[ROCm] Include hsa headers for rocm-triton whl (#129235)

* Include hsa headers for rocm-triton whl

* Update triton pin to release/3.0.x tip

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

---------

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

.ci/docker/aotriton_version.txt

@@ -1,5 +1,5 @@
 0.6b
 manylinux_2_17
-rocm6.1
-7f07e8a1cb1f99627eb6d77f5c0e9295c775f3c7
-77c29fa3f3b614e187d7213d745e989a92708cee2bc6020419ab49019af399d1
+rocm6
+04b5df8c8123f90cba3ede7e971e6fbc6040d506
+3db6ecbc915893ff967abd6e1b43bd5f54949868873be60dc802086c3863e648

.ci/docker/build.sh

@@ -373,13 +373,6 @@ case "$image" in
     CONDA_CMAKE=yes
     EXECUTORCH=yes
     ;;
-  pytorch-linux-jammy-py3.12-halide)
-    CUDA_VERSION=12.4
-    ANACONDA_PYTHON_VERSION=3.12
-    GCC_VERSION=11
-    CONDA_CMAKE=yes
-    HALIDE=yes
-    ;;
   pytorch-linux-focal-linter)
     # TODO: Use 3.9 here because of this issue https://github.com/python/mypy/issues/13627.
     # We will need to update mypy version eventually, but that's for another day. The task
@@ -497,7 +490,6 @@ docker build \
        --build-arg "DOCS=${DOCS}" \
        --build-arg "INDUCTOR_BENCHMARKS=${INDUCTOR_BENCHMARKS}" \
        --build-arg "EXECUTORCH=${EXECUTORCH}" \
-       --build-arg "HALIDE=${HALIDE}" \
        --build-arg "XPU_VERSION=${XPU_VERSION}" \
        --build-arg "ACL=${ACL:-}" \
        --build-arg "SKIP_SCCACHE_INSTALL=${SKIP_SCCACHE_INSTALL:-}" \

.ci/docker/ci_commit_pins/executorch.txt

@@ -1 +1 @@
-c572f9e509b5ec5d56f4d218271e36269bba244f
+d4b3e5cc607e97afdba79dc90f8ef968142f347c

.ci/docker/ci_commit_pins/halide.txt

@@ -1 +0,0 @@
-340136fec6d3ebc73e7a19eba1663e9b0ba8ab2d

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

@@ -1 +1 @@
-1b2f15840e0d70eec50d84c7a0575cb835524def
+aac14a3b93f11d781d1d5ebc5400b15ae8df5185

.ci/docker/ci_commit_pins/triton.txt

@@ -1 +1 @@
-dedb7bdf339a3546896d4820366ca562c586bfa0
+45fff310c891f5a92d55445adf8cc9d29df5841e

.ci/docker/common/install_aotriton.sh

@@ -9,7 +9,7 @@ TARBALL='aotriton.tar.bz2'
 read -d "\n" VER MANYLINUX ROCMBASE PINNED_COMMIT SHA256 < aotriton_version.txt || true
 ARCH=$(uname -m)
 AOTRITON_INSTALL_PREFIX="$1"
-AOTRITON_URL="https://github.com/ROCm/aotriton/releases/download/${VER}/aotriton-${VER}-${MANYLINUX}_${ARCH}-${ROCMBASE}-shared.tar.bz2"
+AOTRITON_URL="https://github.com/ROCm/aotriton/releases/download/${VER}/aotriton-${VER}-${MANYLINUX}_${ARCH}-${ROCMBASE}.tar.bz2"
 
 cd "${AOTRITON_INSTALL_PREFIX}"
 # Must use -L to follow redirects

.ci/docker/common/install_conda.sh

@@ -85,7 +85,7 @@ fi
   else
     CONDA_COMMON_DEPS="astunparse pyyaml mkl=2021.4.0 mkl-include=2021.4.0 setuptools"
 
-    if [ "$ANACONDA_PYTHON_VERSION" = "3.11" ] || [ "$ANACONDA_PYTHON_VERSION" = "3.12" ] || [ "$ANACONDA_PYTHON_VERSION" = "3.13" ]; then
+    if [ "$ANACONDA_PYTHON_VERSION" = "3.11" ] || [ "$ANACONDA_PYTHON_VERSION" = "3.12" ]; then
       conda_install numpy=1.26.0 ${CONDA_COMMON_DEPS}
     else
       conda_install numpy=1.21.2 ${CONDA_COMMON_DEPS}

.ci/docker/common/install_executorch.sh

@@ -37,9 +37,6 @@ install_conda_dependencies() {
 
 install_pip_dependencies() {
   pushd executorch/.ci/docker
-  # Install PyTorch CPU build beforehand to avoid installing the much bigger CUDA
-  # binaries later, ExecuTorch only needs CPU
-  pip_install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
   # Install all Python dependencies
   pip_install -r requirements-ci.txt
   popd
@@ -47,14 +44,13 @@ install_pip_dependencies() {
 
 setup_executorch() {
   pushd executorch
-  # Setup swiftshader and Vulkan SDK which are required to build the Vulkan delegate
-  as_jenkins bash .ci/scripts/setup-vulkan-linux-deps.sh
+  source .ci/scripts/utils.sh
 
-  export PYTHON_EXECUTABLE=python
-  export EXECUTORCH_BUILD_PYBIND=ON
-  export CMAKE_ARGS="-DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON"
+  install_flatc_from_source
+  pip_install .
 
-  as_jenkins .ci/scripts/setup-linux.sh cmake
+  # Make sure that all the newly generate files are owned by Jenkins
+  chown -R jenkins .
   popd
 }
 

.ci/docker/common/install_halide.sh

@@ -1,46 +0,0 @@
-#!/bin/bash
-set -ex
-
-source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
-
-COMMIT=$(get_pinned_commit halide)
-test -n "$COMMIT"
-
-# activate conda to populate CONDA_PREFIX
-test -n "$ANACONDA_PYTHON_VERSION"
-eval "$(conda shell.bash hook)"
-conda activate py_$ANACONDA_PYTHON_VERSION
-
-if [ -n "${UBUNTU_VERSION}" ];then
-    apt update
-    apt-get install -y lld liblld-15-dev libpng-dev libjpeg-dev libgl-dev \
-                  libopenblas-dev libeigen3-dev libatlas-base-dev libzstd-dev
-fi
-
-conda_install numpy scipy imageio cmake ninja
-
-git clone --depth 1 --branch release/16.x --recursive https://github.com/llvm/llvm-project.git
-cmake -DCMAKE_BUILD_TYPE=Release \
-        -DLLVM_ENABLE_PROJECTS="clang" \
-        -DLLVM_TARGETS_TO_BUILD="X86;NVPTX" \
-        -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ASSERTIONS=ON \
-        -DLLVM_ENABLE_EH=ON -DLLVM_ENABLE_RTTI=ON -DLLVM_BUILD_32_BITS=OFF \
-        -S llvm-project/llvm -B llvm-build -G Ninja
-cmake --build llvm-build
-cmake --install llvm-build --prefix llvm-install
-export LLVM_ROOT=`pwd`/llvm-install
-export LLVM_CONFIG=$LLVM_ROOT/bin/llvm-config
-
-git clone https://github.com/halide/Halide.git
-pushd Halide
-git checkout ${COMMIT} && git submodule update --init --recursive
-pip_install -r requirements.txt
-cmake -G Ninja -DCMAKE_BUILD_TYPE=Release -S . -B build
-cmake --build build
-test -e ${CONDA_PREFIX}/lib/python3 || ln -s python${ANACONDA_PYTHON_VERSION} ${CONDA_PREFIX}/lib/python3
-cmake --install build --prefix ${CONDA_PREFIX}
-chown -R jenkins ${CONDA_PREFIX}
-popd
-rm -rf Halide llvm-build llvm-project llvm-install
-
-python -c "import halide"  # check for errors

.ci/docker/common/install_onnx.sh

@@ -33,9 +33,7 @@ pip_install coloredlogs packaging
 pip_install onnxruntime==1.18
 pip_install onnx==1.16.0
 # pip_install "onnxscript@git+https://github.com/microsoft/onnxscript@3e869ef8ccf19b5ebd21c10d3e9c267c9a9fa729" --no-deps
-pip_install onnxscript==0.1.0.dev20240613 --no-deps
-# required by onnxscript
-pip_install ml_dtypes
+pip_install onnxscript==0.1.0.dev20240523 --no-deps
 
 # Cache the transformers model to be used later by ONNX tests. We need to run the transformers
 # package to download the model. By default, the model is cached at ~/.cache/huggingface/hub/

.ci/docker/requirements-ci.txt

@@ -85,10 +85,10 @@ librosa>=0.6.2 ; python_version < "3.11"
 #Pinned versions:
 #test that import:
 
-mypy==1.10.0
+mypy==1.9.0
 # Pin MyPy version because new errors are likely to appear with each release
 #Description: linter
-#Pinned versions: 1.10.0
+#Pinned versions: 1.9.0
 #test that import: test_typing.py, test_type_hints.py
 
 networkx==2.8.8
@@ -306,7 +306,7 @@ pywavelets==1.5.0 ; python_version >= "3.12"
 #Pinned versions: 1.4.1
 #test that import:
 
-lxml==5.0.0
+lxml==5.0.0.
 #Description: This is a requirement of unittest-xml-reporting
 
 # Python-3.9 binaries

.ci/docker/ubuntu-cuda/Dockerfile

@@ -103,14 +103,6 @@ COPY triton_version.txt triton_version.txt
 RUN if [ -n "${TRITON}" ]; then bash ./install_triton.sh; fi
 RUN rm install_triton.sh common_utils.sh triton.txt triton_version.txt
 
-ARG HALIDE
-# Build and install halide
-COPY ./common/install_halide.sh install_halide.sh
-COPY ./common/common_utils.sh common_utils.sh
-COPY ci_commit_pins/halide.txt halide.txt
-RUN if [ -n "${HALIDE}" ]; then bash ./install_halide.sh; fi
-RUN rm install_halide.sh common_utils.sh halide.txt
-
 # Install ccache/sccache (do this last, so we get priority in PATH)
 COPY ./common/install_cache.sh install_cache.sh
 ENV PATH /opt/cache/bin:$PATH

.ci/docker/ubuntu/Dockerfile

@@ -155,14 +155,6 @@ COPY ci_commit_pins/executorch.txt executorch.txt
 RUN if [ -n "${EXECUTORCH}" ]; then bash ./install_executorch.sh; fi
 RUN rm install_executorch.sh common_utils.sh executorch.txt
 
-ARG HALIDE
-# Build and install halide
-COPY ./common/install_halide.sh install_halide.sh
-COPY ./common/common_utils.sh common_utils.sh
-COPY ci_commit_pins/halide.txt halide.txt
-RUN if [ -n "${HALIDE}" ]; then bash ./install_halide.sh; fi
-RUN rm install_halide.sh common_utils.sh halide.txt
-
 ARG ONNX
 # Install ONNX dependencies
 COPY ./common/install_onnx.sh ./common/common_utils.sh ./

.ci/pytorch/build.sh

@@ -230,10 +230,6 @@ if [[ "${BUILD_ENVIRONMENT}" != *android* && "${BUILD_ENVIRONMENT}" != *cuda* ]]
   export BUILD_STATIC_RUNTIME_BENCHMARK=ON
 fi
 
-if [[ "$BUILD_ENVIRONMENT" == *-debug* ]]; then
-  export CMAKE_BUILD_TYPE=RelWithAssert
-fi
-
 # Do not change workspace permissions for ROCm CI jobs
 # as it can leave workspace with bad permissions for cancelled jobs
 if [[ "$BUILD_ENVIRONMENT" != *rocm* ]]; then
@@ -288,26 +284,12 @@ else
         # Which should be backward compatible with Numpy-1.X
         python -mpip install --pre numpy==2.0.0rc1
       fi
-
-      WERROR=1 python setup.py clean
-
-      if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
-        BUILD_LIBTORCH_WHL=1 BUILD_PYTHON_ONLY=0 python setup.py bdist_wheel
-        BUILD_LIBTORCH_WHL=0 BUILD_PYTHON_ONLY=1 python setup.py bdist_wheel --cmake
-      else
-        WERROR=1 python setup.py bdist_wheel
-      fi
+      WERROR=1 python setup.py bdist_wheel
     else
-      python setup.py clean
       if [[ "$BUILD_ENVIRONMENT" == *xla* ]]; then
         source .ci/pytorch/install_cache_xla.sh
       fi
-      if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
-        echo "USE_SPLIT_BUILD cannot be used with xla or rocm"
-        exit 1
-      else
-        python setup.py bdist_wheel
-      fi
+      python setup.py bdist_wheel
     fi
     pip_install_whl "$(echo dist/*.whl)"
 
@@ -346,10 +328,9 @@ else
     CUSTOM_OP_TEST="$PWD/test/custom_operator"
     python --version
     SITE_PACKAGES="$(python -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
-
     mkdir -p "$CUSTOM_OP_BUILD"
     pushd "$CUSTOM_OP_BUILD"
-    cmake "$CUSTOM_OP_TEST" -DCMAKE_PREFIX_PATH="$SITE_PACKAGES/torch;$SITE_PACKAGES" -DPython_EXECUTABLE="$(which python)" \
+    cmake "$CUSTOM_OP_TEST" -DCMAKE_PREFIX_PATH="$SITE_PACKAGES/torch" -DPython_EXECUTABLE="$(which python)" \
           -DCMAKE_MODULE_PATH="$CUSTOM_TEST_MODULE_PATH" -DUSE_ROCM="$CUSTOM_TEST_USE_ROCM"
     make VERBOSE=1
     popd
@@ -362,7 +343,7 @@ else
     SITE_PACKAGES="$(python -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
     mkdir -p "$JIT_HOOK_BUILD"
     pushd "$JIT_HOOK_BUILD"
-    cmake "$JIT_HOOK_TEST" -DCMAKE_PREFIX_PATH="$SITE_PACKAGES/torch;$SITE_PACKAGES" -DPython_EXECUTABLE="$(which python)" \
+    cmake "$JIT_HOOK_TEST" -DCMAKE_PREFIX_PATH="$SITE_PACKAGES/torch" -DPython_EXECUTABLE="$(which python)" \
           -DCMAKE_MODULE_PATH="$CUSTOM_TEST_MODULE_PATH" -DUSE_ROCM="$CUSTOM_TEST_USE_ROCM"
     make VERBOSE=1
     popd
@@ -374,7 +355,7 @@ else
     python --version
     mkdir -p "$CUSTOM_BACKEND_BUILD"
     pushd "$CUSTOM_BACKEND_BUILD"
-    cmake "$CUSTOM_BACKEND_TEST" -DCMAKE_PREFIX_PATH="$SITE_PACKAGES/torch;$SITE_PACKAGES" -DPython_EXECUTABLE="$(which python)" \
+    cmake "$CUSTOM_BACKEND_TEST" -DCMAKE_PREFIX_PATH="$SITE_PACKAGES/torch" -DPython_EXECUTABLE="$(which python)" \
           -DCMAKE_MODULE_PATH="$CUSTOM_TEST_MODULE_PATH" -DUSE_ROCM="$CUSTOM_TEST_USE_ROCM"
     make VERBOSE=1
     popd

.ci/pytorch/common_utils.sh

@@ -56,29 +56,9 @@ function assert_git_not_dirty() {
 function pip_install_whl() {
   # This is used to install PyTorch and other build artifacts wheel locally
   # without using any network connection
-
-  # Convert the input arguments into an array
-  local args=("$@")
-
-  # Check if the first argument contains multiple paths separated by spaces
-  if [[ "${args[0]}" == *" "* ]]; then
-    # Split the string by spaces into an array
-    IFS=' ' read -r -a paths <<< "${args[0]}"
-    # Loop through each path and install individually
-    for path in "${paths[@]}"; do
-      echo "Installing $path"
-      python3 -mpip install --no-index --no-deps "$path"
-    done
-  else
-    # Loop through each argument and install individually
-    for path in "${args[@]}"; do
-      echo "Installing $path"
-      python3 -mpip install --no-index --no-deps "$path"
-    done
-  fi
+  python3 -mpip install --no-index --no-deps "$@"
 }
 
-
 function pip_install() {
   # retry 3 times
   # old versions of pip don't have the "--progress-bar" flag
@@ -198,7 +178,7 @@ function install_torchrec_and_fbgemm() {
 
 function clone_pytorch_xla() {
   if [[ ! -d ./xla ]]; then
-    git clone --recursive --quiet https://github.com/pytorch/xla.git
+    git clone --recursive -b r2.4 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)"
@@ -208,6 +188,28 @@ function clone_pytorch_xla() {
   fi
 }
 
+function checkout_install_torchdeploy() {
+  local commit
+  commit=$(get_pinned_commit multipy)
+  pushd ..
+  git clone --recurse-submodules https://github.com/pytorch/multipy.git
+  pushd multipy
+  git checkout "${commit}"
+  python multipy/runtime/example/generate_examples.py
+  BUILD_CUDA_TESTS=1 pip install -e .
+  popd
+  popd
+}
+
+function test_torch_deploy(){
+ pushd ..
+ pushd multipy
+ ./multipy/runtime/build/test_deploy
+ ./multipy/runtime/build/test_deploy_gpu
+ popd
+ popd
+}
+
 function checkout_install_torchbench() {
   local commit
   commit=$(get_pinned_commit torchbench)
@@ -222,8 +224,6 @@ function checkout_install_torchbench() {
     # to install and test other models
     python install.py --continue_on_fail
   fi
-  echo "Print all dependencies after TorchBench is installed"
-  python -mpip freeze
   popd
 }
 

.ci/pytorch/create_test_cert.py

@@ -6,7 +6,6 @@ from cryptography.hazmat.primitives import hashes, serialization
 from cryptography.hazmat.primitives.asymmetric import rsa
 from cryptography.x509.oid import NameOID
 
-
 temp_dir = mkdtemp()
 print(temp_dir)
 

.ci/pytorch/multigpu-test.sh

@@ -18,9 +18,8 @@ time python test/run_test.py --verbose -i distributed/test_c10d_gloo
 time python test/run_test.py --verbose -i distributed/test_c10d_nccl
 time python test/run_test.py --verbose -i distributed/test_c10d_spawn_gloo
 time python test/run_test.py --verbose -i distributed/test_c10d_spawn_nccl
-time python test/run_test.py --verbose -i distributed/test_compute_comm_reordering
+time python test/run_test.py --verbose -i distributed/test_cuda_p2p
 time python test/run_test.py --verbose -i distributed/test_store
-time python test/run_test.py --verbose -i distributed/test_symmetric_memory
 time python test/run_test.py --verbose -i distributed/test_pg_wrapper
 time python test/run_test.py --verbose -i distributed/rpc/cuda/test_tensorpipe_agent
 # FSDP tests

.ci/pytorch/perf_test/compare_with_baseline.py

@@ -3,7 +3,6 @@ import json
 import math
 import sys
 
-
 parser = argparse.ArgumentParser()
 parser.add_argument(
     "--test-name", dest="test_name", action="store", required=True, help="test name"

.ci/pytorch/perf_test/get_stats.py

@@ -3,7 +3,6 @@ import sys
 
 import numpy
 
-
 sample_data_list = sys.argv[1:]
 sample_data_list = [float(v.strip()) for v in sample_data_list]
 

.ci/pytorch/perf_test/update_commit_hash.py

@@ -1,7 +1,6 @@
 import json
 import sys
 
-
 data_file_path = sys.argv[1]
 commit_hash = sys.argv[2]
 

.ci/pytorch/print_sccache_log.py

@@ -1,6 +1,5 @@
 import sys
 
-
 log_file_path = sys.argv[1]
 
 with open(log_file_path) as f:

.ci/pytorch/test.sh

@@ -249,7 +249,9 @@ fi
 # This tests that the debug asserts are working correctly.
 if [[ "$BUILD_ENVIRONMENT" == *-debug* ]]; then
     echo "We are in debug mode: $BUILD_ENVIRONMENT. Expect the python assertion to fail"
-    (cd test && ! get_exit_code python -c "import torch; torch._C._crash_if_debug_asserts_fail(424242)")
+    # TODO: Enable the check after we setup the build to run debug asserts without having
+    #       to do a full (and slow) debug build
+    # (cd test && ! get_exit_code python -c "import torch; torch._C._crash_if_debug_asserts_fail(424242)")
 elif [[ "$BUILD_ENVIRONMENT" != *-bazel-* ]]; then
     # Noop when debug is disabled. Skip bazel jobs because torch isn't available there yet.
     echo "We are not in debug mode: $BUILD_ENVIRONMENT. Expect the assertion to pass"
@@ -262,6 +264,18 @@ elif [[ $TEST_CONFIG == 'nogpu_AVX512' ]]; then
   export ATEN_CPU_CAPABILITY=avx2
 fi
 
+# temp workarounds for https://github.com/pytorch/pytorch/issues/126692, remove when fixed
+if [[ "$BUILD_ENVIRONMENT" != *-bazel-* ]]; then
+  pushd test
+  CUDA_VERSION=$(python -c "import torch; print(torch.version.cuda)")
+  if [ "$CUDA_VERSION" == "12.4" ]; then
+    ISCUDA124="cu124"
+  else
+    ISCUDA124=""
+  fi
+  popd
+fi
+
 test_python_legacy_jit() {
   time python test/run_test.py --include test_jit_legacy test_jit_fuser_legacy --verbose
   assert_git_not_dirty
@@ -275,9 +289,6 @@ test_python_shard() {
 
   # Bare --include flag is not supported and quoting for lint ends up with flag not being interpreted correctly
   # shellcheck disable=SC2086
-
-  # modify LD_LIBRARY_PATH to ensure it has the conda env.
-  # This set of tests has been shown to be buggy without it for the split-build
   time python test/run_test.py --exclude-jit-executor --exclude-distributed-tests $INCLUDE_CLAUSE --shard "$1" "$NUM_TEST_SHARDS" --verbose $PYTHON_TEST_EXTRA_OPTION
 
   assert_git_not_dirty
@@ -336,31 +347,17 @@ test_inductor_distributed() {
   assert_git_not_dirty
 }
 
-test_inductor_shard() {
-  if [[ -z "$NUM_TEST_SHARDS" ]]; then
-    echo "NUM_TEST_SHARDS must be defined to run a Python test shard"
-    exit 1
-  fi
-
+test_inductor() {
   python tools/dynamo/verify_dynamo.py
-  python test/run_test.py --inductor \
-    --include test_modules test_ops test_ops_gradients test_torch \
-    --shard "$1" "$NUM_TEST_SHARDS" \
-    --verbose
-
+  python test/run_test.py --inductor --include test_modules test_ops test_ops_gradients test_torch --verbose
   # Do not add --inductor for the following inductor unit tests, otherwise we will fail because of nested dynamo state
-  python test/run_test.py \
-    --include inductor/test_torchinductor inductor/test_torchinductor_opinfo inductor/test_aot_inductor \
-    --shard "$1" "$NUM_TEST_SHARDS" \
-    --verbose
-}
+  python test/run_test.py --include inductor/test_torchinductor inductor/test_torchinductor_opinfo inductor/test_aot_inductor --verbose
 
-test_inductor_aoti() {
   # docker build uses bdist_wheel which does not work with test_aot_inductor
   # TODO: need a faster way to build
   if [[ "$BUILD_ENVIRONMENT" != *rocm* ]]; then
-    BUILD_AOT_INDUCTOR_TEST=1 python setup.py develop
-    CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="${TORCH_LIB_DIR}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference
+      BUILD_AOT_INDUCTOR_TEST=1 python setup.py develop
+      CPP_TESTS_DIR="${BUILD_BIN_DIR}" LD_LIBRARY_PATH="${TORCH_LIB_DIR}" python test/run_test.py --cpp --verbose -i cpp/test_aoti_abi_check cpp/test_aoti_inference
   fi
 }
 
@@ -379,7 +376,7 @@ test_inductor_cpp_wrapper_abi_compatible() {
     --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_training.csv"
   python benchmarks/dynamo/check_accuracy.py \
     --actual "$TEST_REPORTS_DIR/inductor_cpp_wrapper_training.csv" \
-    --expected "benchmarks/dynamo/ci_expected_accuracy/inductor_timm_training.csv"
+    --expected "benchmarks/dynamo/ci_expected_accuracy/${ISCUDA124}/inductor_timm_training.csv"
 }
 
 # "Global" flags for inductor benchmarking controlled by TEST_CONFIG
@@ -404,7 +401,7 @@ if [[ "${TEST_CONFIG}" == *dynamic* ]]; then
   DYNAMO_BENCHMARK_FLAGS+=(--dynamic-shapes --dynamic-batch-only)
 fi
 
-if [[ "${TEST_CONFIG}" == *cpu_inductor* || "${TEST_CONFIG}" == *cpu_aot_inductor* ]]; then
+if [[ "${TEST_CONFIG}" == *cpu_inductor* ]]; then
   DYNAMO_BENCHMARK_FLAGS+=(--device cpu)
 else
   DYNAMO_BENCHMARK_FLAGS+=(--device cuda)
@@ -529,10 +526,9 @@ test_single_dynamo_benchmark() {
     test_perf_for_dashboard "$suite" \
       "${DYNAMO_BENCHMARK_FLAGS[@]}" "$@" "${partition_flags[@]}"
   else
-    if [[ "${TEST_CONFIG}" == *aot_inductor* && "${TEST_CONFIG}" != *cpu_aot_inductor* ]]; then
+    if [[ "${TEST_CONFIG}" == *aot_inductor* ]]; then
       # Test AOTInductor with the ABI-compatible mode on CI
       # This can be removed once the ABI-compatible mode becomes default.
-      # For CPU device, we perfer non ABI-compatible mode on CI when testing AOTInductor.
       export TORCHINDUCTOR_ABI_COMPATIBLE=1
     fi
     python "benchmarks/dynamo/$suite.py" \
@@ -542,10 +538,10 @@ test_single_dynamo_benchmark() {
       --output "$TEST_REPORTS_DIR/${name}_${suite}.csv"
     python benchmarks/dynamo/check_accuracy.py \
       --actual "$TEST_REPORTS_DIR/${name}_$suite.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/${TEST_CONFIG}_${name}.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${ISCUDA124}/${TEST_CONFIG}_${name}.csv"
     python benchmarks/dynamo/check_graph_breaks.py \
       --actual "$TEST_REPORTS_DIR/${name}_$suite.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/${TEST_CONFIG}_${name}.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${ISCUDA124}/${TEST_CONFIG}_${name}.csv"
   fi
 }
 
@@ -554,11 +550,6 @@ test_inductor_micro_benchmark() {
   python benchmarks/gpt_fast/benchmark.py --output "${TEST_REPORTS_DIR}/gpt_fast_benchmark.csv"
 }
 
-test_inductor_halide() {
-  python test/run_test.py --include inductor/test_halide.py --verbose
-  assert_git_not_dirty
-}
-
 test_dynamo_benchmark() {
   # Usage: test_dynamo_benchmark huggingface 0
   TEST_REPORTS_DIR=$(pwd)/test/test-reports
@@ -573,15 +564,11 @@ test_dynamo_benchmark() {
   elif [[ "${TEST_CONFIG}" == *perf* ]]; then
     test_single_dynamo_benchmark "dashboard" "$suite" "$shard_id" "$@"
   else
-    if [[ "${TEST_CONFIG}" == *cpu_inductor* || "${TEST_CONFIG}" == *cpu_aot_inductor* ]]; then
-      local dt="float32"
-      if [[ "${TEST_CONFIG}" == *amp* ]]; then
-        dt="amp"
-      fi
+    if [[ "${TEST_CONFIG}" == *cpu_inductor* ]]; then
       if [[ "${TEST_CONFIG}" == *freezing* ]]; then
-        test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --"$dt" --freezing "$@"
+        test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --float32 --freezing "$@"
       else
-        test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --"$dt" "$@"
+        test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --float32 "$@"
       fi
     elif [[ "${TEST_CONFIG}" == *aot_inductor* ]]; then
       test_single_dynamo_benchmark "inference" "$suite" "$shard_id" --inference --bfloat16 "$@"
@@ -605,7 +592,7 @@ test_inductor_torchbench_smoketest_perf() {
     --bfloat16 --inference --inductor --only moco --output "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv"
   python benchmarks/dynamo/check_accuracy.py \
     --actual "$TEST_REPORTS_DIR/inductor_cpp_wrapper_inference.csv" \
-    --expected "benchmarks/dynamo/ci_expected_accuracy/inductor_torchbench_inference.csv"
+    --expected "benchmarks/dynamo/ci_expected_accuracy/${ISCUDA124}/inductor_torchbench_inference.csv"
 
   python benchmarks/dynamo/torchbench.py --device cuda --performance --backend inductor --float16 --training \
     --batch-size-file "$(realpath benchmarks/dynamo/torchbench_models_list.txt)" --only hf_Bert \
@@ -620,8 +607,13 @@ test_inductor_torchbench_smoketest_perf() {
   # https://github.com/pytorch/pytorch/actions/runs/7158691360/job/19491437314,
   # and thus we lower its threshold to reduce flakiness. If this continues to be a problem,
   # we switch to use some other model.
-  # lowering threshold from 4.9 to 4.7 for cu124. Will bump it up after cuda 12.4.0->12.4.1 update
-  python benchmarks/dynamo/check_perf_csv.py -f "$TEST_REPORTS_DIR/inductor_inference_smoketest.csv" -t 4.7
+  # Use 4.7 for cuda 12.4, change back to 4.9 after fixing https://github.com/pytorch/pytorch/issues/126692
+  if [ "$CUDA_VERSION" == "12.4" ]; then
+    THRESHOLD=4.7
+  else
+    THRESHOLD=4.9
+  fi
+  python benchmarks/dynamo/check_perf_csv.py -f "$TEST_REPORTS_DIR/inductor_inference_smoketest.csv" -t $THRESHOLD
 
   # Check memory compression ratio for a few models
   for test in hf_Albert timm_vision_transformer; do
@@ -640,7 +632,7 @@ test_inductor_torchbench_smoketest_perf() {
       --only $test --output "$TEST_REPORTS_DIR/inductor_warm_start_smoketest_$test.csv"
     python benchmarks/dynamo/check_accuracy.py \
       --actual "$TEST_REPORTS_DIR/inductor_warm_start_smoketest_$test.csv" \
-      --expected "benchmarks/dynamo/ci_expected_accuracy/inductor_huggingface_training.csv"
+      --expected "benchmarks/dynamo/ci_expected_accuracy/${ISCUDA124}/inductor_huggingface_training.csv"
   done
 }
 
@@ -1177,21 +1169,15 @@ test_executorch() {
 
   pushd /executorch
 
-  export PYTHON_EXECUTABLE=python
-  export EXECUTORCH_BUILD_PYBIND=ON
-  export CMAKE_ARGS="-DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON"
-
-  # NB: We need to rebuild ExecuTorch runner here because it depends on PyTorch
-  # from the PR
+  # NB: We need to build ExecuTorch runner here and not inside the Docker image
+  # because it depends on PyTorch
   # shellcheck disable=SC1091
-  source .ci/scripts/setup-linux.sh cmake
-
-  echo "Run ExecuTorch unit tests"
-  pytest -v -n auto
-  # shellcheck disable=SC1091
-  LLVM_PROFDATA=llvm-profdata-12 LLVM_COV=llvm-cov-12 bash test/run_oss_cpp_tests.sh
+  source .ci/scripts/utils.sh
+  build_executorch_runner "cmake"
 
   echo "Run ExecuTorch regression tests for some models"
+  # NB: This is a sample model, more can be added here
+  export PYTHON_EXECUTABLE=python
   # TODO(huydhn): Add more coverage here using ExecuTorch's gather models script
   # shellcheck disable=SC1091
   source .ci/scripts/test.sh mv3 cmake xnnpack-quantization-delegation ''
@@ -1251,10 +1237,11 @@ elif [[ "$TEST_CONFIG" == distributed ]]; then
   if [[ "${SHARD_NUMBER}" == 1 ]]; then
     test_rpc
   fi
+elif [[ "$TEST_CONFIG" == deploy ]]; then
+  checkout_install_torchdeploy
+  test_torch_deploy
 elif [[ "${TEST_CONFIG}" == *inductor_distributed* ]]; then
   test_inductor_distributed
-elif [[ "${TEST_CONFIG}" == *inductor-halide* ]]; then
-  test_inductor_halide
 elif [[ "${TEST_CONFIG}" == *inductor-micro-benchmark* ]]; then
   test_inductor_micro_benchmark
 elif [[ "${TEST_CONFIG}" == *huggingface* ]]; then
@@ -1266,14 +1253,13 @@ elif [[ "${TEST_CONFIG}" == *timm* ]]; then
   id=$((SHARD_NUMBER-1))
   test_dynamo_benchmark timm_models "$id"
 elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
-  if [[ "${TEST_CONFIG}" == *cpu_inductor* || "${TEST_CONFIG}" == *cpu_aot_inductor* ]]; then
+  if [[ "${TEST_CONFIG}" == *cpu_inductor* ]]; then
     install_torchaudio cpu
   else
     install_torchaudio cuda
   fi
   install_torchtext
   install_torchvision
-  TORCH_CUDA_ARCH_LIST="8.0;8.6" pip_install git+https://github.com/pytorch/ao.git
   id=$((SHARD_NUMBER-1))
   # https://github.com/opencv/opencv-python/issues/885
   pip_install opencv-python==4.8.0.74
@@ -1292,7 +1278,7 @@ elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
     checkout_install_torchbench
     # Do this after checkout_install_torchbench to ensure we clobber any
     # nightlies that torchbench may pull in
-    if [[ "${TEST_CONFIG}" != *cpu_inductor* && "${TEST_CONFIG}" != *cpu_aot_inductor* ]]; then
+    if [[ "${TEST_CONFIG}" != *cpu_inductor* ]]; then
       install_torchrec_and_fbgemm
     fi
     PYTHONPATH=$(pwd)/torchbench test_dynamo_benchmark torchbench "$id"
@@ -1300,19 +1286,17 @@ elif [[ "${TEST_CONFIG}" == *torchbench* ]]; then
 elif [[ "${TEST_CONFIG}" == *inductor_cpp_wrapper_abi_compatible* ]]; then
   install_torchvision
   test_inductor_cpp_wrapper_abi_compatible
-elif [[ "${TEST_CONFIG}" == *inductor* ]]; then
+elif [[ "${TEST_CONFIG}" == *inductor* && "${SHARD_NUMBER}" == 1 ]]; then
   install_torchvision
-  test_inductor_shard "${SHARD_NUMBER}"
-  if [[ "${SHARD_NUMBER}" == 1 ]]; then
-    test_inductor_aoti
-    test_inductor_distributed
-  fi
-elif [[ "${TEST_CONFIG}" == *dynamo* ]]; then
+  test_inductor
+  test_inductor_distributed
+elif [[ "${TEST_CONFIG}" == *dynamo* && "${SHARD_NUMBER}" == 1 && $NUM_TEST_SHARDS -gt 1 ]]; then
+  install_torchvision
+  test_dynamo_shard 1
+  test_aten
+elif [[ "${TEST_CONFIG}" == *dynamo* && $SHARD_NUMBER -gt 1 && $NUM_TEST_SHARDS -gt 1 ]]; then
   install_torchvision
   test_dynamo_shard "${SHARD_NUMBER}"
-  if [[ "${SHARD_NUMBER}" == 1 ]]; then
-    test_aten
-  fi
 elif [[ "${BUILD_ENVIRONMENT}" == *rocm* && -n "$TESTS_TO_INCLUDE" ]]; then
   install_torchvision
   test_python_shard "$SHARD_NUMBER"

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

@@ -4,7 +4,6 @@ import os
 import subprocess
 import sys
 
-
 COMMON_TESTS = [
     (
         "Checking that torch is available",

.circleci/codegen_validation/normalize_yaml_fragment.py

@@ -5,7 +5,6 @@ import sys
 
 import yaml
 
-
 # Need to import modules that lie on an upward-relative path
 sys.path.append(os.path.join(sys.path[0], ".."))
 

.circleci/scripts/binary_linux_test.sh

@@ -97,16 +97,8 @@ if [[ "$PACKAGE_TYPE" == conda ]]; then
   )
 elif [[ "$PACKAGE_TYPE" != libtorch ]]; then
   if [[ "\$BUILD_ENVIRONMENT" != *s390x* ]]; then
-    if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
-      pkg_no_python="$(ls -1 /final_pkgs/torch_no_python* | sort |tail -1)"
-      pkg_torch="$(ls -1 /final_pkgs/torch-* | sort |tail -1)"
-      # todo: after folder is populated use the pypi_pkg channel instead
-      pip install "\$pkg_no_python" "\$pkg_torch" --index-url "https://download.pytorch.org/whl/\${CHANNEL}/${DESIRED_CUDA}_pypi_pkg"
-      retry pip install -q numpy protobuf typing-extensions
-    else
-      pip install "\$pkg" --index-url "https://download.pytorch.org/whl/\${CHANNEL}/${DESIRED_CUDA}"
-      retry pip install -q numpy protobuf typing-extensions
-    fi
+    pip install "\$pkg" --index-url "https://download.pytorch.org/whl/\${CHANNEL}/${DESIRED_CUDA}"
+    retry pip install -q numpy protobuf typing-extensions
   else
     pip install "\$pkg"
     retry pip install -q numpy protobuf typing-extensions
@@ -118,18 +110,9 @@ if [[ "$PACKAGE_TYPE" == libtorch ]]; then
   cd /tmp/libtorch
 fi
 
-if [[ "$GPU_ARCH_TYPE" == xpu ]]; then
-  # Workaround for __mkl_tmp_MOD unbound variable issue, refer https://github.com/pytorch/pytorch/issues/130543
-  set +u
-  source /opt/intel/oneapi/pytorch-gpu-dev-0.5/oneapi-vars.sh
-fi
-
 # Test the package
 /builder/check_binary.sh
 
-# Clean temp files
-cd /builder && git clean -ffdx
-
 # =================== The above code will be executed inside Docker container ===================
 EOL
 echo

.circleci/scripts/binary_populate_env.sh

@@ -33,9 +33,9 @@ if [[ -z "$DOCKER_IMAGE" ]]; then
   if [[ "$PACKAGE_TYPE" == conda ]]; then
     export DOCKER_IMAGE="pytorch/conda-cuda"
   elif [[ "$DESIRED_CUDA" == cpu ]]; then
-    export DOCKER_IMAGE="pytorch/manylinux:cpu"
+    export DOCKER_IMAGE="pytorch/manylinux-cpu"
   else
-    export DOCKER_IMAGE="pytorch/manylinux-builder:${DESIRED_CUDA:2}"
+    export DOCKER_IMAGE="pytorch/manylinux-cuda${DESIRED_CUDA:2}"
   fi
 fi
 
@@ -77,8 +77,7 @@ TRITON_VERSION=$(cat $PYTORCH_ROOT/.ci/docker/triton_version.txt)
 # Here PYTORCH_EXTRA_INSTALL_REQUIREMENTS is already set for the all the wheel builds hence append TRITON_CONSTRAINT
 TRITON_CONSTRAINT="platform_system == 'Linux' and platform_machine == 'x86_64' and python_version < '3.13'"
 if [[ "$PACKAGE_TYPE" =~ .*wheel.* &&  -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]]; then
-  # Only linux Python < 3.13 are supported wheels for triton
-  TRITON_REQUIREMENT="triton==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"
+  TRITON_REQUIREMENT="pytorch-triton==${TRITON_VERSION}; ${TRITON_CONSTRAINT}"
   if [[ -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*dev.* ]]; then
       TRITON_SHORTHASH=$(cut -c1-10 $PYTORCH_ROOT/.ci/docker/ci_commit_pins/triton.txt)
       TRITON_REQUIREMENT="pytorch-triton==${TRITON_VERSION}+${TRITON_SHORTHASH}; ${TRITON_CONSTRAINT}"
@@ -100,18 +99,30 @@ if [[ "$PACKAGE_TYPE" =~ .*wheel.* && -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_B
     fi
 fi
 
-# Set triton via PYTORCH_EXTRA_INSTALL_REQUIREMENTS for triton xpu package
-if [[ "$PACKAGE_TYPE" =~ .*wheel.* && -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*xpu.* && $(uname) == "Linux" ]]; then
-    TRITON_REQUIREMENT="pytorch-triton-xpu==${TRITON_VERSION}"
-    if [[ -n "$PYTORCH_BUILD_VERSION" && "$PYTORCH_BUILD_VERSION" =~ .*dev.* ]]; then
-        TRITON_SHORTHASH=$(cut -c1-10 $PYTORCH_ROOT/.ci/docker/ci_commit_pins/triton-xpu.txt)
-        TRITON_REQUIREMENT="pytorch-triton-xpu==${TRITON_VERSION}+${TRITON_SHORTHASH}"
-    fi
-    if [[ -z "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]]; then
-        export PYTORCH_EXTRA_INSTALL_REQUIREMENTS="${TRITON_REQUIREMENT}"
-    else
-        export PYTORCH_EXTRA_INSTALL_REQUIREMENTS="${PYTORCH_EXTRA_INSTALL_REQUIREMENTS} | ${TRITON_REQUIREMENT}"
+JAVA_HOME=
+BUILD_JNI=OFF
+if [[ "$PACKAGE_TYPE" == libtorch ]]; then
+  POSSIBLE_JAVA_HOMES=()
+  POSSIBLE_JAVA_HOMES+=(/usr/local)
+  POSSIBLE_JAVA_HOMES+=(/usr/lib/jvm/java-8-openjdk-amd64)
+  POSSIBLE_JAVA_HOMES+=(/Library/Java/JavaVirtualMachines/*.jdk/Contents/Home)
+  # Add the Windows-specific JNI path
+  POSSIBLE_JAVA_HOMES+=("$PWD/pytorch/.circleci/windows-jni/")
+  for JH in "${POSSIBLE_JAVA_HOMES[@]}" ; do
+    if [[ -e "$JH/include/jni.h" ]] ; then
+      # Skip if we're not on Windows but haven't found a JAVA_HOME
+      if [[ "$JH" == "$PWD/pytorch/.circleci/windows-jni/" && "$OSTYPE" != "msys" ]] ; then
+        break
+      fi
+      echo "Found jni.h under $JH"
+      JAVA_HOME="$JH"
+      BUILD_JNI=ON
+      break
     fi
+  done
+  if [ -z "$JAVA_HOME" ]; then
+    echo "Did not find jni.h"
+  fi
 fi
 
 cat >"$envfile" <<EOL
@@ -124,7 +135,6 @@ export DESIRED_PYTHON="${DESIRED_PYTHON:-}"
 export DESIRED_CUDA="$DESIRED_CUDA"
 export LIBTORCH_VARIANT="${LIBTORCH_VARIANT:-}"
 export BUILD_PYTHONLESS="${BUILD_PYTHONLESS:-}"
-export USE_SPLIT_BUILD="${USE_SPLIT_BUILD:-}"
 if [[ "${OSTYPE}" == "msys" ]]; then
   export LIBTORCH_CONFIG="${LIBTORCH_CONFIG:-}"
   if [[ "${LIBTORCH_CONFIG:-}" == 'debug' ]]; then
@@ -148,6 +158,8 @@ export TORCH_CONDA_BUILD_FOLDER='pytorch-nightly'
 export ANACONDA_USER='pytorch'
 
 export USE_FBGEMM=1
+export JAVA_HOME=$JAVA_HOME
+export BUILD_JNI=$BUILD_JNI
 export PIP_UPLOAD_FOLDER="$PIP_UPLOAD_FOLDER"
 export DOCKER_IMAGE="$DOCKER_IMAGE"
 

.circleci/scripts/binary_upload.sh

@@ -25,15 +25,6 @@ if [[ "${DRY_RUN}" = "disabled" ]]; then
   AWS_S3_CP="aws s3 cp"
 fi
 
-if [[ "${USE_SPLIT_BUILD:-false}" == "true" ]]; then
-  UPLOAD_SUBFOLDER="${UPLOAD_SUBFOLDER}_pypi_pkg"
-fi
-
-# this is special build with all dependencies packaged
-if [[ ${BUILD_NAME} == *-full* ]]; then
-  UPLOAD_SUBFOLDER="${UPLOAD_SUBFOLDER}_full"
-fi
-
 # Sleep 2 minutes between retries for conda upload
 retry () {
   "$@"  || (sleep 5m && "$@") || (sleep 5m && "$@") || (sleep 5m && "$@") || (sleep 5m && "$@")

.circleci/scripts/trigger_azure_pipeline.py

@@ -8,7 +8,6 @@ import time
 
 import requests
 
-
 AZURE_PIPELINE_BASE_URL = "https://aiinfra.visualstudio.com/PyTorch/"
 AZURE_DEVOPS_PAT_BASE64 = os.environ.get("AZURE_DEVOPS_PAT_BASE64_SECRET", "")
 PIPELINE_ID = "911"

.clang-tidy

@@ -62,6 +62,4 @@ readability-string-compare,
 '
 HeaderFilterRegex: '^(aten/|c10/|torch/).*$'
 WarningsAsErrors: '*'
-CheckOptions:
-  misc-header-include-cycle.IgnoredFilesList: 'format.h;ivalue.h;custom_class.h;Dict.h;List.h'
 ...

.flake8

@@ -2,7 +2,7 @@
 # NOTE: **Mirror any changes** to this file the [tool.ruff] config in pyproject.toml
 # before we can fully move to use ruff
 enable-extensions = G
-select = B,C,E,F,G,P,SIM1,SIM911,T4,W,B9,TOR0,TOR1,TOR2,TOR9
+select = B,C,E,F,G,P,SIM1,T4,W,B9,TOR0,TOR1,TOR2,TOR9
 max-line-length = 120
 # C408 ignored because we like the dict keyword argument syntax
 # E501 is not flexible enough, we're using B950 instead

.git-blame-ignore-revs

@@ -40,7 +40,3 @@ e6ec0efaf87703c5f889cfc20b29be455885d58d
 a53cda1ddc15336dc1ff0ce1eff2a49cdc5f882e
 # 2024-01-02 clangformat: fused adam #116583
 9dc68d1aa9e554d09344a10fff69f7b50b2d23a0
-# 2024-06-28 enable UFMT in `torch/storage.py`
-d80939e5e9337e8078f11489afefec59fd42f93b
-# 2024-06-28 enable UFMT in `torch.utils.data`
-7cf0b90e49689d45be91aa539fdf54cf2ea8a9a3

.github/actionlint.yaml

@@ -47,5 +47,3 @@ self-hosted-runner:
     - macos-latest-xlarge
     - macos-13-xlarge
     - macos-14-xlarge
-    # Organization-wide Intel hosted XPU runners
-    - linux.idc.xpu

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

@@ -14,14 +14,12 @@ runs:
     - name: Cleans up diskspace
       shell: bash
       run: |
-        set -ex
         diskspace_cutoff=${{ inputs.diskspace-cutoff }}
-        docker_root_dir=$(docker info -f '{{.DockerRootDir}}')
-        diskspace=$(df -H --output=pcent ${docker_root_dir} | sed -n 2p | sed 's/%//' | sed 's/ //')
+        diskspace=$(df -H / --output=pcent | sed -n 2p | sed 's/%//' | sed 's/ //')
         msg="Please file an issue on pytorch/pytorch reporting the faulty runner. Include a link to the runner logs so the runner can be identified"
         if [[ "$diskspace" -ge "$diskspace_cutoff" ]] ; then
             docker system prune -af
-            diskspace_new=$(df -H --output=pcent ${docker_root_dir} | sed -n 2p | sed 's/%//' | sed 's/ //')
+            diskspace_new=$(df -H / --output=pcent | sed -n 2p | sed 's/%//' | sed 's/ //')
             if [[ "$diskspace_new" -gt "$diskspace_cutoff" ]] ; then
                 echo "Error: Available diskspace is less than $diskspace_cutoff percent. Not enough diskspace."
                 echo "$msg"

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

@@ -52,13 +52,6 @@ inputs:
     description: Hugging Face Hub token
     required: false
     default: ""
-  use_split_build:
-    description: |
-      [Experimental] Build a libtorch only wheel and build pytorch such that
-      are built from the libtorch wheel.
-    required: false
-    type: boolean
-    default: false
 outputs:
   docker-image:
     value: ${{ steps.calculate-docker-image.outputs.docker-image }}
@@ -151,7 +144,6 @@ runs:
         DEBUG: ${{ inputs.build-with-debug == 'true' && '1' || '0' }}
         OUR_GITHUB_JOB_ID: ${{ steps.get-job-id.outputs.job-id }}
         HUGGING_FACE_HUB_TOKEN: ${{ inputs.HUGGING_FACE_HUB_TOKEN }}
-        USE_SPLIT_BUILD: ${{ inputs.use_split_build }}
       shell: bash
       run: |
         # detached container should get cleaned up by teardown_ec2_linux
@@ -171,7 +163,6 @@ runs:
           -e PR_LABELS \
           -e OUR_GITHUB_JOB_ID \
           -e HUGGING_FACE_HUB_TOKEN \
-          -e USE_SPLIT_BUILD \
           --env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
           --security-opt seccomp=unconfined \
           --cap-add=SYS_PTRACE \
@@ -192,7 +183,7 @@ runs:
 
     - name: Store PyTorch Build Artifacts on S3
       uses: seemethere/upload-artifact-s3@v5
-      if: inputs.build-generates-artifacts == 'true' && steps.build.outcome != 'skipped' && inputs.use_split_build != 'true'
+      if: inputs.build-generates-artifacts == 'true' && steps.build.outcome != 'skipped'
       with:
         name: ${{ inputs.build-environment }}
         retention-days: 14
@@ -200,16 +191,6 @@ runs:
         path: artifacts.zip
         s3-bucket: ${{ inputs.s3-bucket }}
 
-    - name: Store PyTorch Build Artifacts on S3 for split build
-      uses: seemethere/upload-artifact-s3@v5
-      if: inputs.build-generates-artifacts == 'true' && steps.build.outcome != 'skipped' && inputs.use_split_build == 'true'
-      with:
-        name: ${{ inputs.build-environment }}-experimental-split-build
-        retention-days: 14
-        if-no-files-found: error
-        path: artifacts.zip
-        s3-bucket: ${{ inputs.s3-bucket }}
-
     - name: Upload sccache stats
       if: steps.build.outcome != 'skipped'
       uses: seemethere/upload-artifact-s3@v5

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

@@ -26,7 +26,6 @@ runs:
           -e PYTORCH_FINAL_PACKAGE_DIR \
           -e PYTORCH_ROOT \
           -e SKIP_ALL_TESTS \
-          -e USE_SPLIT_BUILD \
           --tty \
           --detach \
           -v "${GITHUB_WORKSPACE}/pytorch:/pytorch" \
@@ -36,8 +35,7 @@ runs:
           "${DOCKER_IMAGE}"
         )
 
-        echo "CONTAINER_NAME=${container_name}" >> "$GITHUB_ENV"
-        if [[ "${GPU_ARCH_TYPE}" != "rocm" && "${BUILD_ENVIRONMENT}" != "linux-aarch64-binary-manywheel" && "${BUILD_ENVIRONMENT}" != "linux-s390x-binary-manywheel" && "${GPU_ARCH_TYPE}" != "xpu" ]]; then
+        if [[ "${GPU_ARCH_TYPE}" != "rocm" && "${BUILD_ENVIRONMENT}" != "linux-aarch64-binary-manywheel" && "${BUILD_ENVIRONMENT}" != "linux-s390x-binary-manywheel" ]]; then
           # Propagate download.pytorch.org IP to container. This is only needed on Linux non aarch64 runner
           grep download.pytorch.org /etc/hosts | docker exec -i "${container_name}" bash -c "/bin/cat >> /etc/hosts"
         fi
@@ -48,9 +46,10 @@ runs:
         docker exec -t "${container_name}" bash -c "source ${BINARY_ENV_FILE} && bash -x /run.sh"
 
     - name: Cleanup docker
-      if: always() && (env.BUILD_ENVIRONMENT == 'linux-s390x-binary-manywheel' || env.GPU_ARCH_TYPE == 'xpu')
+      if: always() && env.BUILD_ENVIRONMENT == 'linux-s390x-binary-manywheel'
       shell: bash
       run: |
-        # on s390x or xpu stop the container for clean worker stop
+        # on s390x stop the container for clean worker stop
+        # ignore expansion of "docker ps -q" since it could be empty
         # shellcheck disable=SC2046
-        docker stop "${{ env.CONTAINER_NAME }}" || true
+        docker stop $(docker ps -q) || true

.github/ci_commit_pins/audio.txt

@@ -1 +1 @@
-69b2a0adc2ec03ab99990d7e8be3d4510438c148
+b829e936f7cc61b48149f5f957a451a38bf2a178

.github/ci_commit_pins/xla.txt

@@ -1 +1 @@
-6f0b61e5d782913a0fc7743812f2a8e522189111
+r2.4

.github/merge_rules.yaml

@@ -27,9 +27,11 @@
   - third_party/onnx
   - caffe2/python/onnx/**
   approved_by:
+  - BowenBao
   - justinchuby
   - liqunfu
   - shubhambhokare1
+  - thiagocrepaldi
   - titaiwangms
   - wschin
   - xadupre
@@ -242,7 +244,6 @@
   - torch/csrc/xpu/**
   - torch/xpu/**
   - test/xpu/**
-  - test/test_xpu.py
   - third_party/xpu.txt
   - .ci/docker/ci_commit_pins/triton-xpu.txt
   approved_by:
@@ -286,7 +287,6 @@
   - test/cpp/dist_autograd/**
   - test/cpp/rpc/**
   approved_by:
-  - wconstab
   - mrshenli
   - pritamdamania87
   - zhaojuanmao
@@ -313,25 +313,6 @@
   - Lint
   - pull
 
-- name: DCP
-  patterns:
-  - torch/distributed/checkpoint/**
-  approved_by:
-  - LucasLLC
-  - fegin
-  - wz337
-  - saumishr
-  - daulet-askarov
-  - pradeepdfb
-  - kirtiteja
-  - mhorowitz
-  - saiteja64
-  mandatory_checks_name:
-  - EasyCLA
-  - Lint
-  - pull
-
-
 - name: IDEEP
   patterns:
   - third_party/ideep
@@ -395,21 +376,13 @@
 
 - name: CPU inductor
   patterns:
-  - torch/_inductor/mkldnn_ir.py
   - torch/_inductor/mkldnn_lowerings.py
   - torch/_inductor/fx_passes/mkldnn_fusion.py
   - torch/_inductor/fx_passes/quantization.py
-  - torch/_inductor/codegen/cpp_prefix.h
   - torch/_inductor/codegen/cpp.py
-  - torch/_inductor/codegen/cpp_utils.py
-  - torch/_inductor/codegen/cpp_micro_gemm.py
-  - torch/_inductor/codegen/cpp_template_kernel.py
-  - torch/_inductor/codegen/cpp_template.py
-  - torch/_inductor/codegen/cpp_gemm_template.py
   - test/inductor/test_mkldnn_pattern_matcher.py
-  - test/inductor/test_cpu_repro.py
+  - test/inductor/test_cpu_repo.py
   - test/inductor/test_cpu_cpp_wrapper.py
-  - test/inductor/test_cpu_select_algorithm.py
   - aten/src/ATen/cpu/**
   - aten/src/ATen/native/quantized/cpu/**
   - test/quantization/core/test_quantized_op.py

.github/pytorch-probot.yml

@@ -26,4 +26,3 @@ retryable_workflows:
 - windows-binary
 labeler_config: labeler.yml
 label_to_label_config: label_to_label.yml
-mergebot: True

.github/scripts/build_triton_wheel.py

@@ -1,5 +1,4 @@
 #!/usr/bin/env python3
-
 import os
 import shutil
 import sys
@@ -8,17 +7,12 @@ from subprocess import check_call
 from tempfile import TemporaryDirectory
 from typing import Optional
 
-
 SCRIPT_DIR = Path(__file__).parent
 REPO_DIR = SCRIPT_DIR.parent.parent
 
 
-def read_triton_pin(device: str = "cuda") -> str:
-    triton_file = "triton.txt"
-    if device == "rocm":
-        triton_file = "triton-rocm.txt"
-    elif device == "xpu":
-        triton_file = "triton-xpu.txt"
+def read_triton_pin(rocm_hash: bool = False) -> str:
+    triton_file = "triton.txt" if not rocm_hash else "triton-rocm.txt"
     with open(REPO_DIR / ".ci" / "docker" / "ci_commit_pins" / triton_file) as f:
         return f.read().strip()
 
@@ -55,7 +49,7 @@ def build_triton(
     version: str,
     commit_hash: str,
     build_conda: bool = False,
-    device: str = "cuda",
+    build_rocm: bool = False,
     py_version: Optional[str] = None,
     release: bool = False,
 ) -> Path:
@@ -75,14 +69,11 @@ def build_triton(
         triton_basedir = Path(tmpdir) / "triton"
         triton_pythondir = triton_basedir / "python"
         triton_repo = "https://github.com/openai/triton"
-        if device == "rocm":
+        if build_rocm:
             triton_pkg_name = "pytorch-triton-rocm"
-        elif device == "xpu":
-            triton_pkg_name = "pytorch-triton-xpu"
-            triton_repo = "https://github.com/intel/intel-xpu-backend-for-triton"
         else:
             triton_pkg_name = "pytorch-triton"
-        check_call(["git", "clone", triton_repo, "triton"], cwd=tmpdir)
+        check_call(["git", "clone", triton_repo], cwd=tmpdir)
         if release:
             ver, rev, patch = version.split(".")
             check_call(
@@ -149,7 +140,7 @@ def build_triton(
             expected_version=None,
         )
 
-        if device == "rocm":
+        if build_rocm:
             check_call(
                 [f"{SCRIPT_DIR}/amd/package_triton_wheel.sh"],
                 cwd=triton_basedir,
@@ -164,7 +155,7 @@ def build_triton(
         whl_path = next(iter((triton_pythondir / "dist").glob("*.whl")))
         shutil.copy(whl_path, Path.cwd())
 
-        if device == "rocm":
+        if build_rocm:
             check_call(
                 [f"{SCRIPT_DIR}/amd/patch_triton_wheel.sh", Path.cwd()],
                 cwd=triton_basedir,
@@ -179,19 +170,17 @@ def main() -> None:
     parser = ArgumentParser("Build Triton binaries")
     parser.add_argument("--release", action="store_true")
     parser.add_argument("--build-conda", action="store_true")
-    parser.add_argument(
-        "--device", type=str, default="cuda", choices=["cuda", "rocm", "xpu"]
-    )
+    parser.add_argument("--build-rocm", action="store_true")
     parser.add_argument("--py-version", type=str)
     parser.add_argument("--commit-hash", type=str)
     parser.add_argument("--triton-version", type=str, default=read_triton_version())
     args = parser.parse_args()
 
     build_triton(
-        device=args.device,
+        build_rocm=args.build_rocm,
         commit_hash=args.commit_hash
         if args.commit_hash
-        else read_triton_pin(args.device),
+        else read_triton_pin(args.build_rocm),
         version=args.triton_version,
         build_conda=args.build_conda,
         py_version=args.py_version,

.github/scripts/check_labels.py

@@ -5,6 +5,7 @@ import sys
 from typing import Any
 
 from github_utils import gh_delete_comment, gh_post_pr_comment
+
 from gitutils import get_git_remote_name, get_git_repo_dir, GitRepo
 from label_utils import has_required_labels, is_label_err_comment, LABEL_ERR_MSG
 from trymerge import GitHubPR

.github/scripts/cherry_pick.py

@@ -3,10 +3,12 @@
 import json
 import os
 import re
-from typing import Any, cast, Dict, List, Optional
+from typing import Any, Optional
+
 from urllib.error import HTTPError
 
-from github_utils import gh_fetch_url, gh_post_pr_comment, gh_query_issues_by_labels
+from github_utils import gh_fetch_url, gh_post_pr_comment
+
 from gitutils import get_git_remote_name, get_git_repo_dir, GitRepo
 from trymerge import get_pr_commit_sha, GitHubPR
 
@@ -17,7 +19,6 @@ REQUIRES_ISSUE = {
     "critical",
     "fixnewfeature",
 }
-RELEASE_BRANCH_REGEX = re.compile(r"release/(?P<version>.+)")
 
 
 def parse_args() -> Any:
@@ -57,33 +58,6 @@ def get_merge_commit_sha(repo: GitRepo, pr: GitHubPR) -> Optional[str]:
     return commit_sha if pr.is_closed() else None
 
 
-def get_release_version(onto_branch: str) -> Optional[str]:
-    """
-    Return the release version if the target branch is a release branch
-    """
-    m = re.match(RELEASE_BRANCH_REGEX, onto_branch)
-    return m.group("version") if m else ""
-
-
-def get_tracker_issues(
-    org: str, project: str, onto_branch: str
-) -> List[Dict[str, Any]]:
-    """
-    Find the tracker issue from the repo. The tracker issue needs to have the title
-    like [VERSION] Release Tracker following the convention on PyTorch
-    """
-    version = get_release_version(onto_branch)
-    if not version:
-        return []
-
-    tracker_issues = gh_query_issues_by_labels(org, project, labels=["release tracker"])
-    if not tracker_issues:
-        return []
-
-    # Figure out the tracker issue from the list by looking at the title
-    return [issue for issue in tracker_issues if version in issue.get("title", "")]
-
-
 def cherry_pick(
     github_actor: str,
     repo: GitRepo,
@@ -103,49 +77,17 @@ def cherry_pick(
     )
 
     try:
-        org, project = repo.gh_owner_and_name()
-
-        cherry_pick_pr = ""
         if not dry_run:
+            org, project = repo.gh_owner_and_name()
             cherry_pick_pr = submit_pr(repo, pr, cherry_pick_branch, onto_branch)
 
-        tracker_issues_comments = []
-        tracker_issues = get_tracker_issues(org, project, onto_branch)
-        for issue in tracker_issues:
-            issue_number = int(str(issue.get("number", "0")))
-            if not issue_number:
-                continue
+            msg = f"The cherry pick PR is at {cherry_pick_pr}"
+            if fixes:
+                msg += f" and it is linked with issue {fixes}"
+            elif classification in REQUIRES_ISSUE:
+                msg += f" and it is recommended to link a {classification} cherry pick PR with an issue"
 
-            res = cast(
-                Dict[str, Any],
-                post_tracker_issue_comment(
-                    org,
-                    project,
-                    issue_number,
-                    pr.pr_num,
-                    cherry_pick_pr,
-                    classification,
-                    fixes,
-                    dry_run,
-                ),
-            )
-
-            comment_url = res.get("html_url", "")
-            if comment_url:
-                tracker_issues_comments.append(comment_url)
-
-        msg = f"The cherry pick PR is at {cherry_pick_pr}"
-        if fixes:
-            msg += f" and it is linked with issue {fixes}."
-        elif classification in REQUIRES_ISSUE:
-            msg += f" and it is recommended to link a {classification} cherry pick PR with an issue."
-
-        if tracker_issues_comments:
-            msg += " The following tracker issues are updated:\n"
-            for tracker_issues_comment in tracker_issues_comments:
-                msg += f"* {tracker_issues_comment}\n"
-
-        post_pr_comment(org, project, pr.pr_num, msg, dry_run)
+            post_comment(org, project, pr.pr_num, msg)
 
     finally:
         if current_branch:
@@ -217,9 +159,7 @@ def submit_pr(
         raise RuntimeError(msg) from error
 
 
-def post_pr_comment(
-    org: str, project: str, pr_num: int, msg: str, dry_run: bool = False
-) -> List[Dict[str, Any]]:
+def post_comment(org: str, project: str, pr_num: int, msg: str) -> None:
     """
     Post a comment on the PR itself to point to the cherry picking PR when success
     or print the error when failure
@@ -242,35 +182,7 @@ def post_pr_comment(
     comment = "\n".join(
         (f"### Cherry picking #{pr_num}", f"{msg}", "", f"{internal_debugging}")
     )
-    return gh_post_pr_comment(org, project, pr_num, comment, dry_run)
-
-
-def post_tracker_issue_comment(
-    org: str,
-    project: str,
-    issue_num: int,
-    pr_num: int,
-    cherry_pick_pr: str,
-    classification: str,
-    fixes: str,
-    dry_run: bool = False,
-) -> List[Dict[str, Any]]:
-    """
-    Post a comment on the tracker issue (if any) to record the cherry pick
-    """
-    comment = "\n".join(
-        (
-            "Link to landed trunk PR (if applicable):",
-            f"* https://github.com/{org}/{project}/pull/{pr_num}",
-            "",
-            "Link to release branch PR:",
-            f"* {cherry_pick_pr}",
-            "",
-            "Criteria Category:",
-            " - ".join((classification.capitalize(), fixes.capitalize())),
-        )
-    )
-    return gh_post_pr_comment(org, project, issue_num, comment, dry_run)
+    gh_post_pr_comment(org, project, pr_num, comment)
 
 
 def main() -> None:
@@ -302,7 +214,7 @@ def main() -> None:
 
     except RuntimeError as error:
         if not args.dry_run:
-            post_pr_comment(org, project, pr_num, str(error))
+            post_comment(org, project, pr_num, str(error))
         else:
             raise error
 

.github/scripts/close_nonexistent_disable_issues.py

@@ -10,7 +10,6 @@ import requests
 import rockset  # type: ignore[import]
 from gitutils import retries_decorator
 
-
 LOGS_QUERY = """
 with
     shas as (

.github/scripts/collect_ciflow_labels.py

@@ -1,12 +1,10 @@
 #!/usr/bin/env python3
-
 import sys
 from pathlib import Path
 from typing import Any, cast, Dict, List, Set
 
 import yaml
 
-
 GITHUB_DIR = Path(__file__).parent.parent
 
 

.github/scripts/convert_lintrunner_annotations_to_github.py

@@ -1,6 +1,7 @@
 import json
 import subprocess
 import sys
+
 from enum import Enum
 from pathlib import Path
 from typing import NamedTuple, Optional

.github/scripts/delete_old_branches.py

@@ -9,7 +9,6 @@ from typing import Any, Callable, Dict, List, Set
 from github_utils import gh_fetch_json_dict, gh_graphql
 from gitutils import GitRepo
 
-
 SEC_IN_DAY = 24 * 60 * 60
 CLOSED_PR_RETENTION = 30 * SEC_IN_DAY
 NO_PR_RETENTION = 1.5 * 365 * SEC_IN_DAY

.github/scripts/ensure_actions_will_cancel.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 
 import sys
+
 from pathlib import Path
 
 import yaml

.github/scripts/export_pytorch_labels.py

@@ -14,6 +14,7 @@ import json
 from typing import Any
 
 import boto3  # type: ignore[import]
+
 from label_utils import gh_get_labels
 
 

.github/scripts/filter_test_configs.py

@@ -15,7 +15,6 @@ from urllib.request import Request, urlopen
 
 import yaml
 
-
 REENABLE_TEST_REGEX = "(?i)(Close(d|s)?|Resolve(d|s)?|Fix(ed|es)?) (#|https://github.com/pytorch/pytorch/issues/)([0-9]+)"
 
 PREFIX = "test-config/"
@@ -39,9 +38,9 @@ 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=tIl0Qo224T_NDVw0dtG4hU1cZJM97inV"
 # 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=GPyRZRsOo26Gfk_WjAoNNxEMGXkIxIes"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"

.github/scripts/generate_binary_build_matrix.py

@@ -8,13 +8,11 @@ architectures:
     * CPU
     * Latest CUDA
     * Latest ROCM
-    * Latest XPU
 """
 
 import os
 from typing import Dict, List, Optional, Tuple
 
-
 CUDA_ARCHES = ["11.8", "12.1", "12.4"]
 
 
@@ -26,7 +24,6 @@ CUDA_ARCHES_CUDNN_VERSION = {"11.8": "9", "12.1": "9", "12.4": "9"}
 
 ROCM_ARCHES = ["6.0", "6.1"]
 
-XPU_ARCHES = ["xpu"]
 
 CPU_CXX11_ABI_ARCH = ["cpu-cxx11-abi"]
 
@@ -51,7 +48,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-curand-cu11==10.3.0.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusolver-cu11==11.4.1.48; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparse-cu11==11.7.5.86; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu11==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu11==2.20.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvtx-cu11==11.8.86; platform_system == 'Linux' and platform_machine == 'x86_64'"
     ),
     "12.1": (
@@ -64,7 +61,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.20.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'"
     ),
     "12.4": (
@@ -77,7 +74,7 @@ PYTORCH_EXTRA_INSTALL_REQUIREMENTS = {
         "nvidia-curand-cu12==10.3.5.119; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusolver-cu12==11.6.0.99; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-cusparse-cu12==12.3.0.142; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-        "nvidia-nccl-cu12==2.21.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+        "nvidia-nccl-cu12==2.20.5; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvtx-cu12==12.4.99; platform_system == 'Linux' and platform_machine == 'x86_64' | "
         "nvidia-nvjitlink-cu12==12.4.99; platform_system == 'Linux' and platform_machine == 'x86_64'"
     ),
@@ -135,8 +132,6 @@ def arch_type(arch_version: str) -> str:
         return "cuda"
     elif arch_version in ROCM_ARCHES:
         return "rocm"
-    elif arch_version in XPU_ARCHES:
-        return "xpu"
     elif arch_version in CPU_CXX11_ABI_ARCH:
         return "cpu-cxx11-abi"
     elif arch_version in CPU_AARCH64_ARCH:
@@ -161,7 +156,6 @@ WHEEL_CONTAINER_IMAGES = {
         gpu_arch: f"pytorch/manylinux-builder:rocm{gpu_arch}-{DEFAULT_TAG}"
         for gpu_arch in ROCM_ARCHES
     },
-    "xpu": f"pytorch/manylinux2_28-builder:xpu-{DEFAULT_TAG}",
     "cpu": f"pytorch/manylinux-builder:cpu-{DEFAULT_TAG}",
     "cpu-cxx11-abi": f"pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-{DEFAULT_TAG}",
     "cpu-aarch64": f"pytorch/manylinuxaarch64-builder:cpu-aarch64-{DEFAULT_TAG}",
@@ -227,7 +221,6 @@ def translate_desired_cuda(gpu_arch_type: str, gpu_arch_version: str) -> str:
         "cuda": f"cu{gpu_arch_version.replace('.', '')}",
         "cuda-aarch64": "cu124",
         "rocm": f"rocm{gpu_arch_version}",
-        "xpu": "xpu",
     }.get(gpu_arch_type, gpu_arch_version)
 
 
@@ -332,13 +325,13 @@ def generate_wheels_matrix(
         package_type = "manywheel"
 
     if python_versions is None:
-        python_versions = FULL_PYTHON_VERSIONS + ["3.13"]
+        python_versions = FULL_PYTHON_VERSIONS
 
     if arches is None:
         # Define default compute archivectures
         arches = ["cpu"]
         if os == "linux":
-            arches += CPU_CXX11_ABI_ARCH + CUDA_ARCHES + ROCM_ARCHES + XPU_ARCHES
+            arches += CPU_CXX11_ABI_ARCH + CUDA_ARCHES + ROCM_ARCHES
         elif os == "windows":
             arches += CUDA_ARCHES
         elif os == "linux-aarch64":
@@ -361,16 +354,9 @@ def generate_wheels_matrix(
                 or arch_version == "cpu-aarch64"
                 or arch_version == "cpu-s390x"
                 or arch_version == "cuda-aarch64"
-                or arch_version == "xpu"
                 else arch_version
             )
 
-            # TODO: Enable python 3.13 on rocm, xpu, aarch64, windows
-            if (
-                gpu_arch_type in ["rocm", "xpu"] or os != "linux"
-            ) and python_version == "3.13":
-                continue
-
             # 12.1 linux wheels require PYTORCH_EXTRA_INSTALL_REQUIREMENTS to install
             if (
                 arch_version in ["12.4", "12.1", "11.8"]
@@ -400,49 +386,6 @@ def generate_wheels_matrix(
                         ),
                     }
                 )
-                if arch_version != "cuda-aarch64":
-                    ret.append(
-                        {
-                            "python_version": python_version,
-                            "gpu_arch_type": gpu_arch_type,
-                            "gpu_arch_version": gpu_arch_version,
-                            "desired_cuda": translate_desired_cuda(
-                                gpu_arch_type, gpu_arch_version
-                            ),
-                            "use_split_build": "True",
-                            "devtoolset": "",
-                            "container_image": WHEEL_CONTAINER_IMAGES[arch_version],
-                            "package_type": package_type,
-                            "pytorch_extra_install_requirements": (
-                                PYTORCH_EXTRA_INSTALL_REQUIREMENTS[arch_version]  # fmt: skip
-                                if os != "linux-aarch64"
-                                else ""
-                            ),
-                            "build_name": f"{package_type}-py{python_version}-{gpu_arch_type}{gpu_arch_version}-split".replace(  # noqa: B950
-                                ".", "_"
-                            ),
-                        }
-                    )
-                    # Special build building to use on Colab. PyThon 3.10 for 12.1 CUDA
-                    if python_version == "3.10" and arch_version == "12.1":
-                        ret.append(
-                            {
-                                "python_version": python_version,
-                                "gpu_arch_type": gpu_arch_type,
-                                "gpu_arch_version": gpu_arch_version,
-                                "desired_cuda": translate_desired_cuda(
-                                    gpu_arch_type, gpu_arch_version
-                                ),
-                                "use_split_build": "False",
-                                "devtoolset": "",
-                                "container_image": WHEEL_CONTAINER_IMAGES[arch_version],
-                                "package_type": package_type,
-                                "pytorch_extra_install_requirements": "",
-                                "build_name": f"{package_type}-py{python_version}-{gpu_arch_type}{gpu_arch_version}-full".replace(  # noqa: B950
-                                    ".", "_"
-                                ),
-                            }
-                        )
             else:
                 ret.append(
                     {
@@ -453,9 +396,7 @@ def generate_wheels_matrix(
                             gpu_arch_type, gpu_arch_version
                         ),
                         "devtoolset": (
-                            "cxx11-abi"
-                            if arch_version in ["cpu-cxx11-abi", "xpu"]
-                            else ""
+                            "cxx11-abi" if arch_version == "cpu-cxx11-abi" else ""
                         ),
                         "container_image": WHEEL_CONTAINER_IMAGES[arch_version],
                         "package_type": package_type,

.github/scripts/generate_ci_workflows.py

@@ -8,8 +8,8 @@ from typing import Dict, Iterable, List, Literal, Set
 from typing_extensions import TypedDict  # Python 3.11+
 
 import generate_binary_build_matrix  # type: ignore[import]
-import jinja2
 
+import jinja2
 
 Arch = Literal["windows", "linux", "macos"]
 

.github/scripts/generate_docker_release_matrix.py

@@ -16,7 +16,6 @@ from typing import Dict, List
 
 import generate_binary_build_matrix
 
-
 DOCKER_IMAGE_TYPES = ["runtime", "devel"]
 
 

.github/scripts/generate_pytorch_version.py

@@ -4,11 +4,11 @@ import argparse
 import os
 import re
 import subprocess
+
 from datetime import datetime
 from distutils.util import strtobool
 from pathlib import Path
 
-
 LEADING_V_PATTERN = re.compile("^v")
 TRAILING_RC_PATTERN = re.compile("-rc[0-9]*$")
 LEGACY_BASE_VERSION_SUFFIX_PATTERN = re.compile("a0$")

.github/scripts/get_workflow_job_id.py

@@ -11,6 +11,7 @@ import sys
 import time
 import urllib
 import urllib.parse
+
 from typing import Any, Callable, Dict, List, Optional, Tuple
 from urllib.request import Request, urlopen
 

.github/scripts/get_workflow_type.py

@@ -0,0 +1,99 @@
+import json
+from argparse import ArgumentParser
+from typing import Any
+
+from github import Auth, Github
+from github.Issue import Issue
+
+
+WORKFLOW_TYPE_LABEL = "label"
+WORKFLOW_TYPE_RG = "rg"
+WORKFLOW_TYPE_BOTH = "both"
+
+
+def parse_args() -> Any:
+    parser = ArgumentParser("Get dynamic rollout settings")
+    parser.add_argument("--github-token", type=str, required=True, help="GitHub token")
+    parser.add_argument(
+        "--github-repo",
+        type=str,
+        required=False,
+        default="pytorch/test-infra",
+        help="GitHub repo to get the issue",
+    )
+    parser.add_argument(
+        "--github-issue", type=int, required=True, help="GitHub issue umber"
+    )
+    parser.add_argument(
+        "--github-user", type=str, required=True, help="GitHub username"
+    )
+    parser.add_argument(
+        "--github-branch", type=str, required=True, help="Current GitHub branch"
+    )
+
+    return parser.parse_args()
+
+
+def get_gh_client(github_token: str) -> Github:
+    auth = Auth.Token(github_token)
+    return Github(auth=auth)
+
+
+def get_issue(gh: Github, repo: str, issue_num: int) -> Issue:
+    repo = gh.get_repo(repo)
+    return repo.get_issue(number=issue_num)
+
+
+def is_exception_branch(branch: str) -> bool:
+    return branch.split("/")[0] in {"main", "nightly", "release", "landchecks"}
+
+
+def get_workflow_type(issue: Issue, username: str) -> str:
+    user_list = issue.get_comments()[0].body.split("\r\n")
+    try:
+        run_option = issue.get_comments()[1].body.split("\r\n")[0]
+    except Exception as e:
+        run_option = "single"
+
+    if user_list[0] == "!":
+        # Use old runners for everyone
+        return WORKFLOW_TYPE_LABEL
+    elif user_list[1] == "*":
+        if run_option == WORKFLOW_TYPE_BOTH:
+            # Use ARC runners and old runners for everyone
+            return WORKFLOW_TYPE_BOTH
+        else:
+            # Use only ARC runners for everyone
+            return WORKFLOW_TYPE_RG
+    elif username in user_list:
+        if run_option == WORKFLOW_TYPE_BOTH:
+            # Use ARC runners and old runners for a specific user
+            return WORKFLOW_TYPE_BOTH
+        else:
+            # Use only ARC runners for a specific user
+            return WORKFLOW_TYPE_RG
+    else:
+        # Use old runners by default
+        return WORKFLOW_TYPE_LABEL
+
+
+def main() -> None:
+    args = parse_args()
+
+    if is_exception_branch(args.github_branch):
+        output = {"workflow_type": WORKFLOW_TYPE_LABEL}
+    else:
+        try:
+            gh = get_gh_client(args.github_token)
+            issue = get_issue(gh, args.github_repo, args.github_issue)
+
+            output = {"workflow_type": get_workflow_type(issue, args.github_user)}
+        except Exception as e:
+            output = {"workflow_type": WORKFLOW_TYPE_LABEL}
+
+    json_output = json.dumps(output)
+    print(json_output)
+
+
+if __name__ == "__main__":
+    main()

.github/scripts/github_utils.py

@@ -3,6 +3,7 @@
 import json
 import os
 import warnings
+
 from dataclasses import dataclass
 from typing import Any, Callable, cast, Dict, List, Optional, Tuple, Union
 from urllib.error import HTTPError
@@ -201,12 +202,3 @@ def gh_update_pr_state(org: str, repo: str, pr_num: int, state: str = "open") ->
             )
         else:
             raise
-
-
-def gh_query_issues_by_labels(
-    org: str, repo: str, labels: List[str], state: str = "open"
-) -> List[Dict[str, Any]]:
-    url = f"{GITHUB_API_URL}/repos/{org}/{repo}/issues"
-    return gh_fetch_json(
-        url, method="GET", params={"labels": ",".join(labels), "state": state}
-    )

.github/scripts/gitutils.py

@@ -19,7 +19,6 @@ from typing import (
     Union,
 )
 
-
 T = TypeVar("T")
 
 RE_GITHUB_URL_MATCH = re.compile("^https://.*@?github.com/(.+)/(.+)$")

.github/scripts/label_utils.py

@@ -1,12 +1,12 @@
 """GitHub Label Utilities."""
 
 import json
+
 from functools import lru_cache
 from typing import Any, List, Tuple, TYPE_CHECKING, Union
 
 from github_utils import gh_fetch_url_and_headers, GitHubComment
 
-
 # TODO: this is a temp workaround to avoid circular dependencies,
 #       and should be removed once GitHubPR is refactored out of trymerge script.
 if TYPE_CHECKING:

.github/scripts/lintrunner.sh

@@ -29,7 +29,6 @@ python3 -m tools.pyi.gen_pyi \
     --native-functions-path aten/src/ATen/native/native_functions.yaml \
     --tags-path aten/src/ATen/native/tags.yaml \
     --deprecated-functions-path "tools/autograd/deprecated.yaml"
-python3 torch/utils/data/datapipes/gen_pyi.py
 
 RC=0
 # Run lintrunner on all files

.github/scripts/pytest_cache.py

@@ -9,7 +9,6 @@ from pytest_caching_utils import (
     upload_pytest_cache,
 )
 
-
 TEMP_DIR = "./tmp"  # a backup location in case one isn't provided
 
 

.github/scripts/pytest_caching_utils.py

@@ -14,7 +14,6 @@ from file_io_utils import (
     zip_folder,
 )
 
-
 PYTEST_CACHE_KEY_PREFIX = "pytest_cache"
 PYTEST_CACHE_DIR_NAME = ".pytest_cache"
 BUCKET = "gha-artifacts"

.github/scripts/runner_determinator.py

@@ -1,210 +0,0 @@
-# flake8: noqa: G004
-
-import logging
-import os
-from argparse import ArgumentParser
-from logging import LogRecord
-from typing import Any, Iterable
-
-from github import Auth, Github
-from github.Issue import Issue
-
-
-WORKFLOW_LABEL_META = ""  # use meta runners
-WORKFLOW_LABEL_LF = "lf."  # use runners from the linux foundation
-
-GITHUB_OUTPUT = os.getenv("GITHUB_OUTPUT", "")
-GH_OUTPUT_KEY_LABEL_TYPE = "label-type"
-
-
-class ColorFormatter(logging.Formatter):
-    """Color codes the log messages based on the log level"""
-
-    COLORS = {
-        "WARNING": "\033[33m",  # Yellow
-        "ERROR": "\033[31m",  # Red
-        "CRITICAL": "\033[31m",  # Red
-        "INFO": "\033[0m",  # Reset
-        "DEBUG": "\033[0m",  # Reset
-    }
-
-    def format(self, record: LogRecord) -> str:
-        log_color = self.COLORS.get(record.levelname, "\033[0m")  # Default to reset
-        record.msg = f"{log_color}{record.msg}\033[0m"
-        return super().format(record)
-
-
-handler = logging.StreamHandler()
-handler.setFormatter(ColorFormatter(fmt="%(levelname)-8s: %(message)s"))
-
-log = logging.getLogger(os.path.basename(__file__))
-log.addHandler(handler)
-log.setLevel(logging.INFO)
-
-
-def set_github_output(key: str, value: str) -> None:
-    """
-    Defines outputs of the github action that invokes this script
-    """
-    if not GITHUB_OUTPUT:
-        # See https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/ for deprecation notice
-        log.warning(
-            "No env var found for GITHUB_OUTPUT, you must be running this code locally. Falling back to the deprecated print method."
-        )
-        print(f"::set-output name={key}::{value}")
-        return
-
-    with open(GITHUB_OUTPUT, "a") as f:
-        log.info(f"Setting output: {key}='{value}'")
-        f.write(f"{key}={value}\n")
-
-
-def parse_args() -> Any:
-    parser = ArgumentParser("Get dynamic rollout settings")
-    parser.add_argument("--github-token", type=str, required=True, help="GitHub token")
-    parser.add_argument(
-        "--github-issue-repo",
-        type=str,
-        required=False,
-        default="pytorch/test-infra",
-        help="GitHub repo to get the issue",
-    )
-    parser.add_argument(
-        "--github-repo",
-        type=str,
-        required=True,
-        help="GitHub repo where CI is running",
-    )
-    parser.add_argument(
-        "--github-issue", type=int, required=True, help="GitHub issue number"
-    )
-    parser.add_argument(
-        "--github-actor", type=str, required=True, help="GitHub triggering_actor"
-    )
-    parser.add_argument(
-        "--github-issue-owner", type=str, required=True, help="GitHub issue owner"
-    )
-    parser.add_argument(
-        "--github-branch", type=str, required=True, help="Current GitHub branch or tag"
-    )
-    parser.add_argument(
-        "--github-ref-type",
-        type=str,
-        required=True,
-        help="Current GitHub ref type, branch or tag",
-    )
-
-    return parser.parse_args()
-
-
-def get_gh_client(github_token: str) -> Github:
-    auth = Auth.Token(github_token)
-    return Github(auth=auth)
-
-
-def get_issue(gh: Github, repo: str, issue_num: int) -> Issue:
-    repo = gh.get_repo(repo)
-    return repo.get_issue(number=issue_num)
-
-
-def get_potential_pr_author(
-    gh: Github, repo: str, username: str, ref_type: str, ref_name: str
-) -> str:
-    # If the trigger was a new tag added by a bot, this is a ciflow case
-    # Fetch the actual username from the original PR. The PR number is
-    # embedded in the tag name: ciflow/<name>/<pr-number>
-    if username == "pytorch-bot[bot]" and ref_type == "tag":
-        split_tag = ref_name.split("/")
-        if (
-            len(split_tag) == 3
-            and split_tag[0] == "ciflow"
-            and split_tag[2].isnumeric()
-        ):
-            pr_number = split_tag[2]
-            try:
-                repository = gh.get_repo(repo)
-                pull = repository.get_pull(number=int(pr_number))
-            except Exception as e:
-                raise Exception(  # noqa: TRY002
-                    f"issue with pull request {pr_number} from repo {repository}"
-                ) from e
-            return pull.user.login
-    # In all other cases, return the original input username
-    return username
-
-
-def is_exception_branch(branch: str) -> bool:
-    return branch.split("/")[0] in {"main", "nightly", "release", "landchecks"}
-
-
-def get_workflow_type(issue: Issue, workflow_requestors: Iterable[str]) -> str:
-    try:
-        first_comment = issue.get_comments()[0].body.strip("\n\t ")
-
-        if first_comment[0] == "!":
-            log.info("LF Workflows are disabled for everyone. Using meta runners.")
-            return WORKFLOW_LABEL_META
-        elif first_comment[0] == "*":
-            log.info("LF Workflows are enabled for everyone. Using LF runners.")
-            return WORKFLOW_LABEL_LF
-        else:
-            all_opted_in_users = {
-                usr_raw.strip("\n\t@ ") for usr_raw in first_comment.split()
-            }
-            opted_in_requestors = {
-                usr for usr in workflow_requestors if usr in all_opted_in_users
-            }
-            if opted_in_requestors:
-                log.info(
-                    f"LF Workflows are enabled for {', '.join(opted_in_requestors)}. Using LF runners."
-                )
-                return WORKFLOW_LABEL_LF
-            else:
-                log.info(
-                    f"LF Workflows are disabled for {', '.join(workflow_requestors)}. Using meta runners."
-                )
-                return WORKFLOW_LABEL_META
-
-    except Exception as e:
-        log.error(
-            f"Failed to get determine workflow type. Falling back to meta runners. Exception: {e}"
-        )
-        return WORKFLOW_LABEL_META
-
-
-def main() -> None:
-    args = parse_args()
-
-    if args.github_ref_type == "branch" and is_exception_branch(args.github_branch):
-        log.info(f"Exception branch: '{args.github_branch}', using meta runners")
-        label_type = WORKFLOW_LABEL_META
-    else:
-        try:
-            gh = get_gh_client(args.github_token)
-            # The default issue we use - https://github.com/pytorch/test-infra/issues/5132
-            issue = get_issue(gh, args.github_issue_repo, args.github_issue)
-            username = get_potential_pr_author(
-                gh,
-                args.github_repo,
-                args.github_actor,
-                args.github_ref_type,
-                args.github_branch,
-            )
-            label_type = get_workflow_type(
-                issue,
-                (
-                    args.github_issue_owner,
-                    username,
-                ),
-            )
-        except Exception as e:
-            log.error(
-                f"Failed to get issue. Falling back to meta runners. Exception: {e}"
-            )
-            label_type = WORKFLOW_LABEL_META
-
-    set_github_output(GH_OUTPUT_KEY_LABEL_TYPE, label_type)
-
-
-if __name__ == "__main__":
-    main()

.github/scripts/sync_distributed_folder_prototype.sh

@@ -2,7 +2,7 @@
 
 set -eoux pipefail
 
-SYNC_BRANCH=pytorch-stable-prototype
+SYNC_BRANCH=fbcode/pytorch-stable-prototype
 
 git config user.email "fake@example.com"
 git config user.name  "PyTorch Stable Bot"
@@ -11,9 +11,7 @@ git fetch origin main
 git fetch origin "$SYNC_BRANCH"
 git checkout "$SYNC_BRANCH"
 
-# Using a hardcoded SHA here is a massive speedup as we can skip the entire history of the pytorch GitHub repo.
-# This specific SHA was chosen as it was before the "branch point" of the stable branch
-for SHA in $(git log ba3b05fdf37ddbc3c301294d6a560a816335e717..origin/main --pretty="%h" --reverse -- torch/distributed torch/csrc/distributed test/distributed test/cpp/c10d benchmarks/distributed)
+for SHA in $(git log 4333e122d4b74cdf84351ed2907045c6a767b4cd..origin/main --pretty="%h" --reverse -- torch/distributed torch/csrc/distributed test/distributed test/cpp/c10d benchmarks/distributed)
 do
     # `git merge-base --is-ancestor` exits with code 0 if the given SHA is an ancestor, and non-0 otherwise
     if git merge-base --is-ancestor $SHA HEAD || [[ $(git log --grep="(cherry picked from commit $SHA") ]]
@@ -22,12 +20,7 @@ do
         continue
     fi
     echo "Copying $SHA"
-    git cherry-pick -x "$SHA" -X theirs
-    git reset --soft HEAD~1
-    git add torch/distributed torch/csrc/distributed test/distributed test/cpp/c10d benchmarks/distributed
-    git checkout .
-    git commit --reuse-message=HEAD@{1}
-    git clean -f
+    git cherry-pick -x "$SHA"
 done
 
 if [[ "${WITH_PUSH}" == true ]]; then

.github/scripts/tag_docker_images_for_release.py

@@ -41,7 +41,7 @@ def main() -> None:
     )
 
     options = parser.parse_args()
-    tagged_images: Dict[str, bool] = {}
+    tagged_images: Dict[str, bool] = dict()
     platform_images = [
         generate_binary_build_matrix.WHEEL_CONTAINER_IMAGES,
         generate_binary_build_matrix.LIBTORCH_CONTAINER_IMAGES,

.github/scripts/td_llm_indexer.sh

@@ -7,7 +7,6 @@ cd llm-target-determinator
 pip install -q -r requirements.txt
 cd ../codellama
 pip install -e .
-pip install numpy==1.26.0
 
 # Run indexer
 cd ../llm-target-determinator

.github/scripts/test_trymerge.py

@@ -17,7 +17,9 @@ from unittest import main, mock, skip, TestCase
 from urllib.error import HTTPError
 
 from github_utils import gh_graphql
+
 from gitutils import get_git_remote_name, get_git_repo_dir, GitRepo
+
 from trymerge import (
     categorize_checks,
     DRCI_CHECKRUN_NAME,
@@ -37,7 +39,6 @@ from trymerge import (
     validate_revert,
 )
 
-
 if "GIT_REMOTE_URL" not in os.environ:
     os.environ["GIT_REMOTE_URL"] = "https://github.com/pytorch/pytorch"
 
@@ -179,9 +180,6 @@ def mock_gh_get_info() -> Any:
     return {
         "closed": False,
         "isCrossRepository": False,
-        "headRefName": "foo",
-        "baseRefName": "bar",
-        "baseRepository": {"defaultBranchRef": {"name": "bar"}},
         "files": {"nodes": [], "pageInfo": {"hasNextPage": False}},
         "changedFiles": 0,
     }
@@ -396,7 +394,6 @@ class TestTryMerge(TestCase):
         # self.assertGreater(len(pr.get_checkrun_conclusions()), 3)
         self.assertGreater(pr.get_commit_count(), 60)
 
-    @skip("GitHub doesn't keep this data anymore")
     def test_gql_retrieve_checksuites(self, *args: Any) -> None:
         "Fetch comments and conclusions for PR with 60 commits"
         pr = GitHubPR("pytorch", "pytorch", 94787)
@@ -894,24 +891,6 @@ class TestBypassFailures(TestCase):
         self.assertTrue(len(ignorable["FLAKY"]) == 1)
         self.assertTrue(len(ignorable["BROKEN_TRUNK"]) == 0)
 
-    def test_ignore_failures_older_run_same_workflow(self, *args: Any) -> None:
-        pr = GitHubPR("pytorch", "pytorch", 129013)
-        checks = pr.get_checkrun_conclusions()
-        checks = get_classifications(
-            pr.pr_num,
-            pr.project,
-            checks,
-            [],
-        )
-        pending, failed, ignorable = categorize_checks(
-            checks,
-            list(checks.keys()),
-        )
-        self.assertTrue(len(pending) == 0)
-        self.assertTrue(len(failed) == 0)
-        self.assertTrue(len(ignorable["FLAKY"]) == 2)
-        self.assertTrue(len(ignorable["UNSTABLE"]) == 13)
-
     @mock.patch("trymerge.read_merge_rules", side_effect=xla_merge_rules)
     def test_dont_ignore_flaky_failures(self, *args: Any) -> None:
         """
@@ -1040,7 +1019,7 @@ class TestGitHubPRGhstackDependencies(TestCase):
         )
 
     @skip(
-        reason="This test is run against a mutable PR that has changed, so it no longer works. The test should be changed"
+        reason="This test is run against a mutalbe PR that has changed, so it no longer works. The test should be changed"
     )
     @mock.patch("trymerge.read_merge_rules")
     @mock.patch("trymerge.GitRepo")

.github/scripts/trymerge.py

@@ -45,6 +45,7 @@ from github_utils import (
     gh_update_pr_state,
     GitHubComment,
 )
+
 from gitutils import (
     are_ghstack_branches_in_sync,
     get_git_remote_name,
@@ -61,7 +62,6 @@ from label_utils import (
 )
 from trymerge_explainer import get_revert_message, TryMergeExplainer
 
-
 # labels
 MERGE_IN_PROGRESS_LABEL = "merging"
 MERGE_COMPLETE_LABEL = "merged"
@@ -81,10 +81,9 @@ JobNameToStateDict = Dict[str, JobCheckState]
 
 
 class WorkflowCheckState:
-    def __init__(self, name: str, url: str, run_id: int, status: Optional[str]):
+    def __init__(self, name: str, url: str, status: Optional[str]):
         self.name: str = name
         self.url: str = url
-        self.run_id: int = run_id
         self.status: Optional[str] = status
         self.jobs: JobNameToStateDict = {}
 
@@ -123,7 +122,6 @@ fragment PRCheckSuites on CheckSuiteConnection {
       workflowRun {
         workflow {
           name
-          databaseId
         }
         databaseId
         url
@@ -514,7 +512,7 @@ def add_workflow_conclusions(
     workflows: Dict[str, WorkflowCheckState] = {}
 
     # for the jobs that don't have a workflow
-    no_workflow_obj: WorkflowCheckState = WorkflowCheckState("", "", 0, None)
+    no_workflow_obj: WorkflowCheckState = WorkflowCheckState("", "", None)
 
     def add_conclusions(edges: Any) -> None:
         for edge_idx, edge in enumerate(edges):
@@ -525,30 +523,18 @@ def add_workflow_conclusions(
             workflow_obj: WorkflowCheckState = no_workflow_obj
 
             if workflow_run is not None:
-                # This is the usual workflow run ID we see on GitHub
-                workflow_run_id = workflow_run["databaseId"]
-                # While this is the metadata name and ID of the workflow itself
                 workflow_name = workflow_run["workflow"]["name"]
-                workflow_id = workflow_run["workflow"]["databaseId"]
-
                 workflow_conclusion = node["conclusion"]
                 # Do not override existing status with cancelled
                 if workflow_conclusion == "CANCELLED" and workflow_name in workflows:
                     continue
-
-                # Only keep the latest workflow run for each workflow, heuristically,
-                # it's the run with largest run ID
-                if (
-                    workflow_id not in workflows
-                    or workflows[workflow_id].run_id < workflow_run_id
-                ):
-                    workflows[workflow_id] = WorkflowCheckState(
+                if workflow_name not in workflows:
+                    workflows[workflow_name] = WorkflowCheckState(
                         name=workflow_name,
                         status=workflow_conclusion,
                         url=workflow_run["url"],
-                        run_id=workflow_run_id,
                     )
-                workflow_obj = workflows[workflow_id]
+                workflow_obj = workflows[workflow_name]
 
             while checkruns is not None:
                 for checkrun_node in checkruns["nodes"]:
@@ -586,12 +572,12 @@ def add_workflow_conclusions(
     # the jobs in but don't put the workflow in.  We care more about the jobs in
     # the workflow that ran than the container workflow.
     res: JobNameToStateDict = {}
-    for workflow in workflows.values():
+    for workflow_name, workflow in workflows.items():
         if len(workflow.jobs) > 0:
             for job_name, job in workflow.jobs.items():
                 res[job_name] = job
         else:
-            res[workflow.name] = JobCheckState(
+            res[workflow_name] = JobCheckState(
                 workflow.name,
                 workflow.url,
                 workflow.status,
@@ -1177,6 +1163,7 @@ class GitHubPR:
             # Finally, upload the record to Rockset. The list of pending and failed
             # checks are at the time of the merge
             save_merge_record(
+                collection=ROCKSET_MERGES_COLLECTION,
                 comment_id=comment_id,
                 pr_num=self.pr_num,
                 owner=self.org,
@@ -1192,8 +1179,10 @@ class GitHubPR:
                 merge_base_sha=self.get_merge_base(),
                 merge_commit_sha=merge_commit_sha,
                 is_failed=False,
+                dry_run=dry_run,
                 skip_mandatory_checks=skip_mandatory_checks,
                 ignore_current=bool(ignore_current_checks),
+                workspace=ROCKSET_MERGES_WORKSPACE,
             )
         else:
             print("Missing comment ID or PR number, couldn't upload to Rockset")
@@ -1500,6 +1489,7 @@ def checks_to_markdown_bullets(
 
 @retries_decorator()
 def save_merge_record(
+    collection: str,
     comment_id: int,
     pr_num: int,
     owner: str,
@@ -1515,44 +1505,59 @@ def save_merge_record(
     merge_base_sha: str,
     merge_commit_sha: str = "",
     is_failed: bool = False,
+    dry_run: bool = False,
     skip_mandatory_checks: bool = False,
     ignore_current: bool = False,
     error: str = "",
+    workspace: str = "commons",
 ) -> None:
     """
-    This saves the merge records as a json, which can later be uploaded to s3
+    This saves the merge records into Rockset, so we can query them (for fun and profit)
     """
+    if dry_run:
+        # Decide not to save the record to Rockset if dry-run is set to not pollute
+        # the collection
+        return
 
-    # Prepare the record to be written into Rockset
-    data = [
-        {
-            "comment_id": comment_id,
-            "pr_num": pr_num,
-            "owner": owner,
-            "project": project,
-            "author": author,
-            "pending_checks": pending_checks,
-            "failed_checks": failed_checks,
-            "ignore_current_checks": ignore_current_checks,
-            "broken_trunk_checks": broken_trunk_checks,
-            "flaky_checks": flaky_checks,
-            "unstable_checks": unstable_checks,
-            "last_commit_sha": last_commit_sha,
-            "merge_base_sha": merge_base_sha,
-            "merge_commit_sha": merge_commit_sha,
-            "is_failed": is_failed,
-            "skip_mandatory_checks": skip_mandatory_checks,
-            "ignore_current": ignore_current,
-            "error": error,
-            # This is a unique identifier for the record for deduping purposes
-            # in rockset.  Any unique string would work
-            "_id": f"{project}-{pr_num}-{comment_id}-{os.environ.get('GITHUB_RUN_ID')}",
-        }
-    ]
-    repo_root = Path(__file__).resolve().parent.parent.parent
+    try:
+        import rockset  # type: ignore[import]
 
-    with open(repo_root / "merge_record.json", "w") as f:
-        json.dump(data, f)
+        # Prepare the record to be written into Rockset
+        data = [
+            {
+                "comment_id": comment_id,
+                "pr_num": pr_num,
+                "owner": owner,
+                "project": project,
+                "author": author,
+                "pending_checks": pending_checks,
+                "failed_checks": failed_checks,
+                "ignore_current_checks": ignore_current_checks,
+                "broken_trunk_checks": broken_trunk_checks,
+                "flaky_checks": flaky_checks,
+                "unstable_checks": unstable_checks,
+                "last_commit_sha": last_commit_sha,
+                "merge_base_sha": merge_base_sha,
+                "merge_commit_sha": merge_commit_sha,
+                "is_failed": is_failed,
+                "skip_mandatory_checks": skip_mandatory_checks,
+                "ignore_current": ignore_current,
+                "error": error,
+            }
+        ]
+
+        client = rockset.RocksetClient(
+            host="api.usw2a1.rockset.com", api_key=os.environ["ROCKSET_API_KEY"]
+        )
+        client.Documents.add_documents(
+            collection=collection,
+            data=data,
+            workspace=workspace,
+        )
+
+    except ModuleNotFoundError:
+        print("Rockset is missing, no record will be saved")
+        return
 
 
 @retries_decorator(rc=[])
@@ -2325,15 +2330,6 @@ def main() -> None:
             dry_run=args.dry_run,
         )
         return
-    if not pr.is_ghstack_pr() and pr.base_ref() != pr.default_branch():
-        gh_post_pr_comment(
-            org,
-            project,
-            args.pr_num,
-            f"PR targets {pr.base_ref()} rather than {pr.default_branch()}, refusing merge request",
-            dry_run=args.dry_run,
-        )
-        return
 
     if args.check_mergeability:
         if pr.is_ghstack_pr():
@@ -2369,6 +2365,7 @@ def main() -> None:
             # list of pending and failed checks here, but they are not really
             # needed at the moment
             save_merge_record(
+                collection=ROCKSET_MERGES_COLLECTION,
                 comment_id=args.comment_id,
                 pr_num=args.pr_num,
                 owner=org,
@@ -2383,9 +2380,11 @@ def main() -> None:
                 last_commit_sha=pr.last_commit().get("oid", ""),
                 merge_base_sha=pr.get_merge_base(),
                 is_failed=True,
+                dry_run=args.dry_run,
                 skip_mandatory_checks=args.force,
                 ignore_current=args.ignore_current,
                 error=str(e),
+                workspace=ROCKSET_MERGES_WORKSPACE,
             )
         else:
             print("Missing comment ID or PR number, couldn't upload to Rockset")

.github/scripts/tryrebase.py

@@ -11,7 +11,6 @@ from github_utils import gh_post_pr_comment as gh_post_comment
 from gitutils import get_git_remote_name, get_git_repo_dir, GitRepo
 from trymerge import GitHubPR
 
-
 SAME_SHA_ERROR = (
     "\n```\nAborting rebase because rebasing the branch resulted in the same sha as the target branch.\n"
     + "This usually happens because the PR has already been merged.  Please rebase locally and push.\n```"

.github/templates/common.yml.j2

@@ -8,7 +8,7 @@
 # NOTE: If testing pytorch/builder changes you can change this variable to change what pytorch/builder reference
 #       the binary builds will check out
 {%- set builder_repo = "pytorch/builder" -%}
-{%- set builder_branch = "main" -%}
+{%- set builder_branch = "release/2.4" -%}
 
 {%- macro concurrency(build_environment) -%}
 concurrency:

.github/templates/linux_binary_build_workflow.yml.j2

@@ -81,7 +81,7 @@ jobs:
   !{{ config["build_name"] }}-test:  # Testing
     if: ${{ github.repository_owner == 'pytorch' }}
     needs: !{{ config["build_name"] }}-build
-    {%- if config["gpu_arch_type"] not in ["rocm", "xpu"] %}
+    {%- if config["gpu_arch_type"] != "rocm" %}
     uses: ./.github/workflows/_binary-test-linux.yml
     with:!{{ upload.binary_env_as_input(config) }}
       build_name: !{{ config["build_name"] }}
@@ -101,40 +101,6 @@ jobs:
       {%- endif %}
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-    {%- elif config["gpu_arch_type"] == "xpu" %}
-    runs-on: linux.idc.xpu
-    timeout-minutes: !{{ common.timeout_minutes }}
-    !{{ upload.binary_env(config) }}
-    permissions:
-      id-token: write
-      contents: read
-    steps:
-      - name: Setup XPU
-        uses: ./.github/actions/setup-xpu
-      - name: configure aws credentials
-        id: aws_creds
-        uses: aws-actions/configure-aws-credentials@v1.7.0
-        with:
-          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
-          aws-region: us-east-1
-      - name: Login to Amazon ECR
-        id: login-ecr
-        uses: aws-actions/amazon-ecr-login@v2
-      - uses: !{{ common.download_artifact_action }}
-        name: Download Build Artifacts
-        with:
-          name: !{{ config["build_name"] }}
-          path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
-      - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
-        with:
-          docker-image: !{{ config["container_image"] }}
-      - name: Test Pytorch binary
-        uses: ./pytorch/.github/actions/test-pytorch-binary
-      - name: Teardown XPU
-        uses: ./.github/actions/teardown-xpu
     {%- else %}
     runs-on: linux.rocm.gpu
     timeout-minutes: !{{ common.timeout_minutes }}
@@ -147,8 +113,8 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ runner.temp }}/artifacts/"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: ROCm set GPU_FLAG
         run: |
           echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"

.github/templates/macos_binary_build_workflow.yml.j2

@@ -81,8 +81,8 @@ jobs:
           elif [ -d "/Applications/Xcode_13.3.1.app" ]; then
             echo "DEVELOPER_DIR=/Applications/Xcode_13.3.1.app/Contents/Developer" >> "${GITHUB_ENV}"
           fi
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: Install sccache (only for non-forked PRs, and pushes to trunk)
         uses: nick-fields/retry@v2.8.2
         if: ${{ github.event_name == 'push' || github.event.pull_request.head.repo.full_name == github.repository }}

.github/templates/upload.yml.j2

@@ -30,9 +30,6 @@
   {%- if config["devtoolset"] %}
       DESIRED_DEVTOOLSET: !{{ config["devtoolset"] }}
   {%- endif %}
-  {%- if config.use_split_build is defined %}
-      use_split_build: !{{ config["use_split_build"] }}
-  {%- endif %}
 {%- endif %}
 {%- if config["package_type"] == "libtorch" %}
   {%- if config["libtorch_config"] %}
@@ -47,7 +44,6 @@
       # without this value pip does not get installed for some reason
       DESIRED_PYTHON: "3.8"
   {%- endif %}
-
 {%- else %}
       DESIRED_PYTHON: "!{{ config["python_version"] }}"
 {%- endif %}

.github/templates/windows_binary_build_workflow.yml.j2

@@ -65,8 +65,8 @@ jobs:
     steps:
       !{{ common.setup_ec2_windows() }}
       !{{ set_runner_specific_vars() }}
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: Populate binary env
         shell: bash
         run: |
@@ -105,8 +105,8 @@ jobs:
         with:
           name: !{{ config["build_name"] }}
           path: "${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
-      !{{ common.checkout(deep_clone=False, directory="pytorch") }}
-      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch) }}
+      !{{ common.checkout(deep_clone=False, directory="pytorch", checkout_pr_head=False) }}
+      !{{ common.checkout(deep_clone=False, directory="builder", repository=common.builder_repo, branch=common.builder_branch, checkout_pr_head=False) }}
       - name: Populate binary env
         shell: bash
         run: |

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

@@ -37,7 +37,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -59,25 +59,25 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         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/2.4
 
       - 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/2.4
         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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -141,5 +141,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_android-full-build-test.yml

@@ -37,7 +37,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -59,25 +59,25 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         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/2.4
 
       - 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/2.4
         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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -186,5 +186,5 @@ jobs:
         if: always()
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

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

@@ -27,11 +27,6 @@ on:
         type: string
         description: |
           A JSON description of what configs to run later on.
-      runner:
-        required: false
-        type: string
-        default: "linux.large"
-        description: Runner type
 
 env:
   GIT_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
@@ -39,7 +34,7 @@ env:
 jobs:
   filter:
     if: github.repository_owner == 'pytorch'
-    runs-on: ${{ inputs.runner }}
+    runs-on: [self-hosted, linux.large]
     outputs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
       is-test-matrix-empty: ${{ steps.filter.outputs.is-test-matrix-empty }}
@@ -47,7 +42,7 @@ jobs:
       reenabled-issues: ${{ steps.filter.outputs.reenabled-issues }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -69,25 +64,25 @@ 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/2.4
         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/2.4
 
       - 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/2.4
         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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -97,7 +92,7 @@ jobs:
         run: echo "IN_ARC_RUNNER=$([ -f /.inarc ] && echo true || echo false)" >> "$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/2.4
         if: ${{ inputs.cuda-version != 'cpu' && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
       - name: Output disk space left
@@ -206,5 +201,5 @@ 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/2.4
         if: always()

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

@@ -21,13 +21,6 @@ on:
         default: 210
         type: number
         description: timeout for the job
-      use_split_build:
-        description: |
-          [Experimental] Build a libtorch only wheel and build pytorch such that
-          are built from the libtorch wheel.
-        required: false
-        type: boolean
-        default: false
       ALPINE_IMAGE:
         required: false
         type: string
@@ -117,7 +110,6 @@ jobs:
       PR_NUMBER: ${{ github.event.pull_request.number }}
       PYTORCH_FINAL_PACKAGE_DIR: /artifacts
       SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
-      USE_SPLIT_BUILD: ${{ inputs.use_split_build }}
     steps:
       - name: Make the env permanent during this workflow (but not the secrets)
         shell: bash
@@ -145,7 +137,6 @@ jobs:
             echo "PR_NUMBER=${{ env.PR_NUMBER }}"
             echo "PYTORCH_FINAL_PACKAGE_DIR=${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
             echo "SHA1=${{ env.SHA1 }}"
-            echo "USE_SPLIT_BUILD=${{ env.use_split_build }}"
           } >> "${GITHUB_ENV} }}"
 
       - name: List the env
@@ -154,13 +145,13 @@ 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/2.4
         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/2.4
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -190,7 +181,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
           quiet-checkout: true
@@ -204,7 +194,7 @@ jobs:
       - name: Checkout pytorch/builder to builder dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -230,7 +220,7 @@ jobs:
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && inputs.build_environment != 'linux-s390x-binary-manywheel' }}
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.4
         with:
           docker-image: ${{ inputs.DOCKER_IMAGE }}
 
@@ -255,7 +245,6 @@ jobs:
             -e PYTORCH_ROOT \
             -e SKIP_ALL_TESTS \
             -e PYTORCH_EXTRA_INSTALL_REQUIREMENTS \
-            -e USE_SPLIT_BUILD \
             --tty \
             --detach \
             -v "${GITHUB_WORKSPACE}/pytorch:/pytorch" \
@@ -288,7 +277,7 @@ 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/2.4
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

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

@@ -63,13 +63,6 @@ on:
         required: true
         type: string
         description: Hardware to run this job on. Valid values are linux.4xlarge, linux.4xlarge.nvidia.gpu, linux.arm64.2xlarge, and linux.rocm.gpu
-      use_split_build:
-        description: |
-          [Experimental] Build a libtorch only wheel and build pytorch such that
-          are built from the libtorch wheel.
-        required: false
-        type: boolean
-        default: false
     secrets:
       github-token:
         required: true
@@ -104,7 +97,6 @@ jobs:
       PR_NUMBER: ${{ github.event.pull_request.number }}
       PYTORCH_FINAL_PACKAGE_DIR: /artifacts
       SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
-      USE_SPLIT_BUILD: ${{ inputs.use_split_build }}
     steps:
       - name: Make the env permanent during this workflow (but not the secrets)
         shell: bash
@@ -132,19 +124,18 @@ jobs:
             echo "PR_NUMBER=${{ env.PR_NUMBER }}"
             echo "PYTORCH_FINAL_PACKAGE_DIR=${{ env.PYTORCH_FINAL_PACKAGE_DIR }}"
             echo "SHA1=${{ env.SHA1 }}"
-            echo "USE_SPLIT_BUILD=${{ env.USE_SPLIT_BUILD }}"
           } >> "${GITHUB_ENV} }}"
 
       - 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/2.4
         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/2.4
         with:
           no-sudo: ${{ inputs.build_environment == 'linux-aarch64-binary-manywheel' || inputs.build_environment == 'linux-s390x-binary-manywheel' }}
 
@@ -167,7 +158,6 @@ jobs:
       - name: Checkout PyTorch to pytorch dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
           submodules: recursive
           path: pytorch
 
@@ -180,7 +170,7 @@ jobs:
       - name: Checkout pytorch/builder to builder dir
         uses: malfet/checkout@silent-checkout
         with:
-          ref: main
+          ref: release/2.4
           submodules: recursive
           repository: pytorch/builder
           path: builder
@@ -211,12 +201,12 @@ jobs:
           path: "${{ runner.temp }}/artifacts/"
 
       - name: Install nvidia driver, nvidia-docker runtime, set GPU_FLAG
-        uses: pytorch/test-infra/.github/actions/setup-nvidia@main
+        uses: pytorch/test-infra/.github/actions/setup-nvidia@release/2.4
         if: ${{ inputs.GPU_ARCH_TYPE == 'cuda' && steps.filter.outputs.is-test-matrix-empty == 'False' }}
 
       - name: Pull Docker image
         if: ${{ steps.filter.outputs.is-test-matrix-empty == 'False' && 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/2.4
         with:
           docker-image: ${{ inputs.DOCKER_IMAGE }}
 
@@ -226,7 +216,7 @@ 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/2.4
 
       - name: Chown workspace
         if: always() && inputs.build_environment != 'linux-s390x-binary-manywheel'

.github/workflows/_binary-upload.yml

@@ -55,13 +55,6 @@ on:
         required: false
         type: string
         description: Desired python version
-      use_split_build:
-        description: |
-          [Experimental] Build a libtorch only wheel and build pytorch such that
-          are built from the libtorch wheel.
-        required: false
-        type: boolean
-        default: false
     secrets:
       github-token:
         required: true
@@ -100,10 +93,9 @@ jobs:
       PR_NUMBER: ${{ github.event.pull_request.number }}
       PYTORCH_FINAL_PACKAGE_DIR: /artifacts
       SHA1: ${{ github.event.pull_request.head.sha || github.sha }}
-      USE_SPLIT_BUILD: ${{ inputs.use_split_build }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 

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

@@ -23,7 +23,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -44,7 +44,7 @@ jobs:
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Set up JDK 8
         uses: actions/setup-java@v3
@@ -53,7 +53,7 @@ jobs:
           distribution: 'temurin'
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: 3.8
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}

.github/workflows/_docs.yml

@@ -80,7 +80,7 @@ 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/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -91,7 +91,7 @@ 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/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -106,12 +106,12 @@ 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/2.4
         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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -218,5 +218,5 @@ 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/2.4
         if: always()

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

@@ -46,7 +46,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -80,7 +80,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Populate CI build options
         shell: bash
@@ -102,7 +102,7 @@ jobs:
             brew install libtool
 
       - name: Setup miniconda for iOS
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: "3.9"
           environment-file: .github/requirements/conda-env-iOS.txt

.github/workflows/_linux-build-label.yml

@@ -56,13 +56,6 @@ on:
         required: false
         type: string
         default: ""
-      use_split_build:
-        description: |
-          [Experimental] Build a libtorch only wheel and build pytorch such that
-          are built from the libtorch wheel.
-        required: false
-        type: boolean
-        default: false
     secrets:
       HUGGING_FACE_HUB_TOKEN:
         required: false
@@ -88,7 +81,7 @@ jobs:
       test-matrix: ${{ steps.linux-build.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/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
@@ -97,7 +90,7 @@ 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/2.4
 
       - name: Linux Build
         id: linux-build
@@ -114,4 +107,3 @@ jobs:
           aws-role-to-assume: ${{ inputs.aws-role-to-assume }}
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
-          use_split_build: ${{ inputs.use_split_build }}

.github/workflows/_linux-build-rg.yml

@@ -0,0 +1,105 @@
+name: linux-build-rg
+
+on:
+  workflow_call:
+    inputs:
+      build-environment:
+        required: true
+        type: string
+        description: Top-level label for what's being built/tested.
+      docker-image-name:
+        required: true
+        type: string
+        description: Name of the base docker image to build with.
+      build-generates-artifacts:
+        required: false
+        type: boolean
+        default: true
+        description: If set, upload generated build artifacts.
+      build-with-debug:
+        required: false
+        type: boolean
+        default: false
+        description: If set, build in debug mode.
+      sync-tag:
+        required: false
+        type: string
+        default: ""
+        description: |
+          If this is set, our linter will use this to make sure that every other
+          job with the same `sync-tag` is identical.
+      cuda-arch-list:
+        required: false
+        type: string
+        default: "5.2"
+        description: |
+          List of CUDA architectures CI build should target.
+      runner-group:
+        required: false
+        type: string
+        default: "arc-lf-linux.2xlarge"
+        description: Runner group to select group type
+      test-matrix:
+        required: false
+        type: string
+        description: |
+          An option JSON description of what test configs to run later on. This
+          is moved here from the Linux test workflow so that we can apply filter
+          logic using test-config labels earlier and skip unnecessary builds
+      s3-bucket:
+        description: S3 bucket to download artifact
+        required: false
+        type: string
+        default: "gha-artifacts"
+      aws-role-to-assume:
+        description: role to assume for downloading artifacts
+        required: false
+        type: string
+        default: ""
+    secrets:
+      HUGGING_FACE_HUB_TOKEN:
+        required: false
+        description: |
+          HF Auth token to avoid rate limits when downloading models or datasets from hub
+
+    outputs:
+      docker-image:
+        value: ${{ jobs.build.outputs.docker-image }}
+        description: The docker image containing the built PyTorch.
+      test-matrix:
+        value: ${{ jobs.build.outputs.test-matrix }}
+        description: An optional JSON description of what test configs to run later on.
+
+jobs:
+  build:
+    # Don't run on forked repos
+    if: github.repository_owner == 'pytorch'
+    runs-on:
+      group: ${{ inputs.runner-group }}
+    timeout-minutes: 240
+    outputs:
+      docker-image: ${{ steps.linux-build.outputs.docker-image }}
+      test-matrix: ${{ steps.linux-build.outputs.test-matrix }}
+    steps:
+      # [pytorch repo ref]
+      # Use a pytorch/pytorch reference instead of a reference to the local
+      # checkout because when we run this action we don't *have* a local
+      # checkout. In other cases you should prefer a local checkout.
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
+
+      - name: Linux Build
+        id: linux-build
+        uses: ./.github/actions/linux-build
+        with:
+          build-environment: ${{ inputs.build-environment }}
+          docker-image-name: ${{ inputs.docker-image-name }}
+          build-generates-artifacts: ${{ inputs.build-generates-artifacts }}
+          build-with-debug: ${{ inputs.build-with-debug }}
+          sync-tag: ${{ inputs.sync-tag }}
+          cuda-arch-list: ${{ inputs.cuda-arch-list }}
+          test-matrix: ${{ inputs.test-matrix }}
+          s3-bucket: ${{ inputs.s3-bucket }}
+          aws-role-to-assume: ${{ inputs.aws-role-to-assume }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}

.github/workflows/_linux-build.yml

@@ -39,7 +39,7 @@ on:
         type: string
         default: "linux.2xlarge"
         description: |
-          Label of the runner this job should run on.
+          List of CUDA architectures CI build should target.
       test-matrix:
         required: false
         type: string
@@ -64,14 +64,6 @@ on:
         required: false
         type: string
         default: ""
-      use_split_build:
-        description: |
-          [Experimental] Build a libtorch only wheel and build pytorch such that
-          are built from the libtorch wheel.
-        required: false
-        type: boolean
-        default: false
-
     secrets:
       HUGGING_FACE_HUB_TOKEN:
         required: false
@@ -98,7 +90,7 @@ 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/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
 
@@ -107,7 +99,7 @@ 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/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -122,7 +114,7 @@ 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/2.4
         with:
           docker-image-name: ${{ inputs.docker-image-name }}
 
@@ -136,7 +128,7 @@ 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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -189,7 +181,6 @@ jobs:
           DEBUG: ${{ inputs.build-with-debug && '1' || '0' }}
           OUR_GITHUB_JOB_ID: ${{ steps.get-job-id.outputs.job-id }}
           HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
-          USE_SPLIT_BUILD: ${{ inputs.use_split_build }}
         run: |
           # detached container should get cleaned up by teardown_ec2_linux
           container_name=$(docker run \
@@ -208,7 +199,6 @@ jobs:
             -e PR_LABELS \
             -e OUR_GITHUB_JOB_ID \
             -e HUGGING_FACE_HUB_TOKEN \
-            -e USE_SPLIT_BUILD \
             --env-file="/tmp/github_env_${GITHUB_RUN_ID}" \
             --security-opt seccomp=unconfined \
             --cap-add=SYS_PTRACE \
@@ -228,7 +218,7 @@ jobs:
 
       - name: Store PyTorch Build Artifacts on S3
         uses: seemethere/upload-artifact-s3@v5
-        if: inputs.build-generates-artifacts && steps.build.outcome != 'skipped' && inputs.use_split_build != 'true'
+        if: inputs.build-generates-artifacts && steps.build.outcome != 'skipped'
         with:
           name: ${{ inputs.build-environment }}
           retention-days: 14
@@ -236,16 +226,6 @@ jobs:
           path: artifacts.zip
           s3-bucket: ${{ inputs.s3-bucket }}
 
-      - name: Store PyTorch Build Artifacts on S3
-        uses: seemethere/upload-artifact-s3@v5
-        if: inputs.build-generates-artifacts && steps.build.outcome != 'skipped' && inputs.use_split_build == 'true'
-        with:
-          name: ${{ inputs.build-environment }}-experimental-split-build
-          retention-days: 14
-          if-no-files-found: error
-          path: artifacts.zip
-          s3-bucket: ${{ inputs.s3-bucket }}
-
       - name: Upload sccache stats
         if: steps.build.outcome != 'skipped'
         uses: seemethere/upload-artifact-s3@v5
@@ -258,5 +238,5 @@ jobs:
           s3-bucket: ${{ inputs.s3-bucket }}
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()

.github/workflows/_linux-test-label.yml

@@ -0,0 +1,85 @@
+name: linux-test-rg
+
+on:
+  workflow_call:
+    inputs:
+      build-environment:
+        required: true
+        type: string
+        description: Top-level label for what's being built/tested.
+      test-matrix:
+        required: true
+        type: string
+        description: JSON description of what test configs to run.
+      docker-image:
+        required: true
+        type: string
+        description: Docker image to run in.
+      sync-tag:
+        required: false
+        type: string
+        default: ""
+        description: |
+          If this is set, our linter will use this to make sure that every other
+          job with the same `sync-tag` is identical.
+      timeout-minutes:
+        required: false
+        type: number
+        default: 240
+        description: |
+          Set the maximum (in minutes) how long the workflow should take to finish
+      use-gha:
+        required: false
+        type: string
+        default: ""
+        description: If set to any value, upload to GHA. Otherwise upload to S3.
+      dashboard-tag:
+        required: false
+        type: string
+        default: ""
+      s3-bucket:
+        description: S3 bucket to download artifact
+        required: false
+        type: string
+        default: "gha-artifacts"
+      aws-role-to-assume:
+        description: role to assume for downloading artifacts
+        required: false
+        type: string
+        default: ""
+    secrets:
+      HUGGING_FACE_HUB_TOKEN:
+        required: false
+        description: |
+          HF Auth token to avoid rate limits when downloading models or datasets from hub
+
+env:
+  GIT_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
+
+jobs:
+  test:
+    # Don't run on forked repos or empty test matrix
+    if: github.repository_owner == 'pytorch' && toJSON(fromJSON(inputs.test-matrix).include) != '[]'
+    strategy:
+      matrix: ${{ fromJSON(inputs.test-matrix) }}
+      fail-fast: false
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
+    steps:
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
+
+      - name: Linux Test
+        id: linux-test
+        uses: ./.github/actions/linux-test
+        with:
+          build-environment: ${{ inputs.build-environment }}
+          test-matrix: ${{ inputs.test-matrix }}
+          docker-image: ${{ inputs.docker-image }}
+          sync-tag: ${{ inputs.sync-tag }}
+          use-gha: ${{ inputs.use-gha }}
+          dashboard-tag: ${{ inputs.dashboard-tag }}
+          s3-bucket: ${{ inputs.s3-bucket }}
+          aws-role-to-assume: ${{ inputs.aws-role-to-assume }}
+          HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/_linux-test-rg.yml

@@ -0,0 +1,86 @@
+name: linux-test-label
+
+on:
+  workflow_call:
+    inputs:
+      build-environment:
+        required: true
+        type: string
+        description: Top-level label for what's being built/tested.
+      test-matrix:
+        required: true
+        type: string
+        description: JSON description of what test configs to run.
+      docker-image:
+        required: true
+        type: string
+        description: Docker image to run in.
+      sync-tag:
+        required: false
+        type: string
+        default: ""
+        description: |
+          If this is set, our linter will use this to make sure that every other
+          job with the same `sync-tag` is identical.
+      timeout-minutes:
+        required: false
+        type: number
+        default: 240
+        description: |
+          Set the maximum (in minutes) how long the workflow should take to finish
+      use-gha:
+        required: false
+        type: string
+        default: ""
+        description: If set to any value, upload to GHA. Otherwise upload to S3.
+      dashboard-tag:
+        required: false
+        type: string
+        default: ""
+      s3-bucket:
+        description: S3 bucket to download artifact
+        required: false
+        type: string
+        default: "gha-artifacts"
+      aws-role-to-assume:
+        description: role to assume for downloading artifacts
+        required: false
+        type: string
+        default: ""
+    secrets:
+      HUGGING_FACE_HUB_TOKEN:
+        required: false
+        description: |
+          HF Auth token to avoid rate limits when downloading models or datasets from hub
+
+env:
+  GIT_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
+
+jobs:
+  test:
+    # Don't run on forked repos or empty test matrix
+    if: github.repository_owner == 'pytorch' && toJSON(fromJSON(inputs.test-matrix).include) != '[]'
+    strategy:
+      matrix: ${{ fromJSON(inputs.test-matrix) }}
+      fail-fast: false
+    runs-on:
+      group: ${{ matrix.runner }}
+    timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
+    steps:
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
+
+      - name: Linux Test
+        id: linux-test
+        uses: ./.github/actions/linux-test
+        with:
+          build-environment: ${{ inputs.build-environment }}
+          test-matrix: ${{ inputs.test-matrix }}
+          docker-image: ${{ inputs.docker-image }}
+          sync-tag: ${{ inputs.sync-tag }}
+          use-gha: ${{ inputs.use-gha }}
+          dashboard-tag: ${{ inputs.dashboard-tag }}
+          s3-bucket: ${{ inputs.s3-bucket }}
+          aws-role-to-assume: ${{ inputs.aws-role-to-assume }}
+          HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/_linux-test.yml

@@ -67,7 +67,7 @@ jobs:
     timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
     steps:
       - name: Setup SSH (Click me for login details)
-        uses: pytorch/test-infra/.github/actions/setup-ssh@main
+        uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.4
         if: ${{ !contains(matrix.runner, 'gcp.a100') }}
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
@@ -76,7 +76,7 @@ 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/2.4
 
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
@@ -91,7 +91,7 @@ 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/2.4
         with:
           docker-image-name: ${{ inputs.docker-image }}
 
@@ -105,7 +105,7 @@ 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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 
@@ -116,7 +116,7 @@ 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/2.4
         if: ${{ contains(inputs.build-environment, 'cuda') && !contains(matrix.config, 'nogpu') && steps.check_arc_runner.outputs.IN_ARC_RUNNER == 'false' }}
 
       - name: Lock NVIDIA A100 40GB Frequency
@@ -333,7 +333,7 @@ jobs:
           path: ./**/core.[1-9]*
 
       - name: Teardown Linux
-        uses: pytorch/test-infra/.github/actions/teardown-linux@main
+        uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.4
         if: always()
 
       # NB: We are currently having an intermittent GPU-related issue on G5 runners with

.github/workflows/_mac-build.yml

@@ -71,11 +71,11 @@ jobs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
     steps:
       - name: Clean up disk space before running MacOS workflow
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Set xcode version
         env:
@@ -87,7 +87,7 @@ jobs:
 
       - name: Setup miniconda
         if: inputs.environment-file == ''
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -97,7 +97,7 @@ jobs:
       # environment even though the arch is x86-64
       - name: Setup miniconda using the provided environment file
         if: inputs.environment-file != ''
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: ${{ inputs.environment-file }}
@@ -207,4 +207,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4

.github/workflows/_mac-test-mps.yml

@@ -40,7 +40,7 @@ jobs:
       reenabled-issues: ${{ steps.filter.outputs.reenabled-issues }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           submodules: false
 
@@ -81,7 +81,7 @@ jobs:
           use-gha: true
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -159,4 +159,4 @@ jobs:
       - name: Clean up disk space
         if: always()
         continue-on-error: true
-        uses: pytorch/test-infra/.github/actions/check-disk-space@main
+        uses: pytorch/test-infra/.github/actions/check-disk-space@release/2.4

.github/workflows/_mac-test.yml

@@ -74,11 +74,11 @@ 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/2.4
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Download build artifacts
         uses: ./.github/actions/download-build-artifacts
@@ -93,7 +93,7 @@ jobs:
           use-gha: true
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: ${{ inputs.python-version }}
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
@@ -216,4 +216,4 @@ 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/2.4

.github/workflows/_rocm-test.yml

@@ -58,7 +58,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 
@@ -80,12 +80,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
 

.github/workflows/_run_android_tests.yml

@@ -23,7 +23,7 @@ jobs:
       keep-going: ${{ steps.filter.outputs.keep-going }}
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           fetch-depth: 1
           submodules: false
@@ -54,10 +54,10 @@ jobs:
       SUPPORT_ABI: '${{ matrix.support_abi }}'
     steps:
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup miniconda
-        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.4
         with:
           python-version: 3.8
           environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}.txt

.github/workflows/_runner-determinator.yml

@@ -3,273 +3,39 @@ name: Check whether the workflow owner can use ARC runners
 on:
   workflow_call:
     inputs:
-      triggering_actor:
+      user_name:
         required: true
         type: string
-        description: The triggering_actor for the workflow. Use github.triggering_actor
-      issue_owner:
-        required: true
-        type: string
-        description: The owner of the issue. Use github.event.pull_request.user.login || github.event.issue.user.login
+        description: The name of the workflow owner.
       curr_branch:
         required: true
         type: string
-        description: Current branch or tag.
-      curr_ref_type:
-        required: false
-        type: string
-        default: branch
-        description: The value of "github.ref_type", "branch" or "tag"
+        description: Current branch.
       issue_number:
         required: false
         type: string
         default: "5132"
-        description: |
-          Fetch's GitHub Issue from pytorch/test-infra
-          Example: https://github.com/pytorch/test-infra/issues/5132
 
     outputs:
-      label-type:
+      workflow-type:
         description: Type of runners to use
-        value: ${{ jobs.runner-determinator.outputs.label-type }}
+        value: ${{ jobs.runner-determinator.outputs.workflow-type }}
 
 jobs:
   runner-determinator:
-    runs-on: ubuntu-latest
+    runs-on: linux.4xlarge
     outputs:
-      label-type: ${{ steps.set-condition.outputs.label-type }}
+      workflow-type: ${{ steps.set-condition.outputs.workflow-type }}
     env:
       GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       ISSUE_NUMBER: ${{ inputs.issue_number }}
-      TRIGGERING_ACTOR: ${{ inputs.triggering_actor }}
-      ISSUE_OWNER: ${{ inputs.issue_owner }}
+      USERNAME: ${{ inputs.user_name }}
     steps:
-      # - name: Checkout PyTorch
-      #   uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
-      #   with:
-      #     fetch-depth: 1
-      #     submodules: true
-
-      # TODO: Remove the hardcoded step below
-      # Hardcoding below is temporary for testing ALI runners
-      # This file below should match the script found in .github/scripts/runner_determinator.py
-      - name: Hardcode runner-determinator script
-        id: hardcode-script
-        run: |
-          cat <<EOF > runner_determinator.py
-          # flake8: noqa: G004
-
-          import logging
-          import os
-          from argparse import ArgumentParser
-          from logging import LogRecord
-          from typing import Any, Iterable
-
-          from github import Auth, Github
-          from github.Issue import Issue
-
-
-          WORKFLOW_LABEL_META = ""  # use meta runners
-          WORKFLOW_LABEL_LF = "lf."  # use runners from the linux foundation
-
-          GITHUB_OUTPUT = os.getenv("GITHUB_OUTPUT", "")
-          GH_OUTPUT_KEY_LABEL_TYPE = "label-type"
-
-
-          class ColorFormatter(logging.Formatter):
-              """Color codes the log messages based on the log level"""
-
-              COLORS = {
-                  "WARNING": "\033[33m",  # Yellow
-                  "ERROR": "\033[31m",  # Red
-                  "CRITICAL": "\033[31m",  # Red
-                  "INFO": "\033[0m",  # Reset
-                  "DEBUG": "\033[0m",  # Reset
-              }
-
-              def format(self, record: LogRecord) -> str:
-                  log_color = self.COLORS.get(record.levelname, "\033[0m")  # Default to reset
-                  record.msg = f"{log_color}{record.msg}\033[0m"
-                  return super().format(record)
-
-
-          handler = logging.StreamHandler()
-          handler.setFormatter(ColorFormatter(fmt="%(levelname)-8s: %(message)s"))
-
-          log = logging.getLogger(os.path.basename(__file__))
-          log.addHandler(handler)
-          log.setLevel(logging.INFO)
-
-
-          def set_github_output(key: str, value: str) -> None:
-              """
-              Defines outputs of the github action that invokes this script
-              """
-              if not GITHUB_OUTPUT:
-                  # See https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/ for deprecation notice
-                  log.warning(
-                      "No env var found for GITHUB_OUTPUT, you must be running this code locally. Falling back to the deprecated print method."
-                  )
-                  print(f"::set-output name={key}::{value}")
-                  return
-
-              with open(GITHUB_OUTPUT, "a") as f:
-                  log.info(f"Setting output: {key}='{value}'")
-                  f.write(f"{key}={value}\n")
-
-
-          def parse_args() -> Any:
-              parser = ArgumentParser("Get dynamic rollout settings")
-              parser.add_argument("--github-token", type=str, required=True, help="GitHub token")
-              parser.add_argument(
-                  "--github-issue-repo",
-                  type=str,
-                  required=False,
-                  default="pytorch/test-infra",
-                  help="GitHub repo to get the issue",
-              )
-              parser.add_argument(
-                  "--github-repo",
-                  type=str,
-                  required=True,
-                  help="GitHub repo where CI is running",
-              )
-              parser.add_argument(
-                  "--github-issue", type=int, required=True, help="GitHub issue number"
-              )
-              parser.add_argument(
-                  "--github-actor", type=str, required=True, help="GitHub triggering_actor"
-              )
-              parser.add_argument(
-                  "--github-issue-owner", type=str, required=True, help="GitHub issue owner"
-              )
-              parser.add_argument(
-                  "--github-branch", type=str, required=True, help="Current GitHub branch or tag"
-              )
-              parser.add_argument(
-                  "--github-ref-type",
-                  type=str,
-                  required=True,
-                  help="Current GitHub ref type, branch or tag",
-              )
-
-              return parser.parse_args()
-
-
-          def get_gh_client(github_token: str) -> Github:
-              auth = Auth.Token(github_token)
-              return Github(auth=auth)
-
-
-          def get_issue(gh: Github, repo: str, issue_num: int) -> Issue:
-              repo = gh.get_repo(repo)
-              return repo.get_issue(number=issue_num)
-
-
-          def get_potential_pr_author(
-              gh: Github, repo: str, username: str, ref_type: str, ref_name: str
-          ) -> str:
-              # If the trigger was a new tag added by a bot, this is a ciflow case
-              # Fetch the actual username from the original PR. The PR number is
-              # embedded in the tag name: ciflow/<name>/<pr-number>
-              if username == "pytorch-bot[bot]" and ref_type == "tag":
-                  split_tag = ref_name.split("/")
-                  if (
-                      len(split_tag) == 3
-                      and split_tag[0] == "ciflow"
-                      and split_tag[2].isnumeric()
-                  ):
-                      pr_number = split_tag[2]
-                      try:
-                          repository = gh.get_repo(repo)
-                          pull = repository.get_pull(number=int(pr_number))
-                      except Exception as e:
-                          raise Exception(  # noqa: TRY002
-                              f"issue with pull request {pr_number} from repo {repository}"
-                          ) from e
-                      return pull.user.login
-              # In all other cases, return the original input username
-              return username
-
-
-          def is_exception_branch(branch: str) -> bool:
-              return branch.split("/")[0] in {"main", "nightly", "release", "landchecks"}
-
-
-          def get_workflow_type(issue: Issue, workflow_requestors: Iterable[str]) -> str:
-              try:
-                  first_comment = issue.get_comments()[0].body.strip("\n\t ")
-
-                  if first_comment[0] == "!":
-                      log.info("LF Workflows are disabled for everyone. Using meta runners.")
-                      return WORKFLOW_LABEL_META
-                  elif first_comment[0] == "*":
-                      log.info("LF Workflows are enabled for everyone. Using LF runners.")
-                      return WORKFLOW_LABEL_LF
-                  else:
-                      all_opted_in_users = {
-                          usr_raw.strip("\n\t@ ") for usr_raw in first_comment.split()
-                      }
-                      opted_in_requestors = {
-                          usr for usr in workflow_requestors if usr in all_opted_in_users
-                      }
-                      if opted_in_requestors:
-                          log.info(
-                              f"LF Workflows are enabled for {', '.join(opted_in_requestors)}. Using LF runners."
-                          )
-                          return WORKFLOW_LABEL_LF
-                      else:
-                          log.info(
-                              f"LF Workflows are disabled for {', '.join(workflow_requestors)}. Using meta runners."
-                          )
-                          return WORKFLOW_LABEL_META
-
-              except Exception as e:
-                  log.error(
-                      f"Failed to get determine workflow type. Falling back to meta runners. Exception: {e}"
-                  )
-                  return WORKFLOW_LABEL_META
-
-
-          def main() -> None:
-              args = parse_args()
-
-              if args.github_ref_type == "branch" and is_exception_branch(args.github_branch):
-                  log.info(f"Exception branch: '{args.github_branch}', using meta runners")
-                  label_type = WORKFLOW_LABEL_META
-              else:
-                  try:
-                      gh = get_gh_client(args.github_token)
-                      # The default issue we use - https://github.com/pytorch/test-infra/issues/5132
-                      issue = get_issue(gh, args.github_issue_repo, args.github_issue)
-                      username = get_potential_pr_author(
-                          gh,
-                          args.github_repo,
-                          args.github_actor,
-                          args.github_ref_type,
-                          args.github_branch,
-                      )
-                      label_type = get_workflow_type(
-                          issue,
-                          (
-                              args.github_issue_owner,
-                              username,
-                          ),
-                      )
-                  except Exception as e:
-                      log.error(
-                          f"Failed to get issue. Falling back to meta runners. Exception: {e}"
-                      )
-                      label_type = WORKFLOW_LABEL_META
-
-              set_github_output(GH_OUTPUT_KEY_LABEL_TYPE, label_type)
-
-
-          if __name__ == "__main__":
-              main()
-          EOF
-
-          cat runner_determinator.py
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
+        with:
+          fetch-depth: 1
+          submodules: true
 
       - name: Install dependencies
         run: python3 -m pip install urllib3==1.26.18 PyGithub==2.3.0
@@ -278,14 +44,15 @@ jobs:
         id: set-condition
         run: |
           curr_branch="${{ inputs.curr_branch }}"
-          curr_ref_type="${{ inputs.curr_ref_type }}"
           echo "Current branch is '$curr_branch'"
 
-          python3 runner_determinator.py \
+          output="$(python3 .github/scripts/get_workflow_type.py \
             --github-token "$GITHUB_TOKEN" \
             --github-issue "$ISSUE_NUMBER" \
             --github-branch "$curr_branch" \
-            --github-actor "$TRIGGERING_ACTOR" \
-            --github-issue-owner "$ISSUE_OWNER" \
-            --github-ref-type "$curr_ref_type" \
-            --github-repo "$GITHUB_REPOSITORY"
+            --github-user "$USERNAME")"
+
+          echo "Output: '${output}'"
+
+          WORKFLOW_TYPE=$(echo "${output}" | jq -r '.workflow_type')
+          echo "workflow-type=$WORKFLOW_TYPE" >> "$GITHUB_OUTPUT"

.github/workflows/_win-build.yml

@@ -30,12 +30,6 @@ on:
           An option JSON description of what test configs to run later on. This
           is moved here from the Linux test workflow so that we can apply filter
           logic using test-config labels earlier and skip unnecessary builds
-      runner:
-        required: false
-        type: string
-        default: "windows.4xlarge.nonephemeral"
-        description: |
-          Label of the runner this job should run on.
 
     outputs:
       test-matrix:
@@ -49,13 +43,10 @@ jobs:
   build:
     # Don't run on forked repos.
     if: github.repository_owner == 'pytorch'
-    runs-on: ${{ inputs.runner }}
+    runs-on: [self-hosted, windows.4xlarge.nonephemeral]
     timeout-minutes: 240
     outputs:
       test-matrix: ${{ steps.filter.outputs.test-matrix }}
-    defaults:
-      run:
-        shell: bash
     steps:
       # Duplicated in win-test because this MUST go before a checkout
       - name: Enable git symlinks on Windows and disable fsmonitor daemon
@@ -69,10 +60,10 @@ jobs:
           git config --global core.fsmonitor false
 
       - name: Clean up leftover processes on non-ephemeral Windows runner
-        uses: pytorch/test-infra/.github/actions/cleanup-runner@main
+        uses: pytorch/test-infra/.github/actions/cleanup-runner@release/2.4
 
       - 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/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -87,7 +78,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 
@@ -98,7 +89,6 @@ jobs:
 
       - name: Parse ref
         id: parse-ref
-        shell: bash
         run: python3 .github/scripts/parse_ref.py
 
       - name: Get workflow job id

.github/workflows/_win-test.yml

@@ -41,9 +41,6 @@ jobs:
       fail-fast: false
     runs-on: ${{ matrix.runner }}
     timeout-minutes: ${{ matrix.mem_leak_check == 'mem_leak_check' && 600 || inputs.timeout-minutes }}
-    defaults:
-      run:
-        shell: bash
     steps:
       # Duplicated in win-build because this MUST go before a checkout
       - name: Enable git symlinks on Windows and disable fsmonitor daemon
@@ -57,10 +54,10 @@ jobs:
           git config --global core.fsmonitor false
 
       - name: Clean up leftover processes on non-ephemeral Windows runner
-        uses: pytorch/test-infra/.github/actions/cleanup-runner@main
+        uses: pytorch/test-infra/.github/actions/cleanup-runner@release/2.4
 
       - 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/2.4
         with:
           github-secret: ${{ secrets.GITHUB_TOKEN }}
           instructions: |
@@ -76,7 +73,7 @@ jobs:
 
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
         with:
           no-sudo: true
 
@@ -227,7 +224,6 @@ jobs:
 
       - name: Parse ref
         id: parse-ref
-        shell: bash
         run: python3 .github/scripts/parse_ref.py
 
       - name: Uninstall PyTorch

.github/workflows/_xpu-test.yml

@@ -54,7 +54,7 @@ jobs:
     steps:
       # [see note: pytorch repo ref]
       - name: Checkout PyTorch
-        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.4
 
       - name: Setup XPU
         uses: ./.github/actions/setup-xpu
@@ -72,12 +72,12 @@ jobs:
 
       - name: Calculate docker image
         id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@release/2.4
         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/2.4
         with:
           docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}