Update (base update)

[ghstack-poisoned]

.ci/aarch64_linux/build_aarch64_wheel.py

@@ -1004,7 +1004,7 @@ if __name__ == "__main__":
         install_condaforge_python(host, args.python_version)
         sys.exit(0)
 
-    python_version = args.python_version if args.python_version is not None else "3.9"
+    python_version = args.python_version if args.python_version is not None else "3.10"
 
     if args.use_torch_from_pypi:
         configure_system(host, compiler=args.compiler, python_version=python_version)

.ci/docker/almalinux/Dockerfile

@@ -69,7 +69,8 @@ RUN bash ./install_cuda.sh 13.0
 ENV DESIRED_CUDA=13.0
 
 FROM ${ROCM_IMAGE} as rocm
-ENV PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+ARG PYTORCH_ROCM_ARCH
+ENV PYTORCH_ROCM_ARCH ${PYTORCH_ROCM_ARCH}
 ADD ./common/install_mkl.sh install_mkl.sh
 RUN bash ./install_mkl.sh && rm install_mkl.sh
 ENV MKLROOT /opt/intel

.ci/docker/almalinux/build.sh

@@ -36,6 +36,12 @@ case ${DOCKER_TAG_PREFIX} in
     ;;
   rocm*)
     BASE_TARGET=rocm
+    PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+    # add gfx950 conditionally starting in ROCm 7.0
+    if [[ "$ROCM_VERSION" == *"7.0"* ]]; then
+        PYTORCH_ROCM_ARCH="${PYTORCH_ROCM_ARCH};gfx950"
+    fi
+    EXTRA_BUILD_ARGS="${EXTRA_BUILD_ARGS} --build-arg PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH}"
     ;;
   *)
     echo "ERROR: Unknown docker tag ${DOCKER_TAG_PREFIX}"

.ci/docker/common/install_rocm_magma.sh

@@ -12,8 +12,8 @@ function do_install() {
 
     rocm_version_nodot=${rocm_version//./}
 
-    # Version 2.7.2 + ROCm related updates
-    MAGMA_VERSION=a1625ff4d9bc362906bd01f805dbbe12612953f6
+    # https://github.com/icl-utk-edu/magma/pull/65
+    MAGMA_VERSION=d6e4117bc88e73f06d26c6c2e14f064e8fc3d1ec
     magma_archive="magma-rocm${rocm_version_nodot}-${MAGMA_VERSION}-1.tar.bz2"
 
     rocm_dir="/opt/rocm"

.ci/docker/libtorch/build.sh

@@ -40,12 +40,16 @@ case ${DOCKER_TAG_PREFIX} in
         ;;
     rocm*)
         # we want the patch version of 6.4 instead
-        if [[ $(ver $GPU_ARCH_VERSION) -eq $(ver 6.4) ]]; then
+        if [[ "$GPU_ARCH_VERSION" == *"6.4"* ]]; then
             GPU_ARCH_VERSION="${GPU_ARCH_VERSION}.2"
         fi
         BASE_TARGET=rocm
         GPU_IMAGE=rocm/dev-ubuntu-22.04:${GPU_ARCH_VERSION}-complete
         PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+        # add gfx950 conditionally starting in ROCm 7.0
+        if [[ "$GPU_ARCH_VERSION" == *"7.0"* ]]; then
+            PYTORCH_ROCM_ARCH="${PYTORCH_ROCM_ARCH};gfx950"
+        fi
         DOCKER_GPU_BUILD_ARG="--build-arg PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH} --build-arg ROCM_VERSION=${GPU_ARCH_VERSION}"
         ;;
     *)

.ci/docker/manywheel/build.sh

@@ -82,7 +82,7 @@ case ${image} in
         ;;
     manylinux2_28-builder:rocm*)
         # we want the patch version of 6.4 instead
-        if [[ $(ver $GPU_ARCH_VERSION) -eq $(ver 6.4) ]]; then
+        if [[ "$GPU_ARCH_VERSION" == *"6.4"* ]]; then
             GPU_ARCH_VERSION="${GPU_ARCH_VERSION}.2"
         fi
         TARGET=rocm_final
@@ -90,6 +90,10 @@ case ${image} in
         DEVTOOLSET_VERSION="11"
         GPU_IMAGE=rocm/dev-almalinux-8:${GPU_ARCH_VERSION}-complete
         PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201"
+        # add gfx950 conditionally starting in ROCm 7.0
+        if [[ "$GPU_ARCH_VERSION" == *"7.0"* ]]; then
+            PYTORCH_ROCM_ARCH="${PYTORCH_ROCM_ARCH};gfx950"
+        fi
         DOCKER_GPU_BUILD_ARG="--build-arg ROCM_VERSION=${GPU_ARCH_VERSION} --build-arg PYTORCH_ROCM_ARCH=${PYTORCH_ROCM_ARCH} --build-arg DEVTOOLSET_VERSION=${DEVTOOLSET_VERSION}"
         ;;
     manylinux2_28-builder:xpu)

.ci/docker/requirements-ci.txt

@@ -112,8 +112,6 @@ ninja==1.11.1.3
 #Pinned versions: 1.11.1.3
 #test that import: run_test.py, test_cpp_extensions_aot.py,test_determination.py
 
-numba==0.49.0 ; python_version < "3.9" and platform_machine != "s390x"
-numba==0.55.2 ; python_version == "3.9" and platform_machine != "s390x"
 numba==0.55.2 ; python_version == "3.10" and platform_machine != "s390x"
 numba==0.60.0 ; python_version == "3.12" and platform_machine != "s390x"
 #Description: Just-In-Time Compiler for Numerical Functions
@@ -134,7 +132,7 @@ numba==0.60.0 ; python_version == "3.12" and platform_machine != "s390x"
 #test_nn.py, test_namedtensor.py, test_linalg.py, test_jit_cuda_fuser.py,
 #test_jit.py, test_indexing.py, test_datapipe.py, test_dataloader.py,
 #test_binary_ufuncs.py
-numpy==1.22.4; python_version == "3.9" or python_version == "3.10"
+numpy==1.22.4; python_version == "3.10"
 numpy==1.26.2; python_version == "3.11" or python_version == "3.12"
 numpy==2.1.2; python_version >= "3.13"
 
@@ -326,8 +324,6 @@ pywavelets==1.7.0 ; python_version >= "3.12"
 lxml==5.3.0
 #Description: This is a requirement of unittest-xml-reporting
 
-# Python-3.9 binaries
-
 PyGithub==2.3.0
 
 sympy==1.13.3

.ci/docker/requirements-docs.txt

@@ -1,8 +1,15 @@
 sphinx==5.3.0
 #Description: This is used to generate PyTorch docs
 #Pinned versions: 5.3.0
--e git+https://github.com/pytorch/pytorch_sphinx_theme.git@d53b0ffb9b1cda68260693ea98f3483823c88d8e#egg=pytorch_sphinx_theme2
 
+standard-imghdr==3.13.0; python_version >= "3.13"
+#Description: This is needed by Sphinx, so it needs to be added here.
+# The reasons are as follows:
+# 1) This module has been removed from the Python standard library since Python 3.13(https://peps.python.org/pep-0594/#imghdr);
+# 2) The current version of Sphinx (5.3.0) is not compatible with Python 3.13.
+# Once Sphinx is upgraded to a version compatible with Python 3.13 or later, we can remove this dependency.
+
+-e git+https://github.com/pytorch/pytorch_sphinx_theme.git@d53b0ffb9b1cda68260693ea98f3483823c88d8e#egg=pytorch_sphinx_theme2
 # TODO: sphinxcontrib.katex 0.9.0 adds a local KaTeX server to speed up pre-rendering
 # but it doesn't seem to work and hangs around idly. The initial thought that it is probably
 # something related to Docker setup. We can investigate this later.

.ci/magma-rocm/Makefile

@@ -1,11 +1,11 @@
 SHELL=/usr/bin/env bash
 
 DOCKER_CMD ?= docker
-DESIRED_ROCM ?= 6.4
+DESIRED_ROCM ?= 7.0
 DESIRED_ROCM_SHORT = $(subst .,,$(DESIRED_ROCM))
 PACKAGE_NAME = magma-rocm
 # inherit this from underlying docker image, do not pass this env var to docker
-#PYTORCH_ROCM_ARCH ?= gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201
+#PYTORCH_ROCM_ARCH ?= gfx900;gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201
 
 DOCKER_RUN = set -eou pipefail; ${DOCKER_CMD} run --rm -i \
 	-v $(shell git rev-parse --show-toplevel)/.ci:/builder \
@@ -16,6 +16,7 @@ DOCKER_RUN = set -eou pipefail; ${DOCKER_CMD} run --rm -i \
 	magma-rocm/build_magma.sh
 
 .PHONY: all
+all: magma-rocm70
 all: magma-rocm64
 all: magma-rocm63
 
@@ -24,6 +25,11 @@ clean:
 	$(RM) -r magma-*
 	$(RM) -r output
 
+.PHONY: magma-rocm70
+magma-rocm70: DESIRED_ROCM := 7.0
+magma-rocm70:
+	$(DOCKER_RUN)
+
 .PHONY: magma-rocm64
 magma-rocm64: DESIRED_ROCM := 6.4
 magma-rocm64:

.ci/magma-rocm/build_magma.sh

@@ -6,8 +6,8 @@ set -eou pipefail
 # The script expects DESIRED_CUDA and PACKAGE_NAME to be set
 ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 
-# Version 2.7.2 + ROCm related updates
-MAGMA_VERSION=a1625ff4d9bc362906bd01f805dbbe12612953f6
+# https://github.com/icl-utk-edu/magma/pull/65
+MAGMA_VERSION=d6e4117bc88e73f06d26c6c2e14f064e8fc3d1ec
 
 # Folders for the build
 PACKAGE_FILES=${ROOT_DIR}/magma-rocm/package_files # metadata
@@ -20,7 +20,7 @@ mkdir -p ${PACKAGE_DIR} ${PACKAGE_OUTPUT}/linux-64 ${PACKAGE_BUILD} ${PACKAGE_RE
 
 # Fetch magma sources and verify checksum
 pushd ${PACKAGE_DIR}
-git clone https://bitbucket.org/icl/magma.git
+git clone https://github.com/jeffdaily/magma
 pushd magma
 git checkout ${MAGMA_VERSION}
 popd

.ci/pytorch/cpp_doc_push_script.sh

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

.ci/pytorch/macos-build.sh

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

.ci/pytorch/macos-test.sh

@@ -13,13 +13,9 @@ if [[ ! $(python -c "import torch; print(int(torch.backends.openmp.is_available(
 fi
 popd
 
-python -mpip install -r requirements.txt
-
 # enable debug asserts in serialization
 export TORCH_SERIALIZATION_DEBUG=1
 
-python -mpip install --no-input -r requirements.txt
-
 setup_test_python() {
   # The CircleCI worker hostname doesn't resolve to an address.
   # This environment variable makes ProcessGroupGloo default to
@@ -59,7 +55,7 @@ test_python_shard() {
 
   setup_test_python
 
-  time python test/run_test.py --verbose --exclude-jit-executor --exclude-distributed-tests --shard "$1" "$NUM_TEST_SHARDS"
+  time python test/run_test.py --verbose --exclude-jit-executor --exclude-distributed-tests --exclude-quantization-tests --shard "$1" "$NUM_TEST_SHARDS"
 
   assert_git_not_dirty
 }

.ci/pytorch/test.sh

@@ -322,14 +322,14 @@ test_python_shard() {
 
   # 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 --upload-artifacts-while-running
+  time python test/run_test.py --exclude-jit-executor --exclude-distributed-tests --exclude-quantization-tests $INCLUDE_CLAUSE --shard "$1" "$NUM_TEST_SHARDS" --verbose $PYTHON_TEST_EXTRA_OPTION --upload-artifacts-while-running
 
   assert_git_not_dirty
 }
 
 test_python() {
   # shellcheck disable=SC2086
-  time python test/run_test.py --exclude-jit-executor --exclude-distributed-tests $INCLUDE_CLAUSE --verbose $PYTHON_TEST_EXTRA_OPTION
+  time python test/run_test.py --exclude-jit-executor --exclude-distributed-tests --exclude-quantization-tests $INCLUDE_CLAUSE --verbose $PYTHON_TEST_EXTRA_OPTION
   assert_git_not_dirty
 }
 
@@ -390,6 +390,7 @@ test_dynamo_wrapped_shard() {
     --exclude-distributed-tests \
     --exclude-torch-export-tests \
     --exclude-aot-dispatch-tests \
+    --exclude-quantization-tests \
     --shard "$1" "$NUM_TEST_SHARDS" \
     --verbose \
     --upload-artifacts-while-running
@@ -1162,6 +1163,12 @@ test_distributed() {
   fi
 }
 
+test_quantization() {
+  echo "Testing quantization"
+
+  python test/test_quantization.py
+}
+
 test_rpc() {
   echo "Testing RPC C++ tests"
   # NB: the ending test_rpc must match the current function name for the current
@@ -1579,7 +1586,7 @@ test_executorch() {
 test_linux_aarch64() {
   python test/run_test.py --include test_modules test_mkldnn test_mkldnn_fusion test_openmp test_torch test_dynamic_shapes \
         test_transformers test_multiprocessing test_numpy_interop test_autograd test_binary_ufuncs test_complex test_spectral_ops \
-        test_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops test_ops \
+        test_foreach test_reductions test_unary_ufuncs test_tensor_creation_ops test_ops profiler/test_memory_profiler \
         distributed/elastic/timer/api_test distributed/elastic/timer/local_timer_example distributed/elastic/timer/local_timer_test \
         --shard "$SHARD_NUMBER" "$NUM_TEST_SHARDS" --verbose
 
@@ -1655,6 +1662,8 @@ elif [[ "${TEST_CONFIG}" == *executorch* ]]; then
   test_executorch
 elif [[ "$TEST_CONFIG" == 'jit_legacy' ]]; then
   test_python_legacy_jit
+elif [[ "$TEST_CONFIG" == 'quantization' ]]; then
+  test_quantization
 elif [[ "${BUILD_ENVIRONMENT}" == *libtorch* ]]; then
   # TODO: run some C++ tests
   echo "no-op at the moment"
@@ -1785,6 +1794,8 @@ elif [[ "${TEST_CONFIG}" == h100_distributed ]]; then
   test_h100_distributed
 elif [[ "${TEST_CONFIG}" == "h100-symm-mem" ]]; then
   test_h100_symm_mem
+elif [[ "${TEST_CONFIG}" == "b200-symm-mem" ]]; then
+  test_h100_symm_mem
 elif [[ "${TEST_CONFIG}" == h100_cutlass_backend ]]; then
   test_h100_cutlass_backend
 else

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

@@ -25,7 +25,7 @@ echo Copying over test times file
 robocopy /E "%PYTORCH_FINAL_PACKAGE_DIR_WIN%\.additional_ci_files" "%PROJECT_DIR_WIN%\.additional_ci_files"
 
 echo Run nn tests
-python run_test.py --exclude-jit-executor --exclude-distributed-tests --shard "%SHARD_NUMBER%" "%NUM_TEST_SHARDS%" --verbose
+python run_test.py --exclude-jit-executor --exclude-distributed-tests --exclude-quantization-tests --shard "%SHARD_NUMBER%" "%NUM_TEST_SHARDS%" --verbose
 if ERRORLEVEL 1 goto fail
 
 popd

.ci/pytorch/windows/internal/static_lib_test.bat

@@ -63,7 +63,7 @@ if errorlevel 1 exit /b 1
 call %CONDA_HOME%\condabin\activate.bat testenv
 if errorlevel 1 exit /b 1
 
-call conda install  -y -q -c conda-forge libuv=1.39
+call conda install  -y -q -c conda-forge libuv=1.51
 call conda install -y -q intel-openmp
 
 echo "install and test libtorch"

.ci/wheel/build_wheel.sh

@@ -177,8 +177,7 @@ source ~/${desired_python}-build/bin/activate
 retry pip install "${PINNED_PACKAGES[@]}" -r "${pytorch_rootdir}/requirements.txt"
 retry brew install libomp
 
-# For USE_DISTRIBUTED=1 on macOS, this enables gloo, which needs libuv, which
-# is build as part of tensorpipe submodule
+# For USE_DISTRIBUTED=1 on macOS, need libuv, which is build as part of tensorpipe submodule
 export USE_DISTRIBUTED=1
 
 export USE_MKLDNN=OFF

.clang-tidy

@@ -69,6 +69,8 @@ readability-string-compare,
 '
 HeaderFilterRegex: '^(aten/|c10/|torch/).*$'
 WarningsAsErrors: '*'
+LineFilter:
+  - name: '/usr/include/.*'
 CheckOptions:
   cppcoreguidelines-special-member-functions.AllowSoleDefaultDtor: true
   cppcoreguidelines-special-member-functions.AllowImplicitlyDeletedCopyOrMove: true

.github/actionlint.yaml

@@ -22,6 +22,9 @@ self-hosted-runner:
     - linux.arm64.m7g.4xlarge
     - linux.arm64.m7g.4xlarge.ephemeral
     - linux.arm64.r7g.12xlarge.memory
+    - linux.aws.h100
+    - linux.aws.h100.4
+    - linux.aws.h100.8
     - linux.4xlarge.nvidia.gpu
     - linux.8xlarge.nvidia.gpu
     - linux.16xlarge.nvidia.gpu

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

@@ -59,7 +59,7 @@ runs:
         set -x
 
         # Create new py_tmp env with python-version
-        ${CONDA} create -y -n py_tmp python=${PYTHON_VERSION} intel-openmp
+        ${CONDA} create -y -n py_tmp python=${PYTHON_VERSION} intel-openmp libuv
 
         PYTHON3=$(${CONDA_RUN} -n py_tmp which python3)
         EXIT_CODE=$?

.github/ci_commit_pins/vllm.txt

@@ -1 +1 @@
-090197034faf3b193c4467cedeb9281e3078892d
+8bf8f4582208ac7af230512ff5f3ac1dc36d5222

.github/merge_rules.yaml

@@ -525,6 +525,21 @@
   - Lint
   - pull
 
+- name: typechecking
+  patterns:
+  - 'pyrefly.toml'
+  - 'mypy.ini'
+  - 'mypy-strict.ini'
+  approved_by:
+  - lolpack
+  - maggiemoss
+  - ndmitchell
+  - kinto0
+  mandatory_checks_name:
+  - EasyCLA
+  - Lint
+  - pull
+
 - name: superuser
   patterns:
   - '*'

.github/pytorch-probot.yml

@@ -1,42 +1,44 @@
 tracking_issue: 24422
 ciflow_tracking_issue: 64124
 ciflow_push_tags:
+- ciflow/b200
+- ciflow/b200-symm-mem
 - ciflow/binaries
 - ciflow/binaries_libtorch
 - ciflow/binaries_wheel
-- ciflow/triton_binaries
+- ciflow/h100
+- ciflow/h100-cutlass-backend
+- ciflow/h100-distributed
+- ciflow/h100-symm-mem
 - ciflow/inductor
-- ciflow/inductor-periodic
-- ciflow/inductor-rocm
-- ciflow/inductor-perf-test-nightly-rocm
-- ciflow/inductor-perf-compare
+- ciflow/inductor-cu126
 - ciflow/inductor-micro-benchmark
 - ciflow/inductor-micro-benchmark-cpu-x86
+- ciflow/inductor-perf-compare
+- ciflow/inductor-perf-test-nightly-rocm
 - ciflow/inductor-perf-test-nightly-x86-zen
-- ciflow/inductor-cu126
+- ciflow/inductor-periodic
+- ciflow/inductor-rocm
 - ciflow/linux-aarch64
 - ciflow/mps
 - ciflow/nightly
+- ciflow/op-benchmark
 - ciflow/periodic
 - ciflow/periodic-rocm-mi300
+- ciflow/pull
+- ciflow/quantization-periodic
+- ciflow/riscv64
 - ciflow/rocm
 - ciflow/rocm-mi300
 - ciflow/s390
-- ciflow/riscv64
 - ciflow/slow
+- ciflow/torchbench
+- ciflow/triton_binaries
 - ciflow/trunk
 - ciflow/unstable
-- ciflow/xpu
 - ciflow/vllm
-- ciflow/torchbench
-- ciflow/op-benchmark
-- ciflow/pull
-- ciflow/h100
-- ciflow/h100-distributed
 - ciflow/win-arm64
-- ciflow/h100-symm-mem
-- ciflow/h100-cutlass-backend
-- ciflow/b200
+- ciflow/xpu
 retryable_workflows:
 - pull
 - trunk
@@ -45,4 +47,4 @@ retryable_workflows:
 - inductor-A100-perf-nightly
 labeler_config: labeler.yml
 label_to_label_config: label_to_label.yml
-mergebot: True
+mergebot: true

.github/scripts/generate_binary_build_matrix.py

@@ -30,7 +30,7 @@ CUDA_ARCHES_CUDNN_VERSION = {
 }
 
 # NOTE: Please also update the ROCm sources in `PIP_SOURCES` in tools/nightly.py when changing this
-ROCM_ARCHES = ["6.3", "6.4"]
+ROCM_ARCHES = ["6.4", "7.0"]
 
 XPU_ARCHES = ["xpu"]
 

.github/scripts/generate_ci_workflows.py

@@ -155,7 +155,7 @@ LINUX_BINARY_SMOKE_WORKFLOWS = [
         package_type="manywheel",
         build_configs=generate_binary_build_matrix.generate_wheels_matrix(
             OperatingSystem.LINUX,
-            arches=["12.8"],
+            arches=["13.0"],
             python_versions=["3.12"],
         ),
         branches="main",

.github/templates/linux_binary_build_workflow.yml.j2

@@ -71,12 +71,15 @@ jobs:
     with:!{{ upload.binary_env_as_input(config) }}
       {%- if "aarch64" in build_environment %}
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       {%- elif "s390x" in build_environment %}
       runs_on: linux.s390x
       ALPINE_IMAGE: "docker.io/s390x/alpine"
       timeout-minutes: 420
+      {%- elif config["gpu_arch_type"] == "rocm" %}
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
       {%- elif "conda" in build_environment and config["gpu_arch_type"] == "cuda" %}
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.24xlarge.ephemeral

.github/workflows/_docs.yml

@@ -67,7 +67,7 @@ jobs:
             # an OOM issue when running the job, so this upgrades the runner from 4xlarge
             # to the next available tier of 12xlarge. So much memory just to generate cpp
             # doc
-            runner: ${{ inputs.runner_prefix }}linux.12xlarge
+            runner: ${{ inputs.runner_prefix }}linux.12xlarge.memory
             # TODO: Nightly cpp docs take longer and longer to finish (more than 3h now)
             # Let's try to figure out how this can be improved
             timeout-minutes: 360

.github/workflows/b200-symm-mem.yml

@@ -0,0 +1,60 @@
+name: Limited CI for symmetric memory tests on B200
+
+on:
+  pull_request:
+    paths:
+      - .github/workflows/b200-symm-mem.yml
+  workflow_dispatch:
+  push:
+    tags:
+      - ciflow/b200-symm-mem/*
+  schedule:
+    - cron: 22 8 * * *  # about 1:22am PDT
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
+  cancel-in-progress: true
+
+permissions:
+  id-token: write
+  contents: read
+
+jobs:
+
+  get-label-type:
+    if: github.repository_owner == 'pytorch'
+    name: get-label-type
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    with:
+      triggering_actor: ${{ github.triggering_actor }}
+      issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
+      curr_branch: ${{ github.head_ref || github.ref_name }}
+      curr_ref_type: ${{ github.ref_type }}
+
+  linux-jammy-cuda12_8-py3_10-gcc11-sm100-build-symm:
+    name: linux-jammy-cuda12.8-py3.10-gcc11-sm100-symm
+    uses: ./.github/workflows/_linux-build.yml
+    needs: get-label-type
+    with:
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      runner: linux.12xlarge.memory
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm100-symm
+      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
+      cuda-arch-list: '10.0'
+      test-matrix: |
+        { include: [
+          { config: "b200-symm-mem", shard: 1, num_shards: 1, runner: "linux.dgx.b200.8" },
+        ]}
+    secrets: inherit
+
+  linux-jammy-cuda12_8-py3_10-gcc11-sm100-test:
+    name: linux-jammy-cuda12.8-py3.10-gcc11-sm100-symm
+    uses: ./.github/workflows/_linux-test.yml
+    needs:
+      - linux-jammy-cuda12_8-py3_10-gcc11-sm100-build-symm
+    with:
+      build-environment: linux-jammy-cuda12.8-py3.10-gcc11-sm100-symm
+      docker-image: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm100-build-symm.outputs.docker-image }}
+      test-matrix: ${{ needs.linux-jammy-cuda12_8-py3_10-gcc11-sm100-build-symm.outputs.test-matrix }}
+      aws-role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
+    secrets: inherit

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

@@ -36,7 +36,7 @@ jobs:
     runs-on: linux.9xlarge.ephemeral
     strategy:
       matrix:
-        tag: ["cuda12.6", "cuda12.8", "cuda12.9", "cuda13.0", "rocm6.3", "rocm6.4", "cpu"]
+        tag: ["cuda12.6", "cuda12.8", "cuda12.9", "cuda13.0", "rocm6.3", "rocm6.4", "rocm7.0", "cpu"]
     steps:
       - name: Build docker image
         uses: pytorch/pytorch/.github/actions/binary-docker-build@main

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

@@ -52,8 +52,8 @@ jobs:
           { tag: "cuda12.9" },
           { tag: "cuda12.8" },
           { tag: "cuda12.6" },
-          { tag: "rocm6.3"  },
           { tag: "rocm6.4"  },
+          { tag: "rocm7.0"  },
           { tag: "cpu"      },
         ]
     steps:

.github/workflows/build-magma-rocm-linux.yml

@@ -34,7 +34,7 @@ jobs:
       id-token: write
     strategy:
       matrix:
-        rocm_version: ["64", "63"]
+        rocm_version: ["70", "64"]
     steps:
       - name: Checkout PyTorch
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

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

@@ -52,8 +52,8 @@ jobs:
           { name: "manylinuxaarch64-builder",       tag: "cuda13.0",          runner: "linux.arm64.2xlarge.ephemeral" },
           { name: "manylinuxaarch64-builder",       tag: "cuda12.8",          runner: "linux.arm64.2xlarge.ephemeral" },
           { name: "manylinuxaarch64-builder",       tag: "cuda12.6",          runner: "linux.arm64.2xlarge.ephemeral" },
-          { name: "manylinux2_28-builder",          tag: "rocm6.3",           runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "rocm6.4",           runner: "linux.9xlarge.ephemeral" },
+          { name: "manylinux2_28-builder",          tag: "rocm7.0",           runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28-builder",          tag: "cpu",               runner: "linux.9xlarge.ephemeral" },
           { name: "manylinux2_28_aarch64-builder",  tag: "cpu-aarch64",       runner: "linux.arm64.2xlarge.ephemeral" },
           { name: "manylinuxcxx11-abi-builder",     tag: "cpu-cxx11-abi",     runner: "linux.9xlarge.ephemeral" },

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

@@ -50,12 +50,12 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        py_vers: [ "3.9", "3.10", "3.11", "3.12", "3.13", "3.13t", "3.14", "3.14t" ]
+        py_vers: [ "3.10", "3.11", "3.12", "3.13", "3.13t", "3.14", "3.14t" ]
         device: ["cuda", "rocm", "xpu", "aarch64"]
         docker-image: ["pytorch/manylinux2_28-builder:cpu"]
         include:
           - device: "rocm"
-            rocm_version: "6.4"
+            rocm_version: "7.0"
             runs_on: "${{ needs.get-label-type.outputs.label-type }}linux.4xlarge"
           - device: "cuda"
             rocm_version: ""
@@ -108,9 +108,6 @@ jobs:
 
           # Determine python executable for given version
           case $PY_VERS in
-          3.9)
-            PYTHON_EXECUTABLE=/opt/python/cp39-cp39/bin/python
-            ;;
           3.10)
             PYTHON_EXECUTABLE=/opt/python/cp310-cp310/bin/python
             ;;
@@ -194,7 +191,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        py_vers: [ "3.9", "3.10", "3.11", "3.12", "3.13", "3.13t", "3.14", "3.14t" ]
+        py_vers: [ "3.10", "3.11", "3.12", "3.13", "3.13t", "3.14", "3.14t" ]
         device: ["xpu"]
     timeout-minutes: 40
     env:

.github/workflows/create_release.yml

@@ -35,6 +35,7 @@ jobs:
       contents: write
     outputs:
       pt_release_name: ${{ steps.release_name.outputs.pt_release_name }}
+      pt_pep517_release_name: ${{ steps.release_name.outputs.pt_pep517_release_name }}
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
@@ -53,8 +54,12 @@ jobs:
           tag_or_branch="${tag_or_branch#refs/heads/}"
           # replace directory separators with _ in branch name
           tag_or_branch="${tag_or_branch//\//_}"
-          echo "PT_RELEASE_NAME=pytorch-$tag_or_branch" >> "$GITHUB_ENV"
-          echo "PT_RELEASE_FILE=pytorch-$tag_or_branch.tar.gz" >> "$GITHUB_ENV"
+          torch_version="$(python -c 'from tools.generate_torch_version import get_torch_version; print(get_torch_version())')"
+          {
+            echo "PT_RELEASE_NAME=pytorch-$tag_or_branch";
+            echo "PT_RELEASE_FILE=pytorch-$tag_or_branch.tar.gz";
+            echo "PT_PEP517_RELEASE_FILE=torch-${torch_version}.tar.gz";
+          } >> "$GITHUB_ENV"
       - name: Checkout optional submodules
         run: python3 tools/optional_submodules.py
       - name: Copy docs requirements for inclusion
@@ -64,30 +69,47 @@ jobs:
           cp .ci/docker/requirements-docs.txt docs/requirements.txt
       - name: Create source distribution
         run: |
-            # Create new folder with specified name so extracting the archive yields that
-            rm -rf "/tmp/$PT_RELEASE_NAME"
-            cp -r "$PWD" "/tmp/$PT_RELEASE_NAME"
-            mv "/tmp/$PT_RELEASE_NAME" .
-            # Cleanup
-            rm -rf "$PT_RELEASE_NAME"/{.circleci,.ci}
-            find "$PT_RELEASE_NAME" -name '.git*' -exec rm -rv {} \; || true
-            # Create archive
-            tar -czf "$PT_RELEASE_FILE" "$PT_RELEASE_NAME"
-            echo "Created source archive $PT_RELEASE_FILE with content: $(ls -a "$PT_RELEASE_NAME")"
+          # Create new folder with specified name so extracting the archive yields that
+          rm -rf "/tmp/$PT_RELEASE_NAME"
+          cp -r "$PWD" "/tmp/$PT_RELEASE_NAME"
+          mv "/tmp/$PT_RELEASE_NAME" .
+          # Cleanup
+          rm -rf "$PT_RELEASE_NAME"/{.circleci,.ci}
+          find "$PT_RELEASE_NAME" -name '.git*' -exec rm -rv {} \; || true
+          # Create archive
+          tar -czf "$PT_RELEASE_FILE" "$PT_RELEASE_NAME"
+          echo "Created source archive $PT_RELEASE_FILE with content: $(ls -a "$PT_RELEASE_NAME")"
+      - name: Create PEP 517 compatible source distribution
+        run: |
+          pip install build==1.2.2.post1 || exit 1
+          python -m build --sdist || exit 1
+          cd dist || exit 1
       - name: Upload source distribution for release
         if: ${{ github.event_name == 'release' }}
         uses: softprops/action-gh-release@da05d552573ad5aba039eaac05058a918a7bf631 # v2.2.2
         with:
-          files: ${{env.PT_RELEASE_FILE}}
-      - name: Upload source distribution to GHA artifacts for release tags
+          files: |
+            ${{ env.PT_RELEASE_FILE }}
+            ${{ env.PT_PEP517_RELEASE_FILE }}
+      - name: Upload source distribution to GHA artifacts  # for release tags
         if: ${{ github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v') && contains(github.ref, 'rc') }}
         uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874 # v4.4.0
         with:
           name: ${{ env.PT_RELEASE_FILE }}
           path: ${{ env.PT_RELEASE_FILE }}
+      - name: Upload PEP 517 source distribution to GHA artifacts  # for release tags
+        if: ${{ github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v') && contains(github.ref, 'rc') }}
+        uses: actions/upload-artifact@50769540e7f4bd5e21e526ee35c689e35e0d6874 # v4.4.0
+        with:
+          name: ${{ env.PT_PEP517_RELEASE_FILE }}
+          path: dist/${{ env.PT_PEP517_RELEASE_FILE }}
       - name: Set output
         id: release_name
-        run: echo "pt_release_name=${{ env.PT_RELEASE_NAME }}.tar.gz" >> "${GITHUB_OUTPUT}"
+        run: |
+          {
+            echo "pt_release_name=${{ env.PT_RELEASE_FILE }}";
+            echo "pt_pep517_release_name=${{ env.PT_PEP517_RELEASE_FILE }}";
+          } >> "${GITHUB_OUTPUT}"
 
   upload_source_code_to_s3:
     if: ${{ github.repository == 'pytorch/pytorch' && github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v') && contains(github.ref, 'rc') }}
@@ -103,6 +125,9 @@ jobs:
       - uses: actions/download-artifact@65a9edc5881444af0b9093a5e628f2fe47ea3b2e # v4.1.7
         with:
           name: ${{ needs.release.outputs.pt_release_name }}
+      - uses: actions/download-artifact@65a9edc5881444af0b9093a5e628f2fe47ea3b2e # v4.1.7
+        with:
+          name: ${{ needs.release.outputs.pt_pep517_release_name }}
       - name: Configure AWS credentials(PyTorch account)
         uses: aws-actions/configure-aws-credentials@ececac1a45f3b08a01d2dd070d28d111c5fe6722 # v4.1.0
         with:
@@ -113,7 +138,9 @@ jobs:
           s3-bucket: pytorch
           s3-prefix: source_code/test
           if-no-files-found: warn
-          path: ${{ needs.release.outputs.pt_release_name }}
+          path: |
+            ${{ needs.release.outputs.pt_release_name }}
+            ${{ needs.release.outputs.pt_pep517_release_name }}
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name }}

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

@@ -62,7 +62,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -128,7 +128,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -174,7 +174,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -220,7 +220,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
@@ -265,7 +265,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.11"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -331,7 +331,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.11"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -377,7 +377,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.11"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -423,7 +423,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.11"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
@@ -468,7 +468,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.12"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -534,7 +534,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.12"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -580,7 +580,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.12"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -626,7 +626,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.12"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_12-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
@@ -671,7 +671,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.13"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -737,7 +737,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.13"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -783,7 +783,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.13"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -829,7 +829,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.13"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
@@ -874,7 +874,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.13t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13t-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -940,7 +940,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.13t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13t-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -986,7 +986,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.13t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13t-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -1032,7 +1032,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.13t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_13t-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
@@ -1077,7 +1077,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.14"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -1143,7 +1143,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.14"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -1189,7 +1189,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.14"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -1235,7 +1235,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.14"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel
@@ -1280,7 +1280,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cpu-aarch64
       DESIRED_PYTHON: "3.14t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14t-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
@@ -1346,7 +1346,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.6
       DESIRED_PYTHON: "3.14t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14t-cuda-aarch64-12_6
       build_environment: linux-aarch64-binary-manywheel
@@ -1392,7 +1392,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda12.8
       DESIRED_PYTHON: "3.14t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14t-cuda-aarch64-12_8
       build_environment: linux-aarch64-binary-manywheel
@@ -1438,7 +1438,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.14t"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      runs_on: linux.arm64.m7g.4xlarge.ephemeral
+      runs_on: linux.arm64.r7g.12xlarge.memory
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_14t-cuda-aarch64-13_0
       build_environment: linux-aarch64-binary-manywheel

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

@@ -316,120 +316,6 @@ jobs:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
 
-  libtorch-rocm6_3-shared-with-deps-release-build:
-    if: ${{ github.repository_owner == 'pytorch' }}
-    uses: ./.github/workflows/_binary-build-linux.yml
-    needs: get-label-type
-    with:
-      PYTORCH_ROOT: /pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: rocm6.3
-      GPU_ARCH_VERSION: "6.3"
-      GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: libtorch-cxx11-builder
-      DOCKER_IMAGE_TAG_PREFIX: rocm6.3
-      LIBTORCH_CONFIG: release
-      LIBTORCH_VARIANT: shared-with-deps
-      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: libtorch-rocm6_3-shared-with-deps-release
-      build_environment: linux-binary-libtorch
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-  libtorch-rocm6_3-shared-with-deps-release-test:  # Testing
-    if: ${{ github.repository_owner == 'pytorch' }}
-    needs:
-      - libtorch-rocm6_3-shared-with-deps-release-build
-      - get-label-type
-    runs-on: linux.rocm.gpu.mi250
-    timeout-minutes: 240
-    env:
-      PYTORCH_ROOT: /pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: rocm6.3
-      GPU_ARCH_VERSION: "6.3"
-      GPU_ARCH_TYPE: rocm
-      SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: libtorch-cxx11-builder
-      DOCKER_IMAGE_TAG_PREFIX: rocm6.3
-      LIBTORCH_CONFIG: release
-      LIBTORCH_VARIANT: shared-with-deps
-    steps:
-      - name: Setup ROCm
-        uses: ./.github/actions/setup-rocm
-      - uses: actions/download-artifact@v4.1.7
-        name: Download Build Artifacts
-        with:
-          name: libtorch-rocm6_3-shared-with-deps-release
-          path: "${{ runner.temp }}/artifacts/"
-      - name: Checkout PyTorch
-        uses: actions/checkout@v4
-        with:
-          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-          submodules: recursive
-          path: pytorch
-          show-progress: false
-      - name: Clean PyTorch checkout
-        run: |
-          # Remove any artifacts from the previous checkouts
-          git clean -fxd
-        working-directory: pytorch
-      - name: ROCm set GPU_FLAG
-        run: |
-          echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"
-      - name: configure aws credentials
-        id: aws_creds
-        if: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') }}
-        uses: aws-actions/configure-aws-credentials@v4
-        with:
-          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
-          aws-region: us-east-1
-          role-duration-seconds: 18000
-      - name: Calculate docker image
-        id: calculate-docker-image
-        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
-        with:
-          docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
-          docker-image-name: libtorch-cxx11-builder
-          custom-tag-prefix: rocm6.3
-          docker-build-dir: .ci/docker
-          working-directory: pytorch
-      - name: Pull Docker image
-        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
-        with:
-          docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
-      - name: Test Pytorch binary
-        uses: ./pytorch/.github/actions/test-pytorch-binary
-        env:
-          DOCKER_IMAGE: ${{ steps.calculate-docker-image.outputs.docker-image }}
-      - name: Teardown ROCm
-        uses: ./.github/actions/teardown-rocm
-  libtorch-rocm6_3-shared-with-deps-release-upload:  # Uploading
-    if: ${{ github.repository_owner == 'pytorch' }}
-    permissions:
-      id-token: write
-      contents: read
-    needs: libtorch-rocm6_3-shared-with-deps-release-test
-    with:
-      PYTORCH_ROOT: /pytorch
-      PACKAGE_TYPE: libtorch
-      # TODO: This is a legacy variable that we eventually want to get rid of in
-      #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: rocm6.3
-      GPU_ARCH_VERSION: "6.3"
-      GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: libtorch-cxx11-builder
-      DOCKER_IMAGE_TAG_PREFIX: rocm6.3
-      LIBTORCH_CONFIG: release
-      LIBTORCH_VARIANT: shared-with-deps
-      build_name: libtorch-rocm6_3-shared-with-deps-release
-    secrets:
-      github-token: ${{ secrets.GITHUB_TOKEN }}
-    uses: ./.github/workflows/_binary-upload.yml
-
   libtorch-rocm6_4-shared-with-deps-release-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
@@ -447,6 +333,7 @@ jobs:
       LIBTORCH_CONFIG: release
       LIBTORCH_VARIANT: shared-with-deps
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
       build_name: libtorch-rocm6_4-shared-with-deps-release
       build_environment: linux-binary-libtorch
     secrets:
@@ -543,3 +430,118 @@ jobs:
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
     uses: ./.github/workflows/_binary-upload.yml
+
+  libtorch-rocm7_0-shared-with-deps-release-build:
+    if: ${{ github.repository_owner == 'pytorch' }}
+    uses: ./.github/workflows/_binary-build-linux.yml
+    needs: get-label-type
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: libtorch
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: rocm7.0
+      GPU_ARCH_VERSION: "7.0"
+      GPU_ARCH_TYPE: rocm
+      DOCKER_IMAGE: libtorch-cxx11-builder
+      DOCKER_IMAGE_TAG_PREFIX: rocm7.0
+      LIBTORCH_CONFIG: release
+      LIBTORCH_VARIANT: shared-with-deps
+      runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
+      build_name: libtorch-rocm7_0-shared-with-deps-release
+      build_environment: linux-binary-libtorch
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+  libtorch-rocm7_0-shared-with-deps-release-test:  # Testing
+    if: ${{ github.repository_owner == 'pytorch' }}
+    needs:
+      - libtorch-rocm7_0-shared-with-deps-release-build
+      - get-label-type
+    runs-on: linux.rocm.gpu.mi250
+    timeout-minutes: 240
+    env:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: libtorch
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: rocm7.0
+      GPU_ARCH_VERSION: "7.0"
+      GPU_ARCH_TYPE: rocm
+      SKIP_ALL_TESTS: 1
+      DOCKER_IMAGE: libtorch-cxx11-builder
+      DOCKER_IMAGE_TAG_PREFIX: rocm7.0
+      LIBTORCH_CONFIG: release
+      LIBTORCH_VARIANT: shared-with-deps
+    steps:
+      - name: Setup ROCm
+        uses: ./.github/actions/setup-rocm
+      - uses: actions/download-artifact@v4.1.7
+        name: Download Build Artifacts
+        with:
+          name: libtorch-rocm7_0-shared-with-deps-release
+          path: "${{ runner.temp }}/artifacts/"
+      - name: Checkout PyTorch
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+          submodules: recursive
+          path: pytorch
+          show-progress: false
+      - name: Clean PyTorch checkout
+        run: |
+          # Remove any artifacts from the previous checkouts
+          git clean -fxd
+        working-directory: pytorch
+      - name: ROCm set GPU_FLAG
+        run: |
+          echo "GPU_FLAG=--device=/dev/mem --device=/dev/kfd --device=/dev/dri --group-add video --group-add daemon" >> "${GITHUB_ENV}"
+      - name: configure aws credentials
+        id: aws_creds
+        if: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') }}
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_s3_and_ecr_read_only
+          aws-region: us-east-1
+          role-duration-seconds: 18000
+      - name: Calculate docker image
+        id: calculate-docker-image
+        uses: pytorch/test-infra/.github/actions/calculate-docker-image@main
+        with:
+          docker-registry: ${{ startsWith(github.event.ref, 'refs/tags/ciflow/') && '308535385114.dkr.ecr.us-east-1.amazonaws.com' || 'docker.io' }}
+          docker-image-name: libtorch-cxx11-builder
+          custom-tag-prefix: rocm7.0
+          docker-build-dir: .ci/docker
+          working-directory: pytorch
+      - name: Pull Docker image
+        uses: pytorch/test-infra/.github/actions/pull-docker-image@main
+        with:
+          docker-image: ${{ steps.calculate-docker-image.outputs.docker-image }}
+      - name: Test Pytorch binary
+        uses: ./pytorch/.github/actions/test-pytorch-binary
+        env:
+          DOCKER_IMAGE: ${{ steps.calculate-docker-image.outputs.docker-image }}
+      - name: Teardown ROCm
+        uses: ./.github/actions/teardown-rocm
+  libtorch-rocm7_0-shared-with-deps-release-upload:  # Uploading
+    if: ${{ github.repository_owner == 'pytorch' }}
+    permissions:
+      id-token: write
+      contents: read
+    needs: libtorch-rocm7_0-shared-with-deps-release-test
+    with:
+      PYTORCH_ROOT: /pytorch
+      PACKAGE_TYPE: libtorch
+      # TODO: This is a legacy variable that we eventually want to get rid of in
+      #       favor of GPU_ARCH_VERSION
+      DESIRED_CUDA: rocm7.0
+      GPU_ARCH_VERSION: "7.0"
+      GPU_ARCH_TYPE: rocm
+      DOCKER_IMAGE: libtorch-cxx11-builder
+      DOCKER_IMAGE_TAG_PREFIX: rocm7.0
+      LIBTORCH_CONFIG: release
+      LIBTORCH_VARIANT: shared-with-deps
+      build_name: libtorch-rocm7_0-shared-with-deps-release
+    secrets:
+      github-token: ${{ secrets.GITHUB_TOKEN }}
+    uses: ./.github/workflows/_binary-upload.yml

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

@@ -42,7 +42,7 @@ jobs:
       issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
       curr_branch: ${{ github.head_ref || github.ref_name }}
       curr_ref_type: ${{ github.ref_type }}
-  manywheel-py3_12-cuda12_8-build:
+  manywheel-py3_12-cuda13_0-build:
     if: ${{ github.repository_owner == 'pytorch' }}
     uses: ./.github/workflows/_binary-build-linux.yml
     needs: get-label-type
@@ -51,22 +51,22 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu128
-      GPU_ARCH_VERSION: "12.8"
+      DESIRED_CUDA: cu130
+      GPU_ARCH_VERSION: "13.0"
       GPU_ARCH_TYPE: cuda
       DOCKER_IMAGE: manylinux2_28-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.12"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
-      build_name: manywheel-py3_12-cuda12_8
+      build_name: manywheel-py3_12-cuda13_0
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.8.93; platform_system == 'Linux' | nvidia-cuda-runtime-cu12==12.8.90; platform_system == 'Linux' | nvidia-cuda-cupti-cu12==12.8.90; platform_system == 'Linux' | nvidia-cudnn-cu12==9.10.2.21; platform_system == 'Linux' | nvidia-cublas-cu12==12.8.4.1; platform_system == 'Linux' | nvidia-cufft-cu12==11.3.3.83; platform_system == 'Linux' | nvidia-curand-cu12==10.3.9.90; platform_system == 'Linux' | nvidia-cusolver-cu12==11.7.3.90; platform_system == 'Linux' | nvidia-cusparse-cu12==12.5.8.93; platform_system == 'Linux' | nvidia-cusparselt-cu12==0.7.1; platform_system == 'Linux' | nvidia-nccl-cu12==2.27.5; platform_system == 'Linux' | nvidia-nvshmem-cu12==3.3.24; platform_system == 'Linux' | nvidia-nvtx-cu12==12.8.90; platform_system == 'Linux' | nvidia-nvjitlink-cu12==12.8.93; platform_system == 'Linux' | nvidia-cufile-cu12==1.13.1.3; platform_system == 'Linux'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc==13.0.48; platform_system == 'Linux' | nvidia-cuda-runtime==13.0.48; platform_system == 'Linux' | nvidia-cuda-cupti==13.0.48; platform_system == 'Linux' | nvidia-cudnn-cu13==9.13.0.50; platform_system == 'Linux' | nvidia-cublas==13.0.0.19; platform_system == 'Linux' | nvidia-cufft==12.0.0.15; platform_system == 'Linux' | nvidia-curand==10.4.0.35; platform_system == 'Linux' | nvidia-cusolver==12.0.3.29; platform_system == 'Linux' | nvidia-cusparse==12.6.2.49; platform_system == 'Linux' | nvidia-cusparselt-cu13==0.8.0; platform_system == 'Linux' | nvidia-nccl-cu13==2.27.7; platform_system == 'Linux' | nvidia-nvshmem-cu13==3.3.24; platform_system == 'Linux' | nvidia-nvtx==13.0.39; platform_system == 'Linux' | nvidia-nvjitlink==13.0.39; platform_system == 'Linux' | nvidia-cufile==1.15.0.42; platform_system == 'Linux'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
-  manywheel-py3_12-cuda12_8-test:  # Testing
+  manywheel-py3_12-cuda13_0-test:  # Testing
     if: ${{ github.repository_owner == 'pytorch' }}
     needs:
-      - manywheel-py3_12-cuda12_8-build
+      - manywheel-py3_12-cuda13_0-build
       - get-label-type
     uses: ./.github/workflows/_binary-test-linux.yml
     with:
@@ -74,13 +74,13 @@ jobs:
       PACKAGE_TYPE: manywheel
       # TODO: This is a legacy variable that we eventually want to get rid of in
       #       favor of GPU_ARCH_VERSION
-      DESIRED_CUDA: cu128
-      GPU_ARCH_VERSION: "12.8"
+      DESIRED_CUDA: cu130
+      GPU_ARCH_VERSION: "13.0"
       GPU_ARCH_TYPE: cuda
       DOCKER_IMAGE: manylinux2_28-builder
-      DOCKER_IMAGE_TAG_PREFIX: cuda12.8
+      DOCKER_IMAGE_TAG_PREFIX: cuda13.0
       DESIRED_PYTHON: "3.12"
-      build_name: manywheel-py3_12-cuda12_8
+      build_name: manywheel-py3_12-cuda13_0
       build_environment: linux-binary-manywheel
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       runs_on: linux.g4dn.4xlarge.nvidia.gpu # 12.8+ builds need sm_70+ runner

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

@@ -60,6 +60,7 @@ jobs:
       DOCKER_IMAGE_TAG_PREFIX: rocm6.4
       DESIRED_PYTHON: "3.10"
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
+      timeout-minutes: 300
       build_name: manywheel-py3_10-rocm6_4
       build_environment: linux-binary-manywheel-rocm
     secrets:

.github/workflows/pull.yml

@@ -127,6 +127,8 @@ jobs:
     uses: ./.github/workflows/_linux-build.yml
     needs: get-label-type
     with:
+      # More memory is needed to build with asan
+      runner: linux.2xlarge.memory
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       build-environment: linux-jammy-py3.10-clang18-asan
       docker-image-name: ci-image:pytorch-linux-jammy-py3-clang18-asan

.github/workflows/quantization-periodic.yml

@@ -0,0 +1,54 @@
+name: quantization-periodic
+
+on:
+  push:
+    tags:
+      - ciflow/quantization-periodic/*
+  workflow_dispatch:
+  schedule:
+    # run weekly
+    - cron: "45 0 * * 0"
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  cancel-in-progress: true
+
+permissions:
+  id-token: write
+  contents: read
+
+jobs:
+  get-default-label-prefix:
+    name: get-default-label-prefix
+    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
+    if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }}
+    with:
+      triggering_actor: ${{ github.triggering_actor }}
+      issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
+      curr_branch: ${{ github.head_ref || github.ref_name }}
+      curr_ref_type: ${{ github.ref_type }}
+      opt_out_experiments: lf
+
+  periodic-quantization-build:
+    name: periodic-quantization-build
+    uses: ./.github/workflows/_linux-build.yml
+    needs: get-default-label-prefix
+    with:
+      runner_prefix: "${{ needs.get-default-label-prefix.outputs.label-type }}"
+      build-environment: linux-jammy-cuda12.8-cudnn9-py3-gcc11
+      docker-image-name: ci-image:pytorch-linux-jammy-cuda12.8-cudnn9-py3-gcc11
+      cuda-arch-list: '8.9'
+      test-matrix: |
+        { include: [
+          { config: "quantization", shard: 1, num_shards: 1, runner: "${{ needs.get-default-label-prefix.outputs.label-type }}linux.g6.4xlarge.experimental.nvidia.gpu" },
+        ]}
+    secrets: inherit
+  periodic-test-quantization:
+    name: periodic-test-quantization
+    uses: ./.github/workflows/_linux-test.yml
+    needs: periodic-quantization-build
+    with:
+      build-environment: linux-jammy-cuda12.8-cudnn9-py3-gcc11
+      docker-image: ${{ needs.periodic-quantization-build.outputs.docker-image }}
+      test-matrix: ${{ needs.periodic-quantization-build.outputs.test-matrix }}
+    secrets: inherit

.github/workflows/slow.yml

@@ -140,6 +140,8 @@ jobs:
     uses: ./.github/workflows/_linux-build.yml
     needs: get-label-type
     with:
+      # More memory is needed to build with asan
+      runner: linux.2xlarge.memory
       runner_prefix: "${{ needs.get-label-type.outputs.label-type }}"
       build-environment: linux-jammy-py3.10-clang18-asan
       docker-image-name: ci-image:pytorch-linux-jammy-py3-clang18-asan

.gitignore

@@ -82,6 +82,7 @@ torch/return_types.pyi
 torch/nn/functional.pyi
 torch/utils/data/datapipes/datapipe.pyi
 torch/csrc/autograd/generated/*
+torch/csrc/functionalization/generated/*
 torch/csrc/lazy/generated/*.[!m]*
 torch_compile_debug/
 # Listed manually because some files in this directory are not generated

.lintrunner.toml

@@ -49,7 +49,7 @@ init_command = [
     'mccabe==0.7.0',
     'pycodestyle==2.14.0',
     'pyflakes==3.4.0',
-    'torchfix==0.4.0 ; python_version >= "3.9" and python_version < "3.13"',
+    'torchfix==0.4.0 ; python_version >= "3.10" and python_version < "3.13"',
 ]
 
 
@@ -153,7 +153,7 @@ init_command = [
     'python3',
     'tools/linter/adapters/pip_init.py',
     '--dry-run={{DRYRUN}}',
-    'numpy==1.26.4 ; python_version >= "3.9" and python_version <= "3.11"',
+    'numpy==1.26.4 ; python_version >= "3.10" and python_version <= "3.11"',
     'numpy==2.1.0 ; python_version >= "3.12"',
     'expecttest==0.3.0',
     'mypy==1.16.0',
@@ -1453,7 +1453,7 @@ init_command = [
     '--dry-run={{DRYRUN}}',
     'usort==1.0.8.post1',
     'isort==6.0.1',
-    'ruff==0.12.9',  # sync with RUFF
+    'ruff==0.13.1',  # sync with RUFF
 ]
 is_formatter = true
 
@@ -1587,7 +1587,7 @@ init_command = [
     'python3',
     'tools/linter/adapters/pip_init.py',
     '--dry-run={{DRYRUN}}',
-    'ruff==0.12.9',  # sync with PYFMT
+    'ruff==0.13.1',  # sync with PYFMT
 ]
 is_formatter = true
 

BUILD.bazel

@@ -22,6 +22,7 @@ COMMON_COPTS = [
     "-DHAVE_SHM_UNLINK=1",
     "-D_FILE_OFFSET_BITS=64",
     "-DUSE_FBGEMM",
+    "-DUSE_DISTRIBUTED",
     "-DAT_PER_OPERATOR_HEADERS",
     "-DATEN_THREADING=NATIVE",
     "-DNO_CUDNN_DESTROY_HANDLE",
@@ -90,6 +91,8 @@ generated_cpu_cpp = [
     "aten/src/ATen/NativeMetaFunctions.h",
     "aten/src/ATen/RegistrationDeclarations.h",
     "aten/src/ATen/VmapGeneratedPlumbing.h",
+    "aten/src/ATen/ViewMetaClasses.h",
+    "aten/src/ATen/ViewMetaClasses.cpp",
     "aten/src/ATen/core/aten_interned_strings.h",
     "aten/src/ATen/core/enum_tag.h",
     "aten/src/ATen/core/TensorBody.h",
@@ -810,7 +813,7 @@ cc_library(
     name = "torch_python",
     srcs = libtorch_python_core_sources
         + if_cuda(libtorch_python_cuda_sources)
-        + libtorch_python_distributed_sources
+        + if_cuda(libtorch_python_distributed_sources)
         + GENERATED_AUTOGRAD_PYTHON,
     hdrs = glob([
         "torch/csrc/generic/*.cpp",
@@ -832,36 +835,6 @@ pybind_extension(
     ],
 )
 
-cc_library(
-    name = "functorch",
-    hdrs = glob([
-        "functorch/csrc/dim/*.h",
-    ]),
-    srcs = glob([
-        "functorch/csrc/dim/*.cpp",
-    ]),
-    deps = [
-        ":aten_nvrtc",
-        ":torch_python",
-        "@pybind11",
-    ],
-)
-
-pybind_extension(
-    name = "functorch/_C",
-    copts=[
-        "-DTORCH_EXTENSION_NAME=_C"
-    ],
-    srcs = [
-        "functorch/csrc/init_dim_only.cpp",
-    ],
-    deps = [
-        ":functorch",
-        ":torch_python",
-        ":aten_nvrtc",
-    ],
-)
-
 cc_binary(
     name = "torch/bin/torch_shm_manager",
     srcs = [
@@ -902,7 +875,6 @@ py_library(
     ],
     data = [
         ":torch/_C.so",
-        ":functorch/_C.so",
         ":torch/bin/torch_shm_manager",
     ],
 )
@@ -1105,6 +1077,7 @@ test_suite(
         "aten/src/ATen/templates/LazyNonNativeIr.h",
         "aten/src/ATen/templates/RegisterDispatchKey.cpp",
         "aten/src/ATen/templates/RegisterDispatchDefinitions.ini",
+        "aten/src/ATen/templates/ViewMetaClassesPythonBinding.cpp",
         "aten/src/ATen/native/native_functions.yaml",
         "aten/src/ATen/native/tags.yaml",
         "aten/src/ATen/native/ts_native_functions.yaml",

CMakeLists.txt

@@ -180,9 +180,8 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(ppc64le)")
   set(CPU_POWER ON)
 endif()
 
-# For non-supported platforms, turn USE_DISTRIBUTED off by default.
-# NB: USE_DISTRIBUTED simply disables the backend; distributed code
-# still gets built
+# For non-supported platforms, turn USE_DISTRIBUTED off by default. It is not
+# tested and likely won't work without additional changes.
 if(NOT LINUX AND NOT WIN32)
   set(USE_DISTRIBUTED
       OFF
@@ -262,11 +261,11 @@ option(USE_PYTORCH_METAL "Use Metal for PyTorch iOS build" OFF)
 option(USE_PYTORCH_METAL_EXPORT "Export Metal models on MacOSX desktop" OFF)
 option(USE_NATIVE_ARCH "Use -march=native" OFF)
 cmake_dependent_option(USE_MPS "Use MPS for macOS build" ON "MPS_FOUND" OFF)
-option(USE_DISTRIBUTED "Enable default distributed backends" ON)
+option(USE_DISTRIBUTED "Use distributed" ON)
 cmake_dependent_option(USE_NCCL "Use NCCL" ON
                        "USE_DISTRIBUTED;USE_CUDA OR USE_ROCM;UNIX;NOT APPLE" OFF)
 cmake_dependent_option(USE_XCCL "Use XCCL" ON
-                       "USE_DISTRIBUTED;USE_XPU;UNIX;NOT APPLE" OFF)
+                       "USE_XPU;UNIX;NOT APPLE" OFF)
 cmake_dependent_option(USE_RCCL "Use RCCL" ON USE_NCCL OFF)
 cmake_dependent_option(USE_RCCL "Use RCCL" ON "USE_NCCL;NOT WIN32" OFF)
 cmake_dependent_option(USE_STATIC_NCCL "Use static NCCL" OFF "USE_NCCL" OFF)
@@ -438,11 +437,12 @@ if(WIN32)
       PATH_SUFFIXES lib
       NO_DEFAULT_PATH)
     if(NOT libuv_tmp_LIBRARY)
+      set(USE_DISTRIBUTED OFF)
       set(USE_GLOO OFF)
       message(
         WARNING
-          "Libuv is not installed in current conda env. Set USE_GLOO to OFF. "
-          "Please run command 'conda install -c conda-forge libuv=1.39' to install libuv."
+          "Libuv is not installed in current conda env. Set USE_DISTRIBUTED to OFF. "
+          "Please run command 'conda install -c conda-forge libuv=1.51' to install libuv."
       )
     else()
       set(ENV{libuv_ROOT} ${libuv_tmp_LIBRARY}/../../)
@@ -1390,10 +1390,6 @@ endif()
 include(cmake/Summary.cmake)
 caffe2_print_configuration_summary()
 
-if(BUILD_FUNCTORCH)
-  add_subdirectory(functorch)
-endif()
-
 # Parse custom debug info
 if(DEFINED USE_CUSTOM_DEBINFO)
   string(REPLACE ";" " " SOURCE_FILES "${USE_CUSTOM_DEBINFO}")

MANIFEST.in

@@ -1,20 +1,61 @@
 # Reference: https://setuptools.pypa.io/en/latest/userguide/miscellaneous.html
 
-# Include source files in SDist
-include CMakeLists.txt
-include *.bzl *.bazel .bazel* BUILD *.BUILD BUILD.* WORKSPACE
-include BUCK BUCK.*
-include requirements*.txt
-include version.txt
-include [Mm]akefile *.[Mm]akefile [Mm]akefile.*
-include [Dd]ockerfile *.[Dd]ockerfile [Dd]ockerfile.* .dockerignore
+# Include individual top-level files
+include CITATION.cff
+include CODEOWNERS
+include Dockerfile
+include LICENSE
+include MANIFEST.in
+include Makefile
+include NOTICE
+include .bc-linter.yml
+include .clang-format .clang-tidy
+include .cmakelintrc
+include .coveragerc
+include .dockerignore
+include .editorconfig
+include .flake8
+include .gdbinit
+include .lintrunner.toml
+include .lldbinit
+include codex_setup.sh
+include docker.Makefile
+include pyrefly.toml
+include ubsan.supp
+
+# Include bazel and BUCK related files
+include BUILD.bazel BUCK.oss
+include WORKSPACE
+include *.bzl
+include .bazelignore .bazelrc .bazelversion
+
+# Include general configuration files
+include *.ini
+# Include important top-level information
+include *.md
+# Include technical text files at the moment, comprises
+# version.txt, CMakeLists.txt, requirements.txt
+include *.txt
+
+# Include ctags configuration
+include .ctags.d/*.ctags
+
+# Include subfolders completely
+graft .devcontainer
+graft .vscode
 graft android
 graft aten
+graft benchmarks
 graft binaries
 graft c10
 graft caffe2
 graft cmake
+graft docs
 graft functorch
+graft ios
+graft mypy_plugins
+graft scripts
+graft test
 graft third_party
 graft tools
 graft torch
@@ -22,29 +63,37 @@ graft torchgen
 # FIXME: torch-xla build during codegen will fail if include this file in wheel
 exclude torchgen/BUILD.bazel
 
-# Misc files and directories in SDist
-include *.md
-include CITATION.cff
-include LICENSE NOTICE
-include mypy*.ini
-graft benchmarks
-graft docs
-graft mypy_plugins
-graft scripts
+# The following exclusions omit parts from third-party dependencies that
+# contain invalid symlinks[1] and that are not needed for pytorch, such as
+# bindings for unused languages
+prune third_party/flatbuffers/java
+prune third_party/flatbuffers/kotlin
+prune third_party/ittapi/rust
+prune third_party/nccl/pkg/debian
+prune third_party/opentelemetry-cpp/third_party/prometheus-cpp/cmake/project-import-*
+
+# The following document is also an invalid symlink[1] and superfluous
+exclude third_party/flatbuffers/docs/source/CONTRIBUTING.md
+
+# Omit autogenerated code
+prune torchgen/packaged
+
+# Omit caches, compiled, and scm related content
+prune */__pycache__
+prune **/.github
+prune **/.gitlab
+global-exclude *.o *.obj *.so *.dylib *.a *.pxd *.dll *.lib
+global-exclude *.py[cod] *.swp *~
+global-exclude .git .git-blame-ignore-revs .gitattributes .gitignore .gitmodules
+global-exclude .gitlab-ci.yml
 
 # Misc files needed for custom setuptools command
 include .gitignore
 include .gitmodules
 
-# Include test suites in SDist
-graft test
-include pytest.ini
-include .coveragerc
+# [1] Invalid symlinks for the purposes of Python source distributions are,
+# according to the source distribution format[2] links pointing outside the
+# destination directory or links with a `..` component, which is those of
+# concern here.
 
-# Prune generated/compiled files
-prune torchgen/packaged
-prune */__pycache__
-global-exclude *.o *.obj *.so *.a *.dylib *.pxd *.dll *.lib *.py[cod]
-
-prune */.git
-global-exclude .git *~ *.swp
+# [2] https://packaging.python.org/en/latest/specifications/source-distribution-format/#source-distribution-archive-features

README.md

@@ -161,7 +161,7 @@ They require JetPack 4.2 and above, and [@dusty-nv](https://github.com/dusty-nv)
 
 #### Prerequisites
 If you are installing from source, you will need:
-- Python 3.9 or later
+- Python 3.10 or later
 - A compiler that fully supports C++17, such as clang or gcc (gcc 9.4.0 or newer is required, on Linux)
 - Visual Studio or Visual Studio Build Tool (Windows only)
 
@@ -275,7 +275,7 @@ conda install pkg-config libuv
 pip install mkl-static mkl-include
 # Add these packages if torch.distributed is needed.
 # Distributed package support on Windows is a prototype feature and is subject to changes.
-conda install -c conda-forge libuv
+conda install -c conda-forge libuv=1.51
 ```
 
 #### Install PyTorch

aten/src/ATen/ExpandUtils.h

@@ -468,7 +468,7 @@ inline Tensor _sum_to(
       // if we assume no reduction due to unbacked we ensure that at runtime.
       TORCH_MAYBE_SYM_CHECK(
           sym_eq(shape[i - leading_dims], sizes[i]),
-          "non-reduction path was assumed due to unabcked symbols expected those two sizes to be the same:",
+          "non-reduction path was assumed due to unbacked symbols expected those two sizes to be the same:",
           shape[i - leading_dims],
           ", ",
           sizes[i])

aten/src/ATen/FunctionalStorageImpl.cpp

@@ -9,11 +9,6 @@
 
 namespace at::functionalization {
 
-ViewMeta ViewMeta::to_out_idx(int64_t out_idx) {
-  if (out_idx == this->out_index) return *this;
-  return ViewMeta(forward_fn, reverse_fn, has_symbolic_inputs, is_multi_output, is_as_strided, out_idx);
-}
-
 // Note [Functionalization: Alias Removal Part 2]
 // See Note [Functionalization: Alias Removal] for more details.
 // This function applies a single update from one of the views to the StorageImpl.
@@ -42,12 +37,12 @@ ViewMeta ViewMeta::to_out_idx(int64_t out_idx) {
 static const Tensor apply_update(const FunctionalStorageImpl::Update& update, const Tensor& base) {
   at::Tensor t = update.new_val;
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(t));
-  if (update.view_metas.empty()) return t;
+  if (update.view_metas.empty()) { return t; }
 
   std::vector<at::Tensor> tmp_values({base});
   tmp_values.reserve(update.view_metas.size());
   for (size_t i = 0; i < update.view_metas.size() - 1; ++i) {
-    at::Tensor next_view = update.view_metas[i].forward_fn(tmp_values.back(), update.view_metas[i].out_index);
+    at::Tensor next_view = update.view_metas[i]->forward(tmp_values.back());
     // NB: We only actually need tmp_values for ops like select/slice/diagonal/squeeze/as_strided
     // All of these ops require additional information to recover the sizes of the original tensor.
     // If need to, we could probably apply this optimization and only bother computing tmp_values
@@ -55,9 +50,8 @@ static const Tensor apply_update(const FunctionalStorageImpl::Update& update, co
     tmp_values.push_back(std::move(next_view));
   }
   for(int64_t i = static_cast<int64_t>(update.view_metas.size()) - 1; i >= 0; --i) {
-    int64_t out_idx = update.view_metas[i].out_index;
     // Each view inverse is implemented in ViewInverses.cpp.
-    t = update.view_metas[i].reverse_fn(tmp_values[i], t, out_idx);
+    t = update.view_metas[i]->reverse(tmp_values[i], t);
   }
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(t));
   return t;
@@ -111,13 +105,13 @@ FunctionalStorageImpl::FunctionalStorageImpl(const Tensor& base)
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(base_));
 }
 
-void FunctionalStorageImpl::add_update(const Tensor& updated_val, const std::vector<ViewMeta>& metas) {
+void FunctionalStorageImpl::add_update(const Tensor& updated_val, const std::vector<std::shared_ptr<ViewMeta>>& metas) {
   TORCH_CHECK(!frozen_, "cannot mutate tensors with frozen storage");
 
   if (metas.size() > 1) {
     for (size_t i = 1; i < metas.size(); ++i) {
       // Skipping this check for XLA. Would be good to add it back, but it is failing XLA CI
-      TORCH_CHECK(updated_val.device().type() == c10::DeviceType::XLA || !metas[i].is_as_strided,
+      TORCH_CHECK(updated_val.device().type() == c10::DeviceType::XLA || !metas[i]->is_as_strided,
 "During torch.compile, encountered a mutation on a view chain of length ", metas.size(), ", where view ", i,
 " was an as_strided() call. as_strided() is non-compositional, and therefore is not possible to functionalize properly today,"
 "so this behavior is banned in compile. As a workaround, you can either remove the mutation from the model code, or you "

aten/src/ATen/FunctionalStorageImpl.h

@@ -8,44 +8,89 @@ namespace at::functionalization {
 
 // See Note [Functionalization Pass In Core]
 
+enum class InverseReturnMode {
+  /// Specifies that functional inverses should always return a view.
+  AlwaysView,
+  /// Specifies that functional inverses should always return a non-view / copy.
+  NeverView,
+  /// Specifies that functional inverses should return a view unless a (copying)
+  /// scatter
+  /// inverse exists, in which case that will be used instead.
+  /// This avoids as_strided() calls that can be difficult for subclasses to
+  /// handle.
+  ViewOrScatterInverse,
+};
+
+#define FUNCTIONALIZATION_VIEWMETA_NAME(TYPE) \
+  static const char* name() {                 \
+    return #TYPE;                             \
+  }
+
+#define FUNCTIONALIZATION_VIEWMETA_SERIALIZABLE_TUPLE(...) \
+  using SerializableTuple = std::tuple<__VA_ARGS__>
+
 // ViewMeta is a class used by the functionalization pass to navigate between
 // a base tensor and a view tensor.
 // For example, if I call `b = a.view1(...)`
-// the functionalization pass will generate and store a ViewMeta on b that looks
-// like:
+// the functionalization pass will generate and store a ViewMeta specialization
+// for `view1` operation on b that looks like:
 //
-// ViewMeta(
-//   [<captures>](const Tensor& base, int64_t mutated_view_idx) {
-//     return base.view1(...);
-//   },
-//   [<captures>](const at::Tensor& base, const at::Tensor& mutated_view,
-//   int64_t mutated_view_idx) -> at::Tensor {
-//     return at::functionalization::impl::view1_inverse(base, mutated_view,
-//     ...);
+// struct TORCH_API view1_ViewMeta : public ViewMeta {
+//   FUNCTIONALIZATION_VIEWMETA_NAME(view1_ViewMeta);
+//   FUNCTIONALIZATION_VIEWMETA_SERIALIZABLE_TUPLE(
+//       bool /* reapply_views */,
+//       const std::vector<int64_t>&);
+//
+//   view1_ViewMeta(const SerializableTuple& tpl)
+//       : view1_ViewMeta(std::get<0>(tpl), std::get<1>(tpl)) {}
+//
+//   view1_ViewMeta(bool reapply_views, const std::vector<int64_t>& size)
+//       : ViewMeta(/*has_symbolic_inputs=*/false),
+//         reapply_views(reapply_views),
+//         size(size) {}
+//
+//   Tensor forward(const Tensor& base) override {
+//       return base.view1(...);
 //   }
 //
-// The forward_fn lambda describes how to replay view1 on a tensor.
+//   Tensor reverse(const Tensor& base, const Tensor& mutated_view) override {
+//       return at::functionalization::impl::view1_inverse(base, mutated_view,
+//       ...);
+//   }
 //
-// The reverse_fn lambda describes how, given a tensor that is already a view,
+//   SerializableTuple to_serializable_tuple() {
+//     return std::make_tuple(reapply_views, size);
+//   }
+//
+//   bool reapply_views;
+//   std::vector<int64_t> size;
+// };
+//
+// The forward function describes how to replay view1 on a tensor.
+//
+// The reverse function describes how, given a tensor that is already a view,
 // how to get the corresponding base tensor. See Note [Functionalization Pass:
 // View Inverses] for details.
+//
+// `SerializedTuple` is a typedef that defines an `std::tuple<...>` type
+// representing the `ViewMeta` instance state. Methods that take in/return such
+// a type are used for supporting pickle serialization.
 struct ViewMeta {
   ViewMeta(
-      std::function<Tensor(const Tensor&, int64_t)> forward,
-      std::function<Tensor(const Tensor&, const Tensor&, int64_t)> reverse,
       bool has_symbolic_inputs,
       bool is_multi_output = false,
       bool is_as_strided = false,
       int64_t out_idx = 0)
-      : forward_fn(std::move(forward)),
-        reverse_fn(std::move(reverse)),
-        out_index(out_idx),
+      : out_index(out_idx),
         is_multi_output(is_multi_output),
         is_as_strided(is_as_strided),
         has_symbolic_inputs(has_symbolic_inputs) {}
 
-  std::function<Tensor(const Tensor&, int64_t)> forward_fn;
-  std::function<Tensor(const Tensor&, const Tensor&, int64_t)> reverse_fn;
+  virtual ~ViewMeta() = default;
+
+  virtual Tensor forward(const Tensor& base) = 0;
+  virtual Tensor reverse(const Tensor& base, const Tensor& mutated_view) = 0;
+
   // See Note [out_idx in ViewMeta]
   int64_t out_index;
 
@@ -57,10 +102,17 @@ struct ViewMeta {
   // Tells us if this view operation has any symbolic inputs
   bool has_symbolic_inputs;
 
-  // Returns a copy of the current ViewMeta, if out_idx matches the current
-  // out_index. Otherwise, returns a new ViewMeta with the same forward/reverse
+  // Returns a new ViewMeta with the same forward/reverse
   // functions, but a new out index.
-  ViewMeta to_out_idx(int64_t out_idx);
+  //
+  // This method should be implemented by those `ViewMeta` that have more than
+  // one output.
+  virtual std::shared_ptr<ViewMeta> to_out_index(int64_t out_index) {
+    TORCH_CHECK_NOT_IMPLEMENTED(
+        false,
+        "ViewMeta::to_out_index not implemented. ",
+        "Likely because there's only one output.");
+  }
 };
 
 // FunctionalStorageImpl is a subclass of StorageImpl used by the
@@ -93,14 +145,14 @@ struct TORCH_API FunctionalStorageImpl : public c10::StorageImpl {
     // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
     const at::Tensor new_val;
     // NOLINTNEXTLINE(cppcoreguidelines-avoid-const-or-ref-data-members)
-    const std::vector<ViewMeta> view_metas;
+    const std::vector<std::shared_ptr<ViewMeta>> view_metas;
   };
 
   explicit FunctionalStorageImpl(const Tensor& value);
 
   void add_update(
       const Tensor& updated_val,
-      const std::vector<ViewMeta>& view_metas);
+      const std::vector<std::shared_ptr<ViewMeta>>& view_metas);
   bool apply_updates();
   const Tensor& base() {
     return base_;

aten/src/ATen/FunctionalTensorWrapper.cpp

@@ -129,17 +129,19 @@ void FunctionalTensorWrapper::freeze_storage() const {
 // - view_value: The output tensor that we need to wrap.
 // - base: The "base" of the view that `view_value` was generated from.
 // See Note [Functionalization: Alias Removal Part 2] for more details on the mutation replay logic.
-FunctionalTensorWrapper::FunctionalTensorWrapper(const Tensor& view_value, const FunctionalTensorWrapper* base, const functionalization::ViewMeta& meta)
-  : c10::TensorImpl(
-      c10::DispatchKeySet(DispatchKey::Functionalize),
-      view_value.dtype(),
-      base->storage().data_ptr().device()
-    ),
-    value_(view_value),
-    is_multi_output_view_(base->is_multi_output_view_ || meta.is_multi_output),
-    was_storage_changed_(base->was_storage_changed_),
-    is_symbolic_(base->is_symbolic_)
-{
+FunctionalTensorWrapper::FunctionalTensorWrapper(
+    const Tensor& view_value,
+    const FunctionalTensorWrapper* base,
+    const std::shared_ptr<functionalization::ViewMeta>& meta)
+    : c10::TensorImpl(
+          c10::DispatchKeySet(DispatchKey::Functionalize),
+          view_value.dtype(),
+          base->storage().data_ptr().device()),
+      value_(view_value),
+      is_multi_output_view_(
+          base->is_multi_output_view_ || meta->is_multi_output),
+      was_storage_changed_(base->was_storage_changed_),
+      is_symbolic_(base->is_symbolic_) {
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(value_));
   TORCH_INTERNAL_ASSERT(!value_.key_set().has(c10::DispatchKey::Functionalize));
   set_constructor_metadata();
@@ -148,11 +150,10 @@ FunctionalTensorWrapper::FunctionalTensorWrapper(const Tensor& view_value, const
       view_metas_ = base->view_metas_;  // copy
   }
   view_metas_.push_back(meta);
-  maybe_mark_symbolic(meta);
+  maybe_mark_symbolic(meta.get());
   storage_ = base->storage_; // alias this tensor's storage with the base tensor's
 }
 
-
 functionalization::FunctionalStorageImpl* FunctionalTensorWrapper::functional_storage_impl() const {
   return static_cast<functionalization::FunctionalStorageImpl*>(storage_.unsafeGetStorageImpl());
 }
@@ -176,18 +177,18 @@ bool FunctionalTensorWrapper::is_up_to_date() const {
 }
 
 // See Note [Functionalization Pass - Inplace View Ops]
-void FunctionalTensorWrapper::mutate_view_meta(const at::functionalization::ViewMeta& meta) {
+void FunctionalTensorWrapper::mutate_view_meta(const std::shared_ptr<at::functionalization::ViewMeta>& meta) {
   view_metas_.push_back(meta);
   // Manually track the fact that this tensor received a metadata mutation!
   has_metadata_mutation_ = true;
   // Mark this tensor as being symbolic if there are any symbolic inputs used by the view operation.
-  maybe_mark_symbolic(meta);
+  maybe_mark_symbolic(meta.get());
   // Note [Functionalization Pass - Inplace View Ops]
   // So, these ops are special - they're mutation AND view ops. They get special codegen.
   // An example is transpose_, e.g. `a.transpose_()`
   // Calling transpose_() should ensure that a gets an alias, and append the new ViewMeta to a's current list of ViewMetas.
   at::AutoDispatchSkipFunctionalize guard;
-  value_ = meta.forward_fn(value_, meta.out_index);
+  value_ = meta->forward(value_);
   TORCH_INTERNAL_ASSERT(!value_.key_set().has(c10::DispatchKey::Functionalize));
 }
 
@@ -368,15 +369,8 @@ void FunctionalTensorWrapper::sync_() {
   regenerate_from_base();
 }
 
-Tensor FunctionalTensorWrapper::apply_view_metas(const Tensor& base) {
-  auto t = base;
-
-  // Reapply views to get the viewed tensor from the base in alias_
-  for (auto& view_meta: view_metas_) {
-    t = view_meta.forward_fn(t, view_meta.out_index);
-  }
-
-  return t;
+const std::vector<std::shared_ptr<functionalization::ViewMeta>>& FunctionalTensorWrapper::view_metas() const {
+  return view_metas_;
 }
 
 void FunctionalTensorWrapper::regenerate_from_base() {
@@ -385,7 +379,7 @@ void FunctionalTensorWrapper::regenerate_from_base() {
   auto t = storage_impl->base();
 
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(t));
-  t = apply_view_metas(t);
+  t = at::functionalization::impl::apply_view_meta_sequence(t, view_metas_);
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(t));
 
   replace_(t, /*from_lazy_regenerate=*/true);
@@ -727,11 +721,11 @@ bool isFunctionalTensor(const std::optional<Tensor>& t) {
 }
 
 bool isFunctionalTensor(const c10::List<::std::optional<Tensor>>& t_list) {
-  if (t_list.empty()) return false;
+  if (t_list.empty()) { return false; }
   auto functional_count = 0;
   for (const auto i : c10::irange(t_list.size())) {
     auto const & e= t_list[i];
-    if (!e.has_value() || !e->defined()) continue;
+    if (!e.has_value() || !e->defined()) { continue; }
     if (isFunctionalTensor(e)) {
       ++functional_count;
     }
@@ -741,10 +735,10 @@ bool isFunctionalTensor(const c10::List<::std::optional<Tensor>>& t_list) {
 
 template <typename T>
 static bool isFunctionalTensorIListRef(c10::IListRef<T> list) {
-  if (list.size() == 0) return false;
+  if (list.size() == 0) { return false; }
   auto functional_count = 0;
   for (const auto& tensor : list) {
-    if (!tensor.defined()) continue;
+    if (!tensor.defined()) { continue; }
     if (isFunctionalTensor(tensor)) {
       ++functional_count;
     }
@@ -762,20 +756,28 @@ void freeze_functional_tensor(const Tensor& tensor) {
   functional_base_impl->freeze_storage();
 }
 
-Tensor create_functional_tensor_with_view_meta(const at::Tensor& view_to_wrap, const at::Tensor& base, functionalization::ViewMeta meta, int64_t out_idx) {
+Tensor create_functional_tensor_with_view_meta(
+    const at::Tensor& view_to_wrap,
+    const at::Tensor& base,
+    const std::shared_ptr<functionalization::ViewMeta>& meta,
+    int64_t out_idx) {
   TORCH_INTERNAL_ASSERT(!at::functionalization::impl::isFunctionalTensor(view_to_wrap));
   TORCH_INTERNAL_ASSERT(at::functionalization::impl::isFunctionalTensor(base));
   auto functional_base_impl = at::functionalization::impl::unsafeGetFunctionalWrapper(base);
+  auto meta_ = meta;
   if (out_idx != 0) {
     // Note [out_idx in ViewMeta]
     // When a view op outputs multiple tensors, each output needs its own separate ViewMeta.
     // Each ViewMeta also tracks the index of the particular output tensor, which is needed in the reverse function.
-    meta = meta.to_out_idx(out_idx);
+    meta_ = meta->to_out_index(out_idx);
   }
-  return at::detail::make_tensor<FunctionalTensorWrapper>(view_to_wrap, functional_base_impl, meta);
+  return at::detail::make_tensor<FunctionalTensorWrapper>(view_to_wrap, functional_base_impl, meta_);
 }
 
-std::vector<Tensor> create_functional_tensor_with_view_meta(ITensorListRef view_to_wrap, const at::Tensor& base, const functionalization::ViewMeta& meta) {
+std::vector<Tensor> create_functional_tensor_with_view_meta(
+    ITensorListRef view_to_wrap,
+    const at::Tensor& base,
+    const std::shared_ptr<functionalization::ViewMeta>& meta) {
   std::vector<Tensor> outputs(view_to_wrap.size());
   int64_t i = 0;
   for (const auto& tensor : view_to_wrap) {
@@ -785,12 +787,22 @@ std::vector<Tensor> create_functional_tensor_with_view_meta(ITensorListRef view_
   return outputs;
 }
 
-void mutate_view_meta(const at::Tensor& self, const functionalization::ViewMeta& meta) {
+void mutate_view_meta(const at::Tensor& self, const std::shared_ptr<functionalization::ViewMeta>& meta) {
   TORCH_INTERNAL_ASSERT(at::functionalization::impl::isFunctionalTensor(self));
   auto self_impl = at::functionalization::impl::unsafeGetFunctionalWrapper(self);
   self_impl->mutate_view_meta(meta);
 }
 
+Tensor apply_view_meta_sequence(
+    const Tensor& base,
+    const std::vector<std::shared_ptr<functionalization::ViewMeta>>& sequence) {
+  Tensor r = base;
+  for (auto& vm : sequence) {
+    r = vm->forward(r);
+  }
+  return r;
+}
+
 // Note [Propagating strides in the functionalization pass]
 // In order to properly compute stride information, the functionalization pass
 // calls each {view} reference implementations with meta tensors.
@@ -884,7 +896,7 @@ void functionalize_op_helper(const c10::OperatorHandle& op, torch::jit::Stack* s
     const auto& ivalue = returns[idx];
     if (ivalue.isTensor()) {
       const auto& t = ivalue.toTensor();
-      if (!t.defined()) continue;
+      if (!t.defined()) { continue; }
       at::functionalization::impl::sync(t);
       auto t_new = c10::IValue(at::functionalization::impl::from_functional_tensor(t));
       (*stack)[returns_begin + idx] = t_new;

aten/src/ATen/FunctionalTensorWrapper.h

@@ -56,7 +56,7 @@ struct TORCH_API FunctionalTensorWrapper : public c10::TensorImpl {
   explicit FunctionalTensorWrapper(
       const Tensor& view_value,
       const FunctionalTensorWrapper* base,
-      const functionalization::ViewMeta& meta);
+      const std::shared_ptr<functionalization::ViewMeta>& meta);
 
   // Get the underlying, actual tensor, that doesn't know anything about
   // functionalization.
@@ -99,17 +99,17 @@ struct TORCH_API FunctionalTensorWrapper : public c10::TensorImpl {
         ->are_all_mutations_under_no_grad_or_inference_mode();
   }
 
-  void maybe_mark_symbolic(const functionalization::ViewMeta& meta) {
-    is_symbolic_ = is_symbolic_ | meta.has_symbolic_inputs;
+  void maybe_mark_symbolic(functionalization::ViewMeta* meta) {
+    is_symbolic_ = is_symbolic_ | meta->has_symbolic_inputs;
   }
 
   bool is_symbolic() const {
     return is_symbolic_;
   }
 
-  // Runs the forward_fn of every ViewMeta collected in the current instance
-  // to some other base.
-  Tensor apply_view_metas(const Tensor& base);
+  // Retrieves the ViewMeta sequence of this tensor.
+  const std::vector<std::shared_ptr<functionalization::ViewMeta>>& view_metas()
+      const;
 
   // Sync's the underlying tensor with its alias, if it's out of date. This
   // involves two steps: 1) Apply any pending updates/mutations to the alias 2)
@@ -146,7 +146,8 @@ struct TORCH_API FunctionalTensorWrapper : public c10::TensorImpl {
   // from the base tensor. This method is used by inplace-view ops like
   // transpose_. It appends a ViewMeta to the existing stack, and refreshes the
   // tensor by replaying the views off of the alias.
-  void mutate_view_meta(const at::functionalization::ViewMeta& meta);
+  void mutate_view_meta(
+      const std::shared_ptr<at::functionalization::ViewMeta>& meta);
 
   // Custom implementation of self.set_(src)
   void set__impl(const FunctionalTensorWrapper* other);
@@ -285,7 +286,7 @@ struct TORCH_API FunctionalTensorWrapper : public c10::TensorImpl {
   bool is_symbolic_ = false;
 
   size_t generation_ = 0;
-  std::vector<at::functionalization::ViewMeta> view_metas_;
+  std::vector<std::shared_ptr<at::functionalization::ViewMeta>> view_metas_;
 
  protected:
   static void copy_tensor_metadata(
@@ -377,16 +378,20 @@ TORCH_API void propagate_xla_data_direct(
 Tensor create_functional_tensor_with_view_meta(
     const Tensor& view_to_wrap,
     const Tensor& base,
-    functionalization::ViewMeta meta,
+    const std::shared_ptr<functionalization::ViewMeta>& meta,
     int64_t out_idx = 0);
 std::vector<Tensor> create_functional_tensor_with_view_meta(
     ITensorListRef view_to_wrap,
     const Tensor& base,
-    const functionalization::ViewMeta& meta);
+    const std::shared_ptr<functionalization::ViewMeta>& meta);
 
 void mutate_view_meta(
     const Tensor& self,
-    const functionalization::ViewMeta& meta);
+    const std::shared_ptr<functionalization::ViewMeta>& meta);
+
+TORCH_API Tensor apply_view_meta_sequence(
+    const Tensor& base,
+    const std::vector<std::shared_ptr<functionalization::ViewMeta>>& sequence);
 
 void set_sizes_strides_offset(const Tensor& out, const Tensor& meta_out);
 void set_sizes_strides_offset(

aten/src/ATen/FunctionalizeFallbackKernel.cpp

@@ -1,3 +1,5 @@
+#include <ATen/FunctionalizeFallbackKernel.h>
+
 #include <ATen/core/dispatch/Dispatcher.h>
 #include <ATen/core/LegacyTypeDispatch.h>
 #include <ATen/EmptyTensor.h>
@@ -7,7 +9,6 @@
 #include <torch/library.h>
 #include <c10/util/irange.h>
 #include <c10/util/strides.h>
-#include <ATen/EmptyTensor.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/ATen.h>
@@ -28,6 +29,31 @@
 #include <utility>
 #endif
 
+namespace at::functionalization {
+
+Tensor resize__ViewMeta::forward(const Tensor& base) {
+  if (reapply_views) {
+    return base.as_strided(size, c10::contiguous_strides(size));
+  } else {
+    return at::as_strided_copy(base, size, c10::contiguous_strides(size));
+  }
+}
+
+Tensor resize__ViewMeta::reverse(const Tensor& base, const Tensor& mutated_view) {
+  return base.as_strided_scatter(
+      mutated_view, size, c10::contiguous_strides(size));
+}
+
+Tensor _unsafe_view_ViewMeta::forward(const Tensor& base) {
+  return at::_unsafe_view_symint(base, size);
+}
+
+Tensor _unsafe_view_ViewMeta::reverse(const Tensor& base, const Tensor& mutated_view) {
+  return at::_unsafe_view_symint(mutated_view, base.sym_sizes());
+}
+
+} // namespace at::functionalization
+
 namespace {
   void functionalizeFallback(const c10::OperatorHandle& op, c10::DispatchKeySet dispatchKeySet [[maybe_unused]], torch::jit::Stack* stack) {
     const auto& schema = op.schema();
@@ -106,7 +132,9 @@ namespace {
       const auto& ivalue = returns[idx];
       if (ivalue.isTensor() && should_wrap_outputs) {
         const auto& t = ivalue.toTensor();
-        if (!t.defined()) continue;
+        if (!t.defined()) {
+          continue;
+        }
         auto t_new = c10::IValue(at::functionalization::impl::to_functional_tensor(t));
         (*stack)[returns_begin + idx] = t_new;
       } else if (ivalue.isTensorList() && should_wrap_outputs) {
@@ -169,19 +197,8 @@ static const at::Tensor & resize__functionalization(c10::DispatchKeySet dispatch
   // The output of resizing is equivalent to taking a slice of a larger tensor.
   // We have to emulate this "slicing" with an as_strided call.
   auto reapply_views = at::functionalization::impl::getFunctionalizationReapplyViewsTLS();
-  at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta(
-    [reapply_views = reapply_views, size = size.vec()](const at::Tensor & base, int64_t mutated_view_idx [[maybe_unused]]) -> at::Tensor {
-      if (reapply_views) {
-        return base.as_strided(size, c10::contiguous_strides(size));
-      } else {
-        return at::as_strided_copy(base, size, c10::contiguous_strides(size));
-      }
-    },
-    [size = size.vec()](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx [[maybe_unused]]) -> at::Tensor {
-      return base.as_strided_scatter(mutated_view, size, c10::contiguous_strides(size));
-    },
-    /*has_symbolic_inputs=*/false
-  );
+  auto view_meta = std::make_shared<at::functionalization::resize__ViewMeta>(
+      reapply_views, size.vec());
   at::functionalization::impl::mutate_view_meta(self, view_meta);
   return self;
 }
@@ -300,17 +317,11 @@ static at::Tensor _unsafe_view_functionalize(const at::Tensor & self, at::SymInt
     tmp_output = at::_unsafe_view_symint(self_, size);
   }
 
-  bool has_symbolic_inputs = std::any_of(size.begin(), size.end(), [=](auto& s) { return s.is_symbolic(); });
-
-  at::functionalization::ViewMeta view_meta = at::functionalization::ViewMeta(
-    [size = size.vec()](const at::Tensor & base, int64_t mutated_view_idx [[maybe_unused]]) -> at::Tensor {
-      return at::_unsafe_view_symint(base, size);
-    },
-    [size = size.vec()](const at::Tensor & base, const at::Tensor & mutated_view, int64_t mutated_view_idx [[maybe_unused]]) -> at::Tensor {
-      return at::_unsafe_view_symint(mutated_view, base.sym_sizes());
-    },
-    /*has_symbolic_inputs=*/has_symbolic_inputs
-  );
+  bool has_symbolic_inputs = std::any_of(
+      size.begin(), size.end(), [=](auto& s) { return s.is_symbolic(); });
+  auto view_meta =
+      std::make_shared<at::functionalization::_unsafe_view_ViewMeta>(
+          has_symbolic_inputs, size.vec());
 
   auto out = at::functionalization::impl::create_functional_tensor_with_view_meta(tmp_output, self, std::move(view_meta));
   // See  Note [Propagating strides in the functionalization pass]

aten/src/ATen/FunctionalizeFallbackKernel.h

@@ -0,0 +1,58 @@
+#pragma once
+
+#include <ATen/FunctionalStorageImpl.h>
+
+namespace at::functionalization {
+
+// `ViewMeta` implementation for `resize_` operation.
+struct TORCH_API resize__ViewMeta : public ViewMeta {
+  FUNCTIONALIZATION_VIEWMETA_NAME(resize__ViewMeta)
+  FUNCTIONALIZATION_VIEWMETA_SERIALIZABLE_TUPLE(
+      bool /* reapply_views */,
+      const std::vector<int64_t>&);
+
+  resize__ViewMeta(const SerializableTuple& tpl)
+      : resize__ViewMeta(std::get<0>(tpl), std::get<1>(tpl)) {}
+
+  resize__ViewMeta(bool reapply_views, const std::vector<int64_t>& size)
+      : ViewMeta(/*has_symbolic_inputs=*/false),
+        reapply_views(reapply_views),
+        size(size) {}
+
+  Tensor forward(const Tensor& base) override;
+  Tensor reverse(const Tensor& base, const Tensor& mutated_view) override;
+
+  SerializableTuple to_serializable_tuple() {
+    return std::make_tuple(reapply_views, size);
+  }
+
+  bool reapply_views;
+  std::vector<int64_t> size;
+};
+
+// `ViewMeta` implementation for `_unsafe_view` operation.
+struct TORCH_API _unsafe_view_ViewMeta : public ViewMeta {
+  FUNCTIONALIZATION_VIEWMETA_NAME(_unsafe_view_ViewMeta)
+  FUNCTIONALIZATION_VIEWMETA_SERIALIZABLE_TUPLE(
+      bool /* has_symbolic_inputs */,
+      const std::vector<c10::SymInt>&);
+
+  _unsafe_view_ViewMeta(const SerializableTuple& tpl)
+      : _unsafe_view_ViewMeta(std::get<0>(tpl), std::get<1>(tpl)) {}
+
+  _unsafe_view_ViewMeta(
+      bool has_symbolic_inputs,
+      const std::vector<c10::SymInt>& size)
+      : ViewMeta(has_symbolic_inputs), size(size) {}
+
+  Tensor forward(const Tensor& base) override;
+  Tensor reverse(const Tensor& base, const Tensor& mutated_view) override;
+
+  SerializableTuple to_serializable_tuple() {
+    return std::make_tuple(has_symbolic_inputs, size);
+  }
+
+  std::vector<c10::SymInt> size;
+};
+
+} // namespace at::functionalization

aten/src/ATen/core/PythonOpRegistrationTrampoline.cpp

@@ -1,32 +1,22 @@
 #include <ATen/core/PythonOpRegistrationTrampoline.h>
+#include <c10/core/impl/PyInterpreterHooks.h>
 
+// TODO: delete this
 namespace at::impl {
 
-// The strategy is that all python interpreters attempt to register themselves
-// as the main interpreter, but only one wins.  Only that interpreter is
-// allowed to interact with the C++ dispatcher.  Furthermore, when we execute
-// logic on that interpreter, we do so hermetically, never setting pyobj field
-// on Tensor.
-
-std::atomic<c10::impl::PyInterpreter*>
-    PythonOpRegistrationTrampoline::interpreter_{nullptr};
+c10::impl::PyInterpreter* PythonOpRegistrationTrampoline::interpreter_ = nullptr;
 
 c10::impl::PyInterpreter* PythonOpRegistrationTrampoline::getInterpreter() {
-  return PythonOpRegistrationTrampoline::interpreter_.load();
+  return c10::impl::getGlobalPyInterpreter();
 }
 
 bool PythonOpRegistrationTrampoline::registerInterpreter(
     c10::impl::PyInterpreter* interp) {
-  c10::impl::PyInterpreter* expected = nullptr;
-  interpreter_.compare_exchange_strong(expected, interp);
-  if (expected != nullptr) {
-    // This is the second (or later) Python interpreter, which means we need
-    // non-trivial hermetic PyObject TLS
-    c10::impl::HermeticPyObjectTLS::init_state();
+  if (interpreter_ != nullptr) {
     return false;
-  } else {
-    return true;
   }
+  interpreter_ = interp;
+  return true;
 }
 
 } // namespace at::impl

aten/src/ATen/core/PythonOpRegistrationTrampoline.h

@@ -2,19 +2,21 @@
 
 #include <ATen/core/dispatch/Dispatcher.h>
 
-// TODO: this can probably live in c10
+// TODO: We can get rid of this
 
 
 namespace at::impl {
 
+// Manages the single Python interpreter instance for PyTorch.
 class TORCH_API PythonOpRegistrationTrampoline final {
-  static std::atomic<c10::impl::PyInterpreter*> interpreter_;
+  static c10::impl::PyInterpreter* interpreter_;
 
 public:
-  //  Returns true if you successfully registered yourself (that means
-  //  you are in the hot seat for doing the operator registrations!)
+  // Register the Python interpreter. Returns true on first registration,
+  // false if an interpreter was already registered.
   static bool registerInterpreter(c10::impl::PyInterpreter*);
 
+  // Returns the registered interpreter via the global PyInterpreter hooks.
   // Returns nullptr if no interpreter has been registered yet.
   static c10::impl::PyInterpreter* getInterpreter();
 };

aten/src/ATen/cpu/vec/vec_quant.h

@@ -149,5 +149,105 @@ static inline void pack_vnni4(
 #endif
 }
 
+// This is a helper function for transpose_pack_vnni4
+// Transform a [4, 16] block (with incontiguous output)
+// Src:
+// a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13 a14 a15 a16
+// b1 b2 b3 b4 b5 b6 b7 b8 b9 b10 b11 b12 b13 b14 b15 b16
+// c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 c15 c16
+// d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16
+// Dst:
+// a1 a2 a3 a4 b1 b2 b3 b4 c1 c2 c3 c4 d1 d2 d3 d4
+// a5 a6 a7 a8 b5 b6 b7 b8 c5 c6 c7 c8 d5 d6 d7 d8
+// a9 a10 a11 a12 b9 b10 b11 b12 c9 c10 c11 c12 d9 d10 d11 d12
+// a13 a14 a15 a16 b13 b14 b15 b16 c13 c14 c15 c16 d13 d14 d15 d16
+template <typename scalar_t, typename = std::enable_if_t<sizeof(scalar_t) == 1>>
+static inline void transpose_vnni4_pad_4x16_block(
+    const scalar_t* src,
+    scalar_t* dst,
+    int64_t ld_src,
+    int64_t ld_dst,
+    int krem = 4) {
+#if defined(CPU_CAPABILITY_AVX512)
+  __m128i r[4];
+  for (int i = 0; i < krem; ++i) {
+    r[i] = _mm_loadu_si128(reinterpret_cast<const __m128i*>(src + i * ld_src));
+  }
+  for (int i = krem; i < 4; ++i) {
+    r[i] = _mm_setzero_si128();
+  }
+
+  // Transpose 4x16 bytes using unpack and shuffle
+  __m128i t0 = _mm_unpacklo_epi32(r[0], r[1]);
+  __m128i t1 = _mm_unpackhi_epi32(r[0], r[1]);
+  __m128i t2 = _mm_unpacklo_epi32(r[2], r[3]);
+  __m128i t3 = _mm_unpackhi_epi32(r[2], r[3]);
+
+  __m128i r0 = _mm_unpacklo_epi64(t0, t2);
+  __m128i r1 = _mm_unpackhi_epi64(t0, t2);
+  __m128i r2 = _mm_unpacklo_epi64(t1, t3);
+  __m128i r3 = _mm_unpackhi_epi64(t1, t3);
+
+  // Store output
+  if (krem == 4) {
+    // normal case
+    _mm_storeu_si128(reinterpret_cast<__m128i*>(dst), r0);
+    _mm_storeu_si128(reinterpret_cast<__m128i*>(dst + ld_dst), r1);
+    _mm_storeu_si128(reinterpret_cast<__m128i*>(dst + ld_dst * 2), r2);
+    _mm_storeu_si128(reinterpret_cast<__m128i*>(dst + ld_dst * 3), r3);
+  } else {
+    // masked case
+    __mmask16 mask = (1ULL << (krem * 4)) - 1;
+    _mm_mask_storeu_epi8(dst, mask, r0);
+    _mm_mask_storeu_epi8(reinterpret_cast<__m128i*>(dst + ld_dst), mask, r1);
+    _mm_mask_storeu_epi8(
+        reinterpret_cast<__m128i*>(dst + ld_dst * 2), mask, r2);
+    _mm_mask_storeu_epi8(
+        reinterpret_cast<__m128i*>(dst + ld_dst * 3), mask, r3);
+  }
+#else
+  TORCH_CHECK(
+      false,
+      "transpose_vnni4_pad_4x16_block is only supported when AVX-512 is supported")
+#endif
+}
+
+// Do the transpose packing fusion with VNNI4
+// Reorder [K, N] → [N/4, K, 4] (VNNI4-style layout for bit8)
+template <typename scalar_t, typename = std::enable_if_t<sizeof(scalar_t) == 1>>
+static inline void transpose_pack_vnni4(
+    const scalar_t* src,
+    scalar_t* dst,
+    int64_t ld_src,
+    int64_t K,
+    int64_t N) {
+#if defined(CPU_CAPABILITY_AVX512)
+  TORCH_CHECK(
+      N % 16 == 0, "N needs to be multiple of 16 for transpose_pack_vnni4");
+  int64_t bk = 0;
+  int64_t _K = K / 4 * 4;
+  for (; bk < _K; bk += 4) {
+    int64_t bn = 0;
+    for (; bn < N; bn += 16) {
+      transpose_vnni4_pad_4x16_block(
+          src + bk * ld_src + bn, dst + bn * K + bk * 4, ld_src, K * 4);
+    }
+  }
+
+  // Handle leftover K rows (< 4)
+  if (K % 4 != 0) {
+    int krem = K - bk;
+    int64_t bn = 0;
+    for (; bn < N; bn += 16) {
+      transpose_vnni4_pad_4x16_block(
+          src + bk * ld_src + bn, dst + bn * K + bk * 4, ld_src, K * 4, krem);
+    }
+  }
+#else
+  TORCH_CHECK(
+      false, "transpose_pack_vnni4 is only supported when AVX-512 is supported")
+#endif
+}
+
 } // namespace CPU_CAPABILITY
 } // namespace at::vec

aten/src/ATen/cuda/detail/CUDAHooks.cpp

@@ -281,6 +281,9 @@ bool CUDAHooks::compiledWithMIOpen() const {
 
 bool CUDAHooks::supportsDilatedConvolutionWithCuDNN() const {
 #if AT_CUDNN_ENABLED()
+  if (!hasCUDA()) {
+    return false;
+  }
   // NOTE: extra parenthesis around numbers disable clang warnings about
   // dead code
   return true;
@@ -291,6 +294,9 @@ bool CUDAHooks::supportsDilatedConvolutionWithCuDNN() const {
 
 bool CUDAHooks::supportsDepthwiseConvolutionWithCuDNN() const {
 #if AT_CUDNN_ENABLED()
+  if (!hasCUDA()) {
+    return false;
+  }
   cudaDeviceProp* prop = at::cuda::getCurrentDeviceProperties();
   // Check for Volta cores
   if (prop->major >= 7) {
@@ -305,6 +311,9 @@ bool CUDAHooks::supportsDepthwiseConvolutionWithCuDNN() const {
 
 bool CUDAHooks::supportsBFloat16ConvolutionWithCuDNNv8() const {
 #if AT_CUDNN_ENABLED()
+  if (!hasCUDA()) {
+    return false;
+  }
   cudaDeviceProp* prop = at::cuda::getCurrentDeviceProperties();
   // Check for Volta cores
   if (prop->major >= 8) {

aten/src/ATen/native/ConvUtils.h

@@ -465,8 +465,11 @@ inline bool mps_conv_use_channels_last(const at::Tensor& input, const at::Tensor
     return false;
   }
 
-  auto fmt = input.suggest_memory_format();
-  return fmt == at::MemoryFormat::ChannelsLast || fmt == at::MemoryFormat::ChannelsLast3d;
+  auto is_channel_last = [](const at::Tensor& t) {
+    auto fmt = t.suggest_memory_format();
+    return fmt == at::MemoryFormat::ChannelsLast || fmt == at::MemoryFormat::ChannelsLast3d;
+  };
+  return is_channel_last(input) || is_channel_last(weight);
 }
 
 } // namespace at::native

aten/src/ATen/native/Convolution.cpp

@@ -32,10 +32,6 @@
 #include <ATen/native/mkldnn/Utils.h>
 #endif
 
-#ifdef USE_MPS
-#include <ATen/mps/MPSDevice.h>
-#endif
-
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
 #include <ATen/NativeFunctions.h>
@@ -410,11 +406,23 @@ struct ConvParams {
   // cudnn and miopen are guaranteed not to be on mobile, and T102591915 / T110194934 suggest
   // that maybe the compiledWithCuDNN() check sometimes segfaults (though I can't imagine how)
 #if !defined(C10_MOBILE)
-    if (!detail::getCUDAHooks().compiledWithCuDNN()) {
+    if (!detail::getCUDAHooks().compiledWithCuDNN() || !input.is_cuda() || !cudnn_enabled) {
       return false;
     }
+    static long cudnn_version = detail::getCUDAHooks().versionCuDNN();
+    // broken on cuDNN 9.8
+    if (cudnn_version >= 90800) {
+      if (cudnn_conv_suggest_memory_format(input, weight) == at::MemoryFormat::Contiguous &&
+          (input.scalar_type() == at::kBFloat16 || input.scalar_type() == at::kHalf) &&
+          weight.dim() == 5) {
+        for (int i = 2; i < weight.dim(); i++) {
+          if (weight.size(i) != 1) {
+            return false;
+          }
+        }
+      }
+    }
     if (needs_64bit_indexing_no_split(input, weight)) {
-      static long cudnn_version = detail::getCUDAHooks().versionCuDNN();
       if (!(cudnn_version >= 90300 && at::native::cudnnv8_enabled_check_debug())) {
         TORCH_WARN_ONCE("cuDNN cannot be used for large non-batch-splittable convolutions"
                         " if the V8 API is not enabled or before cuDNN version 9.3+."
@@ -422,9 +430,6 @@ struct ConvParams {
         return false;
       }
     }
-    if (!input.is_cuda() || !cudnn_enabled) {
-      return false;
-    }
     if (input.scalar_type() == at::kBFloat16 || weight.scalar_type() == at::kBFloat16) {
       if (!(detail::getCUDAHooks().supportsBFloat16ConvolutionWithCuDNNv8() && at::native::cudnnv8_enabled_check_debug())) {
         return false;
@@ -443,16 +448,19 @@ struct ConvParams {
 
   // Use cudnn for FP16 depthwise convolutions
   bool use_cudnn_depthwise(const at::Tensor& input, const at::Tensor& weight) const  {
-    if (!detail::getCUDAHooks().compiledWithCuDNN()) {
+    if (!cudnn_enabled || !detail::getCUDAHooks().compiledWithCuDNN() || !input.is_cuda()) {
       return false;
     }
-    if (cudnn_conv_suggest_memory_format(input, weight) != at::MemoryFormat::Contiguous && use_cudnn(input, weight)) {
-      // always use cudnn_depthwise for channels_last format
-      return true;
-    }
     // native kernel doesn't support 64-bit non-splittable case
-    if (cudnn_enabled && !(canUse32BitIndexMath(input) && canUse32BitIndexMath(weight))) {
+    if (!(canUse32BitIndexMath(input) && canUse32BitIndexMath(weight))) {
       static long cudnn_version = detail::getCUDAHooks().compiledWithCuDNN() ? detail::getCUDAHooks().versionCuDNN() : -1;
+      // TODO(eqy): remove this once cuDNN fixes 64-bit depthwise support, first broken in 9.11x
+      if (cudnn_conv_suggest_memory_format(input, weight) != at::MemoryFormat::Contiguous) {
+        if (cudnn_version < 0 || cudnn_version > 91000) {
+          return false;
+        }
+      }
+
       if (!(cudnn_version >= 90300 && at::native::cudnnv8_enabled_check_debug())) {
         TORCH_WARN_ONCE("cuDNN cannot be used for large non-batch-splittable convolutions"
                         " if the V8 API is not enabled or before cuDNN version 9.3+."
@@ -462,6 +470,10 @@ struct ConvParams {
         return true;
       }
     }
+    if (cudnn_conv_suggest_memory_format(input, weight) != at::MemoryFormat::Contiguous) {
+      // always use cudnn_depthwise for channels_last format
+      return true;
+    }
     if (detail::getCUDAHooks().supportsDepthwiseConvolutionWithCuDNN()) {
       bool kernel_cond =  (use_cudnn(input, weight) &&
                            input.scalar_type() == kHalf && // only for FP16
@@ -1429,12 +1441,8 @@ static inline at::MemoryFormat determine_backend_memory_format(
       }
       break;
     case ConvBackend::Mps:
+    case ConvBackend::MpsTranspose:
       if (mps_conv_use_channels_last(input, weight)) {
-#ifdef USE_MPS
-        if (!mps::is_macos_13_or_newer(mps::MacOSVersion::MACOS_VER_15_0_PLUS)) {
-          break;
-        }
-#endif
         backend_memory_format = (k == 5) ? MemoryFormat::ChannelsLast3d : MemoryFormat::ChannelsLast;
       }
       break;

aten/src/ATen/native/ConvolutionMM3d.cpp

@@ -9,6 +9,7 @@
 #include <ATen/native/TransposeType.h>
 #include <ATen/native/Unfold3d.h>
 #include <c10/util/irange.h>
+#include <c10/util/safe_numerics.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
 #include <ATen/Functions.h>
@@ -174,6 +175,23 @@ static inline void slow_conv3d_shape_check(
   const int64_t input_height = input.size(dim_height);
   const int64_t input_width = input.size(dim_width);
 
+  constexpr int64_t MAX_SAFE_PAD = (1LL << 61);
+
+  TORCH_CHECK_VALUE(
+    pad_height <= MAX_SAFE_PAD,
+    "Padding height too large: pad_height=",
+    pad_height);
+
+  TORCH_CHECK_VALUE(
+    pad_width <= MAX_SAFE_PAD,
+    "Padding width too large: pad_width=",
+    pad_width);
+
+  TORCH_CHECK_VALUE(
+    pad_depth <= MAX_SAFE_PAD,
+    "Padding depth too large: pad_depth=",
+    pad_depth);
+
   const int64_t exact_input_depth = input_depth + 2 * pad_depth;
   const int64_t exact_input_height = input_height + 2 * pad_height;
   const int64_t exact_input_width = input_width + 2 * pad_width;
@@ -221,6 +239,14 @@ static inline void slow_conv3d_shape_check(
       output_width,
       "). Output size is too small");
 
+  uint64_t kernel_product;
+  TORCH_CHECK(
+    !c10::mul_overflows(kernel_height, kernel_width, &kernel_product),
+    "Kernel height x width product is too large: kernel_height=",
+    kernel_height,
+    ", kernel_width=",
+    kernel_width);
+
   if (weight.defined()) {
     int64_t n_input_plane = weight.size(1);
     if (weight.dim() == 2) {

aten/src/ATen/native/Histogram.cpp

@@ -23,6 +23,7 @@
 #include <ATen/ops/linspace.h>
 #endif
 
+#include <cmath>
 #include <numeric>
 #include <tuple>
 #include <vector>
@@ -202,6 +203,46 @@ select_outer_bin_edges(const Tensor& input, std::optional<c10::ArrayRef<double>>
     return std::make_pair(leftmost_edges, rightmost_edges);
 }
 
+
+/* Bin edges correction based on the precision representation.
+ * To maintain the backward compatibility we take max(std::nextafter<>, +1)
+ * and min(std::nextafter<>, -1) for scalar types. For other types +/- 1 as usual.
+ */
+void bins_edges_correction(const ScalarType& t, double &leftmost_edge, double &rightmost_edge)
+{
+#define UPDATE_WITH_LIMIT(real_type, scalartype) \
+  case ScalarType::scalartype:                   \
+    leftmost_edge = std::min(                    \
+        static_cast<double>(                     \
+            std::nexttoward(                     \
+                static_cast<real_type>(leftmost_edge),   \
+                std::numeric_limits<real_type>::lowest() \
+            )                                    \
+        ),                                       \
+        leftmost_edge - 1.                       \
+    );                                           \
+    rightmost_edge = std::max(                   \
+        static_cast<double>(                     \
+            std::nexttoward(                     \
+                static_cast<real_type>(rightmost_edge), \
+                std::numeric_limits<real_type>::max()   \
+            )                                    \
+        ),                                       \
+        rightmost_edge + 1.                      \
+    );                                           \
+    break;
+
+    switch (t) {
+        UPDATE_WITH_LIMIT(double, Double)
+        UPDATE_WITH_LIMIT(float, Float)
+        default:
+            // Fallback to the default behavior for other types
+            leftmost_edge -= 1;
+            rightmost_edge += 1;
+    }
+#undef UPDATE_WITH_LIMIT
+}
+
 /* histc's version of the logic for outermost bin edges.
  */
 std::pair<double, double> histc_select_outer_bin_edges(const Tensor& input,
@@ -216,8 +257,7 @@ std::pair<double, double> histc_select_outer_bin_edges(const Tensor& input,
     }
 
     if (leftmost_edge == rightmost_edge) {
-        leftmost_edge -= 1;
-        rightmost_edge += 1;
+        bins_edges_correction(input.dtype().toScalarType(), leftmost_edge, rightmost_edge);
     }
 
     TORCH_CHECK(!(std::isinf(leftmost_edge) || std::isinf(rightmost_edge) ||

aten/src/ATen/native/TensorShape.cpp

@@ -1,3 +1,5 @@
+#include <ATen/core/ATen_fwd.h>
+#include <c10/core/ScalarType.h>
 #define TORCH_ASSERT_ONLY_METHOD_OPERATORS
 #include <ATen/AccumulateType.h>
 #include <ATen/Dispatch.h>
@@ -1878,19 +1880,18 @@ Tensor repeat(const Tensor& self, IntArrayRef repeats) {
 
   Tensor xtensor = self.expand(padded_size);
 
-  Tensor result;
+  Tensor urtensor;
   if (self.is_quantized()) {
-    result = at::empty_quantized(target_size, self);
+    urtensor = at::empty_quantized(target_size, self);
   } else {
-    result = at::empty(target_size, self.options());
+    urtensor = at::empty(target_size, self.options());
   }
 
   // return an empty tensor if one of the repeat dimensions is zero
   if (zero_tensor) {
-    return result;
+    return urtensor;
   }
 
-  Tensor urtensor = at::alias(result);
   for (const auto i : c10::irange(xtensor.dim())) {
     // can't unfold with step 0, so make sure step is at least 1
     // (it doesn't matter what it is in that case, because the size is 0).
@@ -1900,7 +1901,22 @@ Tensor repeat(const Tensor& self, IntArrayRef repeats) {
 
   urtensor.copy_(xtensor.expand_as(urtensor));
 
-  return result;
+  // Combine the dimensions to produce the target_size.
+  // xtensor dims: [a0, ..., ad-1]
+  // urtensor dims: [a0, ..., ad-1, b0, ..., bd-1]
+  // b dims are produced by unfold.
+  // Transform urtensor to [a0 * b0, ..., ad-1 * bd-1]
+  const int64_t n_dims = xtensor.dim();
+  auto range_a = at::arange(xtensor.dim(), at::TensorOptions(at::kLong));
+  auto range_b = range_a + n_dims;
+  auto stacked = stack({std::move(range_a), std::move(range_b)}, 1).flatten();
+  auto permutation = IntArrayRef(stacked.data_ptr<int64_t>(), n_dims * 2);
+  // Permute from [a0, ..., ad-1, b0, ..., bd-1] to [a0, b0, ..., ad-1, bd-1]
+  urtensor = urtensor.permute(permutation);
+  // Reshape from [a0, b0, ..., ad-1, bd-1] to [a0 * b0, ..., ad-1 * bd-1]
+  urtensor = urtensor.reshape(target_size);
+
+  return urtensor;
 }
 
 Tensor tile_symint(const Tensor& self, SymIntArrayRef reps) {

aten/src/ATen/native/cuda/Copy.cu

@@ -42,6 +42,19 @@ void bfloat16_copy_kernel_cuda(TensorIteratorBase &iter) {
     });
 }
 
+#ifdef USE_ROCM
+void bfloat16tofloat32_copy_kernel_cuda(TensorIteratorBase &iter) {
+    gpu_kernel_nocast(iter, [] GPU_LAMBDA(at::BFloat16 value) {
+        return static_cast<float>(value);
+    });
+}
+void float16tofloat32_copy_kernel_cuda(TensorIteratorBase &iter) {
+    gpu_kernel_nocast(iter, [] GPU_LAMBDA(at::Half value) {
+        return static_cast<float>(value);
+    });
+}
+#endif
+
 void float8_copy_kernel_cuda(TensorIteratorBase &iter) {
   ScalarType dtype = iter.dtype(0);
   ScalarType other_dtype = iter.dtype(1);
@@ -187,7 +200,17 @@ void direct_copy_kernel_cuda(TensorIteratorBase &iter) {
      } else {
        float16_copy_kernel_cuda(iter);
      }
-  } else if (isBitsType(dtype)) {
+  }
+#ifdef USE_ROCM
+  else if ((iter.dtype(1) == kBFloat16 || iter.dtype(1) == kHalf) && dtype == kFloat) {
+    if (iter.dtype(1) == kBFloat16) {
+      bfloat16tofloat32_copy_kernel_cuda(iter);
+    } else {
+      float16tofloat32_copy_kernel_cuda(iter);
+    }
+  }
+#endif
+  else if (isBitsType(dtype)) {
     TORCH_CHECK(dtype == iter.dtype(1), "copy_() does not support casting "
       "bits types to different bits types. Source dtype is ", iter.dtype(1), "target dtype is ", dtype);
     AT_DISPATCH_BIT_TYPES(dtype, "copy_", [&] {

aten/src/ATen/native/mps/kernels/GridSampler.metal

@@ -223,9 +223,6 @@ void grid_sampler_single_element(
     auto input_size = input_sizes[input_dim];
     auto coord = static_cast<opmath_t<T>>(coords[coord_dim]);
 
-    // Interpret nan as -1
-    coord = isnan(coord) ? -1 : coord;
-
     if (!align_corners) {
       // Map unaligned grid space to aligned grid space
       auto corner_alignment_factor = static_cast<opmath_t<T>>(input_size) /

aten/src/ATen/native/mps/operations/Convolution.mm

@@ -52,9 +52,7 @@ static void fill_depthwise_conv_desc(MPSGraphDepthwiseConvolution3DOpDescriptor*
                                      NSUInteger dilationRateInX,
                                      NSUInteger dilationRateInY,
                                      NSUInteger paddingHorizontal,
-                                     NSUInteger paddingVertical,
-                                     c10::MemoryFormat memory_format,
-                                     NSUInteger groups) {
+                                     NSUInteger paddingVertical) {
   descriptor_.strides =
       @[ @1, [[NSNumber alloc] initWithInteger:strideInY], [[NSNumber alloc] initWithInteger:strideInX] ];
   descriptor_.dilationRates =
@@ -103,7 +101,7 @@ static void fill_conv_desc(MPSGraphConvolution2DOpDescriptor* descriptor_,
   descriptor_.groups = groups;
 }
 
-static Tensor _mps_convolution_impl(const Tensor& input_t_,
+static Tensor _mps_convolution_impl(const Tensor& input_t,
                                     const Tensor& weight_t,
                                     const std::optional<Tensor>& bias_opt,
                                     IntArrayRef padding,
@@ -111,12 +109,15 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
                                     IntArrayRef dilation,
                                     int64_t groups,
                                     std::optional<IntArrayRef> input_shape) {
-  const bool is_macOS_15_0_or_newer = is_macos_13_or_newer(MacOSVersion::MACOS_VER_15_0_PLUS);
-  Tensor input_t = input_t_;
-  bool is3DConv = input_t.dim() == 5;
-  if (!is_macOS_15_0_or_newer || is3DConv) {
-    input_t = input_t.contiguous();
-  }
+  constexpr auto kChannelsLast = MemoryFormat::ChannelsLast;
+  constexpr auto kContiguous = MemoryFormat::Contiguous;
+  const bool is_macos_15_plus = is_macos_13_or_newer(MacOSVersion::MACOS_VER_15_0_PLUS);
+
+  const bool is3DConv = input_t.dim() == 5;
+  const auto memory_format = input_t.suggest_memory_format();
+  const auto input_suggested_layout = memory_format == kChannelsLast && is_macos_15_plus ? kChannelsLast : kContiguous;
+  const bool is_channels_last = mps_conv_use_channels_last(input_t, weight_t) && !is3DConv;
+  const bool bias_defined = bias_opt ? bias_opt->defined() : false;
 
   TORCH_CHECK(isFloatingType(input_t.scalar_type()), "Convolution is supported only for Floating types");
 
@@ -126,15 +127,6 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
   checkAllSameType(c, {input, weight});
   checkAllSameGPU(c, {input, weight});
 
-  bool bias_defined;
-
-  if (bias_opt == std::nullopt)
-    bias_defined = false;
-  else
-    bias_defined = bias_opt->defined();
-
-  auto memory_format = input_t.suggest_memory_format();
-  bool is_channels_last = (memory_format == at::MemoryFormat::ChannelsLast) && !is3DConv;
   auto output_t =
       at::empty(input_shape.has_value() ? input_shape.value()
                                         : conv_output_size(input->sizes(), weight->sizes(), padding, stride, dilation),
@@ -142,12 +134,18 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
                 std::nullopt,
                 kMPS,
                 std::nullopt,
-                is_macOS_15_0_or_newer ? memory_format : MemoryFormat::Contiguous);
+                is_channels_last ? kChannelsLast : kContiguous);
   if (output_t.numel() == 0) {
     return output_t;
   }
   TensorArg output{output_t, "result", 0};
 
+  // TODO: Remove me when MacOS-14 is no longer supported
+  std::optional<Tensor> output_c;
+  if (!is_macos_15_plus && is_channels_last) {
+    output_c = at::empty_like(output_t, output_t.options().memory_format(kContiguous));
+  }
+
   if (!is_macos_13_or_newer(MacOSVersion::MACOS_VER_15_1_PLUS)) {
     // On macOS < 15.1, MPS convolution kernel does not support output channels > 2^16
     for (auto elem : output_t.sizes()) {
@@ -186,32 +184,22 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
                                   getArrayRefString(dilation),
                                   getArrayRefString(padding),
                                   groups,
-                                  is_channels_last,
+                                  input_suggested_layout == kChannelsLast,
                                   mps::getTensorsStringKey({input_t, weight_t}),
                                   bias_defined,
                                   bias_shape_key);
 
-    MPSShape* inputShape = mps::getMPSShape(input_t, memory_format);
-    MPSShape* outputShape = mps::getMPSShape(output_t, memory_format);
-    MPSNDArray* inputNDArray = nil;
-    MPSNDArray* outputNDArray = nil;
-
-    if (input_t.is_contiguous(memory_format) && output_t.is_contiguous(memory_format) && is_macOS_15_0_or_newer) {
-      inputNDArray = getMPSNDArray(input_t, inputShape);
-      outputNDArray = getMPSNDArray(output_t, outputShape);
-    }
-
+    auto inputShape = mps::getMPSShape(input_t, input_suggested_layout);
+    auto outputShape = mps::getMPSShape(output_t, input_suggested_layout);
     auto cachedGraph = LookUpOrCreateCachedGraph<CachedGraph>(key, [&](auto mpsGraph, auto newCachedGraph) {
-      MPSShape* weightShape = mps::getMPSShape(weight_t);
-      bool isDepthwiseConv = ((groups > 1 && (weightShape[1].intValue == 1)) && inputShape.count >= 4 &&
-                              weightShape.count >= 4 && !is_channels_last);
+      bool isDepthwiseConv =
+          (groups > 1 && weight_t.size(1) == 1) && input_t.dim() >= 4 && weight_t.dim() >= 4 && !is_channels_last;
 
-      MPSGraphTensor* inputTensor =
-          mpsGraphRankedPlaceHolder(mpsGraph, getMPSScalarType(input_t.scalar_type()), inputShape);
-      MPSGraphTensor* weightTensor = mpsGraphRankedPlaceHolder(mpsGraph, weight_t);
-      MPSGraphTensor* outputTensor;
+      auto inputTensor = mpsGraphRankedPlaceHolder(mpsGraph, getMPSScalarType(input_t), inputShape);
+      auto weightTensor = mpsGraphRankedPlaceHolder(mpsGraph, weight_t);
+      MPSGraphTensor* outputTensor = nil;
       if (is3DConv) {
-        MPSGraphConvolution3DOpDescriptor* conv3dDescriptor_ = [[MPSGraphConvolution3DOpDescriptor new] autorelease];
+        auto conv3dDescriptor_ = [[MPSGraphConvolution3DOpDescriptor new] autorelease];
         fill_conv3d_desc(conv3dDescriptor_,
                          stride[2],
                          stride[1],
@@ -229,17 +217,9 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
                                                     descriptor:conv3dDescriptor_
                                                           name:nil];
       } else if (isDepthwiseConv) {
-        MPSGraphDepthwiseConvolution3DOpDescriptor* depthWiseConv3dDescriptor_ =
-            [[MPSGraphDepthwiseConvolution3DOpDescriptor new] autorelease];
-        fill_depthwise_conv_desc(depthWiseConv3dDescriptor_,
-                                 stride[1],
-                                 stride[0],
-                                 dilation[1],
-                                 dilation[0],
-                                 padding[1],
-                                 padding[0],
-                                 memory_format,
-                                 groups);
+        auto depthWiseConv3dDescriptor_ = [[MPSGraphDepthwiseConvolution3DOpDescriptor new] autorelease];
+        fill_depthwise_conv_desc(
+            depthWiseConv3dDescriptor_, stride[1], stride[0], dilation[1], dilation[0], padding[1], padding[0]);
 
         MPSGraphTensor* weightTransposeTensor = [mpsGraph transposeTensor:weightTensor
                                                                 dimension:-3
@@ -258,7 +238,7 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
                        dilation[0],
                        padding[1],
                        padding[0],
-                       memory_format,
+                       input_suggested_layout,
                        groups);
 
         outputTensor = [mpsGraph convolution2DWithSourceTensor:inputTensor
@@ -270,13 +250,6 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
       MPSGraphTensor* biasTensor = nil;
       if (bias_defined) {
         biasTensor = mpsGraphUnrankedPlaceHolder(mpsGraph, getMPSDataType(bias_opt.value()));
-      }
-
-      if (is_channels_last && !is_macOS_15_0_or_newer) {
-        outputTensor = mps::convertNHWCtoNCHW(mpsGraph, outputTensor);
-      }
-
-      if (bias_defined) {
         outputTensor = [mpsGraph additionWithPrimaryTensor:outputTensor secondaryTensor:biasTensor name:nil];
       }
       newCachedGraph->inputTensor_ = inputTensor;
@@ -285,27 +258,26 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
       newCachedGraph->outputTensor_ = outputTensor;
     });
 
-    auto inputPlaceholder = inputNDArray ? Placeholder(cachedGraph->inputTensor_, inputNDArray)
-                                         : Placeholder(cachedGraph->inputTensor_, input_t, inputShape);
-    auto weightsPlaceholder = Placeholder(cachedGraph->weightTensor_, weight_t);
+    auto inputPlaceholder = input_suggested_layout == kContiguous
+        ? Placeholder(cachedGraph->inputTensor_, output_c || is3DConv ? input_t.contiguous() : input_t)
+        : Placeholder(cachedGraph->inputTensor_, getMPSNDArray(input_t, inputShape));
+    auto outputPlaceholder = input_suggested_layout == kContiguous
+        ? Placeholder(cachedGraph->outputTensor_, output_c ? *output_c : output_t)
+        : Placeholder(cachedGraph->outputTensor_, getMPSNDArray(output_t, outputShape));
+    auto weightsPlaceholder = Placeholder(cachedGraph->weightTensor_, output_c ? weight_t.contiguous() : weight_t);
     auto biasPlaceholder = Placeholder();
     // Reshape the bias to be broadcastable with output of conv2d or conv3d
     if (bias_defined) {
       if (is3DConv) {
-        biasPlaceholder = Placeholder(cachedGraph->biasTensor_, (bias_opt.value()).view({1, bias_shape[0], 1, 1, 1}));
+        biasPlaceholder = Placeholder(cachedGraph->biasTensor_, bias_opt->view({1, bias_shape[0], 1, 1, 1}));
+      } else if (input_suggested_layout == kChannelsLast) {
+        biasPlaceholder = Placeholder(cachedGraph->biasTensor_, bias_opt->view({1, 1, 1, bias_shape[0]}));
       } else {
-        if (is_channels_last && is_macOS_15_0_or_newer) {
-          biasPlaceholder = Placeholder(cachedGraph->biasTensor_, (bias_opt.value()).view({1, 1, 1, bias_shape[0]}));
-        } else {
-          biasPlaceholder = Placeholder(cachedGraph->biasTensor_, (bias_opt.value()).view({1, bias_shape[0], 1, 1}));
-        }
+        biasPlaceholder = Placeholder(cachedGraph->biasTensor_, bias_opt->view({1, bias_shape[0], 1, 1}));
       }
     }
-    auto outputPlaceholder = outputNDArray ? Placeholder(cachedGraph->outputTensor_, outputNDArray)
-                                           : Placeholder(cachedGraph->outputTensor_, output_t);
 
-    NSMutableDictionary<MPSGraphTensor*, MPSGraphTensorData*>* feeds =
-        [[[NSMutableDictionary alloc] initWithCapacity:3] autorelease];
+    auto feeds = [[[NSMutableDictionary alloc] initWithCapacity:3] autorelease];
     feeds[inputPlaceholder.getMPSGraphTensor()] = inputPlaceholder.getMPSGraphTensorData();
     feeds[weightsPlaceholder.getMPSGraphTensor()] = weightsPlaceholder.getMPSGraphTensorData();
     if (bias_defined) {
@@ -315,6 +287,10 @@ static Tensor _mps_convolution_impl(const Tensor& input_t_,
     runMPSGraph(stream, cachedGraph->graph(), feeds, outputPlaceholder);
   }
 
+  if (output_c) {
+    output_t.copy_(*output_c);
+  }
+
   return output_t;
 }
 
@@ -351,14 +327,21 @@ static Tensor mps_convolution_backward_input(IntArrayRef input_size,
   TensorArg grad_output{grad_output_t, "grad_output", 1}, weight{weight_t, "weight", 2};
   checkAllSameType(c, {grad_output, weight});
   checkAllSameGPU(c, {grad_output, weight});
-  auto memory_format = grad_output_t.suggest_memory_format();
-  bool is_channels_last = (memory_format == at::MemoryFormat::ChannelsLast) && !is3DConv;
-  auto grad_input_t = at::empty(input_size, grad_output_t.options(), std::nullopt);
+  constexpr auto kChannelsLast = at::MemoryFormat::ChannelsLast;
+  bool is_channels_last = mps_conv_use_channels_last(grad_output_t, weight_t) && !is3DConv;
+  auto grad_input_t =
+      at::empty(input_size, grad_output_t.options(), is_channels_last ? std::optional(kChannelsLast) : std::nullopt);
 
   // Avoid "grad_input" when this is being used as transposed convolution
   TensorArg grad_input{grad_input_t, "result", 0};
   convolution_shape_check(c, grad_input, weight, grad_output, padding, stride, dilation, groups);
 
+  // TODO: Remove me when MacOS-14 is no longer supported
+  std::optional<Tensor> grad_input_c;
+  if (!is_macos_13_or_newer(MacOSVersion::MACOS_VER_15_0_PLUS) && is_channels_last) {
+    grad_input_c = at::empty_like(grad_input_t, grad_input_t.options().memory_format(MemoryFormat::Contiguous));
+  }
+
   // Derive from MPSCachedGraph
   struct CachedGraph : public MPSCachedGraph {
     CachedGraph(MPSGraph* graph) : MPSCachedGraph(graph) {}
@@ -370,7 +353,6 @@ static Tensor mps_convolution_backward_input(IntArrayRef input_size,
   // Add backward with input
   @autoreleasepool {
     MPSStream* stream = getCurrentMPSStream();
-
     MPSShape* mps_input_shape = getMPSShape(input_size);
     std::string key = fmt::format("mps_{}_convolution_backward_input:{}:{}:{}:{}:{}:{}",
                                   is3DConv ? "3d_" : "",
@@ -411,15 +393,8 @@ static Tensor mps_convolution_backward_input(IntArrayRef input_size,
       } else if (isDepthwiseConv) {
         MPSGraphDepthwiseConvolution3DOpDescriptor* depthWiseConv3dDescriptor_ =
             [[MPSGraphDepthwiseConvolution3DOpDescriptor new] autorelease];
-        fill_depthwise_conv_desc(depthWiseConv3dDescriptor_,
-                                 stride[1],
-                                 stride[0],
-                                 dilation[1],
-                                 dilation[0],
-                                 padding[1],
-                                 padding[0],
-                                 at::MemoryFormat::Contiguous,
-                                 groups);
+        fill_depthwise_conv_desc(
+            depthWiseConv3dDescriptor_, stride[1], stride[0], dilation[1], dilation[0], padding[1], padding[0]);
         MPSGraphTensor* weightTransposeTensor = [mpsGraph transposeTensor:weightTensor
                                                                 dimension:-3
                                                             withDimension:-4
@@ -454,14 +429,18 @@ static Tensor mps_convolution_backward_input(IntArrayRef input_size,
       newCachedGraph->gradInputTensor_ = gradInputTensor;
     });
 
-    auto gradOutputPlaceholder = Placeholder(cachedGraph->gradOutputTensor_, grad_output_t);
-    auto weightsPlaceholder = Placeholder(cachedGraph->weightTensor_, weight_t);
-    auto outputPlaceholder = Placeholder(cachedGraph->gradInputTensor_, *grad_input);
+    auto gradOutputPlaceholder =
+        Placeholder(cachedGraph->gradOutputTensor_, grad_input_c ? grad_output_t.contiguous() : grad_output_t);
+    auto weightsPlaceholder = Placeholder(cachedGraph->weightTensor_, grad_input_c ? weight_t.contiguous() : weight_t);
+    auto outputPlaceholder = Placeholder(cachedGraph->gradInputTensor_, grad_input_c ? *grad_input_c : grad_input_t);
 
     auto feeds = dictionaryFromPlaceholders(gradOutputPlaceholder, weightsPlaceholder);
     runMPSGraph(stream, cachedGraph->graph(), feeds, outputPlaceholder);
   }
-  return *grad_input;
+  if (grad_input_c) {
+    grad_input_t.copy_(*grad_input_c);
+  }
+  return grad_input_t;
 }
 
 static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
@@ -474,9 +453,11 @@ static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
                                                bool bias_defined) {
   using namespace at::native::mps;
   using namespace mps;
-  bool is3DConv = input_t.dim() == 5;
+  const bool is3DConv = input_t.dim() == 5;
   TORCH_CHECK(isFloatingType(grad_output_t.scalar_type()), "Convolution is supported only for Floating types");
   CheckedFrom c = "mps_convolution_backward_weights";
+  constexpr auto kChannelsLast = at::MemoryFormat::ChannelsLast;
+  bool is_channels_last = mps_conv_use_channels_last(input_t, grad_output_t) && !is3DConv;
 
   // For uniformity with everything else, although it seems grad_weight
   // would be unambiguous too.
@@ -487,7 +468,8 @@ static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
   checkAllSameGPU(c, {grad_output, input});
 
   auto grad_weight_t =
-      at::empty(weight_size, grad_output_t.scalar_type(), std::nullopt, kMPS, std::nullopt, std::nullopt);
+      at::empty(weight_size, grad_output_t.options(), is_channels_last ? std::optional(kChannelsLast) : std::nullopt);
+
   TensorArg grad_weight{grad_weight_t, "result", 0};
 
   convolution_shape_check(c, input, grad_weight, grad_output, padding, stride, dilation, groups);
@@ -500,16 +482,23 @@ static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
     MPSGraphTensor* gradWeightTensor_ = nil;
   };
 
+  // TODO: Remove me when MacOS-14 is no longer supported
+  std::optional<Tensor> grad_weight_c;
+  if (!is_macos_13_or_newer(MacOSVersion::MACOS_VER_15_0_PLUS) && is_channels_last) {
+    grad_weight_c = at::empty_like(grad_weight_t, grad_weight_t.options().memory_format(MemoryFormat::Contiguous));
+  }
+
   @autoreleasepool {
     MPSStream* stream = getCurrentMPSStream();
 
     MPSShape* mps_weight_shape = getMPSShape(weight_size);
-    std::string key = fmt::format("mps_{}convolution_backward_weights:{}:{}:{}:{}:{}",
+    std::string key = fmt::format("mps_{}convolution_backward_weights:{}:{}:{}:{}:{}:{}",
                                   is3DConv ? "3d_" : "",
                                   getArrayRefString(stride),
                                   getArrayRefString(dilation),
                                   getArrayRefString(padding),
                                   groups,
+                                  is_channels_last,
                                   getTensorsStringKey({grad_output_t, input_t, grad_weight_t}));
     auto cachedGraph = LookUpOrCreateCachedGraph<CachedGraph>(key, [&](auto mpsGraph, auto newCachedGraph) {
       MPSShape* inputShape = getMPSShape(input_t);
@@ -541,15 +530,8 @@ static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
       } else if (isDepthwiseConv) {
         MPSGraphDepthwiseConvolution3DOpDescriptor* depthWiseConv3dDescriptor_ =
             [[MPSGraphDepthwiseConvolution3DOpDescriptor new] autorelease];
-        fill_depthwise_conv_desc(depthWiseConv3dDescriptor_,
-                                 stride[1],
-                                 stride[0],
-                                 dilation[1],
-                                 dilation[0],
-                                 padding[1],
-                                 padding[0],
-                                 at::MemoryFormat::Contiguous,
-                                 groups);
+        fill_depthwise_conv_desc(
+            depthWiseConv3dDescriptor_, stride[1], stride[0], dilation[1], dilation[0], padding[1], padding[0]);
         NSNumber* outputFeatChannelDim = mps_weight_shape[0];
         MPSShape* weightShapeTranspose = @[ @1, outputFeatChannelDim, mps_weight_shape[2], mps_weight_shape[3] ];
         MPSGraphTensor* gradWeightTensorTranspose =
@@ -583,14 +565,19 @@ static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
       newCachedGraph->gradWeightTensor_ = gradWeightTensor;
     });
 
-    auto gradOutputPlaceholder = Placeholder(cachedGraph->gradOutputTensor_, grad_output_t);
-    auto inputPlaceholder = Placeholder(cachedGraph->inputTensor_, input_t);
-    auto outputPlaceholder = Placeholder(cachedGraph->gradWeightTensor_, grad_weight_t);
+    auto gradOutputPlaceholder =
+        Placeholder(cachedGraph->gradOutputTensor_, grad_weight_c ? grad_output_t.contiguous() : grad_output_t);
+    auto inputPlaceholder = Placeholder(cachedGraph->inputTensor_, grad_weight_c ? input_t.contiguous() : input_t);
+    auto outputPlaceholder =
+        Placeholder(cachedGraph->gradWeightTensor_, grad_weight_c ? *grad_weight_c : grad_weight_t);
 
     auto feeds = dictionaryFromPlaceholders(gradOutputPlaceholder, inputPlaceholder);
     runMPSGraph(stream, cachedGraph->graph(), feeds, outputPlaceholder);
   }
 
+  if (grad_weight_c) {
+    grad_weight_t.copy_(*grad_weight_c);
+  }
   return grad_weight_t;
 }
 

aten/src/ATen/native/mps/operations/Unique.mm

@@ -9,11 +9,22 @@
 #else
 #include <ATen/ops/_unique2.h>
 #include <ATen/ops/_unique2_native.h>
+#include <ATen/ops/arange.h>
+#include <ATen/ops/argsort.h>
+#include <ATen/ops/cat.h>
+#include <ATen/ops/cumsum.h>
+#include <ATen/ops/full.h>
+#include <ATen/ops/masked_select.h>
+#include <ATen/ops/nonzero.h>
+#include <ATen/ops/ones.h>
+#include <ATen/ops/ones_like.h>
 #include <ATen/ops/slice.h>
 #include <ATen/ops/unique_consecutive.h>
 #include <ATen/ops/unique_consecutive_native.h>
 #include <ATen/ops/unique_dim_consecutive.h>
 #include <ATen/ops/unique_dim_consecutive_native.h>
+#include <ATen/ops/unique_dim_native.h>
+#include <ATen/ops/zeros.h>
 #endif
 
 namespace at::native {
@@ -305,4 +316,85 @@ std::tuple<Tensor, Tensor, Tensor> _unique2_mps(const Tensor& self,
   return _unique_impl_mps(self, return_inverse, return_counts, false, std::nullopt);
 }
 
+static Tensor lexsort_rows_perm_mps(const Tensor& mat_2d) {
+  const auto rows = mat_2d.size(0), cols = mat_2d.size(1);
+  if (rows <= 1 || cols == 0) {
+    return arange(rows, mat_2d.options().dtype(kLong));
+  }
+
+  auto perm = arange(rows, mat_2d.options().dtype(kLong));
+  for (auto c = cols - 1; c >= 0; --c) {
+    auto keys = mat_2d.select(1, c).index_select(0, perm);
+    const auto idx = argsort(keys, /*dim=*/0, /*descending=*/false);
+    perm = perm.index_select(0, idx);
+  }
+  return perm;
+}
+
+static std::tuple<Tensor, Tensor, Tensor> unique_dim_sorted_mps_impl(const Tensor& self,
+                                                                     int64_t dim,
+                                                                     bool return_inverse,
+                                                                     bool return_counts) {
+  dim = maybe_wrap_dim(dim, self.dim());
+
+  auto sizes = self.sizes().vec();
+  auto num_zero_dims = std::count(sizes.begin(), sizes.end(), (int64_t)0);
+  if (self.size(dim) == 0) {
+    auto output = at::empty(sizes, self.options());
+    auto inverse_indices = at::empty({0}, self.options().dtype(kLong));
+    auto counts = at::empty({0}, self.options().dtype(kLong));
+    return {output, inverse_indices, counts};
+  }
+
+  auto transposed = self.moveaxis(dim, 0);
+  auto orig_sizes = transposed.sizes().vec();
+  auto rows = transposed.size(0);
+  auto input_flat = transposed.contiguous().view({rows, -1});
+
+  auto perm = lexsort_rows_perm_mps(input_flat);
+  auto input_sorted = input_flat.index_select(0, perm);
+
+  Tensor is_unique = at::zeros({rows}, self.options().dtype(kBool));
+  if (rows > 0) {
+    is_unique.narrow(0, 0, 1).fill_(true);
+  }
+  if (rows > 1) {
+    auto a = input_sorted.narrow(0, 1, rows - 1);
+    auto b = input_sorted.narrow(0, 0, rows - 1);
+    auto row_changed = a.ne(b).any(1);
+    is_unique.narrow(0, 1, rows - 1).copy_(row_changed);
+  }
+
+  auto unique_pos = nonzero(is_unique).squeeze(1);
+  auto group_id = cumsum(is_unique.to(kLong), 0).sub(1);
+
+  auto unique_rows_2d = input_sorted.index_select(0, unique_pos);
+
+  Tensor inverse_indices = empty({0}, self.options().dtype(kLong));
+  if (return_inverse) {
+    inverse_indices = empty({rows}, self.options().dtype(kLong));
+    inverse_indices.index_copy_(0, perm, group_id);
+  }
+
+  Tensor counts = empty({0}, self.options().dtype(kLong));
+  if (return_counts) {
+    const auto num_unique = unique_pos.size(0);
+    counts = zeros({num_unique}, self.options().dtype(kLong));
+    counts.scatter_add_(0, group_id, ones_like(group_id, group_id.options().dtype(kLong)));
+  }
+
+  orig_sizes[0] = unique_rows_2d.size(0);
+  auto output = unique_rows_2d.view(orig_sizes).moveaxis(0, dim);
+
+  return std::make_tuple(std::move(output), std::move(inverse_indices), std::move(counts));
+}
+
+std::tuple<Tensor, Tensor, Tensor> unique_dim_mps(const Tensor& self,
+                                                  int64_t dim,
+                                                  const bool /*sorted*/,
+                                                  const bool return_inverse,
+                                                  const bool return_counts) {
+  return unique_dim_sorted_mps_impl(self, dim, return_inverse, return_counts);
+}
+
 } // namespace at::native

aten/src/ATen/native/native_functions.yaml

@@ -1409,7 +1409,7 @@
 - func: _sparse_broadcast_to(Tensor(a) self, int[] size) -> Tensor(a)
   variants: function
   dispatch:
-    SparseCPU, SparseCUDA: sparse_broadcast_to
+    SparseCPU, SparseCUDA, SparseMPS: sparse_broadcast_to
 
 - func: cat(Tensor[] tensors, int dim=0) -> Tensor
   structured_delegate: cat.out
@@ -6450,6 +6450,7 @@
   dispatch:
     CPU: unique_dim_cpu
     CUDA: unique_dim_cuda
+    MPS: unique_dim_mps
   tags: dynamic_output_shape
   autogen: unique_dim.out
 

aten/src/ATen/native/quantized/cpu/qembeddingbag_prepack.cpp

@@ -158,12 +158,46 @@ c10::intrusive_ptr<EmbeddingPackedParamsBase> PackedEmbeddingBagWeight::prepack(
   return packed_ptr;
 }
 
+#ifdef USE_FBGEMM
+namespace {
+/// Number of columns in the rowwise min/max buffer passed to the quantization function(s)
+constexpr int kRowwiseMinMaxNumCols = 2;
+
+bool _validate_rowwise_min_max(
+  const at::Tensor& weight,
+  const std::optional<at::Tensor>& rowwise_min_max_opt) {
+  const auto is_valid_rowwise_min_max = rowwise_min_max_opt.has_value();
+
+  if (is_valid_rowwise_min_max) {
+      TORCH_CHECK(
+        (rowwise_min_max_opt->dim() == 2 &&
+        rowwise_min_max_opt->size(0) == weight.size(0) &&
+        rowwise_min_max_opt->size(1) == kRowwiseMinMaxNumCols),
+        "'rowwise_min_max' must be a 2D tensor with shape [num_rows(weight), 2].");
+  }
+
+  return is_valid_rowwise_min_max;
+}
+
+auto _get_rowwise_min_max_contig(
+  const std::optional<at::Tensor>& rowwise_min_max_opt) {
+    return rowwise_min_max_opt.has_value()
+      ? rowwise_min_max_opt->expect_contiguous(rowwise_min_max_opt->suggest_memory_format())
+      : at::borrow_from_optional_tensor(rowwise_min_max_opt);
+}
+}
+#endif // USE_FBGEMM
+
 namespace at::native {
 
 // Note - This is a temporary pack function for embedding bag which quantizes
 // and packs the float weight tensor. In the next step it will be replaced by a
 // quantize and pack function once we support FP scale and FP zero_point
 //
+// The optional rowwise_min_max argument is to support callers to pass in the min/max
+// values of the weight tensor. If the rowwise_min_max is not provided, the min/max
+// values will be computed from the weight tensor.
+//
 // Python example examining a packed 8bit zero_point and scale:
 //
 // >> x = torch.from_numpy(np.array([[[10, 20], [30, 40]],[[50, 60], [70, 80]]],
@@ -221,7 +255,10 @@ namespace at::native {
 //
 //        [[50.        , 60.00000035],
 //         [70.        , 80.00000035]]])
-Tensor& qembeddingbag_byte_prepack_out(Tensor& output, const Tensor& weight) {
+Tensor& qembeddingbag_byte_prepack_out(
+    Tensor& output,
+    const Tensor& weight,
+    const std::optional<Tensor>& rowwise_min_max_opt) {
   // The "last" dimension of an N-Dimensioned batch of embedding bags is
   // quantization channel. E.g. for a 2D embedding bag, this has
   // [ row, col ] dimensions, for batched of embedding bags, dimensions might be
@@ -256,9 +293,16 @@ Tensor& qembeddingbag_byte_prepack_out(Tensor& output, const Tensor& weight) {
   auto* output_data = output.data_ptr<uint8_t>();
 
 #ifdef USE_FBGEMM
+  // Move these outside of the ifdef when we support non-FBGEMM flow.
+  const auto is_valid_rowwise_min_max = _validate_rowwise_min_max(weight, rowwise_min_max_opt);
+  const auto rowwise_min_max_contig = _get_rowwise_min_max_contig(rowwise_min_max_opt);
+
   if (weight_contig->scalar_type() == at::ScalarType::Half) {
     const auto weight_data =
         static_cast<fbgemm::float16*>(weight_contig->data_ptr());
+    const auto rowwise_min_max_data = is_valid_rowwise_min_max
+        ? static_cast<fbgemm::float16*>(rowwise_min_max_contig->data_ptr())
+        : nullptr;
     at::parallel_for(
         0, embedding_rows, 1, [&](int64_t start_idx, int64_t end_idx) {
           fbgemm::FloatOrHalfToFused8BitRowwiseQuantizedSBFloat<
@@ -266,17 +310,21 @@ Tensor& qembeddingbag_byte_prepack_out(Tensor& output, const Tensor& weight) {
               weight_data + start_idx * embedding_cols,
               end_idx - start_idx,
               embedding_cols,
-              output_data + start_idx * output_columns);
+              output_data + start_idx * output_columns,
+              (is_valid_rowwise_min_max ? (rowwise_min_max_data + start_idx * kRowwiseMinMaxNumCols) : nullptr));
         });
   } else {
     const auto weight_data = weight_contig->data_ptr<float>();
+    const auto rowwise_min_max_data =
+        is_valid_rowwise_min_max ? rowwise_min_max_contig->data_ptr<float>() : nullptr;
     at::parallel_for(
         0, embedding_rows, 1, [&](int64_t start_idx, int64_t end_idx) {
           fbgemm::FloatOrHalfToFused8BitRowwiseQuantizedSBFloat<float>(
               weight_data + start_idx * embedding_cols,
               end_idx - start_idx,
               embedding_cols,
-              output_data + start_idx * output_columns);
+              output_data + start_idx * output_columns,
+              (is_valid_rowwise_min_max ? (rowwise_min_max_data + start_idx * kRowwiseMinMaxNumCols) : nullptr));
         });
   }
 
@@ -326,6 +374,22 @@ Tensor qembeddingbag_byte_prepack(const Tensor& weight) {
   return output;
 }
 
+static Tensor qembeddingbag_byte_prepack_with_rowwise_min_max(
+    const Tensor& weight,
+    const Tensor& rowwise_min_max) {
+  const auto weight_contig =
+      weight.expect_contiguous(weight.suggest_memory_format());
+  Tensor output = at::detail::empty_cpu(
+      {0},
+      at::kByte,
+      weight_contig->layout(),
+      weight_contig->device(),
+      std::nullopt,
+      std::nullopt);
+  qembeddingbag_byte_prepack_out(output, weight, rowwise_min_max);
+  return output;
+}
+
 Tensor qembeddingbag_byte_prepack_meta(const Tensor& weight) {
   const auto weight_contig =
       weight.expect_contiguous(weight.suggest_memory_format());
@@ -335,7 +399,7 @@ Tensor qembeddingbag_byte_prepack_meta(const Tensor& weight) {
       "'embedding_bag_byte_prepack' only support float32 or float16.");
   const auto weight_sizes = weight.sym_sizes();
   const auto cols_dim = weight.ndimension() - 1;
-  const auto embedding_cols = weight_sizes[cols_dim];
+  const auto& embedding_cols = weight_sizes[cols_dim];
   // Add 8 bytes per column to store FP32 scale and zero_point per row.
   const auto output_columns = embedding_cols + 2 * sizeof(float);
 
@@ -359,7 +423,8 @@ Tensor _qembeddingbag_nbit_prepack_helper(
     int bit_width,
     const bool optimized_qparams,
     const int64_t nbins,
-    const double ratio) {
+    const double ratio,
+    const std::optional<Tensor>& rowwise_min_max_opt = std::nullopt) {
   TORCH_CHECK(
       weight.scalar_type() == at::ScalarType::Float ||
           weight.scalar_type() == at::ScalarType::Half,
@@ -401,10 +466,17 @@ Tensor _qembeddingbag_nbit_prepack_helper(
   auto* output_data = output.data_ptr<uint8_t>();
 
 #ifdef USE_FBGEMM
+  // Move these outside of the ifdef when we support non-FBGEMM flow.
+  const auto is_valid_rowwise_min_max = _validate_rowwise_min_max(weight, rowwise_min_max_opt);
+  const auto rowwise_min_max_contig = _get_rowwise_min_max_contig(rowwise_min_max_opt);
+
   if (!optimized_qparams) {
     if (weight_contig.scalar_type() == at::ScalarType::Half) {
       const auto weight_data =
           static_cast<fbgemm::float16*>(weight_contig.data_ptr());
+      const auto rowwise_min_max_data = is_valid_rowwise_min_max
+          ? static_cast<fbgemm::float16*>(rowwise_min_max_contig->data_ptr())
+          : nullptr;
       at::parallel_for(
           0, embedding_rows, 1, [&](int64_t start_idx, int64_t end_idx) {
             fbgemm::FloatOrHalfToFusedNBitRowwiseQuantizedSBHalf<
@@ -413,10 +485,13 @@ Tensor _qembeddingbag_nbit_prepack_helper(
                 weight_data + start_idx * embedding_cols,
                 end_idx - start_idx,
                 static_cast<int>(embedding_cols),
-                output_data + start_idx * output_shape[1]);
+                output_data + start_idx * output_shape[1],
+                (is_valid_rowwise_min_max ? (rowwise_min_max_data + start_idx * kRowwiseMinMaxNumCols) : nullptr));
           });
     } else {
       const auto weight_data = weight_contig.data_ptr<float>();
+      const auto rowwise_min_max_data =
+          is_valid_rowwise_min_max ? rowwise_min_max_contig->data_ptr<float>() : nullptr;
       at::parallel_for(
           0, embedding_rows, 1, [&](int64_t start_idx, int64_t end_idx) {
             fbgemm::FloatOrHalfToFusedNBitRowwiseQuantizedSBHalf<float>(
@@ -424,7 +499,8 @@ Tensor _qembeddingbag_nbit_prepack_helper(
                 weight_data + start_idx * embedding_cols,
                 end_idx - start_idx,
                 static_cast<int>(embedding_cols),
-                output_data + start_idx * output_shape[1]);
+                output_data + start_idx * output_shape[1],
+                (is_valid_rowwise_min_max ? (rowwise_min_max_data + start_idx * kRowwiseMinMaxNumCols) : nullptr));
           });
     }
   } else {
@@ -514,6 +590,16 @@ Tensor qembeddingbag_4bit_prepack(
       weight, 4 /*bit_width*/, optimized_qparams, nbins, ratio);
 }
 
+Tensor qembeddingbag_4bit_prepack_with_rowwise_min_max(
+    const Tensor& weight,
+    const Tensor& rowwise_min_max,
+    const bool optimized_qparams,
+    const int64_t nbins,
+    const double ratio) {
+  return _qembeddingbag_nbit_prepack_helper(
+      weight, 4 /*bit_width*/, optimized_qparams, nbins, ratio, rowwise_min_max);
+}
+
 // Applies 2-bit row-wise quantization by determining the range
 // (maximum - minimum) and bias (minimum value) of each row in the input
 // matrix, and then scaling each element to an 2-bit number between 0 and
@@ -531,6 +617,16 @@ Tensor qembeddingbag_2bit_prepack(
       weight, 2 /*bit_width*/, optimized_qparams, nbins, ratio);
 }
 
+Tensor qembeddingbag_2bit_prepack_with_rowwise_min_max(
+    const Tensor& weight,
+    const Tensor& rowwise_min_max,
+    const bool optimized_qparams,
+    const int64_t nbins,
+    const double ratio) {
+  return _qembeddingbag_nbit_prepack_helper(
+      weight, 2 /*bit_width*/, optimized_qparams, nbins, ratio, rowwise_min_max);
+}
+
 class QEmbeddingPackWeights final {
  public:
   static c10::intrusive_ptr<EmbeddingPackedParamsBase> run(const at::Tensor& weight) {
@@ -542,12 +638,21 @@ TORCH_LIBRARY_IMPL(quantized, CPU, m) {
   m.impl(
       TORCH_SELECTIVE_NAME("quantized::embedding_bag_byte_prepack"),
       TORCH_FN(qembeddingbag_byte_prepack));
+  m.impl(
+      TORCH_SELECTIVE_NAME("quantized::embedding_bag_byte_prepack_with_rowwise_min_max"),
+      TORCH_FN(qembeddingbag_byte_prepack_with_rowwise_min_max));
   m.impl(
       TORCH_SELECTIVE_NAME("quantized::embedding_bag_4bit_prepack"),
       TORCH_FN(qembeddingbag_4bit_prepack));
+  m.impl(
+      TORCH_SELECTIVE_NAME("quantized::embedding_bag_4bit_prepack_with_rowwise_min_max"),
+      TORCH_FN(qembeddingbag_4bit_prepack_with_rowwise_min_max));
   m.impl(
       TORCH_SELECTIVE_NAME("quantized::embedding_bag_2bit_prepack"),
       TORCH_FN(qembeddingbag_2bit_prepack));
+  m.impl(
+      TORCH_SELECTIVE_NAME("quantized::embedding_bag_2bit_prepack_with_rowwise_min_max"),
+      TORCH_FN(qembeddingbag_2bit_prepack_with_rowwise_min_max));
 }
 
 TORCH_LIBRARY_IMPL(quantized, QuantizedCPU, m) {

aten/src/ATen/native/quantized/cpu/qembeddingbag_prepack.h

@@ -3,7 +3,10 @@
 
 namespace at::native {
 
-Tensor& qembeddingbag_byte_prepack_out(Tensor& output, const Tensor& weight);
+Tensor& qembeddingbag_byte_prepack_out(
+    Tensor& output,
+    const Tensor& weight,
+    const std::optional<Tensor>& rowwise_min_max_opt = std::nullopt);
 
 Tensor qembeddingbag_byte_prepack(const Tensor& weight);
 

aten/src/ATen/native/quantized/library.cpp

@@ -121,9 +121,12 @@ TORCH_LIBRARY(quantized, m) {
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_unpack(__torch__.torch.classes.quantized.EmbeddingPackedParamsBase W_prepack) -> Tensor W_origin"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_byte_prepack(Tensor weight) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_byte_unpack(Tensor weight) -> Tensor"), {at::Tag::pt2_compliant_tag});
+  m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_byte_prepack_with_rowwise_min_max(Tensor weight, Tensor rowwise_min_max) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_4bit_prepack(Tensor weight, bool optimized_qparams=False, int nbins=200, float ratio=0.16) -> Tensor"), {at::Tag::pt2_compliant_tag});
+  m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_4bit_prepack_with_rowwise_min_max(Tensor weight, Tensor rowwise_min_max, bool optimized_qparams=False, int nbins=200, float ratio=0.16) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_4bit_unpack(Tensor weight) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_2bit_prepack(Tensor weight, bool optimized_qparams=False, int nbins=200, float ratio=0.16) -> Tensor"), {at::Tag::pt2_compliant_tag});
+  m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_2bit_prepack_with_rowwise_min_max(Tensor weight, Tensor rowwise_min_max, bool optimized_qparams=False, int nbins=200, float ratio=0.16) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_2bit_unpack(Tensor weight) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_byte_rowwise_offsets(Tensor weight, Tensor indices, Tensor? offsets=None, bool scale_grad_by_freq=False, int mode=0, bool pruned_weights=False, Tensor? per_sample_weights=None, Tensor? compressed_indices_mapping=None, bool include_last_offset=False) -> Tensor"), {at::Tag::pt2_compliant_tag});
   m.def(TORCH_SELECTIVE_SCHEMA("quantized::embedding_bag_4bit_rowwise_offsets(Tensor weight, Tensor indices, Tensor? offsets=None, bool scale_grad_by_freq=False, int mode=0, bool pruned_weights=False, Tensor? per_sample_weights=None, Tensor? compressed_indices_mapping=None, bool include_last_offset=False) -> Tensor"), {at::Tag::pt2_compliant_tag});

aten/src/ATen/native/sparse/cuda/cuSPARSELtOps.cpp

@@ -120,7 +120,7 @@ at::Tensor _cslt_compress(const Tensor& sparse_input) {
   // buffer (in bytes)
   size_t orig_m = sparse_input.size(0);
   size_t div = orig_m * sparse_input.itemsize();
-  size_t new_n = (compressed_size + div - 1) / div; // floor
+  size_t new_n = (compressed_size + div - 1) / div; // ceil(s,d) = (s+d-1)/d
   auto compressed_tensor = sparse_input.new_empty({(int64_t)orig_m, (int64_t)new_n});
 
   auto& allocator = *::c10::cuda::CUDACachingAllocator::get();
@@ -155,7 +155,7 @@ std::tuple<at::Tensor, int64_t, int64_t, int64_t, int64_t> _cslt_sparse_mm_impl(
     TORCH_CUDASPARSE_CHECK(cusparseLtInit(&handle));
     handle_initialized = true;
   }
-  // cupsarselt constructs
+  // cuSPARSELt constructs
   cusparseLtMatmulDescriptor_t matmul;
   cusparseLtMatmulPlan_t plan;
   cusparseLtMatmulAlgSelection_t alg_sel;

aten/src/ATen/native/transformers/cuda/sdp_utils.cpp

@@ -176,6 +176,28 @@ bool check_head_dim_size_flash(sdp_params const& params, bool debug) {
     }
     return false;
   }
+  if constexpr(caller_is_meff) {
+    bool is_half = (params.query.dtype() == at::kHalf) ||
+      (params.query.dtype() == at::kBFloat16);
+    const int64_t alignment = is_half ? 8 : 4;
+    if (!(query_size_last % alignment == 0 && query_size_last > 0 &&
+          value_size_last % alignment == 0 && value_size_last > 0)) {
+      if (debug) {
+        TORCH_WARN(
+            "Mem efficient attention requires last dimension of inputs to be divisible by ",
+            alignment,
+            ". ",
+            "Got Query.size(-1): ",
+            query_size_last,
+            ", Key.size(-1): ",
+            params.key.sym_size(-1),
+            ", Value.size(-1): ",
+            params.value.sym_size(-1),
+            " instead.");
+      }
+      return false;
+    }
+  }
   return true;
 }
 
@@ -666,6 +688,15 @@ bool can_use_cudnn_attention(const sdp_params& params, bool debug) {
     TORCH_WARN(CUDNN_VERSION, " cuDNN version too old to use cuDNN Attention (< v9.0.0)");
   }
   return false;
+#endif
+#if defined(CUDNN_VERSION)
+  static auto cudnn_version = cudnnGetVersion();
+  if (params.dropout > 0.0 && cudnn_version > 91100 && cudnn_version < 91400) {
+    if (debug) {
+      TORCH_WARN(CUDNN_VERSION, " cuDNN version does not support droppout in SDPA (9.11 - 9.13).");
+    }
+    return false;
+  }
 #endif
   // Define gate functions that determine if a flash kernel can be ran
   // Replace with std::to_array when we migrate to c++20

aten/src/ATen/native/transformers/hip/flash_attn/aot/mha_all_aot.hip

@@ -462,10 +462,11 @@ mha_varlen_fwd_aot(const at::Tensor &q,  // total_q x num_heads x head_size, tot
     using sdp::aotriton_adapter::mk_aotensor;
     using sdp::aotriton_adapter::mk_aoscalartensor;
     using sdp::aotriton_adapter::mk_philoxtensor;
+    using sdp::aotriton_adapter::mk_atomictensor;
     using sdp::aotriton_adapter::cast_dtype;
     at::Tensor atomic_counter;
     if (is_causal) {
-      atomic_counter = at::zeros({1}, q.options());
+      atomic_counter = at::zeros({1}, q.options().dtype(at::kInt));
     }
     aotriton::TensorView<4> empty_bias(0, {0,0,0,0}, {0,0,0,0}, cast_dtype(q.dtype()));
     auto seed = use_philox_state ? mk_philoxtensor(philox_state.seed_.ptr) : mk_aoscalartensor(seed_t);
@@ -474,7 +475,7 @@ mha_varlen_fwd_aot(const at::Tensor &q,  // total_q x num_heads x head_size, tot
     auto nullscalar = mk_philoxtensor(nullptr);
     auto seed_output = use_philox_state ? mk_philoxtensor(seed_t.data_ptr<int64_t>()) : nullscalar;
     auto offset_output = use_philox_state ? mk_philoxtensor(offset_t.data_ptr<int64_t>()) : nullscalar;
-    auto persistent_counter = is_causal ? mk_philoxtensor(atomic_counter.data_ptr<int64_t>()) : nullscalar;
+    auto persistent_counter = mk_atomictensor(is_causal ? atomic_counter.data_ptr<int32_t>() : nullptr);
     if (uses_swa || AOTRITON_ALWAYS_V3_API) {
 #if AOTRITON_V3_API
       using aotriton::v3::flash::CausalType;

aten/src/ATen/templates/FunctionalInverses.h

@@ -2,22 +2,12 @@
 
 // ${generated_comment}
 
+#include <ATen/FunctionalStorageImpl.h>
 #include <ATen/Tensor.h>
 
 namespace at {
 namespace functionalization {
 
-enum class InverseReturnMode {
-  /// Specifies that functional inverses should always return a view.
-  AlwaysView,
-  /// Specifies that functional inverses should always return a non-view / copy.
-  NeverView,
-  /// Specifies that functional inverses should return a view unless a (copying) scatter
-  /// inverse exists, in which case that will be used instead.
-  /// This avoids as_strided() calls that can be difficult for subclasses to handle.
-  ViewOrScatterInverse,
-};
-
 struct FunctionalInverses {
 
 ${view_inverse_declarations}

aten/src/ATen/templates/RegisterFunctionalization.cpp

@@ -4,7 +4,7 @@
 #include <ATen/core/LegacyTypeDispatch.h>
 #include <ATen/EmptyTensor.h>
 #include <ATen/FunctionalTensorWrapper.h>
-#include <ATen/FunctionalInverses.h>
+#include <ATen/ViewMetaClasses.h>
 #include <ATen/MemoryOverlap.h>
 #include <torch/library.h>
 

aten/src/ATen/templates/ViewMetaClasses.cpp

@@ -0,0 +1,19 @@
+// ${generated_comment}
+
+#include <ATen/FunctionalInverses.h>
+#include <ATen/ViewMetaClasses.h>
+
+#ifndef AT_PER_OPERATOR_HEADERS
+#include <ATen/Operators.h>
+#include <ATen/NativeFunctions.h>
+#else
+${op_headers}
+#endif
+
+namespace at {
+namespace functionalization {
+
+${view_meta_implementations}
+
+} // namespace functionalization
+} // namespace at

aten/src/ATen/templates/ViewMetaClasses.h

@@ -0,0 +1,12 @@
+#define TORCH_ASSERT_ONLY_METHOD_OPERATORS
+// ${generated_comment}
+
+#include <ATen/FunctionalStorageImpl.h>
+
+namespace at {
+namespace functionalization {
+
+${view_meta_declarations}
+
+} // namespace functionalization
+} // namespace at

aten/src/ATen/templates/ViewMetaClassesPythonBinding.cpp

@@ -0,0 +1,11 @@
+#include <ATen/ViewMetaClasses.h>
+#include <torch/csrc/functionalization/Module.h>
+
+namespace torch::functionalization {
+
+void initGenerated(PyObject* module) {
+  auto functionalization = py::handle(module).cast<py::module>();
+  $view_meta_bindings
+}
+
+} // namespace torch::functionalization

aten/src/ATen/test/vec_test_all_types.cpp

@@ -1561,6 +1561,38 @@ namespace {
               << "Failure Details:\nTest Seed to reproduce: " << seed;
         }
     }
+#endif
+#if defined(CPU_CAPABILITY_AVX512)
+    TYPED_TEST(Quantization8BitTests, TransposePackVNNI4) {
+        using VT = ValueType<TypeParam>;
+        constexpr auto K = 197;
+        constexpr auto N = 64;
+        constexpr auto L = K * N;
+        constexpr auto ld_src = N;
+        constexpr auto ld_dst = K * 4;
+        CACHE_ALIGN VT x[L];
+        CACHE_ALIGN VT y[L];
+        CACHE_ALIGN VT ref[L];
+        auto seed = TestSeed();
+        ValueGen<VT> generator(VT(-100), VT(100), seed);
+        for (const auto i : c10::irange(L)) {
+          x[i] = generator.get();
+        }
+        at::vec::transpose_pack_vnni4(x, y, ld_src, K, N);
+        int64_t _N = N / 4;
+        for (int64_t k = 0; k < K; k++) {
+          for(int64_t n = 0; n < _N; n++) {
+            for(int64_t l = 0; l < 4; l++) {
+              ref[n * ld_dst + k * 4 + l] =
+                  c10::load(&(x[k * ld_src + n * 4 + l]));
+            }
+          }
+        }
+        for (const auto i : c10::irange(L)) {
+          ASSERT_EQ(y[i], ref[i])
+              << "Failure Details:\nTest Seed to reproduce: " << seed;
+        }
+    }
 #endif
     TYPED_TEST(FunctionalTests, Map) {
         using vec = TypeParam;

benchmarks/dynamo/check_accuracy.py

@@ -78,6 +78,8 @@ def check_accuracy(actual_csv, expected_csv, expected_filename):
                 "google/gemma-3-4b-it",
                 "openai/whisper-tiny",
                 "Qwen/Qwen3-0.6B",
+                "mistralai/Mistral-7B-Instruct-v0.3",
+                "openai/gpt-oss-20b",
             }
         )
 

benchmarks/dynamo/check_graph_breaks.py

@@ -61,6 +61,8 @@ def check_graph_breaks(actual_csv, expected_csv, expected_filename):
                 "google/gemma-3-4b-it",
                 "openai/whisper-tiny",
                 "Qwen/Qwen3-0.6B",
+                "mistralai/Mistral-7B-Instruct-v0.3",
+                "openai/gpt-oss-20b",
             }
         )
 

benchmarks/dynamo/ci_expected_accuracy/aot_eager_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass,0
 
 
 Qwen/Qwen3-0.6B,pass,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass,0
+
+
+
+openai/gpt-oss-20b,pass,0

benchmarks/dynamo/ci_expected_accuracy/aot_inductor_huggingface_inference.csv

@@ -187,3 +187,11 @@ openai/whisper-tiny,fail_to_run,0
 
 
 Qwen/Qwen3-0.6B,fail_to_run,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,fail_to_run,0
+
+
+
+openai/gpt-oss-20b,fail_to_run,0

benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_amp_freezing_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass_due_to_skip,0
 
 
 Qwen/Qwen3-0.6B,pass_due_to_skip,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass_due_to_skip,0
+
+
+
+openai/gpt-oss-20b,pass_due_to_skip,0

benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_freezing_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass_due_to_skip,0
 
 
 Qwen/Qwen3-0.6B,pass_due_to_skip,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass_due_to_skip,0
+
+
+
+openai/gpt-oss-20b,pass_due_to_skip,0

benchmarks/dynamo/ci_expected_accuracy/cpu_inductor_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass_due_to_skip,0
 
 
 Qwen/Qwen3-0.6B,pass_due_to_skip,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass_due_to_skip,0
+
+
+
+openai/gpt-oss-20b,pass_due_to_skip,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_aot_eager_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass,0
 
 
 Qwen/Qwen3-0.6B,pass,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass,0
+
+
+
+openai/gpt-oss-20b,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass,0
 
 
 Qwen/Qwen3-0.6B,pass,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass,0
+
+
+
+openai/gpt-oss-20b,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamic_cpu_inductor_torchbench_inference.csv

@@ -318,7 +318,7 @@ timm_vovnet,pass,0
 
 
 
-torch_multimodal_clip,pass,3
+torch_multimodal_clip,pass,0
 
 
 

benchmarks/dynamo/ci_expected_accuracy/dynamic_inductor_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass,0
 
 
 Qwen/Qwen3-0.6B,pass,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass,0
+
+
+
+openai/gpt-oss-20b,pass,0

benchmarks/dynamo/ci_expected_accuracy/dynamo_eager_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass,0
 
 
 Qwen/Qwen3-0.6B,pass,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass,0
+
+
+
+openai/gpt-oss-20b,pass,0

benchmarks/dynamo/ci_expected_accuracy/inductor_huggingface_inference.csv

@@ -191,3 +191,11 @@ openai/whisper-tiny,pass,0
 
 
 Qwen/Qwen3-0.6B,pass,0
+
+
+
+mistralai/Mistral-7B-Instruct-v0.3,pass,0
+
+
+
+openai/gpt-oss-20b,pass,0

benchmarks/dynamo/huggingface.yaml

@@ -11,6 +11,8 @@ skip:
     - GPTJForQuestionAnswering
     # Model too big
     - google/gemma-3-4b-it
+    - openai/gpt-oss-20b
+    - mistralai/Mistral-7B-Instruct-v0.3
 
   device:
     cpu:
@@ -19,6 +21,8 @@ skip:
       - google/gemma-3-4b-it
       - openai/whisper-tiny
       - Qwen/Qwen3-0.6B
+      - mistralai/Mistral-7B-Instruct-v0.3
+      - openai/gpt-oss-20b
 
   control_flow:
     - AllenaiLongformerBase
@@ -79,6 +83,8 @@ batch_size:
     google/gemma-3-4b-it: 8
     openai/whisper-tiny: 8
     Qwen/Qwen3-0.6B: 8
+    mistralai/Mistral-7B-Instruct-v0.3: 8
+    openai/gpt-oss-20b: 8
 
 
 tolerance:

benchmarks/dynamo/huggingface_llm_models.py

@@ -99,4 +99,6 @@ HF_LLM_MODELS: dict[str, Benchmark] = {
     "google/gemma-3-4b-it": TextGenerationBenchmark,
     "openai/whisper-tiny": WhisperBenchmark,
     "Qwen/Qwen3-0.6B": TextGenerationBenchmark,
+    "mistralai/Mistral-7B-Instruct-v0.3": TextGenerationBenchmark,
+    "openai/gpt-oss-20b": TextGenerationBenchmark,
 }

benchmarks/dynamo/huggingface_models_list.txt

@@ -51,3 +51,5 @@ google/gemma-2-2b,8
 google/gemma-3-4b-it,8
 openai/whisper-tiny,8
 Qwen/Qwen3-0.6B,8
+mistralai/Mistral-7B-Instruct-v0.3, 8
+openai/gpt-oss-20b, 8

benchmarks/dynamo/pr_time_benchmarks/README.md

@@ -6,4 +6,4 @@
 4. (Optional) flip a flag that you know will change the benchmark and run again with b.txt `PYTHONPATH=./ python benchmarks/[YOUR_BENCHMARK].py a.txt`
 5. Compare `a.txt` and `b.txt` located within the `benchmarks/dynamo/pr_time_benchmarks` folder to make sure things look as you expect
 6. Check in your new benchmark file and submit a new PR
-7. In a few days, if your benchmark is stable, bug Laith Sakka to enable running your benchmark on all PRs. If your a meta employee, you can find the dashboard here: internalfb.com/intern/unidash/dashboard/pt2_diff_time_metrics
+7. In a few days, if your benchmark is stable, bug Laith Sakka to enable running your benchmark on all PRs. If you are a meta employee, you can find the dashboard here: https://internalfb.com/intern/unidash/dashboard/pt2_diff_time_metrics

benchmarks/dynamo/pr_time_benchmarks/benchmarks/runtime_overhead.py

@@ -0,0 +1,111 @@
+import sys
+
+from benchmark_base import BenchmarkBase
+
+import torch
+from torch.autograd.grad_mode import inference_mode
+
+
+class Benchmark(BenchmarkBase):
+    def __init__(self, requires_grad, inference_mode, backward, dynamic):
+        assert not (inference_mode and backward), (
+            "inference_mode and backward cannot be both True"
+        )
+
+        self._requires_grad = requires_grad
+        self._inference_mode = inference_mode
+        self._backward = backward
+
+        super().__init__(
+            category="runtime_overhead",
+            backend="inductor",
+            device="cuda",
+            dynamic=dynamic,
+        )
+
+    def name(self):
+        prefix = f"{self.category()}_{self.backend()}"
+        if self._requires_grad:
+            prefix += "_requires_grad"
+        if self._inference_mode:
+            prefix += "_inference_mode"
+        if self._backward:
+            prefix += "_backward"
+        if self.is_dynamic():
+            prefix += "_dynamic"
+        return prefix
+
+    def description(self):
+        return "runtime of a compiled add1 op small input"
+
+    def _prepare_once(self):
+        torch._dynamo.reset()
+        self.a = torch.ones(2, device=self.device(), requires_grad=self._requires_grad)
+
+        @torch.compile(
+            backend=self.backend(),
+            fullgraph=True,
+            dynamic=self.is_dynamic(),
+        )
+        def add1(a):
+            return a + 1
+
+        self._add1 = add1
+
+        # warmup
+        for _ in range(10):
+            if self._backward:
+                self.forward_val = self._add1(self.a).sum()
+                self.forward_val.backward()
+            else:
+                self._work()
+
+    def _prepare(self):
+        if self._backward:
+            self.forward_val = self._add1(self.a).sum()
+
+    def _work(self):
+        if self._inference_mode:
+            with inference_mode():
+                self._add1(self.a)
+        elif self._backward:
+            self.forward_val.backward()
+        else:
+            self._add1(self.a)
+
+
+def main():
+    result_path = sys.argv[1]
+    all = [
+        Benchmark(
+            requires_grad=False, inference_mode=False, backward=False, dynamic=False
+        ),
+        Benchmark(
+            requires_grad=False, inference_mode=True, backward=False, dynamic=False
+        ),
+        Benchmark(
+            requires_grad=True, inference_mode=False, backward=False, dynamic=False
+        ),
+        Benchmark(
+            requires_grad=True, inference_mode=False, backward=True, dynamic=False
+        ),
+        Benchmark(
+            requires_grad=False, inference_mode=False, backward=False, dynamic=True
+        ),
+        Benchmark(
+            requires_grad=False, inference_mode=True, backward=False, dynamic=True
+        ),
+        Benchmark(
+            requires_grad=True, inference_mode=False, backward=False, dynamic=True
+        ),
+        Benchmark(
+            requires_grad=True, inference_mode=False, backward=True, dynamic=True
+        ),
+    ]
+
+    for benchmark in all:
+        benchmark.enable_instruction_count().collect_all().append_results(result_path)
+
+
+if __name__ == "__main__":
+    main()

buckbuild.bzl

@@ -156,7 +156,7 @@ ROOT = "//" if IS_OSS else "//xplat/caffe2"
 # for targets in subfolders
 ROOT_PATH = "//" if IS_OSS else "//xplat/caffe2/"
 
-C10 = "//c10:c10" if IS_OSS else ("//xplat/caffe2/c10:c10_ovrsource" if is_arvr_mode() else "//xplat/caffe2/c10:c10")
+C10 = "//c10:c10" if IS_OSS else "//xplat/caffe2/c10:c10"
 
 # a dictionary maps third party library name to fbsource and oss target
 THIRD_PARTY_LIBS = {
@@ -391,6 +391,8 @@ def get_aten_generated_files(enabled_backends):
         "CompositeExplicitAutogradFunctions_inl.h",
         "CompositeExplicitAutogradNonFunctionalFunctions.h",
         "CompositeExplicitAutogradNonFunctionalFunctions_inl.h",
+        "ViewMetaClasses.h",
+        "ViewMetaClasses.cpp",
         "VmapGeneratedPlumbing.h",
         "core/ATenOpList.cpp",
         "core/TensorBody.h",
@@ -948,7 +950,6 @@ def define_buck_targets(
             [
                 ("torch/csrc/api/include", "torch/**/*.h"),
                 ("", "torch/csrc/**/*.h"),
-                ("", "torch/csrc/**/*.hpp"),
                 ("", "torch/nativert/**/*.h"),
                 ("", "torch/headeronly/**/*.h"),
                 ("", "torch/script.h"),
@@ -1193,6 +1194,7 @@ def define_buck_targets(
             "NativeMetaFunctions.h": ":gen_aten[NativeMetaFunctions.h]",
             "Operators.h": ":gen_aten[Operators.h]",
             "RedispatchFunctions.h": ":gen_aten[RedispatchFunctions.h]",
+            "ViewMetaClasses.h": ":gen_aten[ViewMetaClasses.h]",
             "core/TensorBody.h": ":gen_aten[core/TensorBody.h]",
             "core/aten_interned_strings.h": ":gen_aten[core/aten_interned_strings.h]",
             "core/enum_tag.h": ":gen_aten[core/enum_tag.h]",
@@ -2048,7 +2050,6 @@ def define_buck_targets(
                 ("", "caffe2/utils/*.h"),
                 ("", "caffe2/core/*.h"),
                 ("", "torch/csrc/*.h"),
-                ("", "torch/csrc/*.hpp"),
                 ("", "torch/csrc/api/include/torch/*.h"),
                 ("", "torch/csrc/autograd/*.h"),
                 ("", "torch/csrc/autograd/*/*.h"),