740fb229660f388feddc288c127ab12c82e67d36
+# 2024-07-17 [BE][Easy][12/19] enforce style for empty lines in import segments in `test/i*/` (#129763)
+aecc746fccc4495313167e3a7f94210daf457e1d
+# 2024-07-18 Revert "[BE][Easy][12/19] enforce style for empty lines in import segments in `test/i*/` (#129763)"
+b732b52f1e4378f8486ceb5e7026be3321c2651c
+# 2024-07-18 [BE][Easy][12/19] enforce style for empty lines in import segments in `test/i*/` (#129763)
+134bc4fc34bb02795aa694e66b132dcea5dde1e1
+# 2024-07-26 [BE][Easy][8/19] enforce style for empty lines in import segments in `test/[k-p]*/` (#129759)
+fbe6f42dcf1834213e0baa87b87529161df3c4d7
+# 2024-07-31 [BE][Easy][14/19] enforce style for empty lines in import segments in `torch/_[a-c]*/` and `torch/_[e-h]*/` and `torch/_[j-z]*/` (#129765)
+e7eeee473c6cb45942e87de5a616b0eb635513d6
+# 2024-07-31 Fix lint after PR #130572 (#132316)
+d72e863b3ecd3de4c8ea00518e110da964583f4f
+# 2024-07-31 [BE][Easy][15/19] enforce style for empty lines in import segments in `torch/_d*/` (#129767)
+e74ba1b34a476b46e76b4e32afe2d481f97e9a47
+# 2024-07-31 [BE][Easy][18/19] enforce style for empty lines in import segments in `torch/d*/` (#129770)
+b25ef91bf158ce459d8654e33c50c8e6ed8db716
+# 2024-07-20 [BE][Easy][13/19] enforce style for empty lines in import segments in `test/j*/` (#129764)
+6ff1e43a416c43cd82b210e22ac47384494c172e
+# 2024-11-01 [Lint] Clang-format all metal kernels (#139530)
+b3ad45733bd908b7358959ca1e1f8d026f4507eb
+# 2024-11-17 [BE][MPS] Apply clang-format to mps headers (#140906)
+99014a297c179862af38ee86bac2051434d3db41
+# 2024-11-27 Apply clang-format for ATen/core/boxing headers (#141105)
+19d01a1ef0c0d65768eb0a5c97a25328eec57fbd
+# 2024-12-05 fix the lint from D66795414 (#142122)
+65c2086d452ae6966ce9d7fb3cb2eef2fd0d2add
+# 2024-12-20 Apply clang-format for ATen/core/dispatch headers (#143620)
+cee06e74eeb54994b97000a02b715a4e63a97951
+# 2024-12-22 Better fix for f-strings in set_linter for py3.12 (#143725)
+eebc93d41eeffb936cbf20c9052e1e813d0cc052
+# 2025-01-04 [mps/BE] Fix linter warning/advice. (#144199)
+0dc1e6be192b260f1c072d70e1b06a3ac8e109fa
+# 2025-01-07 Fix lint in `test_provenance_tracing.py` (#144296)
+61c0a3d1cbaf6420e40ab0f9c9019daa21145e69
+# 2025-01-09 [BE] fix ruff rule E226: add missing whitespace around operator in f-strings (#144415)
+dcc3cf7066b4d8cab63ecb73daf1e36b01220a4e

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -5,7 +5,7 @@ body:
 - type: markdown
   attributes:
     value: >
-      #### Before submitting a bug, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/pytorch/pytorch/issues?q=is%3Aissue+sort%3Acreated-desc+).
+      #### Before submitting a bug, please make sure the issue hasn't been already addressed by searching through [the existing and past issues](https://github.com/pytorch/pytorch/issues?q=is%3Aissue+sort%3Acreated-desc+). Note: Please write your bug report in English to ensure it can be understood and addressed by the development team. If you are filing a bug for torch.compile, please use the [torch.compile issue template](https://github.com/pytorch/pytorch/issues/new?q=sort%3Aupdated-desc+is%3Aissue+is%3Aopen&template=pt2-bug-report.yml).
 - type: textarea
   attributes:
     label: 🐛 Describe the bug

.github/ISSUE_TEMPLATE/disable-ci-jobs.md

@@ -5,7 +5,7 @@ title: "DISABLED [WORKFLOW_NAME] / [PLATFORM_NAME] / [JOB_NAME]"
 labels: "module: ci"
 ---
 
-> For example, DISABLED pull / win-vs2019-cpu-py3 / test (default). Once
+> For example, DISABLED pull / win-vs2022-cpu-py3 / test (default). Once
 > created, the job will be disabled within 15 minutes. You can check the
 > list of disabled jobs at https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json
 

.github/ISSUE_TEMPLATE/documentation.yml

@@ -2,6 +2,10 @@ name: 📚 Documentation
 description: Report an issue related to https://pytorch.org/docs/stable/index.html
 
 body:
+- type: markdown
+  attributes:
+    value: >
+      #### Note: Please report your documentation issue in English to ensure it can be understood and addressed by the development team.
 - type: textarea
   attributes:
     label: 📚 The doc issue

.github/ISSUE_TEMPLATE/feature-request.yml

@@ -2,6 +2,10 @@ name: 🚀 Feature request
 description: Submit a proposal/request for a new PyTorch feature
 
 body:
+- type: markdown
+  attributes:
+    value: >
+      #### Note: Please write your feature request in English to ensure it can be understood and addressed by the development team.
 - type: textarea
   attributes:
     label: 🚀 The feature, motivation and pitch

.github/ISSUE_TEMPLATE/pt2-bug-report.yml

@@ -3,6 +3,10 @@ description: Create a report to help us reproduce and fix the bug
 labels: ["oncall: pt2"]
 
 body:
+  - type: markdown
+    attributes:
+      value: >
+        #### Note: Please write your bug report in English to ensure it can be understood and addressed by the development team.
   - type: markdown
     attributes:
       value: >
@@ -18,6 +22,8 @@ body:
 
         - If comparing eager and torch.compile at fp16/bf16, you should use fp32 as baseline
 
+        - Ensure rng state used to compare results is equivalent. Use `torch._inductor.config.fallback_random=True` and reset the torch rng seed between comparisons
+
         If the above requirements are met, add the label "topic: fuzzer" to your issue.
 
   - type: textarea

.github/actionlint.yaml

@@ -42,8 +42,10 @@ self-hosted-runner:
     - windows.8xlarge.nvidia.gpu
     - windows.8xlarge.nvidia.gpu.nonephemeral
     - windows.g5.4xlarge.nvidia.gpu
-    # Organization-wide AMD hosted MI300 runners
+    # Organization-wide AMD hosted runners
     - linux.rocm.gpu
+    - linux.rocm.gpu.2
+    - linux.rocm.gpu.4
     # Repo-specific Apple hosted  runners
     - macos-m1-ultra
     - macos-m2-14

.github/actions/checkout-pytorch/action.yml

@@ -41,10 +41,10 @@ runs:
         mkdir "${GITHUB_WORKSPACE}"
 
     - name: Checkout PyTorch
-      uses: malfet/checkout@silent-checkout
+      uses: actions/checkout@v4
       with:
         ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
         # --depth=1 for speed, manually fetch history and other refs as necessary
         fetch-depth: ${{ inputs.fetch-depth }}
         submodules: ${{ inputs.submodules }}
-        quiet-checkout: true
+        show-progress: false

.github/actions/diskspace-cleanup/action.yml

@@ -17,6 +17,10 @@ runs:
         set -ex
         diskspace_cutoff=${{ inputs.diskspace-cutoff }}
         docker_root_dir=$(docker info -f '{{.DockerRootDir}}')
+        if [ ! -d "$docker_root_dir" ]; then
+            echo "Docker root directory ($docker_root_dir) does not exist. Skipping disk space check."
+            exit 0
+        fi
         diskspace=$(df -H --output=pcent ${docker_root_dir} | sed -n 2p | sed 's/%//' | sed 's/ //')
         msg="Please file an issue on pytorch/pytorch reporting the faulty runner. Include a link to the runner logs so the runner can be identified"
         if [[ "$diskspace" -ge "$diskspace_cutoff" ]] ; then

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

@@ -5,20 +5,6 @@ description: Set up ROCm host for CI
 runs:
   using: composite
   steps:
-    - name: Set DOCKER_HOST
-      shell: bash
-      run: echo "DOCKER_HOST=unix:///run/user/$(id -u)/docker.sock" >> "${GITHUB_ENV}"
-
-    - name: Remove leftover Docker config file
-      shell: bash
-      continue-on-error: true
-      run: |
-        set -ex
-
-        cat ~/.docker/config.json || true
-        # https://stackoverflow.com/questions/64455468/error-when-logging-into-ecr-with-docker-login-error-saving-credentials-not
-        rm -f ~/.docker/config.json
-
     - name: Stop all running docker containers
       if: always()
       shell: bash
@@ -38,6 +24,12 @@ runs:
         cat /opt/rocm/.info/version || true
         whoami
 
+    - name: Runner health check amdgpu info
+      if: always()
+      shell: bash
+      run: |
+        dpkg -l | grep -E "  amdgpu"
+
     - name: Runner health check rocm-smi
       if: always()
       shell: bash
@@ -68,7 +60,7 @@ runs:
         fi
 
     - name: Runner diskspace health check
-      uses: ./.github/actions/diskspace-cleanup
+      uses: pytorch/pytorch/.github/actions/diskspace-cleanup@main
       if: always()
 
     - name: Runner health check disconnect on failure
@@ -77,14 +69,44 @@ runs:
       run: |
         killall runsvc.sh
 
+    - name: Setup useful environment variables
+      shell: bash
+      run: |
+        RUNNER_ARTIFACT_DIR="${RUNNER_TEMP}/artifacts"
+        rm -rf "${RUNNER_ARTIFACT_DIR}"
+        mkdir -p "${RUNNER_ARTIFACT_DIR}"
+        echo "RUNNER_ARTIFACT_DIR=${RUNNER_ARTIFACT_DIR}" >> "${GITHUB_ENV}"
+
+        RUNNER_TEST_RESULTS_DIR="${RUNNER_TEMP}/test-results"
+        rm -rf "${RUNNER_TEST_RESULTS_DIR}"
+        mkdir -p "${RUNNER_TEST_RESULTS_DIR}"
+        echo "RUNNER_TEST_RESULTS_DIR=${RUNNER_TEST_RESULTS_DIR}" >> "${GITHUB_ENV}"
+
+        RUNNER_DOCS_DIR="${RUNNER_TEMP}/docs"
+        rm -rf "${RUNNER_DOCS_DIR}"
+        mkdir -p "${RUNNER_DOCS_DIR}"
+        echo "RUNNER_DOCS_DIR=${RUNNER_DOCS_DIR}" >> "${GITHUB_ENV}"
+
     - name: Preserve github env variables for use in docker
       shell: bash
       run: |
-        env | grep '^GITHUB' >> "/tmp/github_env_${GITHUB_RUN_ID}"
-        env | grep '^CI' >> "/tmp/github_env_${GITHUB_RUN_ID}"
+        env | grep '^GITHUB' >> "${RUNNER_TEMP}/github_env_${GITHUB_RUN_ID}"
+        env | grep '^CI' >> "${RUNNER_TEMP}/github_env_${GITHUB_RUN_ID}"
 
     - name: ROCm set GPU_FLAG
       shell: bash
       run: |
         # All GPUs are visible to the runner; visibility, if needed, will be set by run_test.py.
-        echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"
+        # Add render group for container creation.
+        render_gid=`cat /etc/group | grep render | cut -d: -f3`
+        # Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
+        if [ -f "/etc/podinfo/gha-render-devices" ]; then
+          DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
+        else
+          DEVICE_FLAG="--device /dev/dri"
+        fi
+        # The --group-add daemon and --group-add bin are needed in the Ubuntu 24.04 and Almalinux OSs respectively.
+        # This is due to the device files (/dev/kfd & /dev/dri) being owned by video group on bare metal.
+        # This video group ID maps to subgid 1 inside the docker image due to the /etc/subgid entries.
+        # The group name corresponding to group ID 1 can change depending on the OS, so both are necessary.
+        echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd $DEVICE_FLAG --group-add video --group-add $render_gid --group-add daemon --group-add bin --cap-add=SYS_PTRACE --security-opt seccomp=unconfined --network=host" >> "${GITHUB_ENV}"

.github/actions/test-pytorch-binary/action.yml

@@ -13,7 +13,6 @@ runs:
         container_name=$(docker run \
           ${GPU_FLAG:-} \
           -e BINARY_ENV_FILE \
-          -e BUILDER_ROOT \
           -e BUILD_ENVIRONMENT \
           -e DESIRED_CUDA \
           -e DESIRED_DEVTOOLSET \

.github/actions/upload-utilization-stats/action.yml

@@ -0,0 +1,56 @@
+name: upload-utilization-stats
+
+description: Upload utilization stats to artifacts
+
+inputs:
+    workflow_run_id:
+      type: string
+      description: 'workflow (run) id of the workflow the test is running'
+      required: True
+    workflow_attempt:
+      type: string
+      description: 'the workflow (run) attempt'
+      required: True
+    workflow_name:
+      description: 'name of the workflow'
+      type: string
+      required: True
+    job_id:
+      type: string
+      description: 'the job (run) id for the test'
+      required: True
+    job_name:
+      type: string
+      description: 'the job name of the test'
+      required: True
+
+runs:
+  using: composite
+  steps:
+    - name: Print Inputs
+      shell: bash
+      run: |
+        echo "workflow_id: ${{inputs.workflow_run_id}}"
+        echo "workflow_attempt: ${{inputs.workflow_attempt}}"
+        echo "workflow_Name: ${{inputs.workflow_name}}"
+        echo "job_id: ${{inputs.job_id}}"
+        echo "job_name:  ${{inputs.job_name}}"
+    - uses: nick-fields/retry@v3.0.0
+      name: Setup dependencies
+      with:
+        shell: bash
+        timeout_minutes: 5
+        max_attempts: 5
+        retry_wait_seconds: 30
+        command: |
+          set -eu
+          python3 -m pip install python-dateutil==2.8.2 boto3==1.35.42 pandas==2.1.3
+    - name: Upload utilizatoin stats to s3
+      shell: bash
+      run: |
+        python3 -m tools.stats.upload_utilization_stats.upload_utilization_stats \
+          --workflow-run-id "${{inputs.workflow_run_id}}" \
+          --workflow-name "${{inputs.workflow_name}}" \
+          --workflow-run-attempt "${{inputs.workflow_attempt}}" \
+          --job-id "${{inputs.job_id}}" \
+          --job-name "${{inputs.job_name}}"

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-332760d4b300f00a0d862e3cfe1495db3b1a14f9
+f084f34bbb743fada85f66b0ed8041387565e69c

.github/ci_commit_pins/fbgemm_rocm.txt

@@ -0,0 +1 @@
+5fb5024118e9bb9decf96c2b0b1a8f0010bf56be

.github/ci_commit_pins/torchbench.txt

@@ -1 +1 @@
-766a5e3a189384659fd35a68c3b17b88c761aaac
+373ffb19dc470f4423a3176a4133f8f4b3cdb5bd

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-73f54ba5bd7fb83d7ba81fe6f5e05fb6ee815d6f
+b2b890e962f5fb6f481e5da2eb4a43bb990d0f1b

.github/labeler.yml

@@ -30,9 +30,9 @@
 - torch/fx/experimental/sym_node.py
 - torch/fx/experimental/validator.py
 - torch/fx/experimental/proxy_tensor.py
-- test/distributed/_tensor/test_dtensor_compile.py
+- test/distributed/tensor/test_dtensor_compile.py
 - test/distributed/tensor/parallel/test_fsdp_2d_parallel.py
-- torch/distributed/_tensor/**
+- torch/distributed/tensor/**
 - torch/distributed/fsdp/**
 - torch/csrc/inductor/**
 - torch/csrc/dynamo/**
@@ -107,3 +107,8 @@
 - torch/csrc/dynamo/compiled_autograd.h
 - torch/_dynamo/compiled_autograd.py
 - torch/inductor/test_compiled_autograd.py
+
+"ciflow/xpu":
+- torch/csrc/inductor/aoti_include/xpu.h
+- torch/csrc/inductor/cpp_wrapper/device_internal/xpu.h
+- torch/csrc/inductor/cpp_wrapper/xpu.h