CUDA aarch64 12.6 and 12.8 builds fix triton constraints (#165022)

CUDA aarch64 12.6 and 12.8 builds fix triton constraints (#165013)

Since we have introduced CUDA aarch64 builds for all cuda versions we need to remove this constraint.
This was missed by https://github.com/pytorch/pytorch/pull/162364

Proper constraint on triton should be:
```
Requires-Dist: triton==3.5.0; platform_system == "Linux"
```

not:
```
Requires-Dist: triton==3.5.0; platform_system == "Linux" and platform_machine == "x86_64"
```

Pull Request resolved: https://github.com/pytorch/pytorch/pull/165013
Approved by: https://github.com/Camyll, https://github.com/nWEIdia, https://github.com/tinglvv

(cherry picked from commit 81dbeb06f4b3eb6c56625ec25d377eb7c7c6c573)

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

.ci/aarch64_linux/aarch64_ci_build.sh

@@ -3,17 +3,20 @@ set -eux -o pipefail
 
 GPU_ARCH_VERSION=${GPU_ARCH_VERSION:-}
 
-if [[ "$GPU_ARCH_VERSION" == *"12.9"* ]]; then
+# Set CUDA architecture lists to match x86 build_cuda.sh
+if [[ "$GPU_ARCH_VERSION" == *"12.6"* ]]; then
+    export TORCH_CUDA_ARCH_LIST="8.0;9.0"
+elif [[ "$GPU_ARCH_VERSION" == *"12.8"* ]]; then
     export TORCH_CUDA_ARCH_LIST="8.0;9.0;10.0;12.0"
-fi
-
-if [[ "$GPU_ARCH_VERSION" == *"13.0"* ]]; then
-    export TORCH_CUDA_ARCH_LIST="8.0;9.0;10.0;11.0;12.0"
+elif [[ "$GPU_ARCH_VERSION" == *"13.0"* ]]; then
+    export TORCH_CUDA_ARCH_LIST="8.0;9.0;10.0;11.0;12.0+PTX"
 fi
 
 # Compress the fatbin with -compress-mode=size for CUDA 13
 if [[ "$DESIRED_CUDA" == *"13"* ]]; then
     export TORCH_NVCC_FLAGS="-compress-mode=size"
+    # Bundle ptxas into the cu13 wheel, see https://github.com/pytorch/pytorch/issues/163801
+    export BUILD_BUNDLE_PTXAS=1
 fi
 
 SCRIPTPATH="$( cd -- "$(dirname "$0")" >/dev/null 2>&1 ; pwd -P )"
@@ -27,7 +30,7 @@ cd /
 # on the mounted pytorch repo
 git config --global --add safe.directory /pytorch
 pip install -r /pytorch/requirements.txt
-pip install auditwheel==6.2.0
+pip install auditwheel==6.2.0 wheel
 if [ "$DESIRED_CUDA" = "cpu" ]; then
     echo "BASE_CUDA_VERSION is not set. Building cpu wheel."
     #USE_PRIORITIZED_TEXT_FOR_LD for enable linker script optimization https://github.com/pytorch/pytorch/pull/121975/files
@@ -35,6 +38,16 @@ if [ "$DESIRED_CUDA" = "cpu" ]; then
 else
     echo "BASE_CUDA_VERSION is set to: $DESIRED_CUDA"
     export USE_SYSTEM_NCCL=1
+
+    # Check if we should use NVIDIA libs from PyPI (similar to x86 build_cuda.sh logic)
+    if [[ -z "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" ]]; then
+        echo "Bundling CUDA libraries with wheel for aarch64."
+    else
+        echo "Using nvidia libs from pypi for aarch64."
+        echo "Updated PYTORCH_EXTRA_INSTALL_REQUIREMENTS for aarch64: $PYTORCH_EXTRA_INSTALL_REQUIREMENTS"
+        export USE_NVIDIA_PYPI_LIBS=1
+    fi
+
     #USE_PRIORITIZED_TEXT_FOR_LD for enable linker script optimization https://github.com/pytorch/pytorch/pull/121975/files
     USE_PRIORITIZED_TEXT_FOR_LD=1 python /pytorch/.ci/aarch64_linux/aarch64_wheel_ci_build.py --enable-mkldnn --enable-cuda
 fi

.ci/aarch64_linux/aarch64_wheel_ci_build.py

@@ -69,83 +69,186 @@ def replace_tag(filename) -> None:
         f.writelines(lines)
 
 
+def patch_library_rpath(
+    folder: str,
+    lib_name: str,
+    use_nvidia_pypi_libs: bool = False,
+    desired_cuda: str = "",
+) -> None:
+    """Apply patchelf to set RPATH for a library in torch/lib"""
+    lib_path = f"{folder}/tmp/torch/lib/{lib_name}"
+
+    if use_nvidia_pypi_libs:
+        # For PyPI NVIDIA libraries, construct CUDA RPATH
+        cuda_rpaths = [
+            "$ORIGIN/../../nvidia/cudnn/lib",
+            "$ORIGIN/../../nvidia/nvshmem/lib",
+            "$ORIGIN/../../nvidia/nccl/lib",
+            "$ORIGIN/../../nvidia/cusparselt/lib",
+        ]
+
+        if "130" in desired_cuda:
+            cuda_rpaths.append("$ORIGIN/../../nvidia/cu13/lib")
+        else:
+            cuda_rpaths.extend(
+                [
+                    "$ORIGIN/../../nvidia/cublas/lib",
+                    "$ORIGIN/../../nvidia/cuda_cupti/lib",
+                    "$ORIGIN/../../nvidia/cuda_nvrtc/lib",
+                    "$ORIGIN/../../nvidia/cuda_runtime/lib",
+                    "$ORIGIN/../../nvidia/cufft/lib",
+                    "$ORIGIN/../../nvidia/curand/lib",
+                    "$ORIGIN/../../nvidia/cusolver/lib",
+                    "$ORIGIN/../../nvidia/cusparse/lib",
+                    "$ORIGIN/../../nvidia/nvtx/lib",
+                    "$ORIGIN/../../nvidia/cufile/lib",
+                ]
+            )
+
+        # Add $ORIGIN for local torch libs
+        rpath = ":".join(cuda_rpaths) + ":$ORIGIN"
+    else:
+        # For bundled libraries, just use $ORIGIN
+        rpath = "$ORIGIN"
+
+    if os.path.exists(lib_path):
+        os.system(
+            f"cd {folder}/tmp/torch/lib/; "
+            f"patchelf --set-rpath '{rpath}' --force-rpath {lib_name}"
+        )
+
+
+def copy_and_patch_library(
+    src_path: str,
+    folder: str,
+    use_nvidia_pypi_libs: bool = False,
+    desired_cuda: str = "",
+) -> None:
+    """Copy a library to torch/lib and patch its RPATH"""
+    if os.path.exists(src_path):
+        lib_name = os.path.basename(src_path)
+        shutil.copy2(src_path, f"{folder}/tmp/torch/lib/{lib_name}")
+        patch_library_rpath(folder, lib_name, use_nvidia_pypi_libs, desired_cuda)
+
+
 def package_cuda_wheel(wheel_path, desired_cuda) -> None:
     """
     Package the cuda wheel libraries
     """
     folder = os.path.dirname(wheel_path)
-    wheelname = os.path.basename(wheel_path)
     os.mkdir(f"{folder}/tmp")
     os.system(f"unzip {wheel_path} -d {folder}/tmp")
-    # Common libraries for all CUDA versions
-    common_libs = [
-        # Non-NVIDIA system libraries
-        "/lib64/libgomp.so.1",
-        "/usr/lib64/libgfortran.so.5",
-        "/acl/build/libarm_compute.so",
-        "/acl/build/libarm_compute_graph.so",
-        # Common CUDA libraries (same for all versions)
-        "/usr/local/lib/libnvpl_lapack_lp64_gomp.so.0",
-        "/usr/local/lib/libnvpl_blas_lp64_gomp.so.0",
-        "/usr/local/lib/libnvpl_lapack_core.so.0",
-        "/usr/local/lib/libnvpl_blas_core.so.0",
-        "/usr/local/cuda/extras/CUPTI/lib64/libnvperf_host.so",
-        "/usr/local/cuda/lib64/libcudnn.so.9",
-        "/usr/local/cuda/lib64/libcusparseLt.so.0",
-        "/usr/local/cuda/lib64/libcurand.so.10",
-        "/usr/local/cuda/lib64/libnccl.so.2",
-        "/usr/local/cuda/lib64/libnvshmem_host.so.3",
-        "/usr/local/cuda/lib64/libcudnn_adv.so.9",
-        "/usr/local/cuda/lib64/libcudnn_cnn.so.9",
-        "/usr/local/cuda/lib64/libcudnn_graph.so.9",
-        "/usr/local/cuda/lib64/libcudnn_ops.so.9",
-        "/usr/local/cuda/lib64/libcudnn_engines_runtime_compiled.so.9",
-        "/usr/local/cuda/lib64/libcudnn_engines_precompiled.so.9",
-        "/usr/local/cuda/lib64/libcudnn_heuristic.so.9",
-        "/usr/local/cuda/lib64/libcufile.so.0",
-        "/usr/local/cuda/lib64/libcufile_rdma.so.1",
-        "/usr/local/cuda/lib64/libcusparse.so.12",
-    ]
+    # Delete original wheel since it will be repackaged
+    os.system(f"rm {wheel_path}")
 
-    # CUDA version-specific libraries
-    if "130" in desired_cuda:
-        version_specific_libs = [
-            "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.13",
-            "/usr/local/cuda/lib64/libcublas.so.13",
-            "/usr/local/cuda/lib64/libcublasLt.so.13",
-            "/usr/local/cuda/lib64/libcudart.so.13",
-            "/usr/local/cuda/lib64/libcufft.so.12",
-            "/usr/local/cuda/lib64/libcusolver.so.12",
-            "/usr/local/cuda/lib64/libnvJitLink.so.13",
-            "/usr/local/cuda/lib64/libnvrtc.so.13",
-            "/usr/local/cuda/lib64/libnvrtc-builtins.so.13.0",
-        ]
-    elif "12" in desired_cuda:
-        # Get the last character for libnvrtc-builtins version (e.g., "129" -> "9")
-        minor_version = desired_cuda[-1]
-        version_specific_libs = [
-            "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12",
-            "/usr/local/cuda/lib64/libcublas.so.12",
-            "/usr/local/cuda/lib64/libcublasLt.so.12",
-            "/usr/local/cuda/lib64/libcudart.so.12",
-            "/usr/local/cuda/lib64/libcufft.so.11",
-            "/usr/local/cuda/lib64/libcusolver.so.11",
-            "/usr/local/cuda/lib64/libnvJitLink.so.12",
-            "/usr/local/cuda/lib64/libnvrtc.so.12",
-            f"/usr/local/cuda/lib64/libnvrtc-builtins.so.12.{minor_version}",
+    # Check if we should use PyPI NVIDIA libraries or bundle system libraries
+    use_nvidia_pypi_libs = os.getenv("USE_NVIDIA_PYPI_LIBS", "0") == "1"
+
+    if use_nvidia_pypi_libs:
+        print("Using nvidia libs from pypi - skipping CUDA library bundling")
+        # For PyPI approach, we don't bundle CUDA libraries - they come from PyPI packages
+        # We only need to bundle non-NVIDIA libraries
+        minimal_libs_to_copy = [
+            "/lib64/libgomp.so.1",
+            "/usr/lib64/libgfortran.so.5",
+            "/acl/build/libarm_compute.so",
+            "/acl/build/libarm_compute_graph.so",
+            "/usr/local/lib/libnvpl_lapack_lp64_gomp.so.0",
+            "/usr/local/lib/libnvpl_blas_lp64_gomp.so.0",
+            "/usr/local/lib/libnvpl_lapack_core.so.0",
+            "/usr/local/lib/libnvpl_blas_core.so.0",
         ]
 
-    # Combine all libraries
-    libs_to_copy = common_libs + version_specific_libs
+        # Copy minimal libraries to unzipped_folder/torch/lib
+        for lib_path in minimal_libs_to_copy:
+            copy_and_patch_library(lib_path, folder, use_nvidia_pypi_libs, desired_cuda)
 
-    # Copy libraries to unzipped_folder/a/lib
-    for lib_path in libs_to_copy:
-        lib_name = os.path.basename(lib_path)
-        shutil.copy2(lib_path, f"{folder}/tmp/torch/lib/{lib_name}")
-        os.system(
-            f"cd {folder}/tmp/torch/lib/; "
-            f"patchelf --set-rpath '$ORIGIN' --force-rpath {folder}/tmp/torch/lib/{lib_name}"
-        )
+        # Patch torch libraries used for searching libraries
+        torch_libs_to_patch = [
+            "libtorch.so",
+            "libtorch_cpu.so",
+            "libtorch_cuda.so",
+            "libtorch_cuda_linalg.so",
+            "libtorch_global_deps.so",
+            "libtorch_python.so",
+            "libtorch_nvshmem.so",
+            "libc10.so",
+            "libc10_cuda.so",
+            "libcaffe2_nvrtc.so",
+            "libshm.so",
+        ]
+        for lib_name in torch_libs_to_patch:
+            patch_library_rpath(folder, lib_name, use_nvidia_pypi_libs, desired_cuda)
+    else:
+        print("Bundling CUDA libraries with wheel")
+        # Original logic for bundling system CUDA libraries
+        # Common libraries for all CUDA versions
+        common_libs = [
+            # Non-NVIDIA system libraries
+            "/lib64/libgomp.so.1",
+            "/usr/lib64/libgfortran.so.5",
+            "/acl/build/libarm_compute.so",
+            "/acl/build/libarm_compute_graph.so",
+            # Common CUDA libraries (same for all versions)
+            "/usr/local/lib/libnvpl_lapack_lp64_gomp.so.0",
+            "/usr/local/lib/libnvpl_blas_lp64_gomp.so.0",
+            "/usr/local/lib/libnvpl_lapack_core.so.0",
+            "/usr/local/lib/libnvpl_blas_core.so.0",
+            "/usr/local/cuda/extras/CUPTI/lib64/libnvperf_host.so",
+            "/usr/local/cuda/lib64/libcudnn.so.9",
+            "/usr/local/cuda/lib64/libcusparseLt.so.0",
+            "/usr/local/cuda/lib64/libcurand.so.10",
+            "/usr/local/cuda/lib64/libnccl.so.2",
+            "/usr/local/cuda/lib64/libnvshmem_host.so.3",
+            "/usr/local/cuda/lib64/libcudnn_adv.so.9",
+            "/usr/local/cuda/lib64/libcudnn_cnn.so.9",
+            "/usr/local/cuda/lib64/libcudnn_graph.so.9",
+            "/usr/local/cuda/lib64/libcudnn_ops.so.9",
+            "/usr/local/cuda/lib64/libcudnn_engines_runtime_compiled.so.9",
+            "/usr/local/cuda/lib64/libcudnn_engines_precompiled.so.9",
+            "/usr/local/cuda/lib64/libcudnn_heuristic.so.9",
+            "/usr/local/cuda/lib64/libcufile.so.0",
+            "/usr/local/cuda/lib64/libcufile_rdma.so.1",
+            "/usr/local/cuda/lib64/libcusparse.so.12",
+        ]
+
+        # CUDA version-specific libraries
+        if "13" in desired_cuda:
+            minor_version = desired_cuda[-1]
+            version_specific_libs = [
+                "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.13",
+                "/usr/local/cuda/lib64/libcublas.so.13",
+                "/usr/local/cuda/lib64/libcublasLt.so.13",
+                "/usr/local/cuda/lib64/libcudart.so.13",
+                "/usr/local/cuda/lib64/libcufft.so.12",
+                "/usr/local/cuda/lib64/libcusolver.so.12",
+                "/usr/local/cuda/lib64/libnvJitLink.so.13",
+                "/usr/local/cuda/lib64/libnvrtc.so.13",
+                f"/usr/local/cuda/lib64/libnvrtc-builtins.so.13.{minor_version}",
+            ]
+        elif "12" in desired_cuda:
+            # Get the last character for libnvrtc-builtins version (e.g., "129" -> "9")
+            minor_version = desired_cuda[-1]
+            version_specific_libs = [
+                "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12",
+                "/usr/local/cuda/lib64/libcublas.so.12",
+                "/usr/local/cuda/lib64/libcublasLt.so.12",
+                "/usr/local/cuda/lib64/libcudart.so.12",
+                "/usr/local/cuda/lib64/libcufft.so.11",
+                "/usr/local/cuda/lib64/libcusolver.so.11",
+                "/usr/local/cuda/lib64/libnvJitLink.so.12",
+                "/usr/local/cuda/lib64/libnvrtc.so.12",
+                f"/usr/local/cuda/lib64/libnvrtc-builtins.so.12.{minor_version}",
+            ]
+        else:
+            raise ValueError(f"Unsupported CUDA version: {desired_cuda}.")
+
+        # Combine all libraries
+        libs_to_copy = common_libs + version_specific_libs
+
+        # Copy libraries to unzipped_folder/torch/lib
+        for lib_path in libs_to_copy:
+            copy_and_patch_library(lib_path, folder, use_nvidia_pypi_libs, desired_cuda)
 
     # Make sure the wheel is tagged with manylinux_2_28
     for f in os.scandir(f"{folder}/tmp/"):
@@ -153,14 +256,8 @@ def package_cuda_wheel(wheel_path, desired_cuda) -> None:
             replace_tag(f"{f.path}/WHEEL")
             break
 
-    os.mkdir(f"{folder}/cuda_wheel")
-    os.system(f"cd {folder}/tmp/; zip -r {folder}/cuda_wheel/{wheelname} *")
-    shutil.move(
-        f"{folder}/cuda_wheel/{wheelname}",
-        f"{folder}/{wheelname}",
-        copy_function=shutil.copy2,
-    )
-    os.system(f"rm -rf {folder}/tmp/ {folder}/cuda_wheel/")
+    os.system(f"wheel pack {folder}/tmp/ -d {folder}")
+    os.system(f"rm -rf {folder}/tmp/")
 
 
 def complete_wheel(folder: str) -> str:
@@ -183,14 +280,7 @@ def complete_wheel(folder: str) -> str:
             f"/{folder}/dist/{repaired_wheel_name}",
         )
     else:
-        repaired_wheel_name = wheel_name.replace(
-            "linux_aarch64", "manylinux_2_28_aarch64"
-        )
-        print(f"Renaming {wheel_name} wheel to {repaired_wheel_name}")
-        os.rename(
-            f"/{folder}/dist/{wheel_name}",
-            f"/{folder}/dist/{repaired_wheel_name}",
-        )
+        repaired_wheel_name = list_dir(f"/{folder}/dist")[0]
 
     print(f"Copying {repaired_wheel_name} to artifacts")
     shutil.copy2(
@@ -232,6 +322,16 @@ if __name__ == "__main__":
     if enable_cuda:
         build_vars += "MAX_JOBS=5 "
 
+        # Handle PyPI NVIDIA libraries vs bundled libraries
+        use_nvidia_pypi_libs = os.getenv("USE_NVIDIA_PYPI_LIBS", "0") == "1"
+        if use_nvidia_pypi_libs:
+            print("Configuring build for PyPI NVIDIA libraries")
+            # Configure for dynamic linking (matching x86 logic)
+            build_vars += "ATEN_STATIC_CUDA=0 USE_CUDA_STATIC_LINK=0 USE_CUPTI_SO=1 "
+        else:
+            print("Configuring build for bundled NVIDIA libraries")
+            # Keep existing static linking approach - already configured above
+
     override_package_version = os.getenv("OVERRIDE_PACKAGE_VERSION")
     desired_cuda = os.getenv("DESIRED_CUDA")
     if override_package_version is not None:

.ci/docker/build.sh

@@ -214,8 +214,7 @@ case "$tag" in
     TRITON=yes
     ;;
   pytorch-linux-jammy-py3-gcc11-inductor-benchmarks)
-    # TODO (huydhn): Upgrade this to Python >= 3.10
-    ANACONDA_PYTHON_VERSION=3.9
+    ANACONDA_PYTHON_VERSION=3.10
     GCC_VERSION=11
     VISION=yes
     KATEX=yes
@@ -263,13 +262,10 @@ case "$tag" in
     TRITON_CPU=yes
     ;;
   pytorch-linux-jammy-linter)
-    # TODO: Use 3.9 here because of this issue https://github.com/python/mypy/issues/13627.
-    # We will need to update mypy version eventually, but that's for another day. The task
-    # would be to upgrade mypy to 1.0.0 with Python 3.11
-    PYTHON_VERSION=3.9
+    PYTHON_VERSION=3.10
     ;;
-  pytorch-linux-jammy-cuda12.8-cudnn9-py3.9-linter)
-    PYTHON_VERSION=3.9
+  pytorch-linux-jammy-cuda12.8-cudnn9-py3.10-linter)
+    PYTHON_VERSION=3.10
     CUDA_VERSION=12.8.1
     ;;
   pytorch-linux-jammy-aarch64-py3.10-gcc11)

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

@@ -1 +1 @@
-d0e80f39c562c70986fc548fa6e5852ad86e16e7
+1b0418a9a454b2b93ab8d71f40e59d2297157fae

.ci/docker/ci_commit_pins/triton.txt

@@ -1 +1 @@
-f7888497a1eb9e98d4c07537f0d0bcfe180d1363
+bbb06c0334a6772b92d24bde54956e675c8c6604

.ci/docker/common/install_cuda.sh

@@ -147,7 +147,7 @@ function install_128 {
 }
 
 function install_130 {
-  CUDNN_VERSION=9.12.0.46
+  CUDNN_VERSION=9.13.0.50
   echo "Installing CUDA 13.0 and cuDNN ${CUDNN_VERSION} and NVSHMEM and NCCL and cuSparseLt-0.7.1"
   # install CUDA 13.0 in the same container
   install_cuda 13.0.0 cuda_13.0.0_580.65.06_linux

.ci/docker/common/patch_libstdc.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+set -xe
+# Script used in Linux x86 and aarch64 CD pipeline
+
+# Workaround for exposing statically linked libstdc++ CXX11 ABI symbols.
+# see: https://github.com/pytorch/pytorch/issues/133437
+LIBNONSHARED=$(gcc -print-file-name=libstdc++_nonshared.a)
+nm -g $LIBNONSHARED | grep " T " | grep recursive_directory_iterator | cut -c 20-  > weaken-symbols.txt
+objcopy --weaken-symbols weaken-symbols.txt $LIBNONSHARED $LIBNONSHARED

.ci/docker/libtorch/Dockerfile

@@ -74,6 +74,14 @@ RUN bash ./install_cuda.sh 13.0
 RUN bash ./install_magma.sh 13.0
 RUN ln -sf /usr/local/cuda-13.0 /usr/local/cuda
 
+# Install libibverbs for libtorch and copy to CUDA directory
+RUN apt-get update -y && \
+    apt-get install -y libibverbs-dev librdmacm-dev && \
+    cp /usr/lib/x86_64-linux-gnu/libmlx5.so* /usr/local/cuda/lib64/ && \
+    cp /usr/lib/x86_64-linux-gnu/librdmacm.so* /usr/local/cuda/lib64/ && \
+    cp /usr/lib/x86_64-linux-gnu/libibverbs.so* /usr/local/cuda/lib64/ && \
+    cp /usr/lib/x86_64-linux-gnu/libnl* /usr/local/cuda/lib64/
+
 FROM cpu as rocm
 ARG ROCM_VERSION
 ARG PYTORCH_ROCM_ARCH

.ci/docker/manywheel/Dockerfile_2_28

@@ -130,7 +130,8 @@ ENV LD_LIBRARY_PATH=/opt/rh/gcc-toolset-${DEVTOOLSET_VERSION}/root/usr/lib64:/op
 RUN for cpython_version in "cp312-cp312" "cp313-cp313" "cp313-cp313t"; do \
     /opt/python/${cpython_version}/bin/python -m pip install setuptools wheel; \
     done;
-
+ADD ./common/patch_libstdc.sh patch_libstdc.sh
+RUN bash ./patch_libstdc.sh && rm patch_libstdc.sh
 
 # cmake-3.18.4 from pip; force in case cmake3 already exists
 RUN yum install -y python3-pip && \

.ci/docker/manywheel/Dockerfile_2_28_aarch64

@@ -71,3 +71,5 @@ RUN rm -rf /opt/python/cp33-cp33m /opt/_internal/cpython-3.3.6
 RUN rm -rf /opt/python/cp34-cp34m /opt/_internal/cpython-3.4.6
 COPY --from=openblas     /opt/OpenBLAS/  /opt/OpenBLAS/
 ENV LD_LIBRARY_PATH=/opt/OpenBLAS/lib:$LD_LIBRARY_PATH
+ADD ./common/patch_libstdc.sh patch_libstdc.sh
+RUN bash ./patch_libstdc.sh && rm patch_libstdc.sh

.ci/docker/manywheel/Dockerfile_cuda_aarch64

@@ -95,3 +95,5 @@ COPY --from=nvpl /opt/nvpl/lib/  /usr/local/lib/
 COPY --from=nvpl /opt/nvpl/include/  /usr/local/include/
 RUN ln -sf /usr/local/cuda-${BASE_CUDA_VERSION} /usr/local/cuda
 ENV PATH=/usr/local/cuda/bin:$PATH
+ADD ./common/patch_libstdc.sh patch_libstdc.sh
+RUN bash ./patch_libstdc.sh && rm patch_libstdc.sh

.ci/docker/requirements-ci.txt

@@ -93,8 +93,9 @@ librosa==0.10.2 ; python_version == "3.12" and platform_machine != "s390x"
 #Pinned versions:
 #test that import:
 
-mypy==1.16.0
+mypy==1.16.0 ; platform_system != "Windows"
 # Pin MyPy version because new errors are likely to appear with each release
+# Skip on Windows as lots of type annotations are POSIX specific
 #Description: linter
 #Pinned versions: 1.16.0
 #test that import: test_typing.py, test_type_hints.py

.ci/docker/requirements-docs.txt

@@ -1,7 +1,7 @@
 sphinx==5.3.0
 #Description: This is used to generate PyTorch docs
 #Pinned versions: 5.3.0
--e git+https://github.com/pytorch/pytorch_sphinx_theme.git@1657ad2fc1acdc98aa719eebecbb0128a7c13ce4#egg=pytorch_sphinx_theme2
+-e git+https://github.com/pytorch/pytorch_sphinx_theme.git@71e55749be14ceb56e7f8211a9fb649866b87ad4#egg=pytorch_sphinx_theme2
 
 # TODO: sphinxcontrib.katex 0.9.0 adds a local KaTeX server to speed up pre-rendering
 # but it doesn't seem to work and hangs around idly. The initial thought that it is probably

.ci/docker/triton_version.txt

@@ -1 +1 @@
-3.4.0
+3.5.0

.ci/docker/triton_xpu_version.txt

@@ -1 +1 @@
-3.4.0
+3.5.0

.ci/libtorch/build.sh

@@ -7,4 +7,4 @@ set -ex
 
 SCRIPTPATH="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
 
-USE_NVSHMEM=0 USE_CUSPARSELT=0 BUILD_PYTHONLESS=1 DESIRED_PYTHON="3.9" ${SCRIPTPATH}/../manywheel/build.sh
+USE_NVSHMEM=0 USE_CUSPARSELT=0 BUILD_PYTHONLESS=1 DESIRED_PYTHON="3.10" ${SCRIPTPATH}/../manywheel/build.sh

.ci/lumen_cli/cli/lib/core/vllm/lib.py

@@ -41,7 +41,6 @@ def sample_vllm_test_library():
                 "pytest -v -s basic_correctness/test_cumem.py",
                 "pytest -v -s basic_correctness/test_basic_correctness.py",
                 "pytest -v -s basic_correctness/test_cpu_offload.py",
-                "VLLM_TEST_ENABLE_ARTIFICIAL_PREEMPT=1 pytest -v -s basic_correctness/test_preemption.py",
             ],
         },
         "vllm_basic_models_test": {
@@ -68,16 +67,12 @@ def sample_vllm_test_library():
                         "-v",
                         "-s",
                         "entrypoints/llm",
-                        "--ignore=entrypoints/llm/test_lazy_outlines.py",
                         "--ignore=entrypoints/llm/test_generate.py",
-                        "--ignore=entrypoints/llm/test_generate_multiple_loras.py",
                         "--ignore=entrypoints/llm/test_collective_rpc.py",
                     ]
                 ),
-                "pytest -v -s entrypoints/llm/test_lazy_outlines.py",
-                "pytest -v -s entrypoints/llm/test_generate.py ",
-                "pytest -v -s entrypoints/llm/test_generate_multiple_loras.py",
-                "VLLM_USE_V1=0 pytest -v -s entrypoints/offline_mode",
+                "pytest -v -s entrypoints/llm/test_generate.py",
+                "pytest -v -s entrypoints/offline_mode",
             ],
         },
         "vllm_regression_test": {
@@ -97,14 +92,24 @@ def sample_vllm_test_library():
             "num_gpus": 4,
             "steps": [
                 "pytest -v -s -x lora/test_chatglm3_tp.py",
-                "echo $VLLM_WORKER_MULTIPROC_METHOD",
                 "pytest -v -s -x lora/test_llama_tp.py",
-                "pytest -v -s -x lora/test_multi_loras_with_tp.py",
+                "pytest -v -s -x lora/test_llm_with_multi_loras.py",
             ],
         },
-        "vllm_lora_280_failure_test": {
-            "title": "LoRA 280 failure test",
-            "id": "vllm_lora_280_failure_test",
+        "vllm_distributed_test_28_failure_test": {
+            "title": "Distributed Tests (2 GPUs) pytorch 2.8 release failure",
+            "id": "vllm_distributed_test_28_failure_test",
+            "env_vars": {
+                "VLLM_WORKER_MULTIPROC_METHOD": "spawn",
+            },
+            "num_gpus": 4,
+            "steps": [
+                "pytest -v -s distributed/test_sequence_parallel.py",
+            ],
+        },
+        "vllm_lora_28_failure_test": {
+            "title": "LoRA pytorch 2.8 failure test",
+            "id": "vllm_lora_28_failure_test",
             "steps": ["pytest -v lora/test_quant_model.py"],
         },
         "vllm_multi_model_processor_test": {
@@ -115,6 +120,15 @@ def sample_vllm_test_library():
                 "pytest -v -s models/multimodal/processing --ignore models/multimodal/processing/test_tensor_schema.py",
             ],
         },
+        "vllm_multi_model_test_28_failure_test": {
+            "title": "Multi-Model Test (Failed 2.8 release)",
+            "id": "vllm_multi_model_test_28_failure_test",
+            "package_install": ["git+https://github.com/TIGER-AI-Lab/Mantis.git"],
+            "steps": [
+                "pytest -v -s models/multimodal/generation/test_voxtral.py",
+                "pytest -v -s models/multimodal/pooling",
+            ],
+        },
         "vllm_pytorch_compilation_unit_tests": {
             "title": "PyTorch Compilation Unit Tests",
             "id": "vllm_pytorch_compilation_unit_tests",
@@ -129,6 +143,28 @@ def sample_vllm_test_library():
                 "pytest -v -s compile/test_decorator.py",
             ],
         },
+        "vllm_languagde_model_test_extended_generation_28_failure_test": {
+            "title": "Language Models Test (Extended Generation) 2.8 release failure",
+            "id": "vllm_languagde_model_test_extended_generation_28_failure_test",
+            "package_install": [
+                "--no-build-isolation",
+                "git+https://github.com/Dao-AILab/causal-conv1d@v1.5.0.post8",
+            ],
+            "steps": [
+                "pytest -v -s models/language/generation/test_mistral.py",
+            ],
+        },
+        "vllm_distributed_test_2_gpu_28_failure_test": {
+            "title": "Distributed Tests (2 GPUs) pytorch 2.8 release failure",
+            "id": "vllm_distributed_test_2_gpu_28_failure_test",
+            "env_vars": {
+                "VLLM_WORKER_MULTIPROC_METHOD": "spawn",
+            },
+            "num_gpus": 4,
+            "steps": [
+                "pytest -v -s distributed/test_sequence_parallel.py",
+            ],
+        },
         # TODO(elainewy):need to add g6 with 4 gpus to run this test
         "vllm_lora_test": {
             "title": "LoRA Test %N",

.ci/lumen_cli/cli/lib/core/vllm/vllm_test.py

@@ -104,20 +104,26 @@ class VllmTestRunner(BaseRunner):
         main function to run vllm test
         """
         self.prepare()
-        with working_directory(self.work_directory):
-            if self.test_type == TestInpuType.TEST_PLAN:
-                if self.num_shards > 1:
-                    run_test_plan(
-                        self.test_plan,
-                        "vllm",
-                        sample_vllm_test_library(),
-                        self.shard_id,
-                        self.num_shards,
-                    )
+        try:
+            with working_directory(self.work_directory):
+                if self.test_type == TestInpuType.TEST_PLAN:
+                    if self.num_shards > 1:
+                        run_test_plan(
+                            self.test_plan,
+                            "vllm",
+                            sample_vllm_test_library(),
+                            self.shard_id,
+                            self.num_shards,
+                        )
+                    else:
+                        run_test_plan(
+                            self.test_plan, "vllm", sample_vllm_test_library()
+                        )
                 else:
-                    run_test_plan(self.test_plan, "vllm", sample_vllm_test_library())
-            else:
-                raise ValueError(f"Unknown test type {self.test_type}")
+                    raise ValueError(f"Unknown test type {self.test_type}")
+        finally:
+            # double check the torches are not overridden by other packages
+            check_versions()
 
     def _install_wheels(self, params: VllmTestParameters):
         logger.info("Running vllm test with inputs: %s", params)

.ci/manywheel/build_cuda.sh

@@ -124,6 +124,7 @@ if [[ $CUDA_VERSION == 12* || $CUDA_VERSION == 13* ]]; then
     fi
     if [[ -z "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" ]]; then
         echo "Bundling with cudnn and cublas."
+
         DEPS_LIST+=(
             "/usr/local/cuda/lib64/libcudnn_adv.so.9"
             "/usr/local/cuda/lib64/libcudnn_cnn.so.9"
@@ -133,16 +134,11 @@ if [[ $CUDA_VERSION == 12* || $CUDA_VERSION == 13* ]]; then
             "/usr/local/cuda/lib64/libcudnn_engines_precompiled.so.9"
             "/usr/local/cuda/lib64/libcudnn_heuristic.so.9"
             "/usr/local/cuda/lib64/libcudnn.so.9"
-            "/usr/local/cuda/lib64/libcublas.so.12"
-            "/usr/local/cuda/lib64/libcublasLt.so.12"
             "/usr/local/cuda/lib64/libcusparseLt.so.0"
-            "/usr/local/cuda/lib64/libcudart.so.12"
-            "/usr/local/cuda/lib64/libnvrtc.so.12"
             "/usr/local/cuda/lib64/libnvrtc-builtins.so"
             "/usr/local/cuda/lib64/libcufile.so.0"
             "/usr/local/cuda/lib64/libcufile_rdma.so.1"
             "/usr/local/cuda/lib64/libnvshmem_host.so.3"
-            "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12"
             "/usr/local/cuda/extras/CUPTI/lib64/libnvperf_host.so"
         )
         DEPS_SONAME+=(
@@ -154,22 +150,56 @@ if [[ $CUDA_VERSION == 12* || $CUDA_VERSION == 13* ]]; then
             "libcudnn_engines_precompiled.so.9"
             "libcudnn_heuristic.so.9"
             "libcudnn.so.9"
-            "libcublas.so.12"
-            "libcublasLt.so.12"
             "libcusparseLt.so.0"
-            "libcudart.so.12"
-            "libnvrtc.so.12"
             "libnvrtc-builtins.so"
             "libnvshmem_host.so.3"
             "libcufile.so.0"
             "libcufile_rdma.so.1"
-            "libcupti.so.12"
             "libnvperf_host.so"
         )
         # Add libnvToolsExt only if CUDA version is not 12.9
-        if [[ $CUDA_VERSION != 12.9* ]]; then
-            DEPS_LIST+=("/usr/local/cuda/lib64/libnvToolsExt.so.1")
-            DEPS_SONAME+=("libnvToolsExt.so.1")
+        if [[ $CUDA_VERSION == 13* ]]; then
+            DEPS_LIST+=(
+                "/usr/local/cuda/lib64/libcublas.so.13"
+                "/usr/local/cuda/lib64/libcublasLt.so.13"
+                "/usr/local/cuda/lib64/libcudart.so.13"
+                "/usr/local/cuda/lib64/libnvrtc.so.13"
+                "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.13"
+                "/usr/local/cuda/lib64/libibverbs.so.1"
+                "/usr/local/cuda/lib64/librdmacm.so.1"
+                "/usr/local/cuda/lib64/libmlx5.so.1"
+                "/usr/local/cuda/lib64/libnl-3.so.200"
+                "/usr/local/cuda/lib64/libnl-route-3.so.200")
+            DEPS_SONAME+=(
+                "libcublas.so.13"
+                "libcublasLt.so.13"
+                "libcudart.so.13"
+                "libnvrtc.so.13"
+                "libcupti.so.13"
+                "libibverbs.so.1"
+                "librdmacm.so.1"
+                "libmlx5.so.1"
+                "libnl-3.so.200"
+                "libnl-route-3.so.200")
+            export USE_CUPTI_SO=1
+            export ATEN_STATIC_CUDA=0
+            export USE_CUDA_STATIC_LINK=0
+            export USE_CUFILE=0
+        else
+            DEPS_LIST+=(
+                "/usr/local/cuda/lib64/libnvToolsExt.so.1"
+                "/usr/local/cuda/lib64/libcublas.so.12"
+                "/usr/local/cuda/lib64/libcublasLt.so.12"
+                "/usr/local/cuda/lib64/libcudart.so.12"
+                "/usr/local/cuda/lib64/libnvrtc.so.12"
+                "/usr/local/cuda/extras/CUPTI/lib64/libcupti.so.12")
+            DEPS_SONAME+=(
+                "libnvToolsExt.so.1"
+                "libcublas.so.12"
+                "libcublasLt.so.12"
+                "libcudart.so.12"
+                "libnvrtc.so.12"
+                "libcupti.so.12")
         fi
     else
         echo "Using nvidia libs from pypi."

.ci/pytorch/check_binary.sh

@@ -67,7 +67,7 @@ fi
 #       wheels with cxx11-abi
 
 echo "Checking that the gcc ABI is what we expect"
-if [[ "$(uname)" != 'Darwin' ]]; then
+if [[ "$(uname)" != 'Darwin' &&  "$(uname -m)" != "s390x" ]]; then
   # We also check that there are cxx11 symbols in libtorch
   #
   echo "Checking that symbols in libtorch.so have the right gcc abi"

.ci/pytorch/common_utils.sh

@@ -284,7 +284,7 @@ function install_torchrec_and_fbgemm() {
 
 function clone_pytorch_xla() {
   if [[ ! -d ./xla ]]; then
-    git clone --recursive --quiet https://github.com/pytorch/xla.git
+    git clone --recursive -b r2.9 https://github.com/pytorch/xla.git
     pushd xla
     # pin the xla hash so that we don't get broken by changes to xla
     git checkout "$(cat ../.github/ci_commit_pins/xla.txt)"

.ci/pytorch/cpp_doc_push_script.sh

@@ -58,7 +58,7 @@ time python tools/setup_helpers/generate_code.py \
 
 # Build the docs
 pushd docs/cpp
-time make VERBOSE=1 html -j
+time make VERBOSE=1 html
 
 popd
 popd

.ci/pytorch/macos-build.sh

@@ -35,10 +35,11 @@ fi
 
 print_cmake_info
 if [[ ${BUILD_ENVIRONMENT} == *"distributed"* ]]; then
-  USE_OPENMP=1 WERROR=1 python setup.py bdist_wheel
+  # Needed for inductor benchmarks, as lots of HF networks make `torch.distribtued` calls
+  USE_DISTRIBUTED=1 USE_OPENMP=1 WERROR=1 python setup.py bdist_wheel
 else
-  # NB: we always build with distributed; USE_DISTRIBUTED turns off all
-  # backends (specifically the gloo backend), so test that this case works too
+  # Explicitly set USE_DISTRIBUTED=0 to align with the default build config on mac. This also serves as the sole CI config that tests
+  # that building with USE_DISTRIBUTED=0 works at all. See https://github.com/pytorch/pytorch/issues/86448
   USE_DISTRIBUTED=0 USE_OPENMP=1 MACOSX_DEPLOYMENT_TARGET=11.0 WERROR=1 BUILD_TEST=OFF USE_PYTORCH_METAL=1 python setup.py bdist_wheel --plat-name macosx_11_0_arm64
 fi
 if which sccache > /dev/null; then

.ci/pytorch/macos-test.sh

@@ -13,13 +13,9 @@ if [[ ! $(python -c "import torch; print(int(torch.backends.openmp.is_available(
 fi
 popd
 
-python -mpip install -r requirements.txt
-
 # enable debug asserts in serialization
 export TORCH_SERIALIZATION_DEBUG=1
 
-python -mpip install --no-input -r requirements.txt
-
 setup_test_python() {
   # The CircleCI worker hostname doesn't resolve to an address.
   # This environment variable makes ProcessGroupGloo default to
@@ -199,7 +195,7 @@ torchbench_setup_macos() {
   git checkout "$(cat ../.github/ci_commit_pins/vision.txt)"
   git submodule update --init --recursive
   python setup.py clean
-  python setup.py develop
+  python -m pip install -e . -v --no-build-isolation
   popd
 
   pushd torchaudio
@@ -208,7 +204,7 @@ torchbench_setup_macos() {
   git submodule update --init --recursive
   python setup.py clean
   #TODO: Remove me, when figure out how to make TorchAudio find brew installed openmp
-  USE_OPENMP=0 python setup.py develop
+  USE_OPENMP=0 python -m pip install -e . -v --no-build-isolation
   popd
 
   checkout_install_torchbench

.ci/pytorch/numba-cuda-13.patch

@@ -0,0 +1,25 @@
+From 6e08c9d08e9de59c7af28b720289debbbd384764 Mon Sep 17 00:00:00 2001
+From: Michael Wang <13521008+isVoid@users.noreply.github.com>
+Date: Tue, 1 Apr 2025 17:28:05 -0700
+Subject: [PATCH] Avoid bumping certain driver API to avoid future breakage
+ (#185)
+
+Co-authored-by: isVoid <isVoid@users.noreply.github.com>
+---
+ numba_cuda/numba/cuda/cudadrv/driver.py | 3 +++
+ 1 file changed, 3 insertions(+)
+
+diff --git a/numba_cuda/numba/cuda/cudadrv/driver.py b/numba_cuda/numba/cuda/cudadrv/driver.py
+index 1641bf77..233e9ed7 100644
+--- a/numba_cuda/numba/cuda/cudadrv/driver.py
++++ b/numba_cuda/numba/cuda/cudadrv/driver.py
+@@ -365,6 +365,9 @@ def _find_api(self, fname):
+         else:
+             variants = ('_v2', '')
+ 
++        if fname in ("cuCtxGetDevice", "cuCtxSynchronize"):
++            return getattr(self.lib, fname)
++
+         for variant in variants:
+             try:
+                 return getattr(self.lib, f'{fname}{variant}')

.ci/pytorch/smoke_test/check_binary_symbols.py

@@ -32,6 +32,9 @@ LIBTORCH_NAMESPACE_LIST = (
     "torch::",
 )
 
+# Patterns for detecting statically linked libstdc++ symbols
+STATICALLY_LINKED_CXX11_ABI = [re.compile(r".*recursive_directory_iterator.*")]
+
 
 def _apply_libtorch_symbols(symbols):
     return [
@@ -53,12 +56,17 @@ def get_symbols(lib: str) -> list[tuple[str, str, str]]:
     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], symbol_type: str | None = None
+) -> list[str]:
     def _grep_symbols(
         symbols: list[tuple[str, str, str]], patterns: list[Any]
     ) -> list[str]:
         rc = []
         for _s_addr, _s_type, s_name in symbols:
+            # Filter by symbol type if specified
+            if symbol_type and _s_type != symbol_type:
+                continue
             for pattern in patterns:
                 if pattern.match(s_name):
                     rc.append(s_name)
@@ -80,6 +88,18 @@ def grep_symbols(lib: str, patterns: list[Any]) -> list[str]:
         return functools.reduce(list.__add__, (x.result() for x in tasks), [])
 
 
+def check_lib_statically_linked_libstdc_cxx_abi_symbols(lib: str) -> None:
+    cxx11_statically_linked_symbols = grep_symbols(
+        lib, STATICALLY_LINKED_CXX11_ABI, symbol_type="T"
+    )
+    num_statically_linked_symbols = len(cxx11_statically_linked_symbols)
+    print(f"num_statically_linked_symbols (T): {num_statically_linked_symbols}")
+    if num_statically_linked_symbols > 0:
+        raise RuntimeError(
+            f"Found statically linked libstdc++ symbols (recursive_directory_iterator): {cxx11_statically_linked_symbols[:100]}"
+        )
+
+
 def check_lib_symbols_for_abi_correctness(lib: str) -> None:
     print(f"lib: {lib}")
     cxx11_symbols = grep_symbols(lib, LIBTORCH_CXX11_PATTERNS)
@@ -107,6 +127,7 @@ def main() -> None:
 
     libtorch_cpu_path = str(install_root / "lib" / "libtorch_cpu.so")
     check_lib_symbols_for_abi_correctness(libtorch_cpu_path)
+    check_lib_statically_linked_libstdc_cxx_abi_symbols(libtorch_cpu_path)
 
 
 if __name__ == "__main__":

.ci/pytorch/test.sh

@@ -32,6 +32,16 @@ if [[ "$BUILD_ENVIRONMENT" != *rocm* && "$BUILD_ENVIRONMENT" != *s390x* && -d /v
   git config --global --add safe.directory /var/lib/jenkins/workspace
 fi
 
+
+# Patch numba to avoid CUDA-13 crash, see https://github.com/pytorch/pytorch/issues/162878
+NUMBA_CUDA_DIR=$(python -c "import os;import numba.cuda; print(os.path.dirname(numba.cuda.__file__))" 2>/dev/null || true)
+if [ -n "$NUMBA_CUDA_DIR" ]; then
+  NUMBA_PATCH="$(dirname "$(realpath "${BASH_SOURCE[0]}")")/numba-cuda-13.patch"
+  pushd "$NUMBA_CUDA_DIR"
+  patch -p4 <"$NUMBA_PATCH"
+  popd
+fi
+
 echo "Environment variables:"
 env
 
@@ -1614,6 +1624,25 @@ test_operator_benchmark() {
       --expected "expected_ci_operator_benchmark_eager_float32_cpu.csv"
 }
 
+test_operator_microbenchmark() {
+  TEST_REPORTS_DIR=$(pwd)/test/test-reports
+  mkdir -p "$TEST_REPORTS_DIR"
+  TEST_DIR=$(pwd)
+
+  cd benchmarks/operator_benchmark/pt_extension
+  python -m pip install .
+
+  cd "${TEST_DIR}"/benchmarks/operator_benchmark
+
+  for OP_BENCHMARK_TESTS in matmul mm addmm bmm; do
+    $TASKSET python -m pt.${OP_BENCHMARK_TESTS}_test --tag-filter long \
+      --output-json-for-dashboard "${TEST_REPORTS_DIR}/operator_microbenchmark_${OP_BENCHMARK_TESTS}_compile.json" \
+      --benchmark-name "PyTorch operator microbenchmark" --use-compile
+    $TASKSET python -m pt.${OP_BENCHMARK_TESTS}_test --tag-filter long \
+      --output-json-for-dashboard "${TEST_REPORTS_DIR}/operator_microbenchmark_${OP_BENCHMARK_TESTS}.json" \
+      --benchmark-name "PyTorch operator microbenchmark"
+  done
+}
 
 if ! [[ "${BUILD_ENVIRONMENT}" == *libtorch* || "${BUILD_ENVIRONMENT}" == *-bazel-* ]]; then
   (cd test && python -c "import torch; print(torch.__config__.show())")
@@ -1668,6 +1697,8 @@ elif [[ "${TEST_CONFIG}" == *operator_benchmark* ]]; then
     test_operator_benchmark cpu ${TEST_MODE}
 
   fi
+elif [[ "${TEST_CONFIG}" == *operator_microbenchmark* ]]; then
+  test_operator_microbenchmark
 elif [[ "${TEST_CONFIG}" == *inductor_distributed* ]]; then
   test_inductor_distributed
 elif [[ "${TEST_CONFIG}" == *inductor-halide* ]]; then
@@ -1721,11 +1752,6 @@ elif [[ "${TEST_CONFIG}" == *inductor_cpp_wrapper* ]]; then
 elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
   install_torchvision
   test_inductor_shard "${SHARD_NUMBER}"
-  if [[ "${SHARD_NUMBER}" == 1 ]]; then
-    if [[ "${BUILD_ENVIRONMENT}" != linux-jammy-py3.9-gcc11-build ]]; then
-      test_inductor_distributed
-    fi
-  fi
 elif [[ "${TEST_CONFIG}" == *einops* ]]; then
   test_einops
 elif [[ "${TEST_CONFIG}" == *dynamo_wrapped* ]]; then

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

@@ -137,7 +137,7 @@ sccache --show-stats
 python -c "import os, glob; os.system('python -mpip install --no-index --no-deps ' + glob.glob('dist/*.whl')[0])"
 (
   if "%BUILD_ENVIRONMENT%"=="" (
-    echo NOTE: To run `import torch`, please make sure to activate the conda environment by running `call %CONDA_PARENT_DIR%\Miniconda3\Scripts\activate.bat %CONDA_PARENT_DIR%\Miniconda3` in Command Prompt before running Git Bash.
+    echo NOTE: To run `import torch`, please make sure to activate the conda environment by running `call %CONDA_ROOT_DIR%\Scripts\activate.bat %CONDA_ROOT_DIR%\envs\py_tmp` in Command Prompt before running Git Bash.
   ) else (
     copy /Y "dist\*.whl" "%PYTORCH_FINAL_PACKAGE_DIR%"
 

.ci/pytorch/win-test-helpers/installation-helpers/activate_miniconda3.bat

@@ -3,12 +3,12 @@ if "%BUILD_ENVIRONMENT%"=="" (
 ) else (
   set CONDA_PARENT_DIR=C:\Jenkins
 )
-
+set CONDA_ROOT_DIR=%CONDA_PARENT_DIR%\Miniconda3
 
 :: Be conservative here when rolling out the new AMI with conda. This will try
 :: to install conda as before if it couldn't find the conda installation. This
 :: can be removed eventually after we gain enough confidence in the AMI
-if not exist %CONDA_PARENT_DIR%\Miniconda3 (
+if not exist %CONDA_ROOT_DIR% (
   set INSTALL_FRESH_CONDA=1
 )
 
@@ -17,10 +17,14 @@ if "%INSTALL_FRESH_CONDA%"=="1" (
   if errorlevel 1 exit /b
   if not errorlevel 0 exit /b
 
-  %TMP_DIR_WIN%\Miniconda3-latest-Windows-x86_64.exe /InstallationType=JustMe /RegisterPython=0 /S /AddToPath=0 /D=%CONDA_PARENT_DIR%\Miniconda3
+  %TMP_DIR_WIN%\Miniconda3-latest-Windows-x86_64.exe /InstallationType=JustMe /RegisterPython=0 /S /AddToPath=0 /D=%CONDA_ROOT_DIR%
   if errorlevel 1 exit /b
   if not errorlevel 0 exit /b
 )
 
 :: Activate conda so that we can use its commands, i.e. conda, python, pip
-call %CONDA_PARENT_DIR%\Miniconda3\Scripts\activate.bat %CONDA_PARENT_DIR%\Miniconda3
+call %CONDA_ROOT_DIR%\Scripts\activate.bat %CONDA_ROOT_DIR%
+:: Activate conda so that we can use its commands, i.e. conda, python, pip
+call conda activate py_tmp
+
+call pip install -r .ci/docker/requirements-ci.txt

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

@@ -14,7 +14,7 @@ if not errorlevel 0 exit /b
 :: build\torch. Rather than changing all these references, making a copy of torch folder
 :: from conda to the current workspace is easier. The workspace will be cleaned up after
 :: the job anyway
-xcopy /s %CONDA_PARENT_DIR%\Miniconda3\Lib\site-packages\torch %TMP_DIR_WIN%\build\torch\
+xcopy /s %CONDA_ROOT_DIR%\envs\py_tmp\Lib\site-packages\torch %TMP_DIR_WIN%\build\torch\
 
 pushd .
 if "%VC_VERSION%" == "" (

.ci/pytorch/win-test.sh

@@ -38,7 +38,14 @@ if [[ "$BUILD_ENVIRONMENT" == *cuda* ]]; then
 fi
 
 # TODO: Move both of them to Windows AMI
-python -m pip install pytest-rerunfailures==10.3 pytest-cpp==2.3.0 tensorboard==2.13.0 protobuf==5.29.4 pytest-subtests==0.13.1
+python -m pip install tensorboard==2.13.0 protobuf==5.29.4 pytest-subtests==0.13.1
+
+# Copied from https://github.com/pytorch/test-infra/blob/be01a40157c36cd5a48391fdf44a7bc3ebd4c7e3/aws/ami/windows/scripts/Installers/Install-Pip-Dependencies.ps1#L16 with some adjustments
+# pytest-rerunfailures==10.3 as 10.2 fails with INTERNALERROR> pluggy._manager.PluginValidationError: unknown hook 'pytest_configure_node'
+# scipy from 1.6.3 to 1.10
+# expecttest from 0.1.3 to 0.3.0
+# xdoctest from 1.0.2 to 1.3.0
+python -m pip install "future==0.18.2" "hypothesis==5.35.1" "expecttest==0.3.0" "librosa>=0.6.2" "scipy==1.10.1" "psutil==5.9.1" "pynvml==11.4.1" "pillow==9.2.0" "unittest-xml-reporting<=3.2.0,>=2.0.0" "pytest==7.1.3" "pytest-xdist==2.5.0" "pytest-flakefinder==1.1.0" "pytest-rerunfailures==10.3" "pytest-shard==0.1.2" "sympy==1.11.1" "xdoctest==1.3.0" "pygments==2.12.0" "opt-einsum>=3.3" "networkx==2.8.8" "mpmath==1.2.1" "pytest-cpp==2.3.0" "boto3==1.35.42"
 
 # Install Z3 optional dependency for Windows builds.
 python -m pip install z3-solver==4.15.1.0
@@ -52,9 +59,6 @@ python -m pip install parameterized==0.8.1
 # Install pulp for testing ilps under torch\distributed\_tools
 python -m pip install pulp==2.9.0
 
-# Install expecttest to merge https://github.com/pytorch/pytorch/pull/155308
-python -m pip install expecttest==0.3.0
-
 run_tests() {
     # Run nvidia-smi if available
     for path in '/c/Program Files/NVIDIA Corporation/NVSMI/nvidia-smi.exe' /c/Windows/System32/nvidia-smi.exe; do

.ci/pytorch/windows/cuda128.bat

@@ -37,10 +37,10 @@ IF "%CUDA_PATH_V128%"=="" (
 )
 
 IF "%BUILD_VISION%" == "" (
-    set TORCH_CUDA_ARCH_LIST=6.1;7.0;7.5;8.0;8.6;9.0;10.0;12.0
+    set TORCH_CUDA_ARCH_LIST=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 NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -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%"

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

@@ -1,12 +1,20 @@
-copy "%CUDA_PATH%\bin\cusparse*64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\bin\cublas*64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\bin\cudart*64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\bin\curand*64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\bin\cufft*64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\bin\cusolver*64_*.dll*" pytorch\torch\lib
+
+if %CUDA_VERSION% geq 130 (
+    set "dll_path=bin\x64"
+) else (
+    set "dll_path=bin"
+)
+
+copy "%CUDA_PATH%\%dll_path%\cusparse*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\cublas*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\cudart*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\curand*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\cufft*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\cusolver*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\nvrtc*64_*.dll*" pytorch\torch\lib
+copy "%CUDA_PATH%\%dll_path%\nvJitLink_*.dll*"  pytorch\torch\lib
 
 copy "%CUDA_PATH%\bin\cudnn*64_*.dll*" pytorch\torch\lib
-copy "%CUDA_PATH%\bin\nvrtc*64_*.dll*" pytorch\torch\lib
 copy "%CUDA_PATH%\extras\CUPTI\lib64\cupti64_*.dll*" pytorch\torch\lib
 copy "%CUDA_PATH%\extras\CUPTI\lib64\nvperf_host*.dll*" pytorch\torch\lib
 
@@ -20,8 +28,3 @@ copy "%libuv_ROOT%\bin\uv.dll" pytorch\torch\lib
 if exist "C:\Windows\System32\zlibwapi.dll" (
     copy "C:\Windows\System32\zlibwapi.dll"  pytorch\torch\lib
 )
-
-::copy nvJitLink dll is requires for cuda 12+
-if exist "%CUDA_PATH%\bin\nvJitLink_*.dll*" (
-    copy "%CUDA_PATH%\bin\nvJitLink_*.dll*"  pytorch\torch\lib
-)

.ci/pytorch/windows/internal/driver_update.bat

@@ -1,9 +1,9 @@
-set WIN_DRIVER_VN=528.89
-set "DRIVER_DOWNLOAD_LINK=https://ossci-windows.s3.amazonaws.com/%WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe" & REM @lint-ignore
-curl --retry 3 -kL %DRIVER_DOWNLOAD_LINK% --output %WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe
+set WIN_DRIVER_VN=580.88
+set "DRIVER_DOWNLOAD_LINK=https://ossci-windows.s3.amazonaws.com/%WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe" & REM @lint-ignore
+curl --retry 3 -kL %DRIVER_DOWNLOAD_LINK% --output %WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe
 if errorlevel 1 exit /b 1
 
-start /wait %WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe -s -noreboot
+start /wait %WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe -s -noreboot
 if errorlevel 1 exit /b 1
 
-del %WIN_DRIVER_VN%-data-center-tesla-desktop-winserver-2016-2019-2022-dch-international.exe || ver > NUL
+del %WIN_DRIVER_VN%-data-center-tesla-desktop-win10-win11-64bit-dch-international.exe || ver > NUL

.ci/wheel/build_wheel.sh

@@ -124,19 +124,15 @@ popd
 
 export TH_BINARY_BUILD=1
 export INSTALL_TEST=0 # dont install test binaries into site-packages
-export MACOSX_DEPLOYMENT_TARGET=10.15
+export MACOSX_DEPLOYMENT_TARGET=11.0
 export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
 
-SETUPTOOLS_PINNED_VERSION="==70.1.0"
-PYYAML_PINNED_VERSION="==5.3"
 EXTRA_CONDA_INSTALL_FLAGS=""
 CONDA_ENV_CREATE_FLAGS=""
 RENAME_WHEEL=true
 case $desired_python in
     3.14t)
         echo "Using 3.14 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        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/label/python_rc -c conda-forge"
@@ -145,8 +141,6 @@ case $desired_python in
         ;;
     3.14)
         echo "Using 3.14t deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        PYYAML_PINNED_VERSION=">=6.0.1"
         NUMPY_PINNED_VERSION="==2.1.0"
         EXTRA_CONDA_INSTALL_FLAGS="-c conda-forge/label/python_rc -c conda-forge"
         desired_python="3.14.0rc1"
@@ -154,8 +148,6 @@ case $desired_python in
         ;;
     3.13t)
         echo "Using 3.13 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        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"
@@ -164,37 +156,23 @@ case $desired_python in
         ;;
     3.13)
         echo "Using 3.13 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        PYYAML_PINNED_VERSION=">=6.0.1"
         NUMPY_PINNED_VERSION="==2.1.0"
         ;;
     3.12)
         echo "Using 3.12 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        PYYAML_PINNED_VERSION=">=6.0.1"
         NUMPY_PINNED_VERSION="==2.0.2"
         ;;
     3.11)
         echo "Using 3.11 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        PYYAML_PINNED_VERSION=">=5.3"
         NUMPY_PINNED_VERSION="==2.0.2"
         ;;
     3.10)
         echo "Using 3.10 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        PYYAML_PINNED_VERSION=">=5.3"
-        NUMPY_PINNED_VERSION="==2.0.2"
-        ;;
-    3.9)
-        echo "Using 3.9 deps"
-        SETUPTOOLS_PINNED_VERSION=">=70.1.0"
-        PYYAML_PINNED_VERSION=">=5.3"
         NUMPY_PINNED_VERSION="==2.0.2"
         ;;
     *)
-        echo "Using default deps"
-        NUMPY_PINNED_VERSION="==1.11.3"
+        echo "Unsupported version $desired_python"
+        exit 1
         ;;
 esac
 
@@ -204,8 +182,6 @@ conda create ${EXTRA_CONDA_INSTALL_FLAGS} -yn "$tmp_env_name" python="$desired_p
 source activate "$tmp_env_name"
 
 PINNED_PACKAGES=(
-    "setuptools${SETUPTOOLS_PINNED_VERSION}"
-    "pyyaml${PYYAML_PINNED_VERSION}"
     "numpy${NUMPY_PINNED_VERSION}"
 )
 retry pip install "${PINNED_PACKAGES[@]}" -r "${pytorch_rootdir}/requirements-build.txt"
@@ -213,8 +189,7 @@ pip install requests ninja typing-extensions
 retry pip install -r "${pytorch_rootdir}/requirements.txt" || true
 retry brew install libomp
 
-# For USE_DISTRIBUTED=1 on macOS, this enables gloo, which needs libuv, which
-# is build as part of tensorpipe submodule
+# For USE_DISTRIBUTED=1 on macOS, need libuv, which is build as part of tensorpipe submodule
 export USE_DISTRIBUTED=1
 
 export USE_MKLDNN=OFF
@@ -224,7 +199,7 @@ export BUILD_TEST=OFF
 pushd "$pytorch_rootdir"
 echo "Calling setup.py bdist_wheel at $(date)"
 
-python setup.py bdist_wheel -d "$whl_tmp_dir"
+python setup.py bdist_wheel -d "$whl_tmp_dir" --plat-name ${mac_version}
 
 echo "Finished setup.py bdist_wheel at $(date)"
 

.circleci/scripts/binary_populate_env.sh

@@ -71,14 +71,7 @@ export PYTORCH_BUILD_NUMBER=1
 
 # Set triton version as part of PYTORCH_EXTRA_INSTALL_REQUIREMENTS
 TRITON_VERSION=$(cat $PYTORCH_ROOT/.ci/docker/triton_version.txt)
-
-# Here PYTORCH_EXTRA_INSTALL_REQUIREMENTS is already set for the all the wheel builds hence append TRITON_CONSTRAINT
-TRITON_CONSTRAINT="platform_system == 'Linux' and platform_machine == 'x86_64'"
-
-# CUDA 12.9/13.0 builds have triton for Linux and Linux aarch64 binaries.
-if [[ "$DESIRED_CUDA" == "cu129" ]] || [[ "$DESIRED_CUDA" == "cu130" ]]; then
-  TRITON_CONSTRAINT="platform_system == 'Linux'"
-fi
+TRITON_CONSTRAINT="platform_system == 'Linux'"
 
 if [[ "$PACKAGE_TYPE" =~ .*wheel.* &&  -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" && ! "$PYTORCH_BUILD_VERSION" =~ .*xpu.* ]]; then
   TRITON_REQUIREMENT="triton==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"

.github/actionlint.yaml

@@ -12,7 +12,9 @@ self-hosted-runner:
     - linux.9xlarge.ephemeral
     - am2.linux.9xlarge.ephemeral
     - linux.12xlarge
+    - linux.12xlarge.memory
     - linux.24xlarge
+    - linux.24xlarge.memory
     - linux.24xlarge.ephemeral
     - linux.24xlarge.amd
     - linux.arm64.2xlarge

.github/actions/build-external-packages/action.yml

@@ -4,6 +4,11 @@ name: Build External packages
 description: build external packages for PyTorch
 
 inputs:
+  cuda-version:
+    description: CUDA version to use
+    type: string
+    required: true
+    default: '12.8.1'
   cuda-arch-list:
     description: TORCH_CUDA_ARCH_LIST (e.g., "8.0;8.9;9.0")
     type: string
@@ -44,11 +49,12 @@ runs:
       env:
         SCCACHE_BUCKET: ossci-compiler-cache-circleci-v2
         SCCACHE_REGION: us-east-1
+        CUDA_VERSION: ${{ inputs.cuda-version }}
         TORCH_CUDA_ARCH_LIST: ${{ inputs.cuda-arch-list }}
         BASE_IMAGE: ${{ inputs.docker-image }}
         BUILD_TARGETS: ${{ inputs.build-targets }}
-        PARENT_OUTPUT_DIR: ${{ inputs.output-dir}}
-
+        PARENT_OUTPUT_DIR: ${{ inputs.output-dir }}
+        TORCH_WHEELS_PATH: ${{ inputs.torch-wheel-dir }}
       shell: bash
       run: |
         set -euo pipefail
@@ -69,7 +75,6 @@ runs:
           export OUTPUT_DIR
           echo "Building external package: $target in directory $OUTPUT_DIR"
           python3 -m cli.run build external "$target"
-
         done
 
         END_TIME=$(date +%s)

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

@@ -6,6 +6,12 @@ inputs:
   cuda-version:
     description: which cuda version to install, 'cpu' for none
     required: true
+  python-version:
+    required: false
+    type: string
+    default: "3.10"
+    description: |
+      The python version to be used. Will be 3.10 by default
 
 runs:
   using: composite
@@ -38,18 +44,24 @@ runs:
         CONDA="C:\Jenkins\Miniconda3\condabin\conda.bat"
 
         {
+          echo "CONDA=${CONDA}";
           echo "CONDA_RUN=${CONDA} run --no-capture-output";
           echo "CONDA_BUILD=${CONDA} run conda-build";
           echo "CONDA_INSTALL=${CONDA} install";
         } >> "${GITHUB_ENV}"
 
     - name: Setup Python3
+      env:
+          PYTHON_VERSION: ${{ inputs.python-version }}
       shell: bash
       run: |
         set +e
         set -x
 
-        PYTHON3=$(${CONDA_RUN} which python3)
+        # Create new py_tmp env with python-version
+        ${CONDA} create -y -n py_tmp python=${PYTHON_VERSION} intel-openmp libuv
+
+        PYTHON3=$(${CONDA_RUN} -n py_tmp which python3)
         EXIT_CODE=$?
 
         if [[ "${EXIT_CODE}" == "0" ]]; then
@@ -62,7 +74,7 @@ runs:
           # installation, which is Python 3 based. Its Python is default to Python 3. Further, there
           # is also the Miniconda installation that is Python 2 based, and both can be installed if
           # needed. In both cases, Python binary is just called python
-          PYTHON=$(${CONDA_RUN} which python)
+          PYTHON=$(${CONDA_RUN} -n py_tmp which python)
           EXIT_CODE=$?
 
           if [[ "${EXIT_CODE}" == "0" ]]; then

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-0757bbb660855272f7dd8d31cc84e7c631522805
+27fc2493d383354a008106f22f3be232badee9a1

.github/ci_commit_pins/vllm.txt

@@ -1 +1 @@
-862f2ef893d9751db0a92bd2d4ae0e3d9677872f
+78a47f87ce259a48f0391fa9ae15add05ea7432b

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-763e5b78d4fcd74a9e812256656c075f99d9a781
+r2.9

.github/ci_configs/vllm/Dockerfile.tmp_vllm

@@ -12,54 +12,46 @@ ARG BUILD_BASE_IMAGE=torch-nightly-base
 # by default, it uses devel-ubuntu22.04 official image.
 ARG FINAL_BASE_IMAGE=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04
 
+# The logic is copied from https://github.com/vllm-project/vllm/blob/main/docker/Dockerfile
+ARG GET_PIP_URL="https://bootstrap.pypa.io/get-pip.py"
 
-#################### TORCH NIGHTLY  BASE IMAGE ####################
+
+#################### TORCH NIGHTLY BASE IMAGE ####################
 # A base image for building vLLM with devel ubuntu 22.04, this is mainly used to build vllm in vllm builtkite ci
-From nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 as torch-nightly-base
-ARG CUDA_VERSION=12.8.1
-ARG PYTHON_VERSION=3.12
-ARG TARGETPLATFORM
-ENV DEBIAN_FRONTEND=noninteractive
+FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 as torch-nightly-base
 
-RUN PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && \
-    echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment
+ARG CUDA_VERSION
+ARG PYTHON_VERSION
+ARG GET_PIP_URL
 
-# Install Python and other dependencies if it does not existed
-RUN if ! command -v python3 >/dev/null || ! python3 --version | grep -q "${PYTHON_VERSION}"; then \
-      echo "Installing Python ${PYTHON_VERSION}..." && \
-      echo 'tzdata tzdata/Areas select America' | debconf-set-selections && \
-      echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections && \
-      apt-get update -y && \
-      apt-get install -y ccache software-properties-common git curl sudo && \
-      for i in 1 2 3; do \
-        add-apt-repository -y ppa:deadsnakes/ppa && break || \
-        { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
-      done && \
-      apt-get update -y && \
-      apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv && \
-      update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 && \
-      update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} && \
-      ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config && \
-      curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION}; \
-   else \
-      echo "Python ${PYTHON_VERSION} already present, skipping setup."; \
-   fi \
-   && python3 --version && python3 -m pip --version
+# Install Python and other dependencies
+RUN apt-get update -y \
+    && apt-get install -y ccache software-properties-common git curl wget sudo vim \
+    && add-apt-repository -y ppa:deadsnakes/ppa \
+    && apt-get update -y \
+    && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
+    && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
+    && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \
+    && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
+    && curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION} \
+    && python3 --version && python3 -m pip --version
 
 # Upgrade to GCC 10 to avoid https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92519
 # as it was causing spam when compiling the CUTLASS kernels
 # Ensure gcc >= 10 to avoid CUTLASS issues (bug 92519)
 RUN current_gcc_version=$(gcc -dumpversion | cut -f1 -d.) && \
-    if [ "$current_gcc_version" -lt 10 ]; then \
-      echo "GCC version is $current_gcc_version, installing gcc-10..."; \
-      apt-get update && \
-      apt-get install -y gcc-10 g++-10 && \
-      update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-10 100 && \
-      update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-10 100; \
-    else \
-      echo "GCC version is $current_gcc_version, no need to install gcc-10."; \
-    fi && \
-    gcc --version && g++ --version
+    if command -v apt-get >/dev/null; then \
+        if [ "$current_gcc_version" -lt 10 ]; then \
+            echo "GCC version is $current_gcc_version, installing gcc-10..."; \
+            apt-get update \
+            && apt-get install -y gcc-10 g++-10 \
+            && update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-10 100 \
+            && update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-10 100; \
+        else \
+            echo "GCC version is $current_gcc_version, no need to install gcc-10."; \
+        fi \
+    fi \
+    && gcc --version && g++ --version
 
 # install uv for faster pip installs
 RUN --mount=type=cache,target=/root/.cache/uv \
@@ -79,6 +71,21 @@ ENV UV_LINK_MODE=copy
 FROM ${BUILD_BASE_IMAGE} AS base
 USER root
 
+ARG CUDA_VERSION
+ARG PYTHON_VERSION
+
+# TODO (huydhn): Only work with PyTorch manylinux builder
+ENV PATH="/opt/python/cp312-cp312/bin:${PATH}"
+
+# Install some system dependencies and double check python version
+RUN if command -v apt-get >/dev/null; then \
+        apt-get update -y \
+        && apt-get install -y ccache software-properties-common git curl wget sudo vim; \
+    else \
+        dnf install -y git curl wget sudo vim; \
+    fi \
+    && python3 --version && python3 -m pip --version
+
 # Workaround for https://github.com/openai/triton/issues/2507 and
 # https://github.com/pytorch/pytorch/issues/107960 -- hopefully
 # this won't be needed for future versions of this docker image
@@ -118,17 +125,15 @@ RUN --mount=type=bind,source=${TORCH_WHEELS_PATH},target=/dist \
     if [ -n "$TORCH_WHEELS_PATH" ] && [ "$TORCH_WHEELS_PATH" != "./requirements" ] && [ -d "/dist" ] && ls /dist/torch*.whl >/dev/null 2>&1; then \
         echo "[INFO] Installing torch wheels to build vllm"; \
         torch_whl=$(find /dist -maxdepth 1 -name 'torch-*.whl' -print -quit); \
-        vision_whl=$(find /dist/vision -name 'torchvision*.whl' | head -n1 | xargs); \
-        audio_whl=$(find /dist/audio -name 'torchaudio*.whl' | head -n1 | xargs); \
-        uv pip install --system "${torch_whl}[opt-einsum]"; \
-        uv pip install --system "${vision_whl}"; \
-        uv pip install --system "${audio_whl}"; \
+        vision_whl=$(find /dist -name 'torchvision*.whl' | head -n1 | xargs); \
+        audio_whl=$(find /dist -name 'torchaudio*.whl' | head -n1 | xargs); \
+        uv pip install --system "${torch_whl}[opt-einsum]" "${vision_whl}" "${audio_whl}" /dist/*.whl; \
     elif [ -n "$PINNED_TORCH_VERSION" ]; then \
         echo "[INFO] Installing pinned torch nightly version to build vllm: $PINNED_TORCH_VERSION"; \
-        uv pip install --system "$PINNED_TORCH_VERSION" --index-url https://download.pytorch.org/whl/nightly/cu128; \
+        uv pip install --system "$PINNED_TORCH_VERSION" --index-url https://download.pytorch.org/whl/nightly/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); \
     else \
         echo "[INFO] Installing torch nightly with latest one to build vllm"; \
-        uv pip install --system torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128; \
+        uv pip install --system torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); \
     fi
 
 # Install numba 0.61.2 for cuda environment
@@ -137,12 +142,11 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 
 # Install common dependencies from vllm common.txt
 RUN --mount=type=cache,target=/root/.cache/uv \
-uv pip install --system -r requirements/common.txt
-
+    uv pip install --system -r requirements/common.txt
 
 # Must put before installing xformers, so it can install the correct version of xfomrers.
-ARG exformer_cuda_arch_list='7.5;8.0+PTX;9.0a'
-ENV TORCH_CUDA_ARCH_LIST=${exformer_cuda_arch_list}
+ARG xformers_cuda_arch_list='7.5;8.0+PTX;9.0a'
+ENV TORCH_CUDA_ARCH_LIST=${xformers_cuda_arch_list}
 
 ARG max_jobs=16
 ENV MAX_JOBS=${max_jobs}
@@ -153,8 +157,8 @@ RUN pip freeze | grep -E 'ninja'
 
 # Build xformers with cuda and torch nightly/wheel
 # following official xformers guidance: https://github.com/facebookresearch/xformers#build
-# sha for https://github.com/facebookresearch/xformers/tree/v0.0.31
-ARG XFORMERS_COMMIT=eb0946a363464da96ea40afd1a7f72a907c25497
+# sha for https://github.com/facebookresearch/xformers/tree/v0.0.32.post2
+ARG XFORMERS_COMMIT=5d4b92a5e5a9c6c6d4878283f47d82e17995b468
 ENV CCACHE_DIR=/root/.cache/ccache
 
 RUN --mount=type=cache,target=/root/.cache/ccache \
@@ -188,11 +192,6 @@ RUN pip freeze | grep -E 'torch|xformers|torchvision|torchaudio'
 FROM base AS build
 ARG TARGETPLATFORM
 
-ENV UV_HTTP_TIMEOUT=500
-ENV UV_INDEX_STRATEGY="unsafe-best-match"
-# Use copy mode to avoid hardlink failures with Docker cache mounts
-ENV UV_LINK_MODE=copy
-
 COPY . .
 
 RUN python3 use_existing_torch.py
@@ -251,9 +250,9 @@ RUN --mount=type=cache,target=/root/.cache/ccache \
         python3 setup.py bdist_wheel --dist-dir=vllm-dist --py-limited-api=cp38; \
     fi
 
-RUN echo "[DEBUG] Listing  current directory:" && \
+RUN echo "[INFO] Listing current directory:" && \
     ls -al && \
-    echo "[DEBUG] Showing torch_build_versions.txt content:" && \
+    echo "[INFO] Showing torch_build_versions.txt content:" && \
     cat torch_build_versions.txt
 
 #################### WHEEL BUILD IMAGE ####################
@@ -263,42 +262,40 @@ RUN echo "[DEBUG] Listing  current directory:" && \
 # Setup clean environment for vLLM for test and api server using ubuntu22.04 with AOT flashinfer
 FROM ${FINAL_BASE_IMAGE} AS vllm-base
 USER root
+
+ARG CUDA_VERSION
+ARG PYTHON_VERSION
+ARG GET_PIP_URL
+
+# TODO (huydhn): Only work with PyTorch manylinux builder
+ENV PATH="/opt/python/cp312-cp312/bin:${PATH}"
+
 # prepare for environment starts
 WORKDIR /workspace
 
-RUN PYTHON_VERSION_STR=$(echo ${PYTHON_VERSION} | sed 's/\.//g') && \
-    echo "export PYTHON_VERSION_STR=${PYTHON_VERSION_STR}" >> /etc/environment
-
-# Install Python and other dependencies if it does not existed
-RUN if ! command -v python3 >/dev/null || ! python3 --version | grep -q "${PYTHON_VERSION}"; then \
-      echo "Installing Python ${PYTHON_VERSION}..." && \
-      echo 'tzdata tzdata/Areas select America' | debconf-set-selections && \
-      echo 'tzdata tzdata/Zones/America select Los_Angeles' | debconf-set-selections && \
-      apt-get update -y && \
-      apt-get install -y ccache software-properties-common git curl sudo && \
-      for i in 1 2 3; do \
-        add-apt-repository -y ppa:deadsnakes/ppa && break || \
-        { echo "Attempt $i failed, retrying in 5s..."; sleep 5; }; \
-      done && \
-      apt-get update -y && \
-      apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv && \
-      update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 && \
-      update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} && \
-      ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config && \
-      curl -sS https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION}; \
-   else \
-      echo "Python ${PYTHON_VERSION} already present, skipping setup."; \
-   fi \
-   && python3 --version && python3 -m pip --version
-
+# Install Python and other dependencies
+RUN if command -v apt-get >/dev/null; then \
+        apt-get update -y \
+        && apt-get install -y ccache software-properties-common git curl wget sudo vim \
+        && add-apt-repository -y ppa:deadsnakes/ppa \
+        && apt-get update -y \
+        && apt-get install -y python${PYTHON_VERSION} python${PYTHON_VERSION}-dev python${PYTHON_VERSION}-venv \
+        && update-alternatives --install /usr/bin/python3 python3 /usr/bin/python${PYTHON_VERSION} 1 \
+        && update-alternatives --set python3 /usr/bin/python${PYTHON_VERSION} \
+        && ln -sf /usr/bin/python${PYTHON_VERSION}-config /usr/bin/python3-config \
+        && curl -sS ${GET_PIP_URL} | python${PYTHON_VERSION}; \
+    else \
+        dnf install -y git curl wget sudo vim; \
+    fi \
+    && python3 --version && python3 -m pip --version
 
 # Get the torch versions, and whls used in previous stagtes for consistency
 COPY --from=base /workspace/torch_build_versions.txt ./torch_build_versions.txt
 COPY --from=base /workspace/xformers-dist /wheels/xformers
 COPY --from=build /workspace/vllm-dist /wheels/vllm
-RUN echo "[DEBUG] Listing current directory before torch install step:" && \
+RUN echo "[INFO] Listing current directory before torch install step:" && \
     ls -al && \
-    echo "[DEBUG] Showing torch_build_versions.txt content:" && \
+    echo "[INFO] Showing torch_build_versions.txt content:" && \
     cat torch_build_versions.txt
 
 # Workaround for https://github.com/openai/triton/issues/2507 and
@@ -307,7 +304,6 @@ RUN echo "[DEBUG] Listing current directory before torch install step:" && \
 # or future versions of triton.
 RUN ldconfig /usr/local/cuda-$(echo $CUDA_VERSION | cut -d. -f1,2)/compat/
 
-
 # Install uv for faster pip installs if not existed
 RUN --mount=type=cache,target=/root/.cache/uv \
     if ! python3 -m uv --version > /dev/null 2>&1; then \
@@ -327,15 +323,13 @@ RUN --mount=type=bind,source=${TORCH_WHEELS_PATH},target=/dist \
     --mount=type=cache,target=/root/.cache/uv \
     if [ -n "$TORCH_WHEELS_PATH" ] && [ "$TORCH_WHEELS_PATH" != "./requirements" ] && [ -d "/dist" ] && ls /dist/torch*.whl >/dev/null 2>&1; then \
         torch_whl=$(find /dist -maxdepth 1 -name 'torch-*.whl' -print -quit); \
-        vision_whl=$(find /dist/vision -name 'torchvision*.whl' | head -n1 | xargs); \
-        audio_whl=$(find /dist/audio -name 'torchaudio*.whl' | head -n1 | xargs); \
+        vision_whl=$(find /dist -name 'torchvision*.whl' | head -n1 | xargs); \
+        audio_whl=$(find /dist -name 'torchaudio*.whl' | head -n1 | xargs); \
         echo "[INFO] Use wheels to build : '${torch_whl}' '${audio_whl}' '${vision_whl}'"; \
-        uv pip install --system "${torch_whl}[opt-einsum]"; \
-        uv pip install --system "${vision_whl}"; \
-        uv pip install --system "${audio_whl}"; \
+        uv pip install --system "${torch_whl}[opt-einsum]" "${vision_whl}" "${audio_whl}" /dist/*.whl; \
     else \
         echo "[INFO] Installing torch versions from torch_build_versions.txt"; \
-        uv pip install --system $(cat torch_build_versions.txt | xargs) --index-url https://download.pytorch.org/whl/nightly/cu128; \
+        uv pip install --system $(cat torch_build_versions.txt | xargs) --index-url https://download.pytorch.org/whl/nightly/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.'); \
     fi
 
 # Install the vllm wheel from previous stage
@@ -346,9 +340,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 RUN --mount=type=cache,target=/root/.cache/uv \
     uv pip install --system /wheels/xformers/*.whl --verbose
 
-
 # Build flashinfer from source.
-ARG torch_cuda_arch_list='8.0;8.9;9.0a'
+ARG torch_cuda_arch_list='8.0;8.9;9.0a;10.0a;12.0'
 # install package for build flashinfer
 # see issue: https://github.com/flashinfer-ai/flashinfer/issues/738
 
@@ -416,11 +409,6 @@ RUN --mount=type=cache,target=/root/.cache/uv \
 RUN --mount=type=cache,target=/root/.cache/uv \
     uv pip install --system -r requirements/nightly_torch_test.txt
 
-# Workaround for #17068
-# pinned commit for v2.2.4
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv pip install --system --no-build-isolation "git+https://github.com/state-spaces/mamba@95d8aba8a8c75aedcaa6143713b11e745e7cd0d9#egg=mamba-ssm"
-
 # Logging to confirm the torch versions
 RUN pip freeze | grep -E 'torch|xformers|vllm|flashinfer'
 

.github/scripts/build_triton_wheel.py

@@ -84,6 +84,7 @@ def build_triton(
                 ["git", "checkout", f"release/{ver}.{rev}.x"], cwd=triton_basedir
             )
         else:
+            check_call(["git", "fetch", "origin", commit_hash], cwd=triton_basedir)
             check_call(["git", "checkout", commit_hash], cwd=triton_basedir)
 
         # change built wheel name and version

.github/scripts/filter_test_configs.py

@@ -41,9 +41,9 @@ SUPPORTED_PERIODICAL_MODES: dict[str, Callable[[Optional[str]], bool]] = {
 }
 
 # The link to the published list of disabled jobs
-DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json"
+DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json?versionId=hjktHz2WOejHpxKpkqpDknTt5rMTM9KK"
 # and unstable jobs
-UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json"
+UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json?versionId=wrjdvvQTJxgvMO.rGw5MEuMsj6XbjuV7"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"

.github/scripts/generate_binary_build_matrix.py

@@ -16,18 +16,16 @@ from typing import Optional
 
 
 # NOTE: Please also update the CUDA sources in `PIP_SOURCES` in tools/nightly.py when changing this
-CUDA_ARCHES = ["12.6", "12.8", "12.9", "13.0"]
+CUDA_ARCHES = ["12.6", "12.8", "13.0"]
 CUDA_STABLE = "12.8"
 CUDA_ARCHES_FULL_VERSION = {
     "12.6": "12.6.3",
     "12.8": "12.8.1",
-    "12.9": "12.9.1",
     "13.0": "13.0.0",
 }
 CUDA_ARCHES_CUDNN_VERSION = {
     "12.6": "9",
     "12.8": "9",
-    "12.9": "9",
     "13.0": "9",
 }
 
@@ -40,77 +38,60 @@ CPU_AARCH64_ARCH = ["cpu-aarch64"]
 
 CPU_S390X_ARCH = ["cpu-s390x"]
 
-CUDA_AARCH64_ARCHES = ["12.9-aarch64", "13.0-aarch64"]
+CUDA_AARCH64_ARCHES = ["12.6-aarch64", "12.8-aarch64", "13.0-aarch64"]
 
 
 PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
     "12.6": (
-        "nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux' and platform_machine == 'x86_64'"
+        "nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | "
+        "nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | "
+        "nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | "
+        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | "
+        "nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | "
+        "nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | "
+        "nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | "
+        "nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | "
+        "nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | "
+        "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | "
+        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | "
+        "nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | "
+        "nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | "
+        "nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | "
+        "nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'"
     ),
     "12.8": (
-        "nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux' and platform_machine == 'x86_64'"
-    ),
-    "12.9": (
-        "nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'"
+        "nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | "
+        "nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | "
+        "nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | "
+        "nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | "
+        "nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | "
+        "nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | "
+        "nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | "
+        "nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | "
+        "nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | "
+        "nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | "
+        "nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | "
+        "nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | "
+        "nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | "
+        "nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | "
+        "nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'"
     ),
     "13.0": (
-        "nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'"
+        "nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | "
+        "nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | "
+        "nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | "
+        "nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | "
+        "nvidia-cublas==13.0.0.19; platform_system == 'Linux' | "
+        "nvidia-cufft==12.0.0.15; platform_system == 'Linux' | "
+        "nvidia-curand==10.4.0.35; platform_system == 'Linux' | "
+        "nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | "
+        "nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | "
+        "nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | "
+        "nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | "
+        "nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | "
+        "nvidia-nvtx==13.0.39; platform_system == 'Linux' | "
+        "nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | "
+        "nvidia-cufile==1.15.0.42; platform_system == 'Linux'"
     ),
     "xpu": (
         "intel-cmplr-lib-rt==2025.2.1 | "
@@ -240,8 +221,6 @@ def generate_libtorch_matrix(
         if os == "linux":
             arches += CUDA_ARCHES
             arches += ROCM_ARCHES
-            if "13.0" in arches:
-                arches.remove("13.0")
         elif os == "windows":
             arches += CUDA_ARCHES
     if libtorch_variants is None:
@@ -343,7 +322,7 @@ def generate_wheels_matrix(
             # cuda linux wheels require PYTORCH_EXTRA_INSTALL_REQUIREMENTS to install
 
             if (
-                arch_version in ["13.0", "12.9", "12.8", "12.6"]
+                arch_version in ["13.0", "12.8", "12.6"]
                 and os == "linux"
                 or arch_version in CUDA_AARCH64_ARCHES
             ):
@@ -407,6 +386,5 @@ def generate_wheels_matrix(
 
 
 validate_nccl_dep_consistency("13.0")
-validate_nccl_dep_consistency("12.9")
 validate_nccl_dep_consistency("12.8")
 validate_nccl_dep_consistency("12.6")

.github/scripts/generate_ci_workflows.py

@@ -22,7 +22,7 @@ LABEL_CIFLOW_BINARIES = "ciflow/binaries"
 LABEL_CIFLOW_PERIODIC = "ciflow/periodic"
 LABEL_CIFLOW_BINARIES_LIBTORCH = "ciflow/binaries_libtorch"
 LABEL_CIFLOW_BINARIES_WHEEL = "ciflow/binaries_wheel"
-LABEL_CIFLOW_ROCM = "ciflow/rocm-mi300"
+LABEL_CIFLOW_ROCM = "ciflow/rocm"
 
 
 @dataclass
@@ -135,10 +135,12 @@ ROCM_SMOKE_WORKFLOWS = [
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(
             OperatingSystem.LINUX,
             arches=["6.4"],
-            python_versions=["3.9"],
+            python_versions=["3.10"],
         ),
         ciflow_config=CIFlowConfig(
             labels={
+                LABEL_CIFLOW_BINARIES,
+                LABEL_CIFLOW_BINARIES_WHEEL,
                 LABEL_CIFLOW_ROCM,
             },
             isolated_workflow=True,

.github/templates/common.yml.j2

@@ -32,7 +32,7 @@ concurrency:
 {%- macro setup_ec2_windows() -%}
       !{{ display_ec2_information() }}
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}

.github/templates/linux_binary_build_workflow.yml.j2

@@ -56,7 +56,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -77,6 +77,9 @@ jobs:
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
       timeout-minutes: 420
+      {%- elif config["gpu_arch_type"] == "rocm" %}
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
       {%- elif "conda" in build_environment and config["gpu_arch_type"] == "cuda" %}
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.24xlarge.ephemeral
@@ -135,7 +138,7 @@ jobs:
       contents: read
     steps:
       - name: Setup XPU
-        uses: pytorch/pytorch/.github/actions/setup-xpu@main
+        uses: pytorch/pytorch/.github/actions/setup-xpu@release/2.9
       - name: configure aws credentials
         id: aws_creds
         uses: aws-actions/configure-aws-credentials@v4
@@ -150,10 +153,10 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: !{{ config["container_image"] }}
@@ -161,7 +164,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary
@@ -171,7 +174,7 @@ jobs:
       - name: Teardown XPU
         uses: ./.github/actions/teardown-xpu
     {%- else %}
-    runs-on: linux.rocm.gpu.gfx942.1
+    runs-on: linux.rocm.gpu.mi250
     timeout-minutes: !{{ common.timeout_minutes }}
     !{{ upload.binary_env(config) }}
     steps:
@@ -182,7 +185,7 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       - name: ROCm set GPU_FLAG
         run: |
           echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"
@@ -196,7 +199,7 @@ jobs:
           role-duration-seconds: 18000
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: !{{ config["container_image"] }}
@@ -204,7 +207,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary

.github/templates/macos_binary_build_workflow.yml.j2

@@ -68,12 +68,7 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       - name: Populate binary env
         run: |
           # shellcheck disable=SC1091

.github/templates/upload.yml.j2

@@ -33,7 +33,7 @@
   {%- if is_windows %}
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
   {%- endif %}
 
 {%- else %}

.github/templates/windows_binary_build_workflow.yml.j2

@@ -64,7 +64,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -135,7 +135,7 @@ jobs:
 {%- else %}
       !{{ set_runner_specific_vars() }}
       !{{ common.setup_ec2_windows() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
 {%- endif %}
       - name: Populate binary env
         shell: bash
@@ -211,7 +211,7 @@ jobs:
           "pytorch/.ci/pytorch/windows/arm64/bootstrap_rust.bat"
 {%- else %}
       !{{ common.setup_ec2_windows() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
       !{{ set_runner_specific_vars() }}
 {%- endif %}
       - uses: !{{ common.download_artifact_action }}

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

@@ -47,7 +47,7 @@ jobs:
       reenabled-issues: ${{ steps.filter.outputs.reenabled-issues }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           fetch-depth: 1
           submodules: false
@@ -69,25 +69,25 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -97,7 +97,7 @@ jobs:
         run: echo "IN_CONTAINER_RUNNER=$(if [ -f /.inarc ] || [ -f /.incontainer ]; then echo true ; else echo false; fi)" >> "$GITHUB_OUTPUT"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.9
         if: ${{ inputs.cuda-version != 'cpu' && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' }}
 
       - name: Output disk space left
@@ -209,5 +209,5 @@ jobs:
           file-suffix: bazel-${{ github.job }}_${{ steps.get-job-id.outputs.job-id }}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always()

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

@@ -142,13 +142,13 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         if: inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -178,7 +178,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -213,9 +212,9 @@ jobs:
       - name: Calculate docker image
         id: calculate-docker-image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
-          # If doing this in main or release branch, use docker.io. Otherwise
+          # If doing this in release/2.9 or release branch, use docker.io. Otherwise
           # use ECR
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: ${{ inputs.DOCKER_IMAGE }}
@@ -227,7 +226,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -283,7 +282,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

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

@@ -125,14 +125,14 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         if: inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
         # Setup the environment
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -155,7 +155,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           show-progress: false
           path: pytorch
@@ -186,9 +185,7 @@ jobs:
           path: "${{ runner.temp }}/artifacts/"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
-        with:
-          driver-version: ${{ startsWith(inputs.GPU_ARCH_VERSION, '13') && '580.65.06' || '570.133.07' }}
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.9
         if: ${{ inputs.GPU_ARCH_TYPE == 'cuda' && steps.filter.outputs.is-test-matrix-empty == 'False' }}
 
       - name: configure aws credentials
@@ -203,7 +200,7 @@ jobs:
       - name: Calculate docker image
         id: calculate-docker-image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: ${{ inputs.DOCKER_IMAGE }}
@@ -213,7 +210,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -225,7 +222,7 @@ jobs:
 
       - name: Teardown Linux
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

.github/workflows/_binary-upload.yml

@@ -81,7 +81,7 @@ jobs:
       SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 

.github/workflows/_docs.yml

@@ -67,7 +67,7 @@ jobs:
             # an OOM issue when running the job, so this upgrades the runner from 4xlarge
             # to the next available tier of 12xlarge. So much memory just to generate cpp
             # doc
-            runner: ${{ inputs.runner_prefix }}linux.12xlarge
+            runner: ${{ inputs.runner_prefix }}linux.12xlarge.memory
             # TODO: Nightly cpp docs take longer and longer to finish (more than 3h now)
             # Let's try to figure out how this can be improved
             timeout-minutes: 360
@@ -84,7 +84,7 @@ jobs:
     name: build-docs-${{ matrix.docs_type }}-${{ inputs.push }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -95,7 +95,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -110,12 +110,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -222,5 +222,5 @@ jobs:
           s3-prefix: pytorch/pytorch/${{ github.event.pull_request.number }}/functorchdocs
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always()

.github/workflows/_link_check.yml

@@ -11,7 +11,7 @@ on:
 jobs:
   lint-urls:
     if: ${{ github.event_name != 'pull_request' || !contains(github.event.pull_request.labels.*.name, 'skip-url-lint') }}
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@release/2.9
     with:
       job-name: lint-urls
       timeout: 120
@@ -37,7 +37,7 @@ jobs:
 
   lint-xrefs:
     if: ${{ github.event_name != 'pull_request' || !contains(github.event.pull_request.labels.*.name, 'skip-xref-lint') }}
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@release/2.9
     with:
       job-name: lint-xrefs
       timeout: 60

.github/workflows/_linux-build.yml

@@ -134,7 +134,7 @@ jobs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -147,7 +147,7 @@ jobs:
       # checkout because when we run this action we don't *have* a local
       # checkout. In other cases you should prefer a local checkout.
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 
@@ -183,7 +183,7 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
@@ -199,7 +199,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${tag/:/-}"
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         if: inputs.build-environment != 'linux-s390x-binary-manywheel' && steps.use-old-whl.outputs.reuse != 'true'
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
@@ -457,7 +457,7 @@ jobs:
           artifact_prefix: usage_log_build_${{ steps.get-job-id.outputs.job-id }}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always() && inputs.build-environment != 'linux-s390x-binary-manywheel'
 
       - name: Cleanup docker

.github/workflows/_linux-test.yml

@@ -99,7 +99,7 @@ jobs:
       contents: read
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         if: ${{ !contains(matrix.runner, 'b200') && inputs.build-environment != 'linux-s390x-binary-manywheel' }}
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -108,7 +108,7 @@ jobs:
               docker exec -it $(docker container ps --format '{{.ID}}') bash
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 
@@ -139,7 +139,7 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           docker-image-name: ${{ inputs.docker-image }}
@@ -155,7 +155,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${tag/:/-}"
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
@@ -167,9 +167,9 @@ jobs:
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
         id: install-nvidia-driver
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.9
         with:
-          driver-version: ${{ matrix.config == 'legacy_nvidia_driver' && '525.105.17' || '570.133.07' }}
+          driver-version: ${{ matrix.config == 'legacy_nvidia_driver' && '525.105.17' || '580.82.07' }}
         if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false' && !contains(matrix.runner, 'b200') }}
 
       - name: Setup GPU_FLAG for docker run
@@ -273,6 +273,8 @@ jobs:
           TEST_CONFIG: ${{ matrix.config }}
           SHARD_NUMBER: ${{ matrix.shard }}
           NUM_TEST_SHARDS: ${{ matrix.num_shards }}
+          EXTRA_FLAGS: ${{ matrix.extra_flags || '' }}
+          OP_BENCHMARK_TESTS: ${{ matrix.op_benchmark_tests }}
           REENABLED_ISSUES: ${{ steps.keep-going.outputs.reenabled-issues }}
           CONTINUE_THROUGH_ERROR: ${{ steps.keep-going.outputs.keep-going }}
           VERBOSE_TEST_LOGS: ${{ steps.keep-going.outputs.ci-verbose-test-logs }}
@@ -418,7 +420,7 @@ jobs:
           aws-region: us-east-1
 
       - name: Upload the benchmark results
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@release/2.9
         if: inputs.build-environment != 'linux-s390x-binary-manywheel'
         with:
           benchmark-results-dir: test/test-reports
@@ -476,7 +478,7 @@ jobs:
           workflow_attempt: ${{github.run_attempt}}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always() && steps.check_container_runner.outputs.IN_CONTAINER_RUNNER == 'false'
 
       # NB: We are currently having an intermittent GPU-related issue on G5 runners with

.github/workflows/_mac-build.yml

@@ -67,11 +67,11 @@ jobs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
     steps:
       - name: Clean up disk space before running MacOS workflow
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.9
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
 
       - name: Set xcode version
         env:
@@ -82,7 +82,7 @@ jobs:
           fi
 
       - name: Setup Python
-        uses: pytorch/test-infra/.github/actions/setup-python@main
+        uses: pytorch/test-infra/.github/actions/setup-python@release/2.9
         with:
           python-version: ${{ inputs.python-version }}
           pip-requirements-file: .github/requirements/pip-requirements-macOS.txt
@@ -188,4 +188,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.9

.github/workflows/_mac-test.yml

@@ -105,11 +105,11 @@ jobs:
           done
 
       - name: Clean up disk space before running MacOS workflow
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.9
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
 
       - name: Get workflow job id
         id: get-job-id
@@ -119,7 +119,7 @@ jobs:
           github-token: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Setup Python
-        uses: pytorch/test-infra/.github/actions/setup-python@main
+        uses: pytorch/test-infra/.github/actions/setup-python@release/2.9
         with:
           python-version: ${{ inputs.python-version }}
           pip-requirements-file: .github/requirements/pip-requirements-macOS.txt
@@ -257,7 +257,7 @@ jobs:
           file-suffix: ${{ github.job }}-${{ matrix.config }}-${{ matrix.shard }}-${{ matrix.num_shards }}-${{ matrix.runner }}_${{ steps.get-job-id.outputs.job-id }}
 
       - name: Upload the benchmark results
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@release/2.9
         with:
           benchmark-results-dir: test/test-reports
           dry-run: false
@@ -287,4 +287,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.9

.github/workflows/_rocm-test.yml

@@ -81,7 +81,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 
@@ -113,12 +113,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -330,7 +330,7 @@ jobs:
           aws-region: us-east-1
 
       - name: Upload the benchmark results
-        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@release/2.9
         with:
           benchmark-results-dir: test/test-reports
           dry-run: false

.github/workflows/_runner-determinator.yml

@@ -59,7 +59,7 @@ jobs:
       PR_NUMBER: ${{ github.event.pull_request.number }}
     steps:
       # - name: Checkout PyTorch
-      #   uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+      #   uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
       #   with:
       #     fetch-depth: 1
       #     submodules: true

.github/workflows/_win-build.yml

@@ -85,10 +85,10 @@ jobs:
           git config --global core.fsmonitor false
 
       - name: Clean up leftover processes on non-ephemeral Windows runner
-        uses: pytorch/test-infra/.github/actions/cleanup-runner@main
+        uses: pytorch/test-infra/.github/actions/cleanup-runner@release/2.9
 
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -103,7 +103,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 
@@ -151,7 +151,7 @@ jobs:
           BUILD_WHEEL: 1
           MAX_JOBS: 8
           CUDA_VERSION: ${{ inputs.cuda-version }}
-          PYTHON_VERSION: "3.9"
+          PYTHON_VERSION: "3.10"
           SCCACHE_BUCKET: "ossci-compiler-cache"
           SCCACHE_S3_KEY_PREFIX: ${{ github.workflow }}
           SCCACHE_REGION: us-east-1

.github/workflows/_win-test.yml

@@ -78,10 +78,10 @@ jobs:
           git config --global core.fsmonitor false
 
       - name: Clean up leftover processes on non-ephemeral Windows runner
-        uses: pytorch/test-infra/.github/actions/cleanup-runner@main
+        uses: pytorch/test-infra/.github/actions/cleanup-runner@release/2.9
 
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -97,7 +97,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 
@@ -184,7 +184,7 @@ jobs:
         env:
           USE_CUDA: ${{ inputs.cuda-version != 'cpu' && '1' || '0' }}
           INSTALL_WINDOWS_SDK: 1
-          PYTHON_VERSION: 3.9
+          PYTHON_VERSION: "3.10"
           CONTINUE_THROUGH_ERROR: ${{ steps.keep-going.outputs.keep-going }}
           VERBOSE_TEST_LOGS: ${{ steps.keep-going.outputs.ci-verbose-test-logs }}
           TEST_SHOWLOCALS: ${{ steps.keep-going.outputs.ci-test-showlocals }}

.github/workflows/_xpu-test.yml

@@ -77,7 +77,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
 
       - name: Setup XPU
         uses: ./.github/actions/setup-xpu
@@ -95,7 +95,7 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
@@ -109,7 +109,7 @@ jobs:
           echo "docker pull ghcr.io/pytorch/ci-image:${tag/:/-}"
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 

.github/workflows/build-almalinux-images.yml

@@ -39,7 +39,7 @@ jobs:
         tag: ["cuda12.6", "cuda12.8", "cuda12.9", "cuda13.0", "rocm6.3", "rocm6.4", "cpu"]
     steps:
       - name: Build docker image
-        uses: pytorch/pytorch/.github/actions/binary-docker-build@main
+        uses: pytorch/pytorch/.github/actions/binary-docker-build@release/2.9
         with:
           docker-image-name: almalinux-builder
           custom-tag-prefix: ${{matrix.tag}}

.github/workflows/build-libtorch-images.yml

@@ -32,7 +32,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -58,7 +58,7 @@ jobs:
         ]
     steps:
       - name: Build docker image
-        uses: pytorch/pytorch/.github/actions/binary-docker-build@main
+        uses: pytorch/pytorch/.github/actions/binary-docker-build@release/2.9
         with:
           docker-image-name: libtorch-cxx11-builder
           custom-tag-prefix: ${{ matrix.tag }}

.github/workflows/build-manywheel-images-s390x.yml

@@ -25,7 +25,7 @@ jobs:
     runs-on: linux.s390x
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           submodules: false
           no-sudo: true

.github/workflows/build-manywheel-images.yml

@@ -32,7 +32,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -47,12 +47,11 @@ jobs:
       matrix:
         include: [
           { name: "manylinux2_28-builder",          tag: "cuda13.0",         runner: "linux.9xlarge.ephemeral" },
-          { name: "manylinux2_28-builder",          tag: "cuda12.9",         runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "cuda12.8",          runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "cuda12.6",          runner: "linux.9xlarge.ephemeral" },
           { name: "manylinuxaarch64-builder",       tag: "cuda13.0",          runner: "linux.arm64.2xlarge.ephemeral" },
-          { name: "manylinuxaarch64-builder",       tag: "cuda12.9",          runner: "linux.arm64.2xlarge.ephemeral" },
           { name: "manylinuxaarch64-builder",       tag: "cuda12.8",          runner: "linux.arm64.2xlarge.ephemeral" },
+          { name: "manylinuxaarch64-builder",       tag: "cuda12.6",          runner: "linux.arm64.2xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "rocm6.3",           runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "rocm6.4",           runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "cpu",               runner: "linux.9xlarge.ephemeral" },
@@ -64,7 +63,7 @@ jobs:
     name: ${{ matrix.name }}:${{ matrix.tag }}
     steps:
       - name: Build docker image
-        uses: pytorch/pytorch/.github/actions/binary-docker-build@main
+        uses: pytorch/pytorch/.github/actions/binary-docker-build@release/2.9
         with:
           docker-image-name: ${{ matrix.name }}
           custom-tag-prefix: ${{ matrix.tag }}

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

@@ -3,7 +3,7 @@ name: Build Triton wheels
 on:
   push:
     branches:
-      - main
+      - release/2.9
     tags:
       # NOTE: Binary build pipelines should only get triggered on release candidate builds
       # Release candidate tags look like: v1.11.0-rc1
@@ -36,7 +36,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -74,12 +74,12 @@ jobs:
       PLATFORM: 'manylinux_2_28_x86_64'
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           submodules: false
 
@@ -87,7 +87,7 @@ jobs:
         uses: ./.github/actions/setup-linux
 
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
@@ -184,7 +184,7 @@ jobs:
           path: ${{ runner.temp }}/artifacts/wheelhouse/*
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always()
 
   build-wheel-win:
@@ -217,7 +217,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}

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

@@ -0,0 +1,248 @@
+name: Build vLLM wheels
+
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - .github/workflows/build-vllm-wheel.yml
+      - .github/ci_commit_pins/vllm.txt
+  workflow_dispatch:
+  pull_request:
+    paths:
+      - .github/workflows/build-vllm-wheel.yml
+      - .github/ci_commit_pins/vllm.txt
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  cancel-in-progress: true
+
+jobs:
+  build-wheel:
+    if: github.repository_owner == 'pytorch'
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: [ '3.12' ]
+        # TODO (huydhn): Add cu130 https://github.com/pytorch/pytorch/pull/162000#issuecomment-3261541554
+        device: [ 'cu128', 'cu129' ]
+        runner: [ 'linux.12xlarge.memory' ]
+        include:
+          - device: cu128
+            manylinux-image: 'pytorch/manylinux2_28-builder:cuda12.8'
+          - device: cu129
+            manylinux-image: 'pytorch/manylinux2_28-builder:cuda12.9'
+    name: "Build ${{ matrix.device }} vLLM wheel"
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 480
+    env:
+      PY_VERS: ${{ matrix.python-version }}
+      MANYLINUX_IMAGE: ${{ matrix.manylinux-image }}
+      PLATFORM: 'manylinux_2_28_x86_64'
+      BUILD_DEVICE: ${{ matrix.device }}
+    steps:
+      - name: Setup SSH (Click me for login details)
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
+        with:
+          github-secret: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
+        with:
+          submodules: false
+
+      - name: Setup Linux
+        uses: ./.github/actions/setup-linux
+
+      - name: Get latest PyTorch nightly
+        shell: bash
+        run: |
+          set -eux
+
+          # Keep PyTorch nightly wheel here so that we can install it later during
+          # vLLM build process
+          mkdir -p "${RUNNER_TEMP}/artifacts/"
+
+          container_name=$(docker run \
+            --tty \
+            --detach \
+            -e PLATFORM \
+            -v "${GITHUB_WORKSPACE}:/pytorch" \
+            -v "${RUNNER_TEMP}/artifacts:/artifacts" \
+            -w /artifacts/ \
+            "${MANYLINUX_IMAGE}"
+          )
+
+          # Determine python executable for given version (copied from build-triton-wheel)
+          case $PY_VERS in
+          3.10)
+            PYTHON_EXECUTABLE=/opt/python/cp310-cp310/bin/python
+            ;;
+          3.11)
+            PYTHON_EXECUTABLE=/opt/python/cp311-cp311/bin/python
+            ;;
+          3.12)
+            PYTHON_EXECUTABLE=/opt/python/cp312-cp312/bin/python
+            ;;
+          3.13)
+            PYTHON_EXECUTABLE=/opt/python/cp313-cp313/bin/python
+            ;;
+          3.13t)
+            PYTHON_EXECUTABLE=/opt/python/cp313-cp313t/bin/python
+            ;;
+          3.14)
+            PYTHON_EXECUTABLE=/opt/python/cp314-cp314/bin/python
+            ;;
+          3.14t)
+            PYTHON_EXECUTABLE=/opt/python/cp314-cp314t/bin/python
+            ;;
+          *)
+            echo "Unsupported python version ${PY_VERS}"
+            exit 1
+            ;;
+          esac
+
+          docker exec -t "${container_name}" "${PYTHON_EXECUTABLE}" -mpip install \
+            --pre torch torchvision torchaudio \
+            --index-url "https://download.pytorch.org/whl/nightly/${BUILD_DEVICE}"
+
+          # I wonder if there is a command to both download and install the wheels
+          # in one go
+          docker exec -t "${container_name}" "${PYTHON_EXECUTABLE}" -mpip download \
+            --pre torch torchvision torchaudio \
+            --index-url "https://download.pytorch.org/whl/nightly/${BUILD_DEVICE}"
+
+          # Save this for later
+          echo "PYTHON_EXECUTABLE=${PYTHON_EXECUTABLE}" >> "$GITHUB_ENV"
+          echo "container_name=${container_name}" >> "$GITHUB_ENV"
+
+      - name: Build vLLM wheel
+        uses: ./.github/actions/build-external-packages
+        with:
+          build-targets: vllm
+          docker-image: ${{ env.MANYLINUX_IMAGE }}
+          cuda-arch-list: '8.0;8.9;9.0;10.0;12.0'
+          torch-wheel-dir: ${{ runner.temp }}/artifacts
+          output-dir: ${{ runner.temp }}/artifacts/externals
+
+      - name: Prepare vLLM wheel
+        shell: bash
+        run: |
+          set -eux
+
+          # Get these wheels ready, the vllm renaming logic is copied from its .buildkite/scripts/upload-wheels.sh
+          docker exec -t "${container_name}" bash -c "
+            set -eux
+
+            nightly=\$(unzip -p torch-* '**/METADATA' | grep '^Version: ' | cut -d' ' -f2 | cut -d'.' -f4)
+
+            pushd externals/vllm/wheels
+            for package in xformers flashinfer-python vllm; do
+              pushd \$package
+              auditwheel repair --plat \$PLATFORM *.whl \
+                --exclude libc10* --exclude libtorch* --exclude libcu* --exclude libnv*
+              repair_wheel=\$(find wheelhouse -name *\${PLATFORM}*)
+              repair_wheel=\$(basename \${repair_wheel})
+              popd
+
+              cp \${package}/wheelhouse/\${repair_wheel} .
+              version=\$(unzip -p \$repair_wheel '**/METADATA' | grep '^Version: ' | cut -d' ' -f2)
+
+              if [[ \$package == vllm ]]; then
+                new_wheel=\${repair_wheel/\$version/1.0.0.\$nightly}
+              else
+                major_version=\$(echo \$version | tr '.+' '.' | cut -d'.' -f1-3)
+                new_wheel=\${repair_wheel/\$version/\$major_version.\$nightly}
+              fi
+
+              mv -- \$repair_wheel \$new_wheel
+              rm -rf \$package
+            done
+            popd
+          "
+
+          docker exec -t "${container_name}" chown -R 1000:1000 /artifacts
+
+      - uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874 # v4.4.0
+        with:
+          name: vllm-wheel-${{ matrix.device }}-${{ matrix.python-version }}-${{ env.PLATFORM }}
+          if-no-files-found: error
+          path: ${{ runner.temp }}/artifacts/externals/vllm/wheels/*.whl
+
+      - name: Teardown Linux
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
+        if: always()
+
+  # Copied from build-triton-wheel workflow (mostly)
+  upload-wheel:
+    name: "Upload ${{ matrix.device }} vLLM wheel"
+    needs:
+      - build-wheel
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        device: [ 'cu128', 'cu129' ]
+    env:
+      BUILD_DEVICE: ${{ matrix.device }}
+    permissions:
+      id-token: write
+      contents: read
+    container:
+      image: continuumio/miniconda3:4.12.0
+    environment: ${{ (github.event_name == 'push' && github.event.ref == 'refs/heads/main') && 'nightly-wheel-upload' || '' }}
+    steps:
+      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+
+      - name: Configure AWS credentials(PyTorch account) for main
+        if: ${{ github.event_name == 'push' && github.event.ref == 'refs/heads/main' }}
+        uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
+        with:
+          role-to-assume: arn:aws:iam::749337293305:role/gha_workflow_nightly_build_wheels
+          aws-region: us-east-1
+
+      - name: Configure AWS credentials(PyTorch account) for RC builds
+        if: ${{ github.event_name == 'push' &&  (startsWith(github.event.ref, 'refs/tags/') && !startsWith(github.event.ref, 'refs/tags/ciflow/')) }}
+        uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
+        with:
+          role-to-assume: arn:aws:iam::749337293305:role/gha_workflow_test_build_wheels
+          aws-region: us-east-1
+
+      - name: Download Build Artifacts
+        uses: actions/download-artifact@65a9edc5881444af0b9093a5e628f2fe47ea3b2e # v4.1.7
+        with:
+          # Download all available artifacts
+          path: ${{ runner.temp }}/artifacts-all
+
+      - name: Select Wheel Artifacts
+        shell: bash
+        run: |
+          set -eux
+          mkdir -p "${RUNNER_TEMP}/artifacts/"
+          mv "${RUNNER_TEMP}"/artifacts-all/vllm-wheel-"${BUILD_DEVICE}"-*/* "${RUNNER_TEMP}/artifacts/"
+
+      - name: Set DRY_RUN (only for tagged pushes)
+        if: ${{ github.event_name == 'push' && (github.event.ref == 'refs/heads/main' || startsWith(github.event.ref, 'refs/tags/v')) }}
+        shell: bash
+        run: |
+          echo "DRY_RUN=disabled" >> "$GITHUB_ENV"
+
+      - name: Set UPLOAD_CHANNEL (only for tagged pushes)
+        if: ${{ github.event_name == 'push' && startsWith(github.event.ref, 'refs/tags/v') }}
+        shell: bash
+        run: |
+          set -ex
+
+          if [[ "${GITHUB_REF_NAME}" = *-rc[0-9]* ]]; then
+            echo "UPLOAD_CHANNEL=test" >> "$GITHUB_ENV"
+          fi
+
+      - name: Upload binaries
+        env:
+          PACKAGE_TYPE: wheel
+          UPLOAD_SUBFOLDER: ${{ env.BUILD_DEVICE }}
+          PKG_DIR: ${{ runner.temp }}/artifacts
+        shell: bash
+        run: |
+          set -ex
+          bash .circleci/scripts/binary_upload.sh

.github/workflows/check-labels.yml

@@ -38,7 +38,7 @@ jobs:
     runs-on: linux.24_04.4x
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           submodules: false
           fetch-depth: 1

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

@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           submodules: false
           fetch-depth: 1

.github/workflows/create_release.yml

@@ -19,7 +19,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

.github/workflows/docker-builds.yml

@@ -33,7 +33,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -70,7 +70,7 @@ jobs:
           pytorch-linux-jammy-py3-clang18-asan,
           pytorch-linux-jammy-py3-clang12-onnx,
           pytorch-linux-jammy-linter,
-          pytorch-linux-jammy-cuda12.8-cudnn9-py3.9-linter,
+          pytorch-linux-jammy-cuda12.8-cudnn9-py3.10-linter,
           # Executorch pin needs update
           # pytorch-linux-jammy-py3-clang12-executorch,
           pytorch-linux-jammy-py3.12-triton-cpu,
@@ -96,21 +96,21 @@ jobs:
       # [see note: pytorch repo ref]
       # deep clone (fetch-depth 0) required for git merge-base
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
 
       - name: Build docker image
         id: build-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-image-name: ci-image:${{ matrix.docker-image-name }}
           always-rebuild: true
           push: true
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.build-docker-image.outputs.docker-image }}
 
@@ -141,5 +141,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always()

.github/workflows/docker-cache-mi300.yml

@@ -20,7 +20,7 @@ jobs:
     runs-on: rocm-docker
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           no-sudo: true
 
@@ -39,13 +39,13 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-image-name: ci-image:pytorch-linux-jammy-rocm-n-py3
           push: false
 
       - name: Pull docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 

.github/workflows/docker-release.yml

@@ -37,7 +37,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -52,7 +52,7 @@ jobs:
       matrix: ${{ steps.generate-matrix.outputs.matrix }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.9
         with:
           fetch-depth: 1
           submodules: true
@@ -82,7 +82,7 @@ jobs:
       CUDNN_VERSION: ${{ matrix.cudnn_version }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
       # [see note: pytorch repo ref]
@@ -164,12 +164,12 @@ jobs:
           fi
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.9
         if: always()
 
   validate:
     needs: build
-    uses: pytorch/test-infra/.github/workflows/validate-docker-images.yml@main
+    uses: pytorch/test-infra/.github/workflows/validate-docker-images.yml@release/2.9
     with:
-      channel: nightly
+      channel: test
       ref: main

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

@@ -41,7 +41,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -112,7 +112,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_10-cuda-aarch64-12_9-build:
+  manywheel-py3_10-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -121,39 +121,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_10-cuda-aarch64-12_9
+      build_name: manywheel-py3_10-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_10-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_10-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_10-cuda-aarch64-12_9-build
+    needs: manywheel-py3_10-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.10"
-      build_name: manywheel-py3_10-cuda-aarch64-12_9
+      build_name: manywheel-py3_10-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_10-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.10"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_10-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_10-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_10-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.10"
+      build_name: manywheel-py3_10-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -178,7 +224,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -269,7 +315,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_11-cuda-aarch64-12_9-build:
+  manywheel-py3_11-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -278,39 +324,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.11"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_11-cuda-aarch64-12_9
+      build_name: manywheel-py3_11-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_11-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_11-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_11-cuda-aarch64-12_9-build
+    needs: manywheel-py3_11-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.11"
-      build_name: manywheel-py3_11-cuda-aarch64-12_9
+      build_name: manywheel-py3_11-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_11-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.11"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_11-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_11-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_11-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.11"
+      build_name: manywheel-py3_11-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -335,7 +427,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -426,7 +518,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_12-cuda-aarch64-12_9-build:
+  manywheel-py3_12-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -435,39 +527,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.12"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_12-cuda-aarch64-12_9
+      build_name: manywheel-py3_12-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_12-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_12-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_12-cuda-aarch64-12_9-build
+    needs: manywheel-py3_12-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.12"
-      build_name: manywheel-py3_12-cuda-aarch64-12_9
+      build_name: manywheel-py3_12-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_12-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.12"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_12-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_12-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_12-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.12"
+      build_name: manywheel-py3_12-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -492,7 +630,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -583,7 +721,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_13-cuda-aarch64-12_9-build:
+  manywheel-py3_13-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -592,39 +730,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.13"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_13-cuda-aarch64-12_9
+      build_name: manywheel-py3_13-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_13-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_13-cuda-aarch64-12_9-build
+    needs: manywheel-py3_13-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.13"
-      build_name: manywheel-py3_13-cuda-aarch64-12_9
+      build_name: manywheel-py3_13-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_13-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.13"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_13-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_13-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_13-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.13"
+      build_name: manywheel-py3_13-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -649,7 +833,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -740,7 +924,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_13t-cuda-aarch64-12_9-build:
+  manywheel-py3_13t-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -749,39 +933,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.13t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_13t-cuda-aarch64-12_9
+      build_name: manywheel-py3_13t-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_13t-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_13t-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_13t-cuda-aarch64-12_9-build
+    needs: manywheel-py3_13t-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.13t"
-      build_name: manywheel-py3_13t-cuda-aarch64-12_9
+      build_name: manywheel-py3_13t-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_13t-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.13t"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_13t-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_13t-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_13t-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.13t"
+      build_name: manywheel-py3_13t-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -806,7 +1036,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13t-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -897,7 +1127,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_14-cuda-aarch64-12_9-build:
+  manywheel-py3_14-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -906,39 +1136,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.14"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_14-cuda-aarch64-12_9
+      build_name: manywheel-py3_14-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_14-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_14-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_14-cuda-aarch64-12_9-build
+    needs: manywheel-py3_14-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.14"
-      build_name: manywheel-py3_14-cuda-aarch64-12_9
+      build_name: manywheel-py3_14-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_14-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.14"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_14-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_14-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_14-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.14"
+      build_name: manywheel-py3_14-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -963,7 +1239,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -1054,7 +1330,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  manywheel-py3_14t-cuda-aarch64-12_9-build:
+  manywheel-py3_14t-cuda-aarch64-12_6-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -1063,39 +1339,85 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.14t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.arm64.m7g.4xlarge.ephemeral
       ALPINE_IMAGE: "arm64v8/alpine"
-      build_name: manywheel-py3_14t-cuda-aarch64-12_9
+      build_name: manywheel-py3_14t-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.9.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.4.1.4; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.10.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.5.82; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.10.65; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.9.79; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.9.86; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.14.1.1; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.6.77; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.6.80; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.6.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.0.4; platform_system == 'Linux' | nvidia-curand-cu12==10.3.7.77; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.1.2; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.4.2; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.6.77; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.6.85; platform_system == 'Linux' | nvidia-cufile-cu12==1.11.1.6; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_14t-cuda-aarch64-12_9-upload:  # Uploading
+  manywheel-py3_14t-cuda-aarch64-12_6-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: manywheel-py3_14t-cuda-aarch64-12_9-build
+    needs: manywheel-py3_14t-cuda-aarch64-12_6-build
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9-aarch64"
+      DESIRED_CUDA: cu126
+      GPU_ARCH_VERSION: "12.6-aarch64"
       GPU_ARCH_TYPE: cuda-aarch64
       DOCKER_IMAGE: manylinuxaarch64-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.14t"
-      build_name: manywheel-py3_14t-cuda-aarch64-12_9
+      build_name: manywheel-py3_14t-cuda-aarch64-12_6
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml
+
+  manywheel-py3_14t-cuda-aarch64-12_8-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.14t"
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      ALPINE_IMAGE: "arm64v8/alpine"
+      build_name: manywheel-py3_14t-cuda-aarch64-12_8
+      build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      timeout-minutes: 420
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  manywheel-py3_14t-cuda-aarch64-12_8-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: manywheel-py3_14t-cuda-aarch64-12_8-build
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: manywheel
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: cu128
+      GPU_ARCH_VERSION: "12.8-aarch64"
+      GPU_ARCH_TYPE: cuda-aarch64
+      DOCKER_IMAGE: manylinuxaarch64-builder
+      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DESIRED_PYTHON: "3.14t"
+      build_name: manywheel-py3_14t-cuda-aarch64-12_8
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -1120,7 +1442,7 @@ jobs:
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14t-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu13==9.12.0.46; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand==10.4.0.35; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile==1.15.0.42; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
       timeout-minutes: 420
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}

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

@@ -41,7 +41,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -248,7 +248,7 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  libtorch-cuda12_9-shared-with-deps-release-build:
+  libtorch-cuda13_0-shared-with-deps-release-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -257,22 +257,22 @@ jobs:
       PACKAGE_TYPE: libtorch
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
+      DESIRED_CUDA: cu130
+      GPU_ARCH_VERSION: "13.0"
       GPU_ARCH_TYPE: cuda
       DOCKER_IMAGE: libtorch-cxx11-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       LIBTORCH_CONFIG: release
       LIBTORCH_VARIANT: shared-with-deps
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: libtorch-cuda12_9-shared-with-deps-release
+      build_name: libtorch-cuda13_0-shared-with-deps-release
       build_environment: linux-binary-libtorch
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  libtorch-cuda12_9-shared-with-deps-release-test:  # Testing
+  libtorch-cuda13_0-shared-with-deps-release-test:  # Testing
     if: ${{ github.repository_owner == 'pytorch' }}
     needs:
-      - libtorch-cuda12_9-shared-with-deps-release-build
+      - libtorch-cuda13_0-shared-with-deps-release-build
       - get-label-type
     uses: ./.github/workflows/_binary-test-linux.yml
     with:
@@ -280,38 +280,38 @@ jobs:
       PACKAGE_TYPE: libtorch
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
+      DESIRED_CUDA: cu130
+      GPU_ARCH_VERSION: "13.0"
       GPU_ARCH_TYPE: cuda
       DOCKER_IMAGE: libtorch-cxx11-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       LIBTORCH_CONFIG: release
       LIBTORCH_VARIANT: shared-with-deps
-      build_name: libtorch-cuda12_9-shared-with-deps-release
+      build_name: libtorch-cuda13_0-shared-with-deps-release
       build_environment: linux-binary-libtorch
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.g4dn.4xlarge.nvidia.gpu # 12.8+ builds need sm_70+ runner
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  libtorch-cuda12_9-shared-with-deps-release-upload:  # Uploading
+  libtorch-cuda13_0-shared-with-deps-release-upload:  # Uploading
     if: ${{ github.repository_owner == 'pytorch' }}
     permissions:
       id-token: write
       contents: read
-    needs: libtorch-cuda12_9-shared-with-deps-release-test
+    needs: libtorch-cuda13_0-shared-with-deps-release-test
     with:
       PYTORCH_ROOT: /pytorch
       PACKAGE_TYPE: libtorch
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
+      DESIRED_CUDA: cu130
+      GPU_ARCH_VERSION: "13.0"
       GPU_ARCH_TYPE: cuda
       DOCKER_IMAGE: libtorch-cxx11-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.9
+      DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       LIBTORCH_CONFIG: release
       LIBTORCH_VARIANT: shared-with-deps
-      build_name: libtorch-cuda12_9-shared-with-deps-release
+      build_name: libtorch-cuda13_0-shared-with-deps-release
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
@@ -333,6 +333,7 @@ jobs:
       LIBTORCH_CONFIG: release
       LIBTORCH_VARIANT: shared-with-deps
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
       build_name: libtorch-rocm6_3-shared-with-deps-release
       build_environment: linux-binary-libtorch
     secrets:
@@ -342,7 +343,7 @@ jobs:
     needs:
       - libtorch-rocm6_3-shared-with-deps-release-build
       - get-label-type
-    runs-on: linux.rocm.gpu.gfx942.1
+    runs-on: linux.rocm.gpu.mi250
     timeout-minutes: 240
     env:
       PYTORCH_ROOT: /pytorch
@@ -368,7 +369,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -390,7 +390,7 @@ jobs:
           role-duration-seconds: 18000
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: libtorch-cxx11-builder
@@ -398,7 +398,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary
@@ -447,6 +447,7 @@ jobs:
       LIBTORCH_CONFIG: release
       LIBTORCH_VARIANT: shared-with-deps
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
       build_name: libtorch-rocm6_4-shared-with-deps-release
       build_environment: linux-binary-libtorch
     secrets:
@@ -456,7 +457,7 @@ jobs:
     needs:
       - libtorch-rocm6_4-shared-with-deps-release-build
       - get-label-type
-    runs-on: linux.rocm.gpu.gfx942.1
+    runs-on: linux.rocm.gpu.mi250
     timeout-minutes: 240
     env:
       PYTORCH_ROOT: /pytorch
@@ -482,7 +483,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -504,7 +504,7 @@ jobs:
           role-duration-seconds: 18000
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: libtorch-cxx11-builder
@@ -512,7 +512,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary

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

@@ -36,7 +36,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

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

@@ -36,7 +36,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -60,7 +60,7 @@ jobs:
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       build_name: manywheel-py3_12-cuda12_8
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.20; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_12-cuda12_8-test:  # Testing

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

@@ -10,7 +10,9 @@ on:
     branches:
       - main
     tags:
-      - 'ciflow/rocm-mi300/*'
+      - 'ciflow/binaries/*'
+      - 'ciflow/binaries_wheel/*'
+      - 'ciflow/rocm/*'
   workflow_dispatch:
 
 permissions:
@@ -36,13 +38,13 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
       curr_branch: ${{ github.head_ref || github.ref_name }}
       curr_ref_type: ${{ github.ref_type }}
-  manywheel-py3_9-rocm6_4-build:
+  manywheel-py3_10-rocm6_4-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -56,18 +58,19 @@ jobs:
       GPU_ARCH_TYPE: rocm
       DOCKER_IMAGE: manylinux2_28-builder
       DOCKER_IMAGE_TAG_PREFIX: rocm6.4
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_9-rocm6_4
+      timeout-minutes: 300
+      build_name: manywheel-py3_10-rocm6_4
       build_environment: linux-binary-manywheel-rocm
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_9-rocm6_4-test:  # Testing
+  manywheel-py3_10-rocm6_4-test:  # Testing
     if: ${{ github.repository_owner == 'pytorch' }}
     needs:
-      - manywheel-py3_9-rocm6_4-build
+      - manywheel-py3_10-rocm6_4-build
       - get-label-type
-    runs-on: linux.rocm.gpu.gfx942.1
+    runs-on: linux.rocm.gpu.mi250
     timeout-minutes: 240
     env:
       PYTORCH_ROOT: /pytorch
@@ -80,19 +83,18 @@ jobs:
       SKIP_ALL_TESTS: 1
       DOCKER_IMAGE: manylinux2_28-builder
       DOCKER_IMAGE_TAG_PREFIX: rocm6.4
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Setup ROCm
         uses: ./.github/actions/setup-rocm
       - uses: actions/download-artifact@v4.1.7
         name: Download Build Artifacts
         with:
-          name: manywheel-py3_9-rocm6_4
+          name: manywheel-py3_10-rocm6_4
           path: "${{ runner.temp }}/artifacts/"
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -114,7 +116,7 @@ jobs:
           role-duration-seconds: 18000
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.9
         with:
           docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
           docker-image-name: manylinux2_28-builder
@@ -122,7 +124,7 @@ jobs:
           docker-build-dir: .ci/docker
           working-directory: pytorch
       - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.9
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
       - name: Test Pytorch binary

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

@@ -41,7 +41,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

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

@@ -46,7 +46,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -67,15 +67,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false

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

@@ -63,15 +63,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -208,15 +202,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -353,15 +341,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -498,15 +480,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -643,15 +619,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -788,15 +758,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -933,15 +897,9 @@ jobs:
           chmod +x "${RUNNER_TEMP}/conda.sh"
           /bin/bash "${RUNNER_TEMP}/conda.sh" -b -p "${RUNNER_TEMP}/anaconda"
           echo "${RUNNER_TEMP}/anaconda/bin" >> "${GITHUB_PATH}"
-          if [ -d "/Applications/Xcode_14.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_14.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
-            echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
-          fi
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false

.github/workflows/generated-windows-arm64-binary-libtorch-debug-nightly.yml

@@ -41,7 +41,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -64,7 +64,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Populate binary env
         shell: cmd
@@ -141,7 +141,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Populate binary env
         shell: cmd
@@ -201,7 +201,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cpu-shared-with-deps-debug
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/generated-windows-arm64-binary-libtorch-release-nightly.yml

@@ -41,7 +41,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -64,7 +64,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Populate binary env
         shell: cmd
@@ -141,7 +141,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Populate binary env
         shell: cmd
@@ -201,7 +201,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cpu-shared-with-deps-release
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}

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

@@ -41,7 +41,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

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

@@ -28,7 +28,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -51,7 +51,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -77,7 +77,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -109,7 +109,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -166,7 +165,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -183,7 +182,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -215,7 +214,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false

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

@@ -35,7 +35,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -58,7 +58,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -84,7 +84,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -116,7 +116,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -173,7 +172,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -190,7 +189,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -222,7 +221,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -283,7 +281,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cpu-shared-with-deps-debug
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -306,7 +304,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -332,7 +330,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -364,7 +362,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -422,7 +419,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -439,7 +436,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -471,7 +468,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -533,7 +529,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cuda12_6-shared-with-deps-debug
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -556,7 +552,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -582,7 +578,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -614,7 +610,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -672,7 +667,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -689,7 +684,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -721,7 +716,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -783,261 +777,11 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cuda12_8-shared-with-deps-debug
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
-  libtorch-cuda12_9-shared-with-deps-debug-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs: get-label-type
-    runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
-    timeout-minutes: 360
-    env:
-      PYTORCH_ROOT: ${{ github.workspace }}/pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
-      GPU_ARCH_TYPE: cuda
-      SKIP_ALL_TESTS: 1
-      LIBTORCH_CONFIG: debug
-      LIBTORCH_VARIANT: shared-with-deps
-      # This is a dummy value for libtorch to work correctly with our batch scripts
-      # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
-    steps:
-      # NOTE: These environment variables are put here so that they can be applied on every job equally
-      #       They are also here because setting them at a workflow level doesn't give us access to the
-      #       runner.temp variable, which we need.
-      - name: Populate binary env
-        shell: bash
-        run: |
-          echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
-          echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
-          echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
-      - name: Display EC2 information
-        shell: bash
-        run: |
-          set -euo pipefail
-          function get_ec2_metadata() {
-            # Pulled from instance metadata endpoint for EC2
-            # see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
-            category=$1
-            curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
-          }
-          echo "ami-id: $(get_ec2_metadata ami-id)"
-          echo "instance-id: $(get_ec2_metadata instance-id)"
-          echo "instance-type: $(get_ec2_metadata instance-type)"
-          echo "system info $(uname -a)"
-      - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
-        continue-on-error: true
-        with:
-          github-secret: ${{ secrets.GITHUB_TOKEN }}
-      - name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
-        shell: bash
-        run: |
-          git config --global core.longpaths true
-          git config --global core.symlinks true
-
-          # https://git-scm.com/docs/git-fsmonitor--daemon.  The daemon could lock
-          # the directory on Windows and prevent GHA from checking out as reported
-          # in https://github.com/actions/checkout/issues/1018
-          git config --global core.fsmonitor false
-      # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
-      - name: Enable long paths on Windows
-        shell: powershell
-        run: |
-          Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
-      # Since it's just a defensive command, the workflow should continue even the command fails. This step can be
-      # removed once Windows Defender is removed from the AMI
-      - name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
-        continue-on-error: true
-        shell: powershell
-        run: |
-          Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
-          # Let's both exclude the path and disable Windows Defender completely just to be sure
-          # that it doesn't interfere
-          Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
-      - name: Checkout PyTorch
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-          submodules: recursive
-          path: pytorch
-          show-progress: false
-      - name: Clean PyTorch checkout
-        run: |
-          # Remove any artifacts from the previous checkouts
-          git clean -fxd
-        working-directory: pytorch
-      - name: Populate binary env
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
-      - name: Build PyTorch binary
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
-      - uses: actions/upload-artifact@v4.4.0
-        if: always()
-        with:
-          name: libtorch-cuda12_9-shared-with-deps-debug
-          retention-days: 14
-          if-no-files-found: error
-          path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
-      - name: Wait until all sessions have drained
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        timeout-minutes: 120
-        run: |
-          .github\scripts\wait_for_ssh_to_drain.ps1
-      - name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        run: |
-          .github\scripts\kill_active_ssh_sessions.ps1
-
-  libtorch-cuda12_9-shared-with-deps-debug-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - libtorch-cuda12_9-shared-with-deps-debug-build
-      - get-label-type
-    runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
-    timeout-minutes: 360
-    env:
-      PYTORCH_ROOT: ${{ github.workspace }}/pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
-      GPU_ARCH_TYPE: cuda
-      SKIP_ALL_TESTS: 1
-      LIBTORCH_CONFIG: debug
-      LIBTORCH_VARIANT: shared-with-deps
-      # This is a dummy value for libtorch to work correctly with our batch scripts
-      # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
-    steps:
-      - name: Display EC2 information
-        shell: bash
-        run: |
-          set -euo pipefail
-          function get_ec2_metadata() {
-            # Pulled from instance metadata endpoint for EC2
-            # see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
-            category=$1
-            curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
-          }
-          echo "ami-id: $(get_ec2_metadata ami-id)"
-          echo "instance-id: $(get_ec2_metadata instance-id)"
-          echo "instance-type: $(get_ec2_metadata instance-type)"
-          echo "system info $(uname -a)"
-      - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
-        continue-on-error: true
-        with:
-          github-secret: ${{ secrets.GITHUB_TOKEN }}
-      - name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
-        shell: bash
-        run: |
-          git config --global core.longpaths true
-          git config --global core.symlinks true
-
-          # https://git-scm.com/docs/git-fsmonitor--daemon.  The daemon could lock
-          # the directory on Windows and prevent GHA from checking out as reported
-          # in https://github.com/actions/checkout/issues/1018
-          git config --global core.fsmonitor false
-      # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
-      - name: Enable long paths on Windows
-        shell: powershell
-        run: |
-          Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
-      # Since it's just a defensive command, the workflow should continue even the command fails. This step can be
-      # removed once Windows Defender is removed from the AMI
-      - name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
-        continue-on-error: true
-        shell: powershell
-        run: |
-          Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
-          # Let's both exclude the path and disable Windows Defender completely just to be sure
-          # that it doesn't interfere
-          Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
-      - name: Checkout PyTorch
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-          submodules: recursive
-          path: pytorch
-          show-progress: false
-      - name: Clean PyTorch checkout
-        run: |
-          # Remove any artifacts from the previous checkouts
-          git clean -fxd
-        working-directory: pytorch
-      # NOTE: These environment variables are put here so that they can be applied on every job equally
-      #       They are also here because setting them at a workflow level doesn't give us access to the
-      #       runner.temp variable, which we need.
-      - name: Populate binary env
-        shell: bash
-        run: |
-          echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
-          echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
-          echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
-      - uses: actions/download-artifact@v4.1.7
-        name: Download Build Artifacts
-        with:
-          name: libtorch-cuda12_9-shared-with-deps-debug
-          path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
-      - name: Populate binary env
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
-      - name: Test PyTorch binary
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_windows_test.sh"
-      - name: Wait until all sessions have drained
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        timeout-minutes: 120
-        run: |
-          .github\scripts\wait_for_ssh_to_drain.ps1
-      - name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        run: |
-          .github\scripts\kill_active_ssh_sessions.ps1
-  libtorch-cuda12_9-shared-with-deps-debug-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: libtorch-cuda12_9-shared-with-deps-debug-test
-    with:
-      PYTORCH_ROOT: ${{ github.workspace }}/pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
-      GPU_ARCH_TYPE: cuda
-      LIBTORCH_CONFIG: debug
-      LIBTORCH_VARIANT: shared-with-deps
-      # This is a dummy value for libtorch to work correctly with our batch scripts
-      # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
-      build_name: libtorch-cuda12_9-shared-with-deps-debug
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-    uses: ./.github/workflows/_binary-upload.yml
   libtorch-cuda13_0-shared-with-deps-debug-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     needs: get-label-type
@@ -1056,7 +800,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -1082,7 +826,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -1114,7 +858,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -1172,7 +915,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1189,7 +932,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -1221,7 +964,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -1283,7 +1025,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cuda13_0-shared-with-deps-debug
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}

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

@@ -28,7 +28,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -51,7 +51,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -77,7 +77,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -109,7 +109,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -166,7 +165,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -183,7 +182,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -215,7 +214,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false

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

@@ -35,7 +35,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
@@ -58,7 +58,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -84,7 +84,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -116,7 +116,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -173,7 +172,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -190,7 +189,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -222,7 +221,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -283,7 +281,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cpu-shared-with-deps-release
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -306,7 +304,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -332,7 +330,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -364,7 +362,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -422,7 +419,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -439,7 +436,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -471,7 +468,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -533,7 +529,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cuda12_6-shared-with-deps-release
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -556,7 +552,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -582,7 +578,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -614,7 +610,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -672,7 +667,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -689,7 +684,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -721,7 +716,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -783,261 +777,11 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cuda12_8-shared-with-deps-release
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
-  libtorch-cuda12_9-shared-with-deps-release-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs: get-label-type
-    runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.4xlarge"
-    timeout-minutes: 360
-    env:
-      PYTORCH_ROOT: ${{ github.workspace }}/pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
-      GPU_ARCH_TYPE: cuda
-      SKIP_ALL_TESTS: 1
-      LIBTORCH_CONFIG: release
-      LIBTORCH_VARIANT: shared-with-deps
-      # This is a dummy value for libtorch to work correctly with our batch scripts
-      # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
-    steps:
-      # NOTE: These environment variables are put here so that they can be applied on every job equally
-      #       They are also here because setting them at a workflow level doesn't give us access to the
-      #       runner.temp variable, which we need.
-      - name: Populate binary env
-        shell: bash
-        run: |
-          echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
-          echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
-          echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
-      - name: Display EC2 information
-        shell: bash
-        run: |
-          set -euo pipefail
-          function get_ec2_metadata() {
-            # Pulled from instance metadata endpoint for EC2
-            # see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
-            category=$1
-            curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
-          }
-          echo "ami-id: $(get_ec2_metadata ami-id)"
-          echo "instance-id: $(get_ec2_metadata instance-id)"
-          echo "instance-type: $(get_ec2_metadata instance-type)"
-          echo "system info $(uname -a)"
-      - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
-        continue-on-error: true
-        with:
-          github-secret: ${{ secrets.GITHUB_TOKEN }}
-      - name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
-        shell: bash
-        run: |
-          git config --global core.longpaths true
-          git config --global core.symlinks true
-
-          # https://git-scm.com/docs/git-fsmonitor--daemon.  The daemon could lock
-          # the directory on Windows and prevent GHA from checking out as reported
-          # in https://github.com/actions/checkout/issues/1018
-          git config --global core.fsmonitor false
-      # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
-      - name: Enable long paths on Windows
-        shell: powershell
-        run: |
-          Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
-      # Since it's just a defensive command, the workflow should continue even the command fails. This step can be
-      # removed once Windows Defender is removed from the AMI
-      - name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
-        continue-on-error: true
-        shell: powershell
-        run: |
-          Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
-          # Let's both exclude the path and disable Windows Defender completely just to be sure
-          # that it doesn't interfere
-          Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
-      - name: Checkout PyTorch
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-          submodules: recursive
-          path: pytorch
-          show-progress: false
-      - name: Clean PyTorch checkout
-        run: |
-          # Remove any artifacts from the previous checkouts
-          git clean -fxd
-        working-directory: pytorch
-      - name: Populate binary env
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
-      - name: Build PyTorch binary
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_windows_build.sh"
-      - uses: actions/upload-artifact@v4.4.0
-        if: always()
-        with:
-          name: libtorch-cuda12_9-shared-with-deps-release
-          retention-days: 14
-          if-no-files-found: error
-          path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
-      - name: Wait until all sessions have drained
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        timeout-minutes: 120
-        run: |
-          .github\scripts\wait_for_ssh_to_drain.ps1
-      - name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        run: |
-          .github\scripts\kill_active_ssh_sessions.ps1
-
-  libtorch-cuda12_9-shared-with-deps-release-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - libtorch-cuda12_9-shared-with-deps-release-build
-      - get-label-type
-    runs-on: "${{ needs.get-label-type.outputs.label-type }}windows.g4dn.xlarge"
-    timeout-minutes: 360
-    env:
-      PYTORCH_ROOT: ${{ github.workspace }}/pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
-      GPU_ARCH_TYPE: cuda
-      SKIP_ALL_TESTS: 1
-      LIBTORCH_CONFIG: release
-      LIBTORCH_VARIANT: shared-with-deps
-      # This is a dummy value for libtorch to work correctly with our batch scripts
-      # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
-    steps:
-      - name: Display EC2 information
-        shell: bash
-        run: |
-          set -euo pipefail
-          function get_ec2_metadata() {
-            # Pulled from instance metadata endpoint for EC2
-            # see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instancedata-data-retrieval.html
-            category=$1
-            curl -H "X-aws-ec2-metadata-token: $(curl -s -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 30")" -fsSL "http://169.254.169.254/latest/meta-data/${category}"
-          }
-          echo "ami-id: $(get_ec2_metadata ami-id)"
-          echo "instance-id: $(get_ec2_metadata instance-id)"
-          echo "instance-type: $(get_ec2_metadata instance-type)"
-          echo "system info $(uname -a)"
-      - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
-        continue-on-error: true
-        with:
-          github-secret: ${{ secrets.GITHUB_TOKEN }}
-      - name: Enable git long paths and symlinks on Windows and disable fsmonitor daemon
-        shell: bash
-        run: |
-          git config --global core.longpaths true
-          git config --global core.symlinks true
-
-          # https://git-scm.com/docs/git-fsmonitor--daemon.  The daemon could lock
-          # the directory on Windows and prevent GHA from checking out as reported
-          # in https://github.com/actions/checkout/issues/1018
-          git config --global core.fsmonitor false
-      # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
-      - name: Enable long paths on Windows
-        shell: powershell
-        run: |
-          Set-ItemProperty -Path "HKLM:\\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1
-      # Since it's just a defensive command, the workflow should continue even the command fails. This step can be
-      # removed once Windows Defender is removed from the AMI
-      - name: Disables Windows Defender scheduled and real-time scanning for files in directories used by PyTorch
-        continue-on-error: true
-        shell: powershell
-        run: |
-          Add-MpPreference -ExclusionPath $(Get-Location).tostring(),$Env:TEMP -ErrorAction Ignore
-          # Let's both exclude the path and disable Windows Defender completely just to be sure
-          # that it doesn't interfere
-          Set-MpPreference -DisableRealtimeMonitoring $True -ErrorAction Ignore
-      - name: Checkout PyTorch
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-          submodules: recursive
-          path: pytorch
-          show-progress: false
-      - name: Clean PyTorch checkout
-        run: |
-          # Remove any artifacts from the previous checkouts
-          git clean -fxd
-        working-directory: pytorch
-      # NOTE: These environment variables are put here so that they can be applied on every job equally
-      #       They are also here because setting them at a workflow level doesn't give us access to the
-      #       runner.temp variable, which we need.
-      - name: Populate binary env
-        shell: bash
-        run: |
-          echo "BINARY_ENV_FILE=${RUNNER_TEMP}/env" >> "${GITHUB_ENV}"
-          echo "PYTORCH_FINAL_PACKAGE_DIR=${RUNNER_TEMP}/artifacts" >> "${GITHUB_ENV}"
-          echo "WIN_PACKAGE_WORK_DIR=${RUNNER_TEMP}"
-      - uses: actions/download-artifact@v4.1.7
-        name: Download Build Artifacts
-        with:
-          name: libtorch-cuda12_9-shared-with-deps-release
-          path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
-      - name: Populate binary env
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_populate_env.sh"
-      - name: Test PyTorch binary
-        shell: bash
-        run: |
-          "${PYTORCH_ROOT}/.circleci/scripts/binary_windows_test.sh"
-      - name: Wait until all sessions have drained
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        timeout-minutes: 120
-        run: |
-          .github\scripts\wait_for_ssh_to_drain.ps1
-      - name: Kill active ssh sessions if still around (Useful if workflow was cancelled)
-        shell: powershell
-        working-directory: pytorch
-        if: always()
-        run: |
-          .github\scripts\kill_active_ssh_sessions.ps1
-  libtorch-cuda12_9-shared-with-deps-release-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: libtorch-cuda12_9-shared-with-deps-release-test
-    with:
-      PYTORCH_ROOT: ${{ github.workspace }}/pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu129
-      GPU_ARCH_VERSION: "12.9"
-      GPU_ARCH_TYPE: cuda
-      LIBTORCH_CONFIG: release
-      LIBTORCH_VARIANT: shared-with-deps
-      # This is a dummy value for libtorch to work correctly with our batch scripts
-      # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
-      build_name: libtorch-cuda12_9-shared-with-deps-release
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-    uses: ./.github/workflows/_binary-upload.yml
   libtorch-cuda13_0-shared-with-deps-release-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     needs: get-label-type
@@ -1056,7 +800,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       # NOTE: These environment variables are put here so that they can be applied on every job equally
       #       They are also here because setting them at a workflow level doesn't give us access to the
@@ -1082,7 +826,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -1114,7 +858,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -1172,7 +915,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1189,7 +932,7 @@ jobs:
           echo "instance-type: $(get_ec2_metadata instance-type)"
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.9
         continue-on-error: true
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -1221,7 +964,6 @@ jobs:
       - name: Checkout PyTorch
         uses: actions/checkout@v4
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           show-progress: false
@@ -1283,7 +1025,7 @@ jobs:
       LIBTORCH_VARIANT: shared-with-deps
       # This is a dummy value for libtorch to work correctly with our batch scripts
       # without this value pip does not get installed for some reason
-      DESIRED_PYTHON: "3.9"
+      DESIRED_PYTHON: "3.10"
       build_name: libtorch-cuda13_0-shared-with-deps-release
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/h100-cutlass-backend.yml

@@ -27,7 +27,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

.github/workflows/h100-distributed.yml

@@ -24,7 +24,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

.github/workflows/h100-symm-mem.yml

@@ -24,7 +24,7 @@ jobs:
   get-label-type:
     if: github.repository_owner == 'pytorch'
     name: get-label-type
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     with:
       triggering_actor: ${{ github.triggering_actor }}
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}

.github/workflows/inductor-micro-benchmark.yml

@@ -20,7 +20,7 @@ permissions:
 jobs:
   get-default-label-prefix:
     name: get-default-label-prefix
-    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@release/2.9
     if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }}
     with:
       triggering_actor: ${{ github.triggering_actor }}