[releng] Docker release, Change docker tag. Refactor Push nightly tags step. Move cuda and cudnn version to docker tag rather then name (#116098)

* [releng] Docker Official release make sure cuda version is part of image name (#116070)

Follow up on https://github.com/pytorch/pytorch/pull/115949

Change docker build image name:
``pytorch:2.1.2-devel``-> ``2.1.2-cuda12.1-cudnn8-devel and 2.1.2-cuda11.8-cudnn8-devel``

Ref: https://github.com/orgs/pytorch/packages/container/package/pytorch-nightly

Naming will be same as in https://hub.docker.com/r/pytorch/pytorch/tags
Pull Request resolved: https://github.com/pytorch/pytorch/pull/116070
Approved by: https://github.com/huydhn, https://github.com/seemethere

* [releng] Docker release Refactor Push nightly tags step. Move cuda and cudnn version to docker tag rather then name (#116097)

Follow up after : https://github.com/pytorch/pytorch/pull/116070

This PR does 2 things.

1. Refactor Push nightly tags step, don't need to extract CUDA_VERSION anymore. New tag should be in this format: ``${PYTORCH_VERSION}-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-runtime``
2. Move cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION) from docker name to tag

Pull Request resolved: https://github.com/pytorch/pytorch/pull/116097
Approved by: https://github.com/jeanschmidt

* [release only change] hardcode pytorch 2.1.2 version

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

@@ -1 +1 @@
-05d67b9418cacda0d356c2102d7c1a887948b013
+34f8189eae57a23cc15b4b4f032fe25757e0db8e

.ci/docker/common/install_inductor_benchmark_deps.sh

@@ -7,18 +7,14 @@ source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
 function install_huggingface() {
   local version
   version=$(get_pinned_commit huggingface)
-  pip_install pandas
-  pip_install scipy
-  pip_install z3-solver
+  pip_install pandas==2.0.3
   pip_install "transformers==${version}"
 }
 
 function install_timm() {
   local commit
   commit=$(get_pinned_commit timm)
-  pip_install pandas
-  pip_install scipy
-  pip_install z3-solver
+  pip_install pandas==2.0.3
   pip_install "git+https://github.com/rwightman/pytorch-image-models@${commit}"
 }
 

.ci/docker/common/install_onnx.sh

@@ -4,6 +4,10 @@ set -ex
 
 source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
 
+retry () {
+    "$@" || (sleep 10 && "$@") || (sleep 20 && "$@") || (sleep 40 && "$@")
+}
+
 # A bunch of custom pip dependencies for ONNX
 pip_install \
   beartype==0.10.4 \
@@ -18,22 +22,17 @@ pip_install \
 # onnx-weekly. Otherwise, onnx-weekly could be
 # overwritten by onnx.
 pip_install \
-  onnxruntime==1.15.1 \
   parameterized==0.8.1 \
   pytest-cov==4.0.0 \
   pytest-subtests==0.10.0 \
   tabulate==0.9.0 \
   transformers==4.31.0
 
-# Using 1.15dev branch for the following not yet released features and fixes.
-# - Segfault fix for shape inference.
-# - Inliner to workaround ORT segfault.
-pip_install onnx-weekly==1.15.0.dev20230717
+pip_install coloredlogs packaging
+retry pip_install -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/ --no-cache-dir --no-input ort-nightly==1.16.0.dev20230908001
 
-# TODO: change this when onnx-script is on testPypi
-# pip_install onnxscript-preview==0.1.0.dev20230809 --no-deps
-# NOTE: temp change for CI to run on unpublished onnxscript PR.
-pip_install "onnxscript@git+https://github.com/microsoft/onnxscript@f69be19ebd3f2e0d7efe64b0c7be3329cbab3822" --no-deps
+pip_install onnx==1.14.1
+pip_install onnxscript-preview==0.1.0.dev20230828 --no-deps
 
 # Cache the transformers model to be used later by ONNX tests. We need to run the transformers
 # package to download the model. By default, the model is cached at ~/.cache/huggingface/hub/

.ci/docker/requirements-ci.txt

@@ -271,7 +271,12 @@ pytest-cpp==2.3.0
 #Pinned versions: 2.3.0
 #test that import:
 
-z3-solver
+z3-solver==4.12.2.0
 #Description: The Z3 Theorem Prover Project
 #Pinned versions:
 #test that import:
+
+tensorboard==2.13.0
+#Description: Also included in .ci/docker/requirements-docs.txt
+#Pinned versions:
+#test that import: test_tensorboard

.ci/pytorch/test.sh

@@ -544,6 +544,10 @@ test_without_numpy() {
   python -c "import sys;sys.path.insert(0, 'fake_numpy');from unittest import TestCase;import torch;x=torch.randn(3,3);TestCase().assertRaises(RuntimeError, lambda: x.numpy())"
   # Regression test for https://github.com/pytorch/pytorch/issues/66353
   python -c "import sys;sys.path.insert(0, 'fake_numpy');import torch;print(torch.tensor([torch.tensor(0.), torch.tensor(1.)]))"
+  # Regression test for https://github.com/pytorch/pytorch/issues/109387
+  if [[ "${TEST_CONFIG}" == *dynamo* ]]; then
+    python -c "import sys;sys.path.insert(0, 'fake_numpy');import torch;torch.compile(lambda x:print(x))('Hello World')"
+  fi
   popd
 }
 

.ci/pytorch/win-test.sh

@@ -35,7 +35,7 @@ if [[ "$BUILD_ENVIRONMENT" == *cuda* ]]; then
 fi
 
 # TODO: Move both of them to Windows AMI
-python -m pip install pytest-rerunfailures==10.3 pytest-cpp==2.3.0
+python -m pip install pytest-rerunfailures==10.3 pytest-cpp==2.3.0 tensorboard==2.13.0
 
 # Install Z3 optional dependency for Windows builds.
 python -m pip install z3-solver

.circleci/scripts/binary_linux_test.sh

@@ -90,7 +90,7 @@ if [[ "$PACKAGE_TYPE" == conda ]]; then
       if [[ "\${TORCH_CONDA_BUILD_FOLDER}" == "pytorch-nightly" ]]; then
               PYTORCH_CHANNEL="pytorch-nightly"
       fi
-      retry conda install \${EXTRA_CONDA_FLAGS} -yq -c nvidia -c "\${PYTORCH_CHANNEL}" "pytorch-cuda=\${cu_ver}"
+      retry conda install \${EXTRA_CONDA_FLAGS} -yq -c nvidia -c pytorch-test "pytorch-cuda=\${cu_ver}"
     fi
     conda install \${EXTRA_CONDA_FLAGS} -y "\$pkg" --offline
   )
@@ -98,9 +98,9 @@ elif [[ "$PACKAGE_TYPE" != libtorch ]]; then
   if [[ "$(uname -m)" == aarch64 ]]; then
     # Using "extra-index-url" until all needed aarch64 dependencies are
     # added to "https://download.pytorch.org/whl/nightly/"
-    pip install "\$pkg" --extra-index-url "https://download.pytorch.org/whl/nightly/${DESIRED_CUDA}"
+    pip install "\$pkg" --extra-index-url "https://download.pytorch.org/whl/test/${DESIRED_CUDA}"
   else
-    pip install "\$pkg" --index-url "https://download.pytorch.org/whl/nightly/${DESIRED_CUDA}"
+    pip install "\$pkg" --index-url "https://download.pytorch.org/whl/test/${DESIRED_CUDA}"
   fi
   retry pip install -q numpy protobuf typing-extensions
 fi

.circleci/scripts/binary_populate_env.sh

@@ -77,7 +77,9 @@ else
   export PYTORCH_BUILD_VERSION="${BASE_BUILD_VERSION}+$DESIRED_CUDA"
 fi
 
-if [[ -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]]; then
+# The build with with-pypi-cudnn suffix is only applicabe to
+# pypi small wheel Linux x86 build
+if [[ -n "${PYTORCH_EXTRA_INSTALL_REQUIREMENTS:-}" ]] && [[ "$(uname)" == 'Linux' && "$(uname -m)" == "x86_64" ]]; then
   export PYTORCH_BUILD_VERSION="${PYTORCH_BUILD_VERSION}-with-pypi-cudnn"
 fi
 

.circleci/scripts/binary_upload.sh

@@ -11,7 +11,7 @@ PKG_DIR=${PKG_DIR:-/tmp/workspace/final_pkgs}
 # currently set within `designate_upload_channel`
 UPLOAD_CHANNEL=${UPLOAD_CHANNEL:-nightly}
 # Designates what subfolder to put packages into
-UPLOAD_SUBFOLDER=${UPLOAD_SUBFOLDER:-cpu}
+UPLOAD_SUBFOLDER=${UPLOAD_SUBFOLDER:-}
 UPLOAD_BUCKET="s3://pytorch"
 BACKUP_BUCKET="s3://pytorch-backup"
 BUILD_NAME=${BUILD_NAME:-}
@@ -64,12 +64,17 @@ s3_upload() {
   local pkg_type
   extension="$1"
   pkg_type="$2"
-  s3_dir="${UPLOAD_BUCKET}/${pkg_type}/${UPLOAD_CHANNEL}/${UPLOAD_SUBFOLDER}/"
+  s3_root_dir="${UPLOAD_BUCKET}/${pkg_type}/${UPLOAD_CHANNEL}"
+  if [[ -z ${UPLOAD_SUBFOLDER:-} ]]; then
+    s3_upload_dir="${s3_root_dir}/"
+  else
+    s3_upload_dir="${s3_root_dir}/${UPLOAD_SUBFOLDER}/"
+  fi
   (
     for pkg in ${PKG_DIR}/*.${extension}; do
       (
         set -x
-        ${AWS_S3_CP} --no-progress --acl public-read "${pkg}" "${s3_dir}"
+        ${AWS_S3_CP} --no-progress --acl public-read "${pkg}" "${s3_upload_dir}"
       )
     done
   )
@@ -82,15 +87,17 @@ pip install -q awscli
 case "${PACKAGE_TYPE}" in
   conda)
     conda_upload
-    # Fetch  platform (eg. win-64, linux-64, etc.) from index file
-    # Because there's no actual conda command to read this
-    subdir=$(\
-      tar -xOf ${PKG_DIR}/*.bz2 info/index.json \
-        | grep subdir  \
-        | cut -d ':' -f2 \
-        | sed -e 's/[[:space:]]//' -e 's/"//g' -e 's/,//' \
-    )
-    BACKUP_DIR="conda/${subdir}"
+    for conda_archive in ${PKG_DIR}/*.tar.bz2; do
+      # Fetch  platform (eg. win-64, linux-64, etc.) from index file because
+      # there's no actual conda command to read this
+      subdir=$(\
+        tar -xOf "${conda_archive}" info/index.json \
+          | grep subdir  \
+          | cut -d ':' -f2 \
+          | sed -e 's/[[:space:]]//' -e 's/"//g' -e 's/,//' \
+      )
+      BACKUP_DIR="conda/${subdir}"
+    done
     ;;
   libtorch)
     s3_upload "zip" "libtorch"

.github/merge_rules.yaml

@@ -7,6 +7,7 @@
   - docs/source/onnx.rst
   - docs/source/onnx*
   - docs/source/scripts/onnx/**
+  - docs/source/_static/img/onnx/**
   - scripts/onnx/**
   - test/onnx/**
   - tools/onnx/**

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

@@ -25,3 +25,4 @@ sympy==1.11.1
 pytest-cpp==2.3.0
 rockset==1.0.3
 z3-solver==4.12.2.0
+tensorboard==2.13.0

.github/scripts/build_triton_wheel.py

@@ -60,12 +60,18 @@ def build_triton(
     build_conda: bool = False,
     build_rocm: bool = False,
     py_version: Optional[str] = None,
+    release: bool = False,
 ) -> Path:
     env = os.environ.copy()
     if "MAX_JOBS" not in env:
         max_jobs = os.cpu_count() or 1
         env["MAX_JOBS"] = str(max_jobs)
 
+    if not release:
+        # Nightly binaries include the triton commit hash, i.e. 2.1.0+e6216047b8
+        # while release build should only include the version, i.e. 2.1.0
+        version = f"{version}+{commit_hash[:10]}"
+
     with TemporaryDirectory() as tmpdir:
         triton_basedir = Path(tmpdir) / "triton"
         triton_pythondir = triton_basedir / "python"
@@ -80,7 +86,7 @@ def build_triton(
         if build_conda:
             with open(triton_basedir / "meta.yaml", "w") as meta:
                 print(
-                    f"package:\n  name: torchtriton\n  version: {version}+{commit_hash[:10]}\n",
+                    f"package:\n  name: torchtriton\n  version: {version}\n",
                     file=meta,
                 )
                 print("source:\n  path: .\n", file=meta)
@@ -103,7 +109,7 @@ def build_triton(
 
             patch_init_py(
                 triton_pythondir / "triton" / "__init__.py",
-                version=f"{version}+{commit_hash[:10]}",
+                version=f"{version}",
             )
             if py_version is None:
                 py_version = f"{sys.version_info.major}.{sys.version_info.minor}"
@@ -129,11 +135,11 @@ def build_triton(
         patch_setup_py(
             triton_pythondir / "setup.py",
             name=triton_pkg_name,
-            version=f"{version}+{commit_hash[:10]}",
+            version=f"{version}",
         )
         patch_init_py(
             triton_pythondir / "triton" / "__init__.py",
-            version=f"{version}+{commit_hash[:10]}",
+            version=f"{version}",
         )
 
         if build_rocm:
@@ -157,12 +163,14 @@ def main() -> None:
     from argparse import ArgumentParser
 
     parser = ArgumentParser("Build Triton binaries")
+    parser.add_argument("--release", action="store_true")
     parser.add_argument("--build-conda", action="store_true")
     parser.add_argument("--build-rocm", action="store_true")
     parser.add_argument("--py-version", type=str)
     parser.add_argument("--commit-hash", type=str)
     parser.add_argument("--triton-version", type=str, default=read_triton_version())
     args = parser.parse_args()
+
     build_triton(
         build_rocm=args.build_rocm,
         commit_hash=args.commit_hash
@@ -171,6 +179,7 @@ def main() -> None:
         version=args.triton_version,
         build_conda=args.build_conda,
         py_version=args.py_version,
+        release=args.release,
     )
 
 

.github/scripts/ensure_actions_will_cancel.py

@@ -1,6 +1,5 @@
 #!/usr/bin/env python3
 
-import argparse
 import sys
 
 from pathlib import Path
@@ -10,9 +9,11 @@ import yaml
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent
 WORKFLOWS = REPO_ROOT / ".github" / "workflows"
-EXPECTED_GROUP = (
+EXPECTED_GROUP_PREFIX = (
     "${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}"
-    "-${{ github.event_name == 'workflow_dispatch' }}"
+)
+EXPECTED_GROUP = (
+    EXPECTED_GROUP_PREFIX + "-${{ github.event_name == 'workflow_dispatch' }}"
 )
 
 
@@ -26,15 +27,8 @@ def should_check(filename: Path) -> bool:
 
 
 if __name__ == "__main__":
-    parser = argparse.ArgumentParser(
-        description="Ensure all relevant GitHub actions jobs will be cancelled based on a concurrency key"
-    )
-    args = parser.parse_args()
-
-    files = list(WORKFLOWS.glob("*.yml"))
-
     errors_found = False
-    files = [f for f in files if should_check(f)]
+    files = [f for f in WORKFLOWS.glob("*.yml") if should_check(f)]
     names = set()
     for filename in files:
         with open(filename) as f:
@@ -46,7 +40,18 @@ if __name__ == "__main__":
             errors_found = True
         names.add(name)
         actual = data.get("concurrency", {})
-        if not actual.get("group", "").startswith(EXPECTED_GROUP):
+        if filename.name == "create_release.yml":
+            if not actual.get("group", "").startswith(EXPECTED_GROUP_PREFIX):
+                print(
+                    f"'concurrency' incorrect or not found in '{filename.relative_to(REPO_ROOT)}'",
+                    file=sys.stderr,
+                )
+                print(
+                    f"concurrency group should start with {EXPECTED_GROUP_PREFIX} but found {actual.get('group', None)}",
+                    file=sys.stderr,
+                )
+                errors_found = True
+        elif not actual.get("group", "").startswith(EXPECTED_GROUP):
             print(
                 f"'concurrency' incorrect or not found in '{filename.relative_to(REPO_ROOT)}'",
                 file=sys.stderr,

.github/scripts/filter_test_configs.py

@@ -62,9 +62,10 @@ SUPPORTED_PERIODICAL_MODES: Dict[str, Callable[[Optional[str]], bool]] = {
 }
 
 # The link to the published list of disabled jobs
-DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json"
+# Pinning Disabled and Unstable job to Oct 4, 2023.
+DISABLED_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/disabled-jobs.json?versionId=EniFrNbB6taGjwKyN94j4oqUeeN8ALfI"
 # and unstable jobs
-UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json"
+UNSTABLE_JOBS_URL = "https://ossci-metrics.s3.amazonaws.com/unstable-jobs.json?versionId=2voGK5DSv0Hzvxhc23ChGcOLEBIO2vHf"
 
 # Some constants used to handle disabled and unstable jobs
 JOB_NAME_SEP = "/"

.github/scripts/generate_binary_build_matrix.py

@@ -16,6 +16,12 @@ from typing import Dict, List, Optional, Tuple
 CUDA_ARCHES = ["11.8", "12.1"]
 
 
+CUDA_ARCHES_FULL_VERSION = {"11.8": "11.8.0", "12.1": "12.1.1"}
+
+
+CUDA_ARCHES_CUDNN_VERSION = {"11.8": "8", "12.1": "8"}
+
+
 ROCM_ARCHES = ["5.5", "5.6"]
 
 
@@ -24,6 +30,21 @@ CPU_CXX11_ABI_ARCH = ["cpu-cxx11-abi"]
 
 CPU_AARCH64_ARCH = ["cpu-aarch64"]
 
+PYTORCH_EXTRA_INSTALL_REQUIREMENTS = (
+    "nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "  # noqa: B950
+    "nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "
+    "triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'"
+)
+
 
 def arch_type(arch_version: str) -> str:
     if arch_version in CUDA_ARCHES:
@@ -39,22 +60,19 @@ def arch_type(arch_version: str) -> str:
 
 
 WHEEL_CONTAINER_IMAGES = {
-    **{
-        gpu_arch: f"pytorch/manylinux-builder:cuda{gpu_arch}"
-        for gpu_arch in CUDA_ARCHES
-    },
-    **{
-        gpu_arch: f"pytorch/manylinux-builder:rocm{gpu_arch}"
-        for gpu_arch in ROCM_ARCHES
-    },
-    "cpu": "pytorch/manylinux-builder:cpu",
-    "cpu-cxx11-abi": "pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi",
-    "cpu-aarch64": "pytorch/manylinuxaarch64-builder:cpu-aarch64",
+    "11.8": "pytorch/manylinux-builder:cuda11.8-2.1",
+    "12.1": "pytorch/manylinux-builder:cuda12.1-2.1",
+    "5.5": "pytorch/manylinux-builder:rocm5.5-2.1",
+    "5.6": "pytorch/manylinux-builder:rocm5.6-2.1",
+    "cpu": "pytorch/manylinux-builder:cpu-2.1",
+    "cpu-cxx11-abi": "pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1",
+    "cpu-aarch64": "pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1",
 }
 
 CONDA_CONTAINER_IMAGES = {
-    **{gpu_arch: f"pytorch/conda-builder:cuda{gpu_arch}" for gpu_arch in CUDA_ARCHES},
-    "cpu": "pytorch/conda-builder:cpu",
+    "11.8": "pytorch/conda-builder:cuda11.8-2.1",
+    "12.1": "pytorch/conda-builder:cuda12.1-2.1",
+    "cpu": "pytorch/conda-builder:cpu-2.1",
 }
 
 PRE_CXX11_ABI = "pre-cxx11"
@@ -63,24 +81,46 @@ RELEASE = "release"
 DEBUG = "debug"
 
 LIBTORCH_CONTAINER_IMAGES: Dict[Tuple[str, str], str] = {
-    **{
-        (gpu_arch, PRE_CXX11_ABI): f"pytorch/manylinux-builder:cuda{gpu_arch}"
-        for gpu_arch in CUDA_ARCHES
-    },
-    **{
-        (gpu_arch, CXX11_ABI): f"pytorch/libtorch-cxx11-builder:cuda{gpu_arch}"
-        for gpu_arch in CUDA_ARCHES
-    },
-    **{
-        (gpu_arch, PRE_CXX11_ABI): f"pytorch/manylinux-builder:rocm{gpu_arch}"
-        for gpu_arch in ROCM_ARCHES
-    },
-    **{
-        (gpu_arch, CXX11_ABI): f"pytorch/libtorch-cxx11-builder:rocm{gpu_arch}"
-        for gpu_arch in ROCM_ARCHES
-    },
-    ("cpu", PRE_CXX11_ABI): "pytorch/manylinux-builder:cpu",
-    ("cpu", CXX11_ABI): "pytorch/libtorch-cxx11-builder:cpu",
+    (
+        "11.8",
+        PRE_CXX11_ABI,
+    ): "pytorch/manylinux-builder:cuda11.8-2.1",
+    (
+        "12.1",
+        PRE_CXX11_ABI,
+    ): "pytorch/manylinux-builder:cuda12.1-2.1",
+    (
+        "11.8",
+        CXX11_ABI,
+    ): "pytorch/libtorch-cxx11-builder:cuda11.8-2.1",
+    (
+        "12.1",
+        CXX11_ABI,
+    ): "pytorch/libtorch-cxx11-builder:cuda12.1-2.1",
+    (
+        "5.5",
+        PRE_CXX11_ABI,
+    ): "pytorch/manylinux-builder:rocm5.5-2.1",
+    (
+        "5.6",
+        PRE_CXX11_ABI,
+    ): "pytorch/manylinux-builder:rocm5.6-2.1",
+    (
+        "5.5",
+        CXX11_ABI,
+    ): "pytorch/libtorch-cxx11-builder:rocm5.5-2.1",
+    (
+        "5.6",
+        CXX11_ABI,
+    ): "pytorch/libtorch-cxx11-builder:rocm5.6-2.1",
+    (
+        "cpu",
+        PRE_CXX11_ABI,
+    ): "pytorch/manylinux-builder:cpu-2.1",
+    (
+        "cpu",
+        CXX11_ABI,
+    ): "pytorch/libtorch-cxx11-builder:cpu-2.1",
 }
 
 FULL_PYTHON_VERSIONS = ["3.8", "3.9", "3.10", "3.11"]
@@ -238,17 +278,7 @@ def generate_wheels_matrix(
                         "devtoolset": "",
                         "container_image": WHEEL_CONTAINER_IMAGES[arch_version],
                         "package_type": package_type,
-                        "pytorch_extra_install_requirements": "nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "  # noqa: B950
-                        "nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | "
-                        "nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'",
+                        "pytorch_extra_install_requirements": PYTORCH_EXTRA_INSTALL_REQUIREMENTS,
                         "build_name": f"{package_type}-py{python_version}-{gpu_arch_type}{gpu_arch_version}-with-pypi-cudnn".replace(  # noqa: B950
                             ".", "_"
                         ),
@@ -273,6 +303,9 @@ def generate_wheels_matrix(
                     "build_name": f"{package_type}-py{python_version}-{gpu_arch_type}{gpu_arch_version}".replace(
                         ".", "_"
                     ),
+                    "pytorch_extra_install_requirements": PYTORCH_EXTRA_INSTALL_REQUIREMENTS
+                    if os != "linux"
+                    else "",
                 }
             )
     return ret

.github/scripts/generate_docker_release_matrix.py

@@ -0,0 +1,45 @@
+#!/usr/bin/env python3
+
+"""Generates a matrix for docker releases through github actions
+
+Will output a condensed version of the matrix. Will include fllowing:
+    * CUDA version short
+    * CUDA full verison
+    * CUDNN version short
+    * Image type either runtime or devel
+    * Platform linux/arm64,linux/amd64
+
+"""
+
+import json
+from typing import Dict, List
+
+import generate_binary_build_matrix
+
+DOCKER_IMAGE_TYPES = ["runtime", "devel"]
+
+
+def generate_docker_matrix() -> Dict[str, List[Dict[str, str]]]:
+    ret: List[Dict[str, str]] = []
+    for cuda, version in generate_binary_build_matrix.CUDA_ARCHES_FULL_VERSION.items():
+        for image in DOCKER_IMAGE_TYPES:
+            platform = (
+                "linux/arm64,linux/amd64" if image == "runtime" else "linux/amd64"
+            )
+            ret.append(
+                {
+                    "cuda": cuda,
+                    "cuda_full_version": version,
+                    "cudnn_version": generate_binary_build_matrix.CUDA_ARCHES_CUDNN_VERSION[
+                        cuda
+                    ],
+                    "image_type": image,
+                    "platform": platform,
+                }
+            )
+    return {"include": ret}
+
+
+if __name__ == "__main__":
+    build_matrix = generate_docker_matrix()
+    print(json.dumps(build_matrix))

.github/templates/common.yml.j2

@@ -37,6 +37,7 @@ concurrency:
       !{{ display_ec2_information() }}
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

.github/templates/linux_binary_build_workflow.yml.j2

@@ -55,12 +55,12 @@ jobs:
     uses: ./.github/workflows/_binary-build-linux.yml
     with:!{{ upload.binary_env_as_input(config) }}
       {%- if "aarch64" in build_environment %}
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       {%- endif %}
       build_name: !{{ config["build_name"] }}
       build_environment: !{{ build_environment }}
-      {%- if config.pytorch_extra_install_requirements is defined %}
+      {%- if config.pytorch_extra_install_requirements is defined and config.pytorch_extra_install_requirements|d('')|length > 0  %}
       PYTORCH_EXTRA_INSTALL_REQUIREMENTS: !{{ config.pytorch_extra_install_requirements }}
       {%- endif %}
     secrets:
@@ -74,7 +74,7 @@ jobs:
       build_name: !{{ config["build_name"] }}
       build_environment: !{{ build_environment }}
       {%- if "aarch64" in build_environment %}
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       {%- elif config["gpu_arch_type"] == "rocm" %}
       runs_on: linux.rocm.gpu

.github/templates/macos_binary_build_workflow.yml.j2

@@ -61,6 +61,9 @@ jobs:
     runs-on: macos-12-xl
     timeout-minutes: !{{ common.timeout_minutes }}
     !{{ upload.binary_env(config, true) }}
+    {%- if config.pytorch_extra_install_requirements is defined and config.pytorch_extra_install_requirements|d('')|length > 0  %}
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: !{{ config.pytorch_extra_install_requirements }}
+    {%- endif %}
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}

.github/templates/windows_binary_build_workflow.yml.j2

@@ -59,6 +59,9 @@ jobs:
     runs-on: windows.4xlarge.nonephemeral
     timeout-minutes: !{{ common.timeout_minutes }}
     !{{ upload.binary_env(config, True) }}
+    {%- if config.pytorch_extra_install_requirements is defined and config.pytorch_extra_install_requirements|d('')|length > 0  %}
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: !{{ config.pytorch_extra_install_requirements }}
+    {%- endif %}
     steps:
       !{{ common.setup_ec2_windows() }}
       !{{ set_runner_specific_vars() }}

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

@@ -140,6 +140,7 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.1
+        continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 
@@ -159,10 +160,12 @@ jobs:
       - name: Clean workspace
         shell: bash
         run: |
+          set -eux
+
           rm -rf "${GITHUB_WORKSPACE}"
           mkdir "${GITHUB_WORKSPACE}"
 
-          if [[ inputs.build_environment == 'linux-aarch64-binary-manywheel' ]]; then
+          if [[ ${{ inputs.build_environment }} == 'linux-aarch64-binary-manywheel' ]]; then
             rm -rf "${RUNNER_TEMP}/artifacts"
             mkdir "${RUNNER_TEMP}/artifacts"
           fi

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

@@ -128,6 +128,7 @@ jobs:
 
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.1
+        continue-on-error: true
         with:
           github-secret: ${{ secrets.github-token }}
 

.github/workflows/_binary-upload.yml

@@ -121,7 +121,7 @@ jobs:
         shell: bash -e -l {0}
         run: |
           # reference ends with an RC suffix
-          if [[ ${GITHUB_REF_NAME} = *-rc[0-9]* ]]; then
+          if [[ "${GITHUB_REF_NAME}" = *-rc[0-9]* ]]; then
             echo "UPLOAD_CHANNEL=test" >> "$GITHUB_ENV"
           fi
 

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

@@ -7,14 +7,6 @@ on:
         required: true
         type: string
         description: Top-level label for what's being built/tested.
-      ios-platform:
-        required: true
-        type: string
-        description: Which iOS platform to build for.
-      ios-arch:
-        required: true
-        type: string
-        description: Which iOS arch to build for.
       sync-tag:
         required: false
         type: string
@@ -31,8 +23,6 @@ on:
 env:
   GIT_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
   BUILD_ENVIRONMENT: ${{ inputs.build-environment }}
-  IOS_PLATFORM: ${{ inputs.ios-platform }}
-  IOS_ARCH: ${{ inputs.ios-arch }}
 
 jobs:
   filter:
@@ -63,6 +53,16 @@ jobs:
       matrix: ${{ fromJSON(needs.filter.outputs.test-matrix) }}
       fail-fast: false
     runs-on: ${{ matrix.runner }}
+    env:
+      IOS_PLATFORM: ${{ matrix.ios_platform }}
+      IOS_ARCH: ${{ matrix.ios_arch }}
+      BUILD_LITE_INTERPRETER: ${{ matrix.use_lite_interpreter }}
+      USE_PYTORCH_METAL: ${{ matrix.use_metal }}
+      USE_COREML_DELEGATE: ${{ matrix.use_coreml }}
+      CUSTOM_OP_LIST: ${{ matrix.use_custom_op_list }}
+      # TODO: Bump it to 2.2.0 after cherry pick this or figure out a better way
+      # to get this version instead of hard coding it here
+      PYTORCH_VERSION: 2.1.0
     timeout-minutes: 240
     steps:
       # [see note: pytorch repo ref]
@@ -70,26 +70,13 @@ jobs:
         uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.1
 
       - name: Populate CI build options
+        shell: bash
         run: |
-          # Most builds use the lite interpreter, if certain builds shouldn't
-          # build the lite interpreter this env variable should get over-written
-          # in the following case statement
-          echo "BUILD_LITE_INTERPRETER=1" >> "${GITHUB_ENV}"
+          set -ex
 
-          case ${BUILD_ENVIRONMENT} in
-            *metal*)
-              echo "USE_PYTORCH_METAL=1" >> "${GITHUB_ENV}"
-              ;;
-            *full_jit*)
-              echo "BUILD_LITE_INTERPRETER=0" >> "${GITHUB_ENV}"
-              ;;
-            *custom*)
-              echo "SELECTED_OP_LIST=${GITHUB_WORKSPACE}/ios/TestApp/custom_build/mobilenetv2.yaml" >> "${GITHUB_ENV}"
-              ;;
-            *coreml*)
-              echo "USE_COREML_DELEGATE=1" >> "${GITHUB_ENV}"
-              ;;
-          esac
+          if [ -n "${CUSTOM_OP_LIST:-}" ]; then
+            echo "SELECTED_OP_LIST=${GITHUB_WORKSPACE}/ios/TestApp/custom_build/${CUSTOM_OP_LIST}" >> "${GITHUB_ENV}"
+          fi
 
       - name: Install brew dependencies
         uses: nick-fields/retry@v2.8.2
@@ -116,54 +103,67 @@ jobs:
           retry_wait_seconds: 90
           command: |
             set -x
-            cd ios/TestApp
-            # install fastlane
+
+            pushd ios/TestApp
+            # Install fastlane
             sudo gem install bundler && bundle install
             bundle update fastlane
+            popd
 
-      - name: Build PyTorch Mobile Runtime
+      - name: Build PyTorch mobile runtime
+        shell: bash
         run: |
+          set -eux
           # shellcheck disable=SC1091
           export TCLLIBPATH="/usr/local/lib"
-          python -VV
           ${CONDA_RUN} scripts/build_ios.sh
 
       - name: Build TestApp
-        if: inputs.ios-platform == 'SIMULATOR'
+        if: matrix.ios_platform == 'SIMULATOR'
         timeout-minutes: 15
         run: |
-          # run the ruby build script
+          # Run the ruby build script
           if ! [ -x "$(command -v xcodebuild)" ]; then
             echo 'Error: xcodebuild is not installed.'
             exit 1
           fi
           ruby scripts/xcode_build.rb -i build_ios/install -x ios/TestApp/TestApp.xcodeproj -p "${IOS_PLATFORM}"
 
-      - name: Run Simulator Tests
-        if: inputs.ios-platform == 'SIMULATOR'
+      - name: Run simulator tests
+        if: matrix.ios_platform == 'SIMULATOR'
+        shell: bash
         run: |
+          set -eux
           # shellcheck disable=SC1091
-          # use the pytorch nightly build to generate models
-          ${CONDA_RUN} pip3 install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cpu/torch_nightly.html
-          # generate models for differnet backends
-          cd "${GITHUB_WORKSPACE}/ios/TestApp/benchmark"
+          # Use the pytorch nightly build to generate models
+          ${CONDA_RUN} pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu
+
+          # Generate models for differnet backends
+          pushd "${GITHUB_WORKSPACE}/ios/TestApp/benchmark"
           mkdir -p ../models
+          # NB: Both of the following scripts only export models with lite interpreter
           if [ "${USE_COREML_DELEGATE}" == 1 ]; then
             ${CONDA_RUN} python coreml_backend.py
           else
-            cd "${GITHUB_WORKSPACE}"
+            pushd "${GITHUB_WORKSPACE}"
             ${CONDA_RUN} python test/mobile/model_test/gen_test_model.py ios-test
+            popd
           fi
-          cd "${GITHUB_WORKSPACE}/ios/TestApp/benchmark"
+
           if [ "${BUILD_LITE_INTERPRETER}" == 1 ]; then
             echo "Setting up the TestApp for LiteInterpreter"
             ruby setup.rb --lite 1
           else
+            # Generate some models for JIT without lite interpreter
+            ${CONDA_RUN} python trace_model.py
+
             echo "Setting up the TestApp for Full JIT"
             ruby setup.rb
           fi
-          cd "${GITHUB_WORKSPACE}/ios/TestApp"
-          # instruments -s -devices
+          popd
+
+          pushd "${GITHUB_WORKSPACE}/ios/TestApp"
+          # Instruments -s -devices
           if [ "${BUILD_LITE_INTERPRETER}" == 1 ]; then
             if [ "${USE_COREML_DELEGATE}" == 1 ]; then
               bundle exec fastlane scan --only_testing TestAppTests/TestAppTests/testCoreML
@@ -173,9 +173,282 @@ jobs:
           else
             bundle exec fastlane scan --only_testing TestAppTests/TestAppTests/testFullJIT
           fi
+          popd
 
-      - name: Dump Simulator Tests On a Failure
-        if: failure() && inputs.ios-platform == 'SIMULATOR'
+      - name: Dump simulator tests on failure
+        if: failure() && matrix.ios_platform == 'SIMULATOR'
         run: |
           echo "Simulator Tests Logs:"
           cat /Users/runner/Library/Logs/scan/*.log
+
+      - name: Prepare the build artifacts for upload
+        shell: bash
+        run: |
+          set -eux
+
+          # The structure of the folder is as follows:
+          #
+          # RUNNER_TEMP/
+          # └── IOS_ARCH/
+          #     ├── LICENSE
+          #     ├── install
+          #     │   ├── include
+          #     │   │   └── headers
+          #     │   └── lib
+          #     │       ├── libXNNPACK.a
+          #     │       ├── libc10.a
+          #     │       ├── libclog.a
+          #     │       ├── libcpuinfo.a
+          #     │       ├── libeigen_blas.a
+          #     │       ├── libpthreadpool.a
+          #     │       ├── libpytorch_qnnpack.a
+          #     │       ├── libtorch.a
+          #     │       └── libtorch_cpu.a
+          #     ├── src
+          #     │   └── LibTorch-Lite.h
+          #     └── version.txt
+          SETUP_DIR="${RUNNER_TEMP}/${IOS_ARCH}"
+          mkdir -p "${SETUP_DIR}/src"
+
+          cp -R "${GITHUB_WORKSPACE}/build_ios/install" "${SETUP_DIR}"
+          # Copy the umbrella header and license
+          if [ "${BUILD_LITE_INTERPRETER}" == 1 ]; then
+            cp "${GITHUB_WORKSPACE}/ios/LibTorch-Lite.h" "${SETUP_DIR}/src"
+          else
+            cp "${GITHUB_WORKSPACE}/ios/LibTorch.h" "${SETUP_DIR}/src"
+          fi
+
+          # Copy license and version
+          cp "${GITHUB_WORKSPACE}/LICENSE" "${SETUP_DIR}"
+          echo "${PYTORCH_VERSION}" > "${SETUP_DIR}"/version.txt
+
+          # Save the podspec for the upload job later
+          if [ "${BUILD_LITE_INTERPRETER}" == "1" ]; then
+            DATE=$(date -u +%Y%m%d)
+            cp "${GITHUB_WORKSPACE}"/ios/LibTorch-Lite-Nightly.podspec.template "${SETUP_DIR}"/LibTorch-Lite-Nightly.podspec
+            sed -i '' -e "s/IOS_NIGHTLY_BUILD_VERSION/${PYTORCH_VERSION}.${DATE}/g" "${SETUP_DIR}"/LibTorch-Lite-Nightly.podspec
+
+            cp "${GITHUB_WORKSPACE}"/ios/LibTorch-Lite.podspec.template "${SETUP_DIR}"/LibTorch-Lite.podspec
+            sed -i '' -e "s/IOS_BUILD_VERSION/${PYTORCH_VERSION}/g" "${SETUP_DIR}"/LibTorch-Lite.podspec
+          else
+            # NB: There is no nightly build without lite interpreter atm
+            cp "${GITHUB_WORKSPACE}"/ios/LibTorch.podspec.template "${SETUP_DIR}"/LibTorch.podspec
+            sed -i '' -e "s/IOS_BUILD_VERSION/${PYTORCH_VERSION}/g" "${SETUP_DIR}"/LibTorch.podspec
+          fi
+
+          pushd "${SETUP_DIR}"
+          # NB: It's important to zip all the files before uploading because the GHA will upload
+          # all files sequentially which is both slow and has too many requests. More info is at
+          # https://github.com/actions/upload-artifact#too-many-uploads-resulting-in-429-responses
+          zip -r "${IOS_ARCH}.zip" install src version.txt LICENSE ./*.podspec
+          popd
+
+      - uses: actions/upload-artifact@v3
+        with:
+          name: pytorch-ios-build-artifacts-${{ matrix.ios_arch }}
+          if-no-files-found: error
+          path: ${{ runner.temp }}/${{ matrix.ios_arch }}/${{ matrix.ios_arch }}.zip
+
+  upload-ios-artifacts:
+    # NB: this job run on GitHub MacOS ephemeral runner so that it can use lipo
+    # to create the fat iOS binaries for both x86_64 and arm64
+    runs-on: macos-12
+    needs: build
+    # NB: Only upload release build, if we need it, we could also turn on nightly here
+    environment: ${{ (github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || startsWith(github.event.ref, 'refs/tags/v'))) && 'ios-upload' || '' }}
+    steps:
+      - uses: actions/checkout@v3
+
+      # For awscli S3 upload
+      - uses: actions/setup-python@v4
+        with:
+          python-version: '3.10'
+          cache: pip
+
+      # For cocoapods pod upload
+      - uses: ruby/setup-ruby@v1
+        with:
+          ruby-version: '3.2'
+          bundler-cache: true
+
+      - name: Download arm64 artifacts
+        uses: actions/download-artifact@v3
+        with:
+          name: pytorch-ios-build-artifacts-arm64
+
+      - name: Download x86_64 artifacts
+        uses: actions/download-artifact@v3
+        with:
+          name: pytorch-ios-build-artifacts-x86_64
+
+      - name: Unzip arm64 and x86_64 artifacts
+        shell: bash
+        run: |
+          set -eux
+
+          for ARCH in "arm64" "x86_64"; do
+            TMP_DIR="${RUNNER_TEMP}/${ARCH}"
+            mkdir -p "${TMP_DIR}"
+
+            cp "${ARCH}.zip" "${TMP_DIR}"
+
+            pushd "${TMP_DIR}"
+            unzip -o "${ARCH}.zip"
+            popd
+          done
+
+      - name: Prepare the artifact
+        env:
+          IS_NIGHTLY: ${{ github.event.ref == 'refs/heads/nightly' }}
+        shell: bash
+        working-directory: ${{ runner.temp }}/arm64
+        run: |
+          set -eux
+
+          DEST_DIR="${RUNNER_TEMP}"/ios
+          echo "DEST_DIR=${DEST_DIR}" >> "$GITHUB_ENV"
+
+          # Prepare all the sub directories
+          mkdir -p "${DEST_DIR}"/install/lib
+
+          # Copy header and share files, arm64 or x86_64 both work
+          cp -R install/include "${DEST_DIR}"/install
+          cp -R install/share "${DEST_DIR}"/install
+          # The last dash is important to copy only files under src
+          cp -R src "${DEST_DIR}"
+          cp LICENSE "${DEST_DIR}"
+
+          if [ "${IS_NIGHTLY}" == true ]; then
+            PYTORCH_VERSION=$(cat version.txt)
+            DATE=$(date -u +%Y%m%d)
+            echo "${PYTORCH_VERSION}.${DATE}" > "${DEST_DIR}"/version.txt
+          else
+            cp version.txt "${DEST_DIR}"
+          fi
+          PYTORCH_VERSION=$(cat "${DEST_DIR}"/version.txt)
+          echo "PYTORCH_VERSION=${PYTORCH_VERSION}" >> "$GITHUB_ENV"
+
+          pushd install/lib
+          # shellcheck disable=SC2207
+          LIBRARIES=($(ls ./*.a))
+          popd
+
+          for LIB in "${LIBRARIES[@]}"; do
+            FROM_LIBS=("${RUNNER_TEMP}"/arm64/install/lib/"${LIB}" "${RUNNER_TEMP}"/x86_64/install/lib/"${LIB}")
+            # Create a fat binary for both arm64 and x86_64
+            lipo -create "${FROM_LIBS[@]}" -o "${DEST_DIR}"/install/lib/"${LIB}"
+            # Print the info
+            lipo -i "${DEST_DIR}"/install/lib/"${LIB}"
+          done
+
+          BUILD_LITE_INTERPRETER=1
+          if [ -f "${RUNNER_TEMP}"/arm64/LibTorch.podspec ]; then
+            # If LibTorch.podspec is used instead of LibTorch-Lite.podspec, the artifact is built
+            # without lite interpreter
+            BUILD_LITE_INTERPRETER=0
+          fi
+          echo "BUILD_LITE_INTERPRETER=${BUILD_LITE_INTERPRETER}" >> "$GITHUB_ENV"
+
+      - name: Prepare the podspec
+        env:
+          IS_NIGHTLY: ${{ github.event.ref == 'refs/heads/nightly' }}
+        shell: bash
+        working-directory: ${{ env.DEST_DIR }}
+        run: |
+          set -eux
+
+          ARTIFACT_NAME=libtorch
+          SPEC_NAME=LibTorch
+
+          if [ "${BUILD_LITE_INTERPRETER}" == "1" ]; then
+            ARTIFACT_NAME="${ARTIFACT_NAME}_lite_ios"
+            SPEC_NAME="${SPEC_NAME}-Lite"
+          else
+            ARTIFACT_NAME="${ARTIFACT_NAME}_ios"
+          fi
+
+          if [ "${IS_NIGHTLY}" == true ]; then
+            ARTIFACT_NAME="${ARTIFACT_NAME}_nightly_${PYTORCH_VERSION}.zip"
+            SPEC_NAME="${SPEC_NAME}-Nightly"
+          else
+            ARTIFACT_NAME="${ARTIFACT_NAME}_${PYTORCH_VERSION}.zip"
+          fi
+
+          SPEC_NAME_WITH_VERSION="${SPEC_NAME}-${PYTORCH_VERSION}.podspec"
+          SPEC_NAME="${SPEC_NAME}.podspec"
+
+          # Also copy the spec file
+          cp "${RUNNER_TEMP}"/arm64/"${SPEC_NAME}" "${SPEC_NAME_WITH_VERSION}"
+
+          # NB: It's important to zip all the files before uploading because the GHA will upload
+          # all files sequentially which is both slow and has too many requests. More info is at
+          # https://github.com/actions/upload-artifact#too-many-uploads-resulting-in-429-responses
+          zip -r "${ARTIFACT_NAME}" install src version.txt LICENSE
+
+          {
+            echo "ARTIFACT_NAME=${ARTIFACT_NAME}"
+            echo "SPEC_NAME_WITH_VERSION=${SPEC_NAME_WITH_VERSION}"
+            echo "SPEC_NAME=${SPEC_NAME}"
+          } >> "$GITHUB_ENV"
+
+      - uses: actions/upload-artifact@v3
+        with:
+          name: pytorch-ios-artifacts
+          if-no-files-found: error
+          path: ${{ env.DEST_DIR }}/${{ env.ARTIFACT_NAME }}
+
+      - uses: actions/upload-artifact@v3
+        with:
+          name: pytorch-ios-podspec
+          if-no-files-found: error
+          path: ${{ env.DEST_DIR }}/${{ env.SPEC_NAME_WITH_VERSION }}
+
+      - name: Set DRY_RUN
+        if: ${{ github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || (startsWith(github.event.ref, 'refs/tags/v'))) }}
+        shell: bash
+        run: |
+          echo "DRY_RUN=disabled" >> "$GITHUB_ENV"
+
+      - name: Upload the artifact to S3
+        env:
+          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_PYTORCH_UPLOADER_ACCESS_KEY_ID }}
+          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_PYTORCH_UPLOADER_SECRET_ACCESS_KEY }}
+          IS_NIGHTLY: ${{ github.event.ref == 'refs/heads/nightly' }}
+        shell: bash
+        working-directory: ${{ env.DEST_DIR }}
+        run: |
+          set -eux
+
+          pip install -q awscli==1.29.40
+
+          DRY_RUN=${DRY_RUN:-enabled}
+          AWS_S3_CP="aws s3 cp --dryrun"
+          if [ "${DRY_RUN}" == "disabled" ]; then
+            AWS_S3_CP="aws s3 cp"
+          fi
+
+          if [ "${IS_NIGHTLY}" == true ]; then
+            BUCKET_NAME="ossci-ios-build"
+          else
+            BUCKET_NAME="ossci-ios"
+          fi
+
+          ${AWS_S3_CP} "${ARTIFACT_NAME}" "s3://${BUCKET_NAME}/" --acl public-read
+          ${AWS_S3_CP} "${SPEC_NAME_WITH_VERSION}" "s3://${BUCKET_NAME}/" --acl public-read
+
+      - name: Upload the artifact to cocoapods (nightly only)
+        env:
+          # We need to set this secret to upload to cocoapods. However, we might want
+          # to NOT set this for PROD release so that we can upload the artifacts manually
+          COCOAPODS_TRUNK_TOKEN: ${{ secrets.COCOAPODS_TRUNK_TOKEN || '' }}
+        if: ${{ github.event_name == 'push' && github.event.ref == 'refs/heads/nightly' && env.COCOAPODS_TRUNK_TOKEN != '' }}
+        shell: bash
+        working-directory: ${{ runner.temp }}/arm64
+        run: |
+          set -eux
+
+          gem install cocoapods
+
+          pod trunk me
+          # Upload the spec to cocoapods
+          pod trunk push --verbose --allow-warnings --use-libraries --skip-import-validation "${SPEC_NAME}"

.github/workflows/_run_android_tests.yml

@@ -41,9 +41,17 @@ jobs:
     strategy:
       matrix: ${{ fromJSON(needs.filter.outputs.test-matrix) }}
       fail-fast: false
+    # NB: This job can only run on GitHub Linux runner atm. This is an ok thing though
+    # because that runner is ephemeral and could access upload secrets
     runs-on: ${{ matrix.runner }}
+    env:
+      # GitHub runner installs Android SDK on this path
+      ANDROID_ROOT: /usr/local/lib/android
+      ANDROID_NDK_VERSION: '21.4.7075529'
+      BUILD_LITE_INTERPRETER: ${{ matrix.use_lite_interpreter }}
+      # 4 of them are supported atm: armeabi-v7a, arm64-v8a, x86, x86_64
+      SUPPORT_ABI: '${{ matrix.support_abi }}'
     steps:
-      # [see note: pytorch repo ref]
       - name: Checkout PyTorch
         uses: pytorch/pytorch/.github/actions/checkout-pytorch@release/2.1
 
@@ -51,7 +59,7 @@ jobs:
         uses: pytorch/test-infra/.github/actions/setup-miniconda@release/2.1
         with:
           python-version: 3.8
-          environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}
+          environment-file: .github/requirements/conda-env-${{ runner.os }}-${{ runner.arch }}.txt
 
       - name: Install NDK
         uses: nick-fields/retry@v2.8.2
@@ -60,12 +68,12 @@ jobs:
           max_attempts: 3
           retry_wait_seconds: 90
           command: |
+            set -eux
+
             # Install NDK 21 after GitHub update
             # https://github.com/actions/virtual-environments/issues/5595
-            ANDROID_ROOT="/usr/local/lib/android"
             ANDROID_SDK_ROOT="${ANDROID_ROOT}/sdk"
             ANDROID_NDK="${ANDROID_SDK_ROOT}/ndk-bundle"
-            ANDROID_NDK_VERSION="21.4.7075529"
 
             SDKMANAGER="${ANDROID_SDK_ROOT}/cmdline-tools/latest/bin/sdkmanager"
             # NB: This step downloads and installs NDK, thus it could be flaky.
@@ -86,8 +94,10 @@ jobs:
 
       - name: Build PyTorch Android
         run: |
+          set -eux
+
           echo "CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname "$(which conda)")/../"}" >> "${GITHUB_ENV}"
-          ${CONDA_RUN} ./scripts/build_pytorch_android.sh x86
+          ${CONDA_RUN} ./scripts/build_pytorch_android.sh "${SUPPORT_ABI}"
 
       - name: Run tests
         uses: reactivecircus/android-emulator-runner@v2

.github/workflows/build-android-binaries.yml

@@ -0,0 +1,48 @@
+name: Build Android binaries
+
+on:
+  push:
+    branches:
+      - nightly
+    tags:
+      # NOTE: Binary build pipelines should only get triggered on release candidate builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
+    paths:
+      - .github/workflows/build-android-binaries.yml
+      - .github/workflows/_run_android_tests.yml
+      - android/**
+  pull_request:
+    paths:
+      - .github/workflows/build-android-binaries.yml
+      - .github/workflows/_run_android_tests.yml
+      - android/**
+  # NB: We can use this workflow dispatch to test and build the binaries manually
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  cancel-in-progress: true
+
+jobs:
+  android-build-test:
+    name: android-build-test
+    uses: ./.github/workflows/_run_android_tests.yml
+    with:
+      test-matrix: |
+        { include: [
+          { config: 'default',
+            shard: 1,
+            num_shards: 1,
+            runner: 'ubuntu-20.04-16x',
+            use_lite_interpreter: 1,
+            support_abi: 'armeabi-v7a,arm64-v8a,x86,x86_64',
+          },
+          { config: 'default',
+            shard: 1,
+            num_shards: 1,
+            runner: 'ubuntu-20.04-16x',
+            use_lite_interpreter: 0,
+            support_abi: 'armeabi-v7a,arm64-v8a,x86,x86_64',
+          },
+        ]}

.github/workflows/build-ios-binaries.yml

@@ -0,0 +1,70 @@
+name: Build iOS binaries
+
+on:
+  push:
+    branches:
+      - nightly
+    tags:
+      # NOTE: Binary build pipelines should only get triggered on release candidate builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
+    paths:
+      - .github/workflows/build-ios-binaries.yml
+      - .github/workflows/_ios-build-test.yml
+  pull_request:
+    paths:
+      - .github/workflows/build-ios-binaries.yml
+      - .github/workflows/_ios-build-test.yml
+  # NB: We can use this workflow dispatch to test and build iOS binaries manually
+  workflow_dispatch:
+    inputs:
+      use_lite_interpreter:
+        description: "Use PyTorch lite interpreter?"
+        type: string
+        default: 1
+      use_coreml:
+        description: "Use Apple Core ML?"
+        type: string
+        default: 1
+      use_custom_op_list:
+        description: "Specify the custom ops list to include in the binaries"
+        type: string
+        default: ""
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  cancel-in-progress: true
+
+jobs:
+  # TODO: Figure out how to migrate this job to M1 runner
+  ios-build-test:
+    name: ios-build-test
+    uses: ./.github/workflows/_ios-build-test.yml
+    with:
+      build-environment: ios-build-test
+      sync-tag: ios-build-test
+      test-matrix: |
+        { include: [
+          { config: "default",
+            shard: 1,
+            num_shards: 1,
+            runner: "macos-12",
+            ios_platform: "SIMULATOR",
+            ios_arch: "x86_64",
+            use_lite_interpreter: ${{ inputs.use_lite_interpreter || 1 }},
+            use_metal: 0,
+            use_coreml: ${{ inputs.use_coreml || 1 }},
+            use_custom_op_list: ${{ inputs.use_custom_op_list || '' }}
+          },
+          { config: "default",
+            shard: 1,
+            num_shards: 1,
+            runner: "macos-12",
+            ios_platform: "OS",
+            ios_arch: "arm64",
+            use_lite_interpreter: ${{ inputs.use_lite_interpreter || 1 }},
+            use_metal: 1,
+            use_coreml: ${{ inputs.use_coreml || 1 }},
+            use_custom_op_list: ${{ inputs.use_custom_op_list || '' }}
+          }
+        ]}

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

@@ -4,6 +4,10 @@ on:
   push:
     branches:
       - release/2.1
+    tags:
+      # NOTE: Binary build pipelines should only get triggered on release candidate builds
+      # Release candidate tags look like: v1.11.0-rc1
+      - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
     paths:
       - .github/workflows/build-triton-wheel.yml
       - .github/scripts/build_triton_wheel.py
@@ -61,6 +65,8 @@ jobs:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
       - name: Build Triton wheel
+        env:
+          IS_RELEASE_TAG: ${{ startsWith(github.event.ref, 'refs/tags/v') }}
         run: |
           set -x
           mkdir -p "${RUNNER_TEMP}/artifacts/"
@@ -98,64 +104,75 @@ jobs:
             BUILD_ROCM="--build-rocm"
           fi
 
+          RELEASE=""
+          if [[ "${IS_RELEASE_TAG}" == true ]]; then
+            RELEASE="--release"
+          fi
+
           docker exec -t "${container_name}" yum install -y zlib-devel zip
           docker exec -t "${container_name}" "${PYTHON_EXECUTABLE}"  -m pip install -U setuptools==67.4.0
-          docker exec -t "${container_name}" "${PYTHON_EXECUTABLE}" /pytorch/.github/scripts/build_triton_wheel.py $BUILD_ROCM
+          docker exec -t "${container_name}" "${PYTHON_EXECUTABLE}" /pytorch/.github/scripts/build_triton_wheel.py $BUILD_ROCM $RELEASE
           docker exec -t "${container_name}" chown -R 1000.1000 /artifacts
 
       - uses: actions/upload-artifact@v3
         with:
-          name: "pytorch-triton-wheel-${{ matrix.py_vers }}"
+          # NB: Use the same name here and all wheels can be downloaded by referring to the same artifact
+          name: pytorch-triton-wheel
           if-no-files-found: error
-          path:
-            ${{ runner.temp }}/artifacts/*
+          path: ${{ runner.temp }}/artifacts/*
 
       - name: Teardown Linux
         uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.1
         if: always()
+
   upload-wheel:
-    runs-on: linux.20_04.4x
+    runs-on: ubuntu-22.04
     needs: build-wheel
     container:
       image: continuumio/miniconda3:4.12.0
-    env:
-      GITHUB_TOKEN: ${{ secrets.github-token }}
+    environment: ${{ (github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || startsWith(github.event.ref, 'refs/tags/v'))) && 'conda-aws-upload' || '' }}
     steps:
-      - name: Download Build Artifacts (3.8)
+      - uses: actions/checkout@v3
+
+      - name: Download Build Artifacts
         uses: actions/download-artifact@v3
         with:
-          name: "pytorch-triton-wheel-3.8"
-          path: "${{ runner.temp }}/artifacts/"
-      - name: Download Build Artifacts (3.9)
-        uses: actions/download-artifact@v3
-        with:
-          name: "pytorch-triton-wheel-3.9"
-          path: "${{ runner.temp }}/artifacts/"
-      - name: Download Build Artifacts (3.10)
-        uses: actions/download-artifact@v3
-        with:
-          name: "pytorch-triton-wheel-3.10"
-          path: "${{ runner.temp }}/artifacts/"
-      - name: Download Build Artifacts (3.11)
-        uses: actions/download-artifact@v3
-        with:
-          name: "pytorch-triton-wheel-3.11"
-          path: "${{ runner.temp }}/artifacts/"
-      - name: Upload binaries
-        if: ${{ github.event_name == 'push' && github.event.ref == 'refs/heads/main' }}
-        env:
-          PKG_DIR: "${{ runner.temp }}/artifacts"
-          # When running these on pull_request events these should be blank
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_S3_UPDATE_ACCESS_KEY_ID }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_S3_UPDATE_SECRET_ACCESS_KEY }}
-          UPLOAD_BUCKET: "s3://pytorch"
+          name: pytorch-triton-wheel
+          path: ${{ runner.temp }}/artifacts/
+
+      - name: Set DRY_RUN (only for tagged pushes)
+        if: ${{ github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || (startsWith(github.event.ref, 'refs/tags/v'))) }}
+        shell: bash
         run: |
-            set -ex
-            pip install -q awscli
-            s3_dir="${UPLOAD_BUCKET}/whl/nightly/"
-            for pkg in "${PKG_DIR}/"*.whl; do
-              aws s3 cp --no-progress --acl public-read "${pkg}" "${s3_dir}"
-             done
+          echo "DRY_RUN=disabled" >> "$GITHUB_ENV"
+
+      - name: Set UPLOAD_CHANNEL (only for tagged pushes)
+        if: ${{ github.event_name == 'push' && startsWith(github.event.ref, 'refs/tags/v') }}
+        shell: bash
+        run: |
+          set -ex
+
+          # reference ends with an RC suffix
+          if [[ "${GITHUB_REF_NAME}" = *-rc[0-9]* ]]; then
+            echo "UPLOAD_CHANNEL=test" >> "$GITHUB_ENV"
+          fi
+
+      # NB: This step is gated by DRY_RUN, which is enabled everywhere except nightly and release branches
+      - name: Upload binaries
+        env:
+          PACKAGE_TYPE: wheel
+          # The UPLOAD_SUBFOLDER needs to be empty here so that triton wheels are uploaded
+          # to nightly or test
+          UPLOAD_SUBFOLDER: ""
+          PKG_DIR: ${{ runner.temp }}/artifacts
+          # When running these on pull_request events these should be blank
+          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_PYTORCH_UPLOADER_ACCESS_KEY_ID }}
+          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_PYTORCH_UPLOADER_SECRET_ACCESS_KEY }}
+        shell: bash
+        run: |
+          set -ex
+          bash .circleci/scripts/binary_upload.sh
+
   build-conda:
     name: "Build Triton Conda"
     runs-on: [self-hosted, linux.2xlarge]
@@ -164,11 +181,9 @@ jobs:
       matrix:
         py_vers: [ "3.8", "3.9", "3.10", "3.11" ]
     timeout-minutes: 40
-    environment: ${{ (github.event_name == 'push' && github.event.ref == 'refs/heads/main') && 'conda-aws-upload' || '' }}
     env:
       DOCKER_IMAGE: pytorch/conda-builder:cpu
       PY_VERS: ${{ matrix.py_vers }}
-      ANACONDA_API_TOKEN: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
     steps:
       - name: Setup SSH (Click me for login details)
         uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.1
@@ -189,6 +204,8 @@ jobs:
           docker-image: ${{ env.DOCKER_IMAGE }}
 
       - name: Build Triton conda package
+        env:
+          IS_RELEASE_TAG: ${{ startsWith(github.event.ref, 'refs/tags/v') }}
         run: |
           set -x
           mkdir -p "${RUNNER_TEMP}/artifacts/"
@@ -198,31 +215,76 @@ jobs:
             -v "${GITHUB_WORKSPACE}:/pytorch" \
             -v "${RUNNER_TEMP}/artifacts:/artifacts" \
             -w /artifacts/ \
-            -e ANACONDA_API_TOKEN \
             "${DOCKER_IMAGE}" \
           )
 
+          RELEASE=""
+          if [[ "${IS_RELEASE_TAG}" == true ]]; then
+            RELEASE="--release"
+          fi
+
           docker exec -t "${container_name}" yum install -y llvm11 llvm11-devel llvm11-static llvm11-libs zlib-devel
-          docker exec -t "${container_name}" python /pytorch/.github/scripts/build_triton_wheel.py --build-conda --py-version="${PY_VERS}"
-
-      - name: Upload artifacts to Anaconda
-        if: ${{ github.event_name == 'push' && github.event.ref == 'refs/heads/main' }}
-        run: |
-          container_name=$(docker container ps --format '{{.ID}}')
-          docker exec -t "${container_name}" sh -c "anaconda upload /artifacts/torch*.tar.bz2 -u pytorch-nightly --label main --no-progress --force"
-
-      - name: Chown artifacts
-        run: |
-          container_name=$(docker container ps --format '{{.ID}}')
+          docker exec -t "${container_name}" python /pytorch/.github/scripts/build_triton_wheel.py --build-conda --py-version="${PY_VERS}" $RELEASE
           docker exec -t "${container_name}" chown -R 1000.1000 /artifacts
 
       - uses: actions/upload-artifact@v3
         with:
-          name: "pytorch-triton-conda-${{ matrix.py_vers }}"
+          # NB: Use the same name here and all wheels can be downloaded by referring to the same artifact
+          name: pytorch-triton-conda
           if-no-files-found: error
-          path:
-            ${{ runner.temp }}/artifacts/*
+          path: ${{ runner.temp }}/artifacts/*
 
       - name: Teardown Linux
         uses: pytorch/test-infra/.github/actions/teardown-linux@release/2.1
         if: always()
+
+  upload-conda:
+    runs-on: ubuntu-22.04
+    needs: build-conda
+    container:
+      image: continuumio/miniconda3:4.12.0
+    environment: ${{ (github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || startsWith(github.event.ref, 'refs/tags/v'))) && 'conda-aws-upload' || '' }}
+    steps:
+      - uses: actions/checkout@v3
+
+      - name: Download Build Artifacts
+        uses: actions/download-artifact@v3
+        with:
+          name: pytorch-triton-conda
+          path: ${{ runner.temp }}/artifacts/
+
+      - name: Set DRY_RUN (only for tagged pushes)
+        if: ${{ github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || (startsWith(github.event.ref, 'refs/tags/v'))) }}
+        shell: bash
+        run: |
+          echo "DRY_RUN=disabled" >> "$GITHUB_ENV"
+
+      - name: Set UPLOAD_CHANNEL (only for tagged pushes)
+        if: ${{ github.event_name == 'push' && startsWith(github.event.ref, 'refs/tags/v') }}
+        shell: bash
+        run: |
+          set -ex
+
+          # reference ends with an RC suffix
+          if [[ "${GITHUB_REF_NAME}" = *-rc[0-9]* ]]; then
+            echo "UPLOAD_CHANNEL=test" >> "$GITHUB_ENV"
+          fi
+
+      # NB: This step is gated by DRY_RUN, which is enabled everywhere except nightly and release branches
+      - name: Upload binaries to Anaconda
+        env:
+          PACKAGE_TYPE: conda
+          PKG_DIR: ${{ runner.temp }}/artifacts
+          # When running these on pull_request events these should be blank
+          CONDA_PYTORCHBOT_TOKEN: ${{ secrets.CONDA_PYTORCHBOT_TOKEN }}
+          CONDA_PYTORCHBOT_TOKEN_TEST: ${{ secrets.CONDA_PYTORCHBOT_TOKEN_TEST }}
+        shell: bash
+        run: |
+          set -ex
+
+          if [[ "${UPLOAD_CHANNEL}" = "nightly" ]]; then
+            export ANACONDA_API_TOKEN="${CONDA_PYTORCHBOT_TOKEN}"
+          else
+            export ANACONDA_API_TOKEN="${CONDA_PYTORCHBOT_TOKEN_TEST}"
+          fi
+          bash .circleci/scripts/binary_upload.sh

.github/workflows/create_release.yml

@@ -2,10 +2,9 @@ name: Create Release
 
 on:
   push:
-    tags: ['v*']
     branches:
       - main
-      - nightly
+      - release/*
   release:
     types: [published]
   pull_request:
@@ -31,6 +30,8 @@ jobs:
         run: |
           tag_or_branch="${PT_GITHUB_REF#refs/tags/}"
           tag_or_branch="${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"
       - name: Create source distribution
@@ -52,5 +53,5 @@ jobs:
           files: ${{env.PT_RELEASE_FILE}}
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name }}
   cancel-in-progress: true

.github/workflows/docker-release.yml

@@ -2,6 +2,15 @@ name: Build Official Docker Images
 
 on:
   workflow_dispatch:
+    inputs:
+      channel:
+        description: "Channel to use (nightly, test, release)"
+        required: false
+        type: choice
+        default: test
+        options:
+          - release
+          - test
   pull_request:
     paths:
       - Dockerfile
@@ -15,6 +24,7 @@ on:
       - v[0-9]+.[0-9]+.[0-9]+-rc[0-9]+
       - ciflow/nightly/*
 
+
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
   cancel-in-progress: true
@@ -26,25 +36,43 @@ env:
   DOCKER_REGISTRY: ghcr.io
   NO_BUILD_SUFFIX: true
   USE_BUILDX: 1
-  WITH_PUSH: ${{ github.event_name == 'push' && (github.event.ref == 'refs/heads/nightly' || (startsWith(github.event.ref, 'refs/tags/') && !startsWith(github.event.ref, 'refs/tags/ciflow/'))) }}
+  WITH_PUSH: ${{ inputs.channel == 'release' }}
 
 jobs:
+  generate-matrix:
+    if: github.repository_owner == 'pytorch'
+    runs-on: [self-hosted, linux.large]
+    outputs:
+      matrix: ${{ steps.generate-matrix.outputs.matrix }}
+    steps:
+      - name: Checkout PyTorch
+        uses: pytorch/pytorch/.github/actions/checkout-pytorch@main
+        with:
+          fetch-depth: 1
+          submodules: true
+      - name: Get docker release matrix
+        id: generate-matrix
+        run: |
+          MATRIX_BLOB="$(python3 .github/scripts/generate_docker_release_matrix.py)"
+          echo "${MATRIX_BLOB}"
+          echo "matrix=${MATRIX_BLOB}" >> "${GITHUB_OUTPUT}"
+
   build:
     if: ${{ github.repository == 'pytorch/pytorch' }}
     runs-on: [self-hosted, linux.2xlarge]
     environment: ${{ (github.ref == 'refs/heads/main' || startsWith(github.event.ref, 'refs/tags/v')) && 'docker-build' || '' }}
     timeout-minutes: 240
+    needs: generate-matrix
     strategy:
-      matrix:
-        include:
-          # nvidia specific images don't exist for arm64 so only build the runtime image
-          - image_type: runtime
-            platform: linux/arm64,linux/amd64
-          - image_type: devel
-            platform: linux/amd64
+      matrix: ${{ fromJson(needs.generate-matrix.outputs.matrix) }}
+      fail-fast: false
     env:
       BUILD_IMAGE_TYPE: ${{ matrix.image_type }}
       BUILD_PLATFORMS: ${{ matrix.platform }}
+      CHANNEL: ${{ inputs.channel }}
+      CUDA_VERSION: ${{ matrix.cuda_full_version }}
+      CUDA_VERSION_SHORT: ${{ matrix.cuda }}
+      CUDNN_VERSION: ${{ matrix.cudnn_version }}
     steps:
       - name: Setup SSH (Click me for login details)
         uses: pytorch/test-infra/.github/actions/setup-ssh@release/2.1
@@ -60,7 +88,7 @@ jobs:
       - name: Setup Linux
         uses: ./.github/actions/setup-linux
       - name: Login to GitHub Container Registry
-        if: ${{ env.WITH_PUSH == 'true' }}
+        if: ${{ inputs.channel == 'release' }}
         uses: docker/login-action@v2
         with:
           registry: ghcr.io
@@ -80,14 +108,16 @@ jobs:
           set -eou pipefail
           # To get QEMU binaries in our PATH
           echo "${RUNNER_TEMP}/bin" >> "${GITHUB_PATH}"
-          # Generate PyTorch version to use
-          echo "PYTORCH_VERSION=$(python3 .github/scripts/generate_pytorch_version.py)" >> "${GITHUB_ENV}"
-      - name: Setup nightly specific variables
-        if: ${{ github.event.ref == 'refs/heads/nightly' || startsWith(github.event.ref, 'refs/tags/ciflow/nightly/') }}
+          # Generate PyTorch version to use without suffix
+          echo "PYTORCH_VERSION=2.1.2" >> "${GITHUB_ENV}"
+
+      - name: Setup release specific variables
         run: |
           {
-            echo "DOCKER_IMAGE=pytorch-nightly";
-            echo "INSTALL_CHANNEL=pytorch-nightly";
+            echo "INSTALL_CHANNEL=pytorch-test";
+            if [[ ${CHANNEL} == "release" ]]; then
+              echo "INSTALL_CHANNEL=pytorch";
+            fi
             echo "TRITON_VERSION=$(cut -f 1 .ci/docker/triton_version.txt)+$(cut -c -10 .ci/docker/ci_commit_pins/triton.txt)";
           } >> "${GITHUB_ENV}"
       - name: Run docker build / push
@@ -97,10 +127,11 @@ jobs:
       - name: Push nightly tags
         if: ${{ github.event.ref == 'refs/heads/nightly' && matrix.image_type == 'runtime' }}
         run: |
-          PYTORCH_DOCKER_TAG="${PYTORCH_VERSION}-runtime"
-          CUDA_VERSION=$(python3 -c "import re;print(re.search('CUDA_VERSION\s+=\s+([0-9\.]+)',open('docker.Makefile').read())[1],end='')")
+          PYTORCH_DOCKER_TAG="${PYTORCH_VERSION}-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-runtime"
+
           PYTORCH_NIGHTLY_COMMIT=$(docker run ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_DOCKER_TAG}" \
                                           python -c 'import torch; print(torch.version.git_version[:7],end="")')
+
           docker tag ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_DOCKER_TAG}" \
                  ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}-cu${CUDA_VERSION}"
           docker push ghcr.io/pytorch/pytorch-nightly:"${PYTORCH_NIGHTLY_COMMIT}-cu${CUDA_VERSION}"

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

@@ -47,12 +47,13 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.8"
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_8-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_8-cpu-aarch64-test:  # Testing
@@ -67,11 +68,11 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -86,7 +87,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-aarch64
     secrets:
@@ -107,12 +108,13 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.9"
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_9-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_9-cpu-aarch64-test:  # Testing
@@ -127,11 +129,11 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -146,7 +148,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-aarch64
     secrets:
@@ -167,12 +169,13 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.10"
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_10-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_10-cpu-aarch64-test:  # Testing
@@ -187,11 +190,11 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -206,7 +209,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-aarch64
     secrets:
@@ -227,12 +230,13 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.11"
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
       build_name: manywheel-py3_11-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_11-cpu-aarch64-test:  # Testing
@@ -247,11 +251,11 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
       build_environment: linux-aarch64-binary-manywheel
-      runs_on: linux.t4g.2xlarge
+      runs_on: linux.arm64.2xlarge
       ALPINE_IMAGE: "arm64v8/alpine"
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
@@ -266,7 +270,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu-aarch64
-      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64
+      DOCKER_IMAGE: pytorch/manylinuxaarch64-builder:cpu-aarch64-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-aarch64
     secrets:

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

@@ -47,7 +47,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       build_environment: linux-binary-conda
@@ -65,7 +65,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       build_environment: linux-binary-conda
@@ -83,7 +83,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
     secrets:
@@ -105,7 +105,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda11_8
       build_environment: linux-binary-conda
@@ -124,7 +124,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda11_8
       build_environment: linux-binary-conda
@@ -143,7 +143,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda11_8
     secrets:
@@ -165,7 +165,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_1
       build_environment: linux-binary-conda
@@ -184,7 +184,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_1
       build_environment: linux-binary-conda
@@ -203,7 +203,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cuda12_1
     secrets:
@@ -224,7 +224,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       build_environment: linux-binary-conda
@@ -242,7 +242,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       build_environment: linux-binary-conda
@@ -260,7 +260,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
     secrets:
@@ -282,7 +282,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
       build_environment: linux-binary-conda
@@ -301,7 +301,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
       build_environment: linux-binary-conda
@@ -320,7 +320,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda11_8
     secrets:
@@ -342,7 +342,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
       build_environment: linux-binary-conda
@@ -361,7 +361,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
       build_environment: linux-binary-conda
@@ -380,7 +380,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cuda12_1
     secrets:
@@ -401,7 +401,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       build_environment: linux-binary-conda
@@ -419,7 +419,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       build_environment: linux-binary-conda
@@ -437,7 +437,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
     secrets:
@@ -459,7 +459,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
       build_environment: linux-binary-conda
@@ -478,7 +478,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
       build_environment: linux-binary-conda
@@ -497,7 +497,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda11_8
     secrets:
@@ -519,7 +519,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
       build_environment: linux-binary-conda
@@ -538,7 +538,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
       build_environment: linux-binary-conda
@@ -557,7 +557,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cuda12_1
     secrets:
@@ -578,7 +578,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       build_environment: linux-binary-conda
@@ -596,7 +596,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       build_environment: linux-binary-conda
@@ -614,7 +614,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
     secrets:
@@ -636,7 +636,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
       build_environment: linux-binary-conda
@@ -655,7 +655,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
       build_environment: linux-binary-conda
@@ -674,7 +674,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/conda-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda11_8
     secrets:
@@ -696,7 +696,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
       build_environment: linux-binary-conda
@@ -715,7 +715,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
       build_environment: linux-binary-conda
@@ -734,7 +734,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/conda-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cuda12_1
     secrets:

.github/workflows/generated-linux-binary-libtorch-cxx11-abi-main.yml

@@ -42,7 +42,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -61,7 +61,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi

.github/workflows/generated-linux-binary-libtorch-cxx11-abi-nightly.yml

@@ -47,7 +47,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -66,7 +66,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -85,7 +85,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -107,7 +107,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-without-deps-cxx11-abi
@@ -126,7 +126,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-without-deps-cxx11-abi
@@ -145,7 +145,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-without-deps-cxx11-abi
@@ -167,7 +167,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-with-deps-cxx11-abi
@@ -186,7 +186,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-with-deps-cxx11-abi
@@ -205,7 +205,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-with-deps-cxx11-abi
@@ -227,7 +227,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-without-deps-cxx11-abi
@@ -246,7 +246,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-without-deps-cxx11-abi
@@ -265,7 +265,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-without-deps-cxx11-abi
@@ -288,7 +288,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-with-deps-cxx11-abi
@@ -308,7 +308,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-with-deps-cxx11-abi
@@ -328,7 +328,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-with-deps-cxx11-abi
@@ -351,7 +351,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-without-deps-cxx11-abi
@@ -371,7 +371,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-without-deps-cxx11-abi
@@ -391,7 +391,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-shared-without-deps-cxx11-abi
@@ -414,7 +414,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-static-with-deps-cxx11-abi
@@ -434,7 +434,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-static-with-deps-cxx11-abi
@@ -454,7 +454,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-static-with-deps-cxx11-abi
@@ -477,7 +477,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-static-without-deps-cxx11-abi
@@ -497,7 +497,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-static-without-deps-cxx11-abi
@@ -517,7 +517,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda11_8-static-without-deps-cxx11-abi
@@ -540,7 +540,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-with-deps-cxx11-abi
@@ -560,7 +560,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-with-deps-cxx11-abi
@@ -580,7 +580,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-with-deps-cxx11-abi
@@ -603,7 +603,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-without-deps-cxx11-abi
@@ -623,7 +623,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-without-deps-cxx11-abi
@@ -643,7 +643,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-shared-without-deps-cxx11-abi
@@ -666,7 +666,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-static-with-deps-cxx11-abi
@@ -686,7 +686,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-static-with-deps-cxx11-abi
@@ -706,7 +706,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-static-with-deps-cxx11-abi
@@ -729,7 +729,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-static-without-deps-cxx11-abi
@@ -749,7 +749,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-static-without-deps-cxx11-abi
@@ -769,7 +769,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cuda12_1-static-without-deps-cxx11-abi
@@ -792,7 +792,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_5-shared-with-deps-cxx11-abi
@@ -814,7 +814,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -855,7 +855,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm5.5
+          docker-image: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -872,7 +872,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_5-shared-with-deps-cxx11-abi
@@ -895,7 +895,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_5-static-with-deps-cxx11-abi
@@ -917,7 +917,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -958,7 +958,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm5.5
+          docker-image: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -975,7 +975,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_5-static-with-deps-cxx11-abi
@@ -998,7 +998,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_6-shared-with-deps-cxx11-abi
@@ -1020,7 +1020,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -1061,7 +1061,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm5.6
+          docker-image: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1078,7 +1078,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_6-shared-with-deps-cxx11-abi
@@ -1101,7 +1101,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_6-static-with-deps-cxx11-abi
@@ -1123,7 +1123,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
     steps:
@@ -1164,7 +1164,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/libtorch-cxx11-builder:rocm5.6
+          docker-image: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1181,7 +1181,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-rocm5_6-static-with-deps-cxx11-abi

.github/workflows/generated-linux-binary-libtorch-pre-cxx11-main.yml

@@ -42,7 +42,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -61,7 +61,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11

.github/workflows/generated-linux-binary-libtorch-pre-cxx11-nightly.yml

@@ -47,7 +47,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -66,7 +66,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -85,7 +85,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-with-deps-pre-cxx11
@@ -107,7 +107,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-without-deps-pre-cxx11
@@ -126,7 +126,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-without-deps-pre-cxx11
@@ -145,7 +145,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-shared-without-deps-pre-cxx11
@@ -167,7 +167,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-static-with-deps-pre-cxx11
@@ -186,7 +186,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-static-with-deps-pre-cxx11
@@ -205,7 +205,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-static-with-deps-pre-cxx11
@@ -227,7 +227,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-static-without-deps-pre-cxx11
@@ -246,7 +246,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-static-without-deps-pre-cxx11
@@ -265,7 +265,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cpu-static-without-deps-pre-cxx11
@@ -288,7 +288,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-with-deps-pre-cxx11
@@ -308,7 +308,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-with-deps-pre-cxx11
@@ -328,7 +328,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-with-deps-pre-cxx11
@@ -351,7 +351,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-without-deps-pre-cxx11
@@ -371,7 +371,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-without-deps-pre-cxx11
@@ -391,7 +391,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-shared-without-deps-pre-cxx11
@@ -414,7 +414,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-static-with-deps-pre-cxx11
@@ -434,7 +434,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-static-with-deps-pre-cxx11
@@ -454,7 +454,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-static-with-deps-pre-cxx11
@@ -477,7 +477,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-static-without-deps-pre-cxx11
@@ -497,7 +497,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-static-without-deps-pre-cxx11
@@ -517,7 +517,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda11_8-static-without-deps-pre-cxx11
@@ -540,7 +540,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-with-deps-pre-cxx11
@@ -560,7 +560,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-with-deps-pre-cxx11
@@ -580,7 +580,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-with-deps-pre-cxx11
@@ -603,7 +603,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-without-deps-pre-cxx11
@@ -623,7 +623,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-without-deps-pre-cxx11
@@ -643,7 +643,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-shared-without-deps-pre-cxx11
@@ -666,7 +666,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-static-with-deps-pre-cxx11
@@ -686,7 +686,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-static-with-deps-pre-cxx11
@@ -706,7 +706,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-static-with-deps-pre-cxx11
@@ -729,7 +729,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-static-without-deps-pre-cxx11
@@ -749,7 +749,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-static-without-deps-pre-cxx11
@@ -769,7 +769,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-cuda12_1-static-without-deps-pre-cxx11
@@ -792,7 +792,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_5-shared-with-deps-pre-cxx11
@@ -814,7 +814,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -855,7 +855,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.5
+          docker-image: pytorch/manylinux-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -872,7 +872,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_5-shared-with-deps-pre-cxx11
@@ -895,7 +895,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_5-static-with-deps-pre-cxx11
@@ -917,7 +917,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -958,7 +958,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.5
+          docker-image: pytorch/manylinux-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -975,7 +975,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_5-static-with-deps-pre-cxx11
@@ -998,7 +998,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_6-shared-with-deps-pre-cxx11
@@ -1020,7 +1020,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -1061,7 +1061,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.6
+          docker-image: pytorch/manylinux-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1078,7 +1078,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_6-shared-with-deps-pre-cxx11
@@ -1101,7 +1101,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_6-static-with-deps-pre-cxx11
@@ -1123,7 +1123,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
     steps:
@@ -1164,7 +1164,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.6
+          docker-image: pytorch/manylinux-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1181,7 +1181,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: pre-cxx11
       build_name: libtorch-rocm5_6-static-with-deps-pre-cxx11

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

@@ -43,7 +43,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -62,7 +62,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -82,11 +82,11 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_8-cuda12_1-with-pypi-cudnn-test:  # Testing
@@ -102,7 +102,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel

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

@@ -47,7 +47,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu
       build_environment: linux-binary-manywheel
@@ -65,7 +65,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu
       build_environment: linux-binary-manywheel
@@ -83,7 +83,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu
     secrets:
@@ -104,7 +104,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-cxx11-abi
@@ -123,7 +123,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-cxx11-abi
@@ -142,7 +142,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cpu-cxx11-abi
@@ -165,7 +165,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -184,7 +184,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
       build_environment: linux-binary-manywheel
@@ -203,7 +203,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda11_8
     secrets:
@@ -225,11 +225,11 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_8-cuda12_1-with-pypi-cudnn-test:  # Testing
@@ -245,7 +245,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
@@ -264,7 +264,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1-with-pypi-cudnn
     secrets:
@@ -286,7 +286,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel
@@ -305,7 +305,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
       build_environment: linux-binary-manywheel
@@ -324,7 +324,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-cuda12_1
     secrets:
@@ -346,7 +346,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm5_5
       build_environment: linux-binary-manywheel
@@ -367,7 +367,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.8"
     steps:
       - name: Setup ROCm
@@ -407,7 +407,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.5
+          docker-image: pytorch/manylinux-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -424,7 +424,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm5_5
     secrets:
@@ -446,7 +446,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm5_6
       build_environment: linux-binary-manywheel
@@ -467,7 +467,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.8"
     steps:
       - name: Setup ROCm
@@ -507,7 +507,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.6
+          docker-image: pytorch/manylinux-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -524,7 +524,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.8"
       build_name: manywheel-py3_8-rocm5_6
     secrets:
@@ -545,7 +545,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu
       build_environment: linux-binary-manywheel
@@ -563,7 +563,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu
       build_environment: linux-binary-manywheel
@@ -581,7 +581,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu
     secrets:
@@ -602,7 +602,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-cxx11-abi
@@ -621,7 +621,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-cxx11-abi
@@ -640,7 +640,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cpu-cxx11-abi
@@ -663,7 +663,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda11_8
       build_environment: linux-binary-manywheel
@@ -682,7 +682,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda11_8
       build_environment: linux-binary-manywheel
@@ -701,7 +701,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda11_8
     secrets:
@@ -723,11 +723,11 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_9-cuda12_1-with-pypi-cudnn-test:  # Testing
@@ -743,7 +743,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
@@ -762,7 +762,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1-with-pypi-cudnn
     secrets:
@@ -784,7 +784,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1
       build_environment: linux-binary-manywheel
@@ -803,7 +803,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1
       build_environment: linux-binary-manywheel
@@ -822,7 +822,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-cuda12_1
     secrets:
@@ -844,7 +844,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm5_5
       build_environment: linux-binary-manywheel
@@ -865,7 +865,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.9"
     steps:
       - name: Setup ROCm
@@ -905,7 +905,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.5
+          docker-image: pytorch/manylinux-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -922,7 +922,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm5_5
     secrets:
@@ -944,7 +944,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm5_6
       build_environment: linux-binary-manywheel
@@ -965,7 +965,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.9"
     steps:
       - name: Setup ROCm
@@ -1005,7 +1005,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.6
+          docker-image: pytorch/manylinux-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1022,7 +1022,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.9"
       build_name: manywheel-py3_9-rocm5_6
     secrets:
@@ -1043,7 +1043,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu
       build_environment: linux-binary-manywheel
@@ -1061,7 +1061,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu
       build_environment: linux-binary-manywheel
@@ -1079,7 +1079,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu
     secrets:
@@ -1100,7 +1100,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-cxx11-abi
@@ -1119,7 +1119,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-cxx11-abi
@@ -1138,7 +1138,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cpu-cxx11-abi
@@ -1161,7 +1161,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1180,7 +1180,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1199,7 +1199,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda11_8
     secrets:
@@ -1221,11 +1221,11 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_10-cuda12_1-with-pypi-cudnn-test:  # Testing
@@ -1241,7 +1241,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
@@ -1260,7 +1260,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1-with-pypi-cudnn
     secrets:
@@ -1282,7 +1282,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1301,7 +1301,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1320,7 +1320,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-cuda12_1
     secrets:
@@ -1342,7 +1342,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm5_5
       build_environment: linux-binary-manywheel
@@ -1363,7 +1363,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.10"
     steps:
       - name: Setup ROCm
@@ -1403,7 +1403,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.5
+          docker-image: pytorch/manylinux-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1420,7 +1420,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm5_5
     secrets:
@@ -1442,7 +1442,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm5_6
       build_environment: linux-binary-manywheel
@@ -1463,7 +1463,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.10"
     steps:
       - name: Setup ROCm
@@ -1503,7 +1503,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.6
+          docker-image: pytorch/manylinux-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1520,7 +1520,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.10"
       build_name: manywheel-py3_10-rocm5_6
     secrets:
@@ -1541,7 +1541,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu
       build_environment: linux-binary-manywheel
@@ -1559,7 +1559,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu
       build_environment: linux-binary-manywheel
@@ -1577,7 +1577,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu
     secrets:
@@ -1598,7 +1598,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-cxx11-abi
@@ -1617,7 +1617,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-cxx11-abi
@@ -1636,7 +1636,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu-cxx11-abi
       GPU_ARCH_TYPE: cpu-cxx11-abi
-      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi
+      DOCKER_IMAGE: pytorch/manylinuxcxx11-abi-builder:cpu-cxx11-abi-2.1
       DESIRED_DEVTOOLSET: cxx11-abi
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cpu-cxx11-abi
@@ -1659,7 +1659,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1678,7 +1678,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda11_8
       build_environment: linux-binary-manywheel
@@ -1697,7 +1697,7 @@ jobs:
       DESIRED_CUDA: cu118
       GPU_ARCH_VERSION: 11.8
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda11.8-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda11_8
     secrets:
@@ -1719,11 +1719,11 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
-      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64'
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     secrets:
       github-token: ${{ secrets.GITHUB_TOKEN }}
   manywheel-py3_11-cuda12_1-with-pypi-cudnn-test:  # Testing
@@ -1739,7 +1739,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1-with-pypi-cudnn
       build_environment: linux-binary-manywheel
@@ -1758,7 +1758,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1-with-pypi-cudnn
     secrets:
@@ -1780,7 +1780,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1799,7 +1799,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1
       build_environment: linux-binary-manywheel
@@ -1818,7 +1818,7 @@ jobs:
       DESIRED_CUDA: cu121
       GPU_ARCH_VERSION: 12.1
       GPU_ARCH_TYPE: cuda
-      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1
+      DOCKER_IMAGE: pytorch/manylinux-builder:cuda12.1-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-cuda12_1
     secrets:
@@ -1840,7 +1840,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm5_5
       build_environment: linux-binary-manywheel
@@ -1861,7 +1861,7 @@ jobs:
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.11"
     steps:
       - name: Setup ROCm
@@ -1901,7 +1901,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.5
+          docker-image: pytorch/manylinux-builder:rocm5.5-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -1918,7 +1918,7 @@ jobs:
       DESIRED_CUDA: rocm5.5
       GPU_ARCH_VERSION: 5.5
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.5-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm5_5
     secrets:
@@ -1940,7 +1940,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm5_6
       build_environment: linux-binary-manywheel
@@ -1961,7 +1961,7 @@ jobs:
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
       SKIP_ALL_TESTS: 1
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.11"
     steps:
       - name: Setup ROCm
@@ -2001,7 +2001,7 @@ jobs:
       - name: Pull Docker image
         uses: pytorch/test-infra/.github/actions/pull-docker-image@release/2.1
         with:
-          docker-image: pytorch/manylinux-builder:rocm5.6
+          docker-image: pytorch/manylinux-builder:rocm5.6-2.1
       - name: Test Pytorch binary
         uses: ./pytorch/.github/actions/test-pytorch-binary
       - name: Teardown ROCm
@@ -2018,7 +2018,7 @@ jobs:
       DESIRED_CUDA: rocm5.6
       GPU_ARCH_VERSION: 5.6
       GPU_ARCH_TYPE: rocm
-      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6
+      DOCKER_IMAGE: pytorch/manylinux-builder:rocm5.6-2.1
       DESIRED_PYTHON: "3.11"
       build_name: manywheel-py3_11-rocm5_6
     secrets:

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

@@ -135,7 +135,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       use_s3: False
@@ -246,7 +246,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       use_s3: False
@@ -357,7 +357,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       use_s3: False
@@ -468,7 +468,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       use_s3: False

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

@@ -48,6 +48,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.8"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -135,7 +136,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: wheel-py3_8-cpu
       use_s3: False
@@ -159,6 +160,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.9"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -246,7 +248,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: wheel-py3_9-cpu
       use_s3: False
@@ -270,6 +272,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.10"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -357,7 +360,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: wheel-py3_10-cpu
       use_s3: False
@@ -381,6 +384,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.11"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -468,7 +472,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: wheel-py3_11-cpu
       use_s3: False

.github/workflows/generated-macos-binary-conda-nightly.yml

@@ -133,7 +133,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: conda-py3_8-cpu
       use_s3: False
@@ -244,7 +244,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: conda-py3_9-cpu
       use_s3: False
@@ -355,7 +355,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: conda-py3_10-cpu
       use_s3: False
@@ -466,7 +466,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/conda-builder:cpu
+      DOCKER_IMAGE: pytorch/conda-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: conda-py3_11-cpu
       use_s3: False

.github/workflows/generated-macos-binary-libtorch-cxx11-abi-nightly.yml

@@ -137,7 +137,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-with-deps-cxx11-abi
@@ -253,7 +253,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: shared-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-shared-without-deps-cxx11-abi
@@ -369,7 +369,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-with-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-with-deps-cxx11-abi
@@ -485,7 +485,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu
+      DOCKER_IMAGE: pytorch/libtorch-cxx11-builder:cpu-2.1
       LIBTORCH_VARIANT: static-without-deps
       DESIRED_DEVTOOLSET: cxx11-abi
       build_name: libtorch-cpu-static-without-deps-cxx11-abi

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

@@ -46,6 +46,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.8"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -133,7 +134,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.8"
       build_name: wheel-py3_8-cpu
       use_s3: False
@@ -157,6 +158,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.9"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -244,7 +246,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.9"
       build_name: wheel-py3_9-cpu
       use_s3: False
@@ -268,6 +270,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.10"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -355,7 +358,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.10"
       build_name: wheel-py3_10-cpu
       use_s3: False
@@ -379,6 +382,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.11"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
       # For sccache access (only on non-forked PRs)
       AWS_ACCESS_KEY_ID: ${{ secrets.MACOS_SCCACHE_S3_ACCESS_KEY_ID }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.MACOS_SCCACHE_S3_SECRET_ACCESS_KEY }}
@@ -466,7 +470,7 @@ jobs:
       #       favor of GPU_ARCH_VERSION
       DESIRED_CUDA: cpu
       GPU_ARCH_TYPE: cpu
-      DOCKER_IMAGE: pytorch/manylinux-builder:cpu
+      DOCKER_IMAGE: pytorch/manylinux-builder:cpu-2.1
       DESIRED_PYTHON: "3.11"
       build_name: wheel-py3_11-cpu
       use_s3: False

.github/workflows/generated-windows-binary-conda-nightly.yml

@@ -63,6 +63,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -173,6 +174,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -300,6 +302,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -411,6 +414,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -539,6 +543,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -650,6 +655,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -777,6 +783,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -887,6 +894,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1014,6 +1022,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1125,6 +1134,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1253,6 +1263,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1364,6 +1375,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1491,6 +1503,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1601,6 +1614,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1728,6 +1742,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1839,6 +1854,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1967,6 +1983,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2078,6 +2095,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2205,6 +2223,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2315,6 +2334,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2442,6 +2462,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2553,6 +2574,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2681,6 +2703,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2792,6 +2815,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

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

@@ -60,6 +60,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -174,6 +175,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

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

@@ -67,6 +67,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -181,6 +182,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -315,6 +317,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -429,6 +432,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -563,6 +567,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -677,6 +682,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -811,6 +817,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -925,6 +932,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1060,6 +1068,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1175,6 +1184,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1311,6 +1321,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1426,6 +1437,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1562,6 +1574,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1677,6 +1690,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1813,6 +1827,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1928,6 +1943,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2064,6 +2080,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2179,6 +2196,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2315,6 +2333,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2430,6 +2449,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2566,6 +2586,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2681,6 +2702,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2817,6 +2839,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2932,6 +2955,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

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

@@ -60,6 +60,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -174,6 +175,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

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

@@ -67,6 +67,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -181,6 +182,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -315,6 +317,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -429,6 +432,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -563,6 +567,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -677,6 +682,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -811,6 +817,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -925,6 +932,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1060,6 +1068,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1175,6 +1184,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1311,6 +1321,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1426,6 +1437,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1562,6 +1574,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1677,6 +1690,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1813,6 +1827,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1928,6 +1943,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2064,6 +2080,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2179,6 +2196,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2315,6 +2333,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2430,6 +2449,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2566,6 +2586,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2681,6 +2702,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2817,6 +2839,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2932,6 +2955,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

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

@@ -46,6 +46,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.8"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -63,6 +64,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -173,6 +175,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -283,6 +286,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.8"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -300,6 +304,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -411,6 +416,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -522,6 +528,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.8"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -539,6 +546,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -650,6 +658,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -760,6 +769,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.9"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -777,6 +787,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -887,6 +898,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -997,6 +1009,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.9"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1014,6 +1027,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1125,6 +1139,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1236,6 +1251,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.9"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1253,6 +1269,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1364,6 +1381,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1474,6 +1492,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.10"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1491,6 +1510,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1601,6 +1621,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1711,6 +1732,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.10"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1728,6 +1750,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1839,6 +1862,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -1950,6 +1974,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.10"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -1967,6 +1992,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2078,6 +2104,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2188,6 +2215,7 @@ jobs:
       GPU_ARCH_TYPE: cpu
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.11"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -2205,6 +2233,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2315,6 +2344,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2425,6 +2455,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.11"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -2442,6 +2473,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2553,6 +2585,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2664,6 +2697,7 @@ jobs:
       GPU_ARCH_TYPE: cuda
       SKIP_ALL_TESTS: 1
       DESIRED_PYTHON: "3.11"
+      PYTORCH_EXTRA_INSTALL_REQUIREMENTS: nvidia-cuda-nvrtc-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-runtime-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cuda-cupti-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cudnn-cu12==8.9.2.26; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cublas-cu12==12.1.3.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cufft-cu12==11.0.2.54; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-curand-cu12==10.3.2.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusolver-cu12==11.4.5.107; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-cusparse-cu12==12.1.0.106; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nccl-cu12==2.18.1; platform_system == 'Linux' and platform_machine == 'x86_64' | nvidia-nvtx-cu12==12.1.105; platform_system == 'Linux' and platform_machine == 'x86_64' | triton==2.1.0; platform_system == 'Linux' and platform_machine == 'x86_64'
     steps:
       - name: Display EC2 information
         shell: bash
@@ -2681,6 +2715,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560
@@ -2792,6 +2827,7 @@ jobs:
           echo "system info $(uname -a)"
       - name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
         uses: seemethere/add-github-ssh-key@v1
+        continue-on-error: true
         with:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
       # Needed for binary builds, see: https://github.com/pytorch/pytorch/issues/73339#issuecomment-1058981560

.github/workflows/lint.yml

@@ -21,6 +21,7 @@ jobs:
       docker-image: pytorch-linux-focal-linter
       fetch-depth: 0
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      test-infra-ref: 'release/2.1'
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
@@ -68,6 +69,7 @@ jobs:
       docker-image: pytorch-linux-focal-linter
       fetch-depth: 0
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      test-infra-ref: 'release/2.1'
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
@@ -122,6 +124,7 @@ jobs:
       docker-image: pytorch-linux-focal-linter
       fetch-depth: 0
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      test-infra-ref: 'release/2.1'
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
@@ -157,6 +160,7 @@ jobs:
       docker-image: pytorch-linux-focal-linter
       fetch-depth: 0
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      test-infra-ref: 'release/2.1'
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
@@ -195,6 +199,7 @@ jobs:
       docker-image: pytorch-linux-focal-linter
       fetch-depth: 0
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      test-infra-ref: 'release/2.1'
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")

.github/workflows/periodic.yml

@@ -112,30 +112,38 @@ jobs:
       cuda-version: "11.8"
       test-matrix: ${{ needs.win-vs2019-cuda11_8-py3-build.outputs.test-matrix }}
 
-  ios-12-5-1-x86-64-coreml:
-    name: ios-12-5-1-x86-64-coreml
+  # TODO: Figure out how to migrate this job to M1 runner
+  ios-build-test:
+    name: ios-build-test
     if: github.event_name != 'schedule' || github.event.schedule == '45 0,8,16 * * 1-5' || github.event.schedule == '45 4 * * 0,6'
     uses: ./.github/workflows/_ios-build-test.yml
     with:
-      build-environment: ios-12-5-1-x86-64-coreml
-      ios-platform: SIMULATOR
-      ios-arch: x86_64
+      build-environment: ios-build-test
+      sync-tag: ios-build-test
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 1, runner: "macos-12" },
-        ]}
-
-  ios-12-5-1-arm64-custom-ops:
-    name: ios-12-5-1-arm64-custom-ops
-    if: github.event_name != 'schedule' || github.event.schedule == '45 0,8,16 * * 1-5' || github.event.schedule == '45 4 * * 0,6'
-    uses: ./.github/workflows/_ios-build-test.yml
-    with:
-      build-environment: ios-12-5-1-arm64-custom-ops
-      ios-platform: OS
-      ios-arch: arm64
-      test-matrix: |
-        { include: [
-          { config: "default", shard: 1, num_shards: 1, runner: "macos-12" },
+          { config: "default",
+            shard: 1,
+            num_shards: 1,
+            runner: "macos-12",
+            ios_platform: "SIMULATOR",
+            ios_arch: "x86_64",
+            use_lite_interpreter: 1,
+            use_metal: 0,
+            use_coreml: 1,
+            use_custom_op_list: ""
+          },
+          { config: "default",
+            shard: 1,
+            num_shards: 1,
+            runner: "macos-12",
+            ios_platform: "OS",
+            ios_arch: "arm64",
+            use_lite_interpreter: 1,
+            use_metal: 1,
+            use_coreml: 1,
+            use_custom_op_list: "mobilenetv2.yaml"
+          }
         ]}
 
   buck-build-test:
@@ -180,7 +188,14 @@ jobs:
     with:
       test-matrix: |
         { include: [
-          { config: "default", shard: 1, num_shards: 1, runner: "ubuntu-20.04-16x" },
+          { config: "default",
+            shard: 1,
+            num_shards: 1,
+            runner: "ubuntu-20.04-16x"
+            use_lite_interpreter: 1,
+            # Just set x86 for testing here
+            support_abi: x86,
+          },
         ]}
 
   linux-vulkan-focal-py3_11-clang10-build:

.gitignore

@@ -364,3 +364,7 @@ venv/
 # Log files
 *.log
 sweep/
+
+# Android build artifacts
+android/pytorch_android/.cxx
+android/pytorch_android_torchvision/.cxx

Dockerfile

@@ -63,7 +63,7 @@ RUN --mount=type=cache,target=/opt/ccache \
 
 FROM conda as conda-installs
 ARG PYTHON_VERSION=3.8
-ARG CUDA_VERSION=11.7
+ARG CUDA_VERSION=12.1
 ARG CUDA_CHANNEL=nvidia
 ARG INSTALL_CHANNEL=pytorch-nightly
 # Automatically set by buildx
@@ -73,8 +73,8 @@ ARG TARGETPLATFORM
 
 # On arm64 we can only install wheel packages.
 RUN case ${TARGETPLATFORM} in \
-         "linux/arm64")  pip install --extra-index-url https://download.pytorch.org/whl/cpu/ torch torchvision torchaudio torchtext ;; \
-         *)              /opt/conda/bin/conda install -c "${INSTALL_CHANNEL}" -c "${CUDA_CHANNEL}" -y "python=${PYTHON_VERSION}" pytorch torchvision torchaudio torchtext "pytorch-cuda=$(echo $CUDA_VERSION | cut -d'.' -f 1-2)"  ;; \
+         "linux/arm64")  pip install --extra-index-url https://download.pytorch.org/whl/cpu/ torch torchvision torchaudio ;; \
+         *)              /opt/conda/bin/conda install -c "${INSTALL_CHANNEL}" -c "${CUDA_CHANNEL}" -y "python=${PYTHON_VERSION}" pytorch torchvision torchaudio "pytorch-cuda=$(echo $CUDA_VERSION | cut -d'.' -f 1-2)"  ;; \
     esac && \
     /opt/conda/bin/conda clean -ya
 RUN /opt/conda/bin/pip install torchelastic

android/gradle.properties

@@ -1,6 +1,6 @@
 ABI_FILTERS=armeabi-v7a,arm64-v8a,x86,x86_64
 
-VERSION_NAME=2.1.0-SNAPSHOT
+VERSION_NAME=2.1.0
 GROUP=org.pytorch
 MAVEN_GROUP=org.pytorch
 SONATYPE_STAGING_PROFILE=orgpytorch

android/pytorch_android/build.gradle

@@ -41,6 +41,7 @@ android {
                 println 'Build pytorch_jni'
                 exclude 'org/pytorch/LiteModuleLoader.java'
                 exclude 'org/pytorch/LiteNativePeer.java'
+                exclude 'org/pytorch/LitePyTorchAndroid.java'
               } else {
                 println 'Build pytorch_jni_lite'
               }

android/pytorch_android/host/build.gradle

@@ -17,6 +17,7 @@ sourceSets {
         java {
             srcDir '../src/main/java'
             exclude 'org/pytorch/PyTorchAndroid.java'
+            exclude 'org/pytorch/LitePyTorchAndroid.java'
             exclude 'org/pytorch/LiteModuleLoader.java'
             exclude 'org/pytorch/LiteNativePeer.java'
         }

android/pytorch_android/src/androidTest/java/org/pytorch/PytorchTestBase.java

@@ -10,6 +10,7 @@ import java.io.IOException;
 import java.util.HashMap;
 import java.util.Map;
 import org.junit.Test;
+import org.junit.Ignore;
 
 public abstract class PytorchTestBase {
   private static final String TEST_MODULE_ASSET_NAME = "android_api_module.ptl";
@@ -413,7 +414,10 @@ public abstract class PytorchTestBase {
   }
 
   @Test
+  @Ignore
   public void testSpectralOps() throws IOException {
+    // NB: This model fails without lite interpreter.  The error is as follows:
+    // RuntimeError: stft requires the return_complex parameter be given for real inputs
     runModel("spectral_ops");
   }
 

android/pytorch_android/src/main/cpp/pytorch_jni_common.cpp

@@ -10,12 +10,6 @@
 #include <fbjni/fbjni.h>
 
 #include "pytorch_jni_common.h"
-#if defined(__ANDROID__)
-#ifndef USE_PTHREADPOOL
-#define USE_PTHREADPOOL
-#endif /* USE_PTHREADPOOL */
-#include <caffe2/utils/threadpool/pthreadpool-cpp.h>
-#endif
 
 namespace pytorch_jni {
 
@@ -666,32 +660,4 @@ at::IValue JIValue::JIValueToAtIValue(
       typeCode);
 }
 
-#if defined(__ANDROID__)
-class PyTorchAndroidJni : public facebook::jni::JavaClass<PyTorchAndroidJni> {
- public:
-  constexpr static auto kJavaDescriptor = "Lorg/pytorch/PyTorchAndroid;";
-
-  static void registerNatives() {
-    javaClassStatic()->registerNatives({
-        makeNativeMethod(
-            "nativeSetNumThreads", PyTorchAndroidJni::setNumThreads),
-    });
-  }
-
-  static void setNumThreads(facebook::jni::alias_ref<jclass>, jint numThreads) {
-    caffe2::pthreadpool()->set_thread_count(numThreads);
-  }
-};
-#endif
-
-void common_registerNatives() {
-  static const int once = []() {
-#if defined(__ANDROID__)
-    pytorch_jni::PyTorchAndroidJni::registerNatives();
-#endif
-    return 0;
-  }();
-  ((void)once);
-}
-
 } // namespace pytorch_jni

android/pytorch_android/src/main/cpp/pytorch_jni_jit.cpp

@@ -17,6 +17,11 @@
 #include <android/asset_manager.h>
 #include <android/asset_manager_jni.h>
 #include <android/log.h>
+
+#ifndef USE_PTHREADPOOL
+#define USE_PTHREADPOOL
+#endif /* USE_PTHREADPOOL */
+#include <caffe2/utils/threadpool/pthreadpool-cpp.h>
 #endif
 
 namespace pytorch_jni {
@@ -235,6 +240,34 @@ class PytorchJni : public facebook::jni::HybridClass<PytorchJni> {
   }
 };
 
+#if defined(__ANDROID__)
+class PyTorchAndroidJni : public facebook::jni::JavaClass<PyTorchAndroidJni> {
+ public:
+  constexpr static auto kJavaDescriptor = "Lorg/pytorch/PyTorchAndroid;";
+
+  static void registerNatives() {
+    javaClassStatic()->registerNatives({
+        makeNativeMethod(
+            "nativeSetNumThreads", PyTorchAndroidJni::setNumThreads),
+    });
+  }
+
+  static void setNumThreads(facebook::jni::alias_ref<jclass>, jint numThreads) {
+    caffe2::pthreadpool()->set_thread_count(numThreads);
+  }
+};
+#endif
+
+void common_registerNatives() {
+  static const int once = []() {
+#if defined(__ANDROID__)
+    pytorch_jni::PyTorchAndroidJni::registerNatives();
+#endif
+    return 0;
+  }();
+  ((void)once);
+}
+
 } // namespace pytorch_jni
 
 JNIEXPORT jint JNICALL JNI_OnLoad(JavaVM* vm, void*) {

android/pytorch_android/src/main/cpp/pytorch_jni_lite.cpp

@@ -18,6 +18,11 @@
 #include <android/asset_manager.h>
 #include <android/asset_manager_jni.h>
 #include <android/log.h>
+
+#ifndef USE_PTHREADPOOL
+#define USE_PTHREADPOOL
+#endif /* USE_PTHREADPOOL */
+#include <caffe2/utils/threadpool/pthreadpool-cpp.h>
 #endif
 
 namespace pytorch_jni {
@@ -199,6 +204,34 @@ class PytorchJni : public facebook::jni::HybridClass<PytorchJni> {
   }
 };
 
+#if defined(__ANDROID__)
+class PyTorchAndroidJni : public facebook::jni::JavaClass<PyTorchAndroidJni> {
+ public:
+  constexpr static auto kJavaDescriptor = "Lorg/pytorch/LitePyTorchAndroid;";
+
+  static void registerNatives() {
+    javaClassStatic()->registerNatives({
+        makeNativeMethod(
+            "nativeSetNumThreads", PyTorchAndroidJni::setNumThreads),
+    });
+  }
+
+  static void setNumThreads(facebook::jni::alias_ref<jclass>, jint numThreads) {
+    caffe2::pthreadpool()->set_thread_count(numThreads);
+  }
+};
+#endif
+
+void common_registerNatives() {
+  static const int once = []() {
+#if defined(__ANDROID__)
+    pytorch_jni::PyTorchAndroidJni::registerNatives();
+#endif
+    return 0;
+  }();
+  ((void)once);
+}
+
 } // namespace pytorch_jni
 
 JNIEXPORT jint JNICALL JNI_OnLoad(JavaVM* vm, void*) {

android/pytorch_android/src/main/java/org/pytorch/LitePyTorchAndroid.java

@@ -0,0 +1,50 @@
+package org.pytorch;
+
+import android.content.res.AssetManager;
+import com.facebook.jni.annotations.DoNotStrip;
+import com.facebook.soloader.nativeloader.NativeLoader;
+import com.facebook.soloader.nativeloader.SystemDelegate;
+
+public final class LitePyTorchAndroid {
+    static {
+        if (!NativeLoader.isInitialized()) {
+            NativeLoader.init(new SystemDelegate());
+        }
+        NativeLoader.loadLibrary("pytorch_jni_lite");
+        PyTorchCodegenLoader.loadNativeLibs();
+    }
+
+    /**
+     * Attention: This is not recommended way of loading production modules, as prepackaged assets
+     * increase apk size etc. For production usage consider using loading from file on the disk {@link
+     * org.pytorch.Module#load(String)}.
+     *
+     * <p>This method is meant to use in tests and demos.
+     */
+    public static Module loadModuleFromAsset(
+            final AssetManager assetManager, final String assetName, final Device device) {
+        return new Module(new LiteNativePeer(assetName, assetManager, device));
+    }
+
+    public static Module loadModuleFromAsset(
+            final AssetManager assetManager, final String assetName) {
+        return new Module(new LiteNativePeer(assetName, assetManager, Device.CPU));
+    }
+
+    /**
+     * Globally sets the number of threads used on native side. Attention: Has global effect, all
+     * modules use one thread pool with specified number of threads.
+     *
+     * @param numThreads number of threads, must be positive number.
+     */
+    public static void setNumThreads(int numThreads) {
+        if (numThreads < 1) {
+            throw new IllegalArgumentException("Number of threads cannot be less than 1");
+        }
+
+        nativeSetNumThreads(numThreads);
+    }
+
+    @DoNotStrip
+    private static native void nativeSetNumThreads(int numThreads);
+}

android/pytorch_android/src/main/java/org/pytorch/PyTorchAndroid.java

@@ -10,7 +10,7 @@ public final class PyTorchAndroid {
     if (!NativeLoader.isInitialized()) {
       NativeLoader.init(new SystemDelegate());
     }
-    NativeLoader.loadLibrary("pytorch_jni_lite");
+    NativeLoader.loadLibrary("pytorch_jni");
     PyTorchCodegenLoader.loadNativeLibs();
   }
 

android/test_app/app/build.gradle

@@ -41,6 +41,11 @@ android {
         buildConfigField("long[]", "INPUT_TENSOR_SHAPE", "new long[]{1, 3, 224, 224}")
         buildConfigField("boolean", "NATIVE_BUILD", 'false')
         buildConfigField("boolean", "USE_VULKAN_DEVICE", 'false')
+        buildConfigField(
+                "int",
+                "BUILD_LITE_INTERPRETER",
+                System.env.BUILD_LITE_INTERPRETER != null ? System.env.BUILD_LITE_INTERPRETER : "1"
+        )
         addManifestPlaceholders([APP_NAME: "@string/app_name", MAIN_ACTIVITY: "org.pytorch.testapp.MainActivity"])
     }
     buildTypes {
@@ -63,14 +68,15 @@ android {
         mnet {
             dimension "model"
             applicationIdSuffix ".mnet"
-            buildConfigField("String", "MODULE_ASSET_NAME", "\"mnet.pt\"")
+            buildConfigField("String", "MODULE_ASSET_NAME", "\"mobilenet_v2.ptl\"")
             addManifestPlaceholders([APP_NAME: "MNET"])
             buildConfigField("String", "LOGCAT_TAG", "\"pytorch-mnet\"")
         }
+        // NB: This is not working atm https://github.com/pytorch/pytorch/issues/102966
         mnetVulkan {
             dimension "model"
             applicationIdSuffix ".mnet_vulkan"
-            buildConfigField("String", "MODULE_ASSET_NAME", "\"mnet_vulkan.pt\"")
+            buildConfigField("String", "MODULE_ASSET_NAME", "\"mobilenet_v2_vulkan.ptl\"")
             buildConfigField("boolean", "USE_VULKAN_DEVICE", 'true')
             addManifestPlaceholders([APP_NAME: "MNET_VULKAN"])
             buildConfigField("String", "LOGCAT_TAG", "\"pytorch-mnet-vulkan\"")
@@ -78,7 +84,7 @@ android {
         resnet18 {
             dimension "model"
             applicationIdSuffix ".resnet18"
-            buildConfigField("String", "MODULE_ASSET_NAME", "\"resnet18.pt\"")
+            buildConfigField("String", "MODULE_ASSET_NAME", "\"resnet18.ptl\"")
             addManifestPlaceholders([APP_NAME: "RN18"])
             buildConfigField("String", "LOGCAT_TAG", "\"pytorch-resnet18\"")
         }
@@ -149,8 +155,8 @@ dependencies {
     //nativeBuildImplementation(name: 'pytorch_android_torchvision-release', ext: 'aar')
     //extractForNativeBuild(name: 'pytorch_android-release', ext: 'aar')
 
-    nightlyImplementation 'org.pytorch:pytorch_android:1.12.0-SNAPSHOT'
-    nightlyImplementation 'org.pytorch:pytorch_android_torchvision:1.12.0-SNAPSHOT'
+    nightlyImplementation 'org.pytorch:pytorch_android:2.2.0-SNAPSHOT'
+    nightlyImplementation 'org.pytorch:pytorch_android_torchvision:2.2.0-SNAPSHOT'
 
     aarImplementation(name:'pytorch_android', ext:'aar')
     aarImplementation(name:'pytorch_android_torchvision', ext:'aar')

android/test_app/app/src/main/java/org/pytorch/testapp/MainActivity.java

@@ -1,6 +1,7 @@
 package org.pytorch.testapp;
 
 import android.content.Context;
+import android.content.res.AssetManager;
 import android.os.Bundle;
 import android.os.Handler;
 import android.os.HandlerThread;
@@ -16,6 +17,8 @@ import java.io.FileOutputStream;
 import java.io.IOException;
 import java.io.InputStream;
 import java.io.OutputStream;
+import java.lang.reflect.InvocationTargetException;
+import java.lang.reflect.Method;
 import java.nio.FloatBuffer;
 import org.pytorch.Device;
 import org.pytorch.IValue;
@@ -42,7 +45,13 @@ public class MainActivity extends AppCompatActivity {
       new Runnable() {
         @Override
         public void run() {
-          final Result result = doModuleForward();
+          final Result result;
+          try {
+            result = doModuleForward();
+          } catch (ClassNotFoundException | NoSuchMethodException | IllegalAccessException |
+                   InvocationTargetException e) {
+            throw new RuntimeException(e);
+          }
           runOnUiThread(
               new Runnable() {
                 @Override
@@ -118,7 +127,7 @@ public class MainActivity extends AppCompatActivity {
 
   @WorkerThread
   @Nullable
-  protected Result doModuleForward() {
+  protected Result doModuleForward() throws ClassNotFoundException, IllegalAccessException, NoSuchMethodException, InvocationTargetException {
     if (mModule == null) {
       final long[] shape = BuildConfig.INPUT_TENSOR_SHAPE;
       long numElements = 1;
@@ -129,12 +138,29 @@ public class MainActivity extends AppCompatActivity {
       mInputTensor =
           Tensor.fromBlob(
               mInputTensorBuffer, BuildConfig.INPUT_TENSOR_SHAPE, MemoryFormat.CHANNELS_LAST);
-      PyTorchAndroid.setNumThreads(1);
-      mModule =
-          BuildConfig.USE_VULKAN_DEVICE
-              ? PyTorchAndroid.loadModuleFromAsset(
-                  getAssets(), BuildConfig.MODULE_ASSET_NAME, Device.VULKAN)
-              : PyTorchAndroid.loadModuleFromAsset(getAssets(), BuildConfig.MODULE_ASSET_NAME);
+
+      Class ptAndroid;
+      if (BuildConfig.BUILD_LITE_INTERPRETER == 1) {
+        ptAndroid = Class.forName("org.pytorch.LitePyTorchAndroid");
+      }
+      else {
+        ptAndroid = Class.forName("org.pytorch.PyTorchAndroid");
+      }
+
+      Method setNumThreads = ptAndroid.getMethod("setNumThreads", int.class);
+      setNumThreads.invoke(null,1);
+
+      Method loadModuleFromAsset = ptAndroid.getMethod(
+              "loadModuleFromAsset",
+              AssetManager.class,
+              String.class,
+              Device.class
+              );
+      mModule = (Module) (BuildConfig.USE_VULKAN_DEVICE
+                    ? loadModuleFromAsset.invoke(
+                            null, getAssets(), BuildConfig.MODULE_ASSET_NAME, Device.VULKAN)
+                    : loadModuleFromAsset.invoke(
+                            null, getAssets(), BuildConfig.MODULE_ASSET_NAME, Device.CPU));
     }
 
     final long startTime = SystemClock.elapsedRealtime();

aten/src/ATen/CMakeLists.txt

@@ -125,6 +125,7 @@ file(GLOB native_ao_sparse_h
             "native/ao_sparse/quantized/cpu/*.h")
 file(GLOB native_quantized_h "native/quantized/*.h" "native/quantized/cpu/*.h" "native/quantized/cudnn/*.h")
 file(GLOB native_cpu_h "native/cpu/*.h")
+file(GLOB native_utils_h "native/utils/*.h")
 
 file(GLOB native_cuda_cu "native/cuda/*.cu")
 file(GLOB native_cuda_cpp "native/cuda/*.cpp")
@@ -540,7 +541,7 @@ install(FILES "${CMAKE_CURRENT_BINARY_DIR}/cmake-exports/ATenConfig.cmake"
 
 set(INSTALL_HEADERS ${base_h} ${ATen_CORE_HEADERS})
 if(NOT INTERN_BUILD_MOBILE)
-  list(APPEND INSTALL_HEADERS ${native_h} ${native_cpu_h} ${native_ao_sparse_h} ${native_quantized_h} ${cuda_h} ${native_cuda_h} ${native_hip_h} ${cudnn_h} ${hip_h} ${mps_h} ${native_mps_h} ${miopen_h})
+  list(APPEND INSTALL_HEADERS ${native_h} ${native_cpu_h} ${native_ao_sparse_h} ${native_quantized_h} ${cuda_h} ${native_cuda_h} ${native_hip_h} ${cudnn_h} ${hip_h} ${mps_h} ${native_mps_h} ${native_utils_h} ${miopen_h})
   # Metal
   if(USE_PYTORCH_METAL_EXPORT)
     # Add files needed from exporting metal models(optimized_for_mobile)

aten/src/ATen/Dispatch.h

@@ -371,6 +371,22 @@ inline void deprecated_AT_DISPATCH_ALL_TYPES_AND_HALF_AND_COMPLEX() {}
       AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND3(     \
           SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, __VA_ARGS__))
 
+#define AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND4(    \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, ...) \
+  AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES(__VA_ARGS__)   \
+  AT_DISPATCH_CASE(SCALARTYPE1, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE2, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE3, __VA_ARGS__)                 \
+  AT_DISPATCH_CASE(SCALARTYPE4, __VA_ARGS__)
+
+#define AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND4(                     \
+    SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, TYPE, NAME, ...) \
+  AT_DISPATCH_SWITCH(                                                    \
+      TYPE,                                                              \
+      NAME,                                                              \
+      AT_DISPATCH_CASE_FLOATING_AND_COMPLEX_TYPES_AND4(                  \
+          SCALARTYPE1, SCALARTYPE2, SCALARTYPE3, SCALARTYPE4, __VA_ARGS__))
+
 #define AT_DISPATCH_CASE_INTEGRAL_TYPES(...)          \
   AT_DISPATCH_CASE(at::ScalarType::Byte, __VA_ARGS__) \
   AT_DISPATCH_CASE(at::ScalarType::Char, __VA_ARGS__) \

aten/src/ATen/ExpandUtils.h

@@ -187,17 +187,18 @@ expand_inplace(
 // See NOTE [ ExpandUtils Borrowing ] above for `MaybeOwned` explanation.
 inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>
 expand_outplace(const Tensor& to_expand1, const Tensor& to_expand2) {
-  if (to_expand1.sizes().equals(to_expand2.sizes())) {
+  auto s1 = to_expand1.sym_sizes();
+  auto s2 = to_expand2.sym_sizes();
+  if (s1.equals(s2)) {
     return std::make_tuple(
         c10::MaybeOwned<Tensor>::borrowed(to_expand1),
         c10::MaybeOwned<Tensor>::borrowed(to_expand2));
   }
 
-  auto expanded_size =
-      infer_size_dimvector(to_expand1.sizes(), to_expand2.sizes());
+  auto expanded_size = infer_size_symdimvector(s1, s2);
   return std::make_tuple(
-      c10::MaybeOwned<Tensor>::owned(to_expand1.expand(expanded_size)),
-      c10::MaybeOwned<Tensor>::owned(to_expand2.expand(expanded_size)));
+      c10::MaybeOwned<Tensor>::owned(to_expand1.expand_symint(expanded_size)),
+      c10::MaybeOwned<Tensor>::owned(to_expand2.expand_symint(expanded_size)));
 }
 
 inline std::tuple<c10::MaybeOwned<Tensor>, c10::MaybeOwned<Tensor>>

aten/src/ATen/cuda/CUDAGraph.cpp

@@ -5,9 +5,13 @@
 #include <c10/cuda/CUDACachingAllocator.h>
 #include <c10/cuda/CUDAFunctions.h>
 
+#include <chrono>
+#include <thread>
+
 namespace at::cuda {
 
 static bool _cuda_graphs_debug = false;
+constexpr int kSynchronizeBusyWaitMillis = 10;
 
 MempoolId_t graph_pool_handle() {
 #if !defined(USE_ROCM) || ROCM_VERSION >= 50300
@@ -55,6 +59,25 @@ CaptureId_t capture_sequence_id() {
  * describes memory management for captures.
  */
 
+std::atomic<int> CUDAGraph::pending_event_queries = 0;
+
+// Track any outstanding event queries that could happen e.g., in a NCCL watchdog so that they
+// can be resolved before the capture begins. Note that event queries are not allowed during a
+// graph capture in the default capture mode.
+void CUDAGraph::inc_pending_event_queries() {
+  pending_event_queries++;
+}
+
+void CUDAGraph::dec_pending_event_queries() {
+  TORCH_INTERNAL_ASSERT(pending_event_queries > 0,
+    "Attempted to decrement the number of outstanding events to be queried, but it was <= 0.");
+  pending_event_queries--;
+}
+
+int CUDAGraph::num_pending_event_queries() {
+  return pending_event_queries;
+}
+
 CUDAGraph::CUDAGraph()
   // CUDAStreams may not be default-constructed.
   : capture_stream_(at::cuda::getCurrentCUDAStream()) {
@@ -115,6 +138,15 @@ void CUDAGraph::capture_begin(MempoolId_t pool/*=0*/, cudaStreamCaptureMode capt
   // due to the capture status being updated _after_ a capture had already started.
   c10::cuda::CUDACachingAllocator::beginAllocateStreamToPool(capture_dev_, capture_stream_, mempool_id_);
 
+  // At this point, any NCCL watchdogs should be aware that we are in capture mode
+  // and therefore should not enqueue any additional work that could be event-queried.
+  // We still must wait on any existing work that has not been cleaned up.
+  while (num_pending_event_queries()) {
+    TORCH_WARN_ONCE("Waiting for pending NCCL work to finish before starting graph capture.");
+    std::this_thread::sleep_for(
+      std::chrono::milliseconds(kSynchronizeBusyWaitMillis));
+  }
+
   // cudaStreamCaptureModeGlobal is the most conservative option to
   // prevent potentially unsafe CUDA API calls during capture.  See
   // https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__STREAM.html#group__CUDART__STREAM_1g9d0535d93a214cbf126835257b16ba85

aten/src/ATen/cuda/CUDAGraph.h

@@ -5,6 +5,8 @@
 #include <c10/cuda/CUDAGraphsC10Utils.h>
 #include <c10/cuda/CUDAStream.h>
 
+#include <mutex>
+
 namespace at {
 
 struct CUDAGeneratorImpl;
@@ -19,6 +21,9 @@ struct TORCH_CUDA_CPP_API CUDAGraph {
   CUDAGraph();
   ~CUDAGraph();
 
+  static void inc_pending_event_queries();
+  static void dec_pending_event_queries();
+  static int num_pending_event_queries();
   void capture_begin(MempoolId_t pool={0, 0}, cudaStreamCaptureMode capture_mode = cudaStreamCaptureModeGlobal);
   void capture_end();
   void replay();
@@ -33,6 +38,8 @@ struct TORCH_CUDA_CPP_API CUDAGraph {
   cudaGraphExec_t graph_exec_ = NULL;
 #endif
 
+  static std::atomic<int> pending_event_queries;
+
   // internal states so reset() can do its best cleaning up
   // Set to true in capture_end if cudaStreamEndCapture succeeded
   // Set back to false soon after, when graph_ is consumed by cudaGraphInstantiate

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

@@ -161,8 +161,10 @@ CUDA_STUB3(cuModuleGetFunction, CUfunction *, CUmodule, const char *);
 CUDA_STUB4(cuOccupancyMaxActiveBlocksPerMultiprocessor, int *, CUfunction, int, size_t);
 CUDA_STUB2(cuGetErrorString, CUresult, const char **);
 CUDA_STUB1(cuCtxGetCurrent, CUcontext *);
+CUDA_STUB1(cuCtxSetCurrent, CUcontext);
 CUDA_STUB1(cuModuleUnload, CUmodule);
 CUDA_STUB3(cuDevicePrimaryCtxGetState, CUdevice, unsigned int *, int *);
+CUDA_STUB2(cuDevicePrimaryCtxRetain, CUcontext *, CUdevice);
 CUDA_STUB4(cuLinkCreate, unsigned int, CUjit_option *, void **, CUlinkState *);
 CUDA_STUB3(cuLinkComplete, CUlinkState, void **, size_t *);
 CUDA_STUB3(cuFuncSetAttribute, CUfunction, CUfunction_attribute, int);

aten/src/ATen/cuda/nvrtc_stub/ATenNVRTC.h

@@ -51,8 +51,10 @@ namespace at { namespace cuda {
   _(cuLaunchKernel)                              \
   _(cuLaunchCooperativeKernel)                   \
   _(cuCtxGetCurrent)                             \
+  _(cuCtxSetCurrent)                             \
   _(cuModuleUnload)                              \
   _(cuDevicePrimaryCtxGetState)                  \
+  _(cuDevicePrimaryCtxRetain)                    \
   _(cuLinkCreate)                                \
   _(cuLinkAddData)                               \
   _(cuLinkComplete)                              \

aten/src/ATen/mps/MPSDevice.mm

@@ -93,10 +93,17 @@ MPSDevice::MPSDevice() : _mtl_device(nil), _mtl_indexing_library(nil) {
   NSArray* devices = [MTLCopyAllDevices() autorelease];
   for (unsigned long i = 0; i < [devices count]; i++) {
     id<MTLDevice> device = devices[i];
-    if (![device isLowPower]) { // exclude Intel GPUs
-      _mtl_device = [device retain];
-      break;
+    if ([device isLowPower]) { // exclude Intel GPUs
+      continue;
     }
+    if (![device supportsFamily:MTLGPUFamilyMac2]) {
+      // Exclude devices that does not support Metal 2.0
+      // Virtualised MPS device on MacOS 12.6 should fail this check
+      TORCH_WARN("Skipping device ", [[device name] UTF8String], " that does not support Metal 2.0");
+      continue;
+    }
+    _mtl_device = [device retain];
+    break;
   }
   TORCH_INTERNAL_ASSERT_DEBUG_ONLY(_mtl_device);
 }

aten/src/ATen/mps/MPSStream.mm

@@ -147,9 +147,9 @@ void MPSStream::addCompletedHandler(MTLCommandBufferHandler block) {
 }
 
 void MPSStream::fill(id<MTLBuffer> buffer, uint8_t value, size_t length, size_t offset, SyncType syncType) {
-  TORCH_INTERNAL_ASSERT(length >= offset);
-  if (length == 0)
+  if (length == 0) {
     return;
+  }
   dispatch_sync(_serialQueue, ^() {
     @autoreleasepool {
       endKernelCoalescing();

aten/src/ATen/native/BinaryOps.cpp

@@ -1393,7 +1393,7 @@ Tensor& comparison_op_out(Tensor& result, const Tensor& self, const Tensor& othe
 
 template <typename OutImpl>
 Tensor comparison_op(const Tensor& self, const Tensor& other, OutImpl& out_impl) {
-  Tensor result = at::empty(self.sizes(), self.options().dtype(kBool).device(self.device()));
+  Tensor result = at::empty({0}, self.options().dtype(kBool));
   return out_impl(result, self, other);
 }
 

aten/src/ATen/native/ConvUtils.h

@@ -389,8 +389,9 @@ static inline bool mkldnn_conv_use_channels_last(const at::Tensor& input, const
       (input_memory_format  == at::MemoryFormat::ChannelsLast) ||
       (weight_memory_format == at::MemoryFormat::ChannelsLast);
 
-  // TODO: add channels last 3d support
-  bool can_use_mkldnn_channels_last_3d = false;
+  bool can_use_mkldnn_channels_last_3d =
+      (input_memory_format  == at::MemoryFormat::ChannelsLast3d) ||
+      (weight_memory_format == at::MemoryFormat::ChannelsLast3d);
 
   return can_use_mkldnn_channels_last_2d || can_use_mkldnn_channels_last_3d;
 }

aten/src/ATen/native/Convolution.cpp

@@ -508,9 +508,6 @@ struct ConvParams {
     if (transposed && is_output_padding_big()) {
       return false;
     }
-    if (transposed && groups > 1 && at::symint::size<T>(input, 1) == groups) {
-      return false;
-    }
     if (input.device().is_cpu() && input.scalar_type() == kBFloat16 && mkldnn_bf16_device_check()) {
       return true;
     }

aten/src/ATen/native/Copy.cpp

@@ -253,7 +253,9 @@ static Tensor & copy_impl(Tensor & self, const Tensor & src, bool non_blocking)
       self.storage_offset() == src.storage_offset() &&
       self.strides().equals(src.strides()) &&
       self.sizes().equals(src.sizes()) &&
-      self.scalar_type() == src.scalar_type()
+      self.scalar_type() == src.scalar_type() &&
+      self.is_conj() == src.is_conj() &&
+      self.is_neg() == src.is_neg()
     );
   if (is_same_data) {
     return self;

aten/src/ATen/native/Distance.cpp

@@ -308,16 +308,18 @@ Tensor cosine_similarity(const Tensor& x1_, const Tensor& x2_, int64_t dim, doub
   // We accept integral types (and bools lol) but vector_norm does not
   auto x1_is_int = c10::isIntegralType(x1_.scalar_type(), /*încludeBool=*/true);
   auto x2_is_int = c10::isIntegralType(x2_.scalar_type(), /*încludeBool=*/true);
-  auto x1 = x1_is_int ? x1_.to(commonDtype) : x1_;
-  auto x2 = x2_is_int ? x2_.to(commonDtype) : x2_;
+  auto x1_t = x1_is_int ? x1_.to(commonDtype) : x1_;
+  auto x2_t = x2_is_int ? x2_.to(commonDtype) : x2_;
+  c10::MaybeOwned<Tensor> x1, x2;
+  std::tie(x1, x2) = expand_outplace(x1_t, x2_t);
 
 
   // We want to divide each tensor by its norm first, as it's more numerically stable.
   // This keeps the result between -1.0 and 1.0
   // We clone them, as we're going to modify them in-place
   // This allows the gradients to propagate propertly all the way to x1 and x2
-  auto x1_norm = at::linalg_vector_norm(x1, 2, /*dim=*/dim, /*keepdim=*/true).clone();
-  auto x2_norm = at::linalg_vector_norm(x2, 2, /*dim=*/dim, /*keepdim=*/true).clone();
+  auto x1_norm = at::linalg_vector_norm(*x1, 2, /*dim=*/dim, /*keepdim=*/true).clone();
+  auto x2_norm = at::linalg_vector_norm(*x2, 2, /*dim=*/dim, /*keepdim=*/true).clone();
 
   {
     at::NoGradGuard guard;
@@ -325,7 +327,7 @@ Tensor cosine_similarity(const Tensor& x1_, const Tensor& x2_, int64_t dim, doub
     x2_norm.clamp_min_(eps);
   }
 
-  return ((x1 / x1_norm) * (x2 / x2_norm)).sum(dim);
+  return ((*x1 / x1_norm) * (*x2 / x2_norm)).sum(dim);
 }
 
 }}  // namespace at::native

aten/src/ATen/native/LinearAlgebra.cpp

@@ -1483,12 +1483,14 @@ static void addmm_impl_cpu_(
   // it is faster to call oneDNN matrix multiplication primitive with RHS*LHS
   // that will call then into Arm® Compute Library (ACL) GEMM kernel and also
   // additionally have support for running kernel with BF16 instructions
-  bool apply_heur = apply_mkldnn_matmul_heur(b.sizes()[0], b.sizes()[1], a.sizes()[1]);
-  if (apply_heur && transpose_a && !transpose_b && result.scalar_type() == at::ScalarType::Float) {
-      mkldnn_matmul(b, a, c, beta.to<float>(), alpha.to<float>());
-      // We have dispatched to ACL GEMM for single precision float
-      // so do not need to dispatch to BLAS GEMM below
-      dispatched = true;
+  if (transpose_c) {
+    bool apply_heur = apply_mkldnn_matmul_heur(b.sizes()[0], b.sizes()[1], a.sizes()[1]);
+    if (apply_heur && transpose_a && !transpose_b && result.scalar_type() == at::ScalarType::Float) {
+        mkldnn_matmul(b, a, c, beta.to<float>(), alpha.to<float>());
+        // We have dispatched to ACL GEMM for single precision float
+        // so do not need to dispatch to BLAS GEMM below
+        dispatched = true;
+    }
   }
 #endif
 

aten/src/ATen/native/ReduceOps.cpp

@@ -2075,7 +2075,27 @@ bool cpu_equal(const Tensor& self, const Tensor& other) {
       && self.layout() == other.layout()
       && self.is_neg() == other.is_neg()
       && self.is_conj() == other.is_conj()) {
-    return true;
+    if (c10::isIntegralType(self.scalar_type(), /*includeBool=*/true)) {
+      return true;
+    }
+    std::atomic<bool> result{true};
+    auto iter = TensorIteratorConfig().add_input(self).build();
+    AT_DISPATCH_FLOATING_AND_COMPLEX_TYPES_AND2(kHalf, kBFloat16, iter.input_dtype(), "equal_notnan_cpu", [&] {
+      iter.for_each([&](char** data, const int64_t *strides, int64_t dim_size) {
+        if (!result) {
+            return;
+        }
+        char* self_data = data[0];
+        for (C10_UNUSED const auto i : c10::irange(dim_size)) {
+          if (isnan_(c10::load<scalar_t>(self_data))) {
+            result = false;
+            return;
+          }
+          self_data += strides[0];
+        }
+      });
+    });
+    return result.load();
   }
 
   std::atomic<bool> result{true};

aten/src/ATen/native/cpu/BinaryOpsKernel.cpp

@@ -101,7 +101,7 @@ void mul_kernel(TensorIteratorBase& iter) {
           using comp_t = c10::complex<float>;
           return comp_t{a} * comp_t{b};
         });
-  } else if (iter.is_scalar(2) && at::isReducedFloatingType(dtype)) {
+  } else if (iter.is_scalar(2) && iter.data_ptr(2) != nullptr && at::isReducedFloatingType(dtype)) {
     AT_DISPATCH_REDUCED_FLOATING_TYPES(dtype, "mul_cpu_reduced_float", [&]() {
       using opmath_t = at::opmath_type<scalar_t>;
       opmath_t b = iter.original_scalar_value<opmath_t>(2);
@@ -125,7 +125,7 @@ void mul_kernel(TensorIteratorBase& iter) {
 
 void div_true_kernel(TensorIteratorBase& iter) {
   const auto dtype = iter.common_dtype();
-  if (iter.is_scalar(2) && at::isReducedFloatingType(dtype)) {
+  if (iter.is_scalar(2) && iter.data_ptr(2) != nullptr && at::isReducedFloatingType(dtype)) {
     AT_DISPATCH_REDUCED_FLOATING_TYPES(dtype, "div_cpu_reduced_float", [&]() {
       using opmath_t = at::opmath_type<scalar_t>;
       opmath_t b = iter.original_scalar_value<opmath_t>(2);
@@ -162,19 +162,28 @@ void div_trunc_kernel(TensorIteratorBase& iter) {
         return a / b;
       });
     });
-  } else if (iter.is_scalar(2) && at::isReducedFloatingType(dtype)) {
-    AT_DISPATCH_REDUCED_FLOATING_TYPES(dtype, "div_trunc_cpu_reduced_float", [&]() {
-      using opmath_t = at::opmath_type<scalar_t>;
-      opmath_t b = iter.original_scalar_value<opmath_t>(2);
-      iter.remove_operand(2);
-      cpu_kernel_vec(iter,
-        [=](scalar_t a) __ubsan_ignore_float_divide_by_zero__ -> scalar_t {
-          return std::trunc(static_cast<opmath_t>(a) / b);
-        },
-        [=](Vectorized<scalar_t> a) {
-          return binary_op_scalar(a, b, [](const Vectorized<opmath_t>& x, const Vectorized<opmath_t>& y) { return (x / y).trunc(); });
+  } else if (iter.is_scalar(2) && iter.data_ptr(2) != nullptr && at::isReducedFloatingType(dtype)) {
+    AT_DISPATCH_REDUCED_FLOATING_TYPES(
+        dtype, "div_trunc_cpu_reduced_float", [&]() {
+          using opmath_t = at::opmath_type<scalar_t>;
+          opmath_t b = iter.original_scalar_value<opmath_t>(2);
+          iter.remove_operand(2);
+          cpu_kernel_vec(
+              iter,
+              [=](scalar_t a)
+                  __ubsan_ignore_float_divide_by_zero__ -> scalar_t {
+                    return std::trunc(static_cast<opmath_t>(a) / b);
+                  },
+              [=](Vectorized<scalar_t> a) {
+                return binary_op_scalar(
+                    a,
+                    b,
+                    [](const Vectorized<opmath_t>& x,
+                       const Vectorized<opmath_t>& y) {
+                      return (x / y).trunc();
+                    });
+              });
         });
-    });
   } else {
     AT_DISPATCH_FLOATING_TYPES_AND2(kBFloat16, kHalf, dtype, "div_trunc_cpu", [&]() {
       cpu_kernel_vec(iter,
@@ -223,20 +232,25 @@ void div_floor_kernel(TensorIteratorBase& iter) {
     });
   } else {
     // See NOTE: [Floor Division in Python]
-    if (iter.is_scalar(2) && at::isReducedFloatingType(dtype)) {
-      AT_DISPATCH_REDUCED_FLOATING_TYPES(dtype, "div_floor_cpu_reduced_float", [&]() {
-        using opmath_t = at::opmath_type<scalar_t>;
-        opmath_t b = iter.original_scalar_value<opmath_t>(2);
-        iter.remove_operand(2);
-        using vec_t = Vectorized<opmath_t>;
-        cpu_kernel_vec(iter,
-          [=](scalar_t a) -> scalar_t {
-            return div_floor_floating(static_cast<opmath_t>(a), b);
-          },
-          [=](Vectorized<scalar_t> a) {
-            return binary_op_scalar(a, b, [](const vec_t& x, const vec_t& y) { return div_floor_floating_vec(x, y); });
+    if (iter.is_scalar(2) && iter.data_ptr(2) != nullptr && at::isReducedFloatingType(dtype)) {
+      AT_DISPATCH_REDUCED_FLOATING_TYPES(
+          dtype, "div_floor_cpu_reduced_float", [&]() {
+            using opmath_t = at::opmath_type<scalar_t>;
+            opmath_t b = iter.original_scalar_value<opmath_t>(2);
+            iter.remove_operand(2);
+            using vec_t = Vectorized<opmath_t>;
+            cpu_kernel_vec(
+                iter,
+                [=](scalar_t a) -> scalar_t {
+                  return div_floor_floating(static_cast<opmath_t>(a), b);
+                },
+                [=](Vectorized<scalar_t> a) {
+                  return binary_op_scalar(
+                      a, b, [](const vec_t& x, const vec_t& y) {
+                        return div_floor_floating_vec(x, y);
+                      });
+                });
           });
-      });
     } else {
       AT_DISPATCH_FLOATING_TYPES_AND2(kBFloat16, kHalf, dtype, "div_floor_cpu", [&]() {
         using vec_t = Vectorized<scalar_t>;

aten/src/ATen/native/cuda/SpectralOps.cpp

@@ -10,6 +10,7 @@
 #include <ATen/native/SpectralOpsUtils.h>
 #include <ATen/native/cuda/CuFFTUtils.h>
 #include <ATen/native/cuda/CuFFTPlanCache.h>
+#include <ATen/cuda/nvrtc_stub/ATenNVRTC.h>
 #include <c10/util/irange.h>
 
 #ifndef AT_PER_OPERATOR_HEADERS
@@ -27,7 +28,6 @@
 #include <cufftXt.h>
 
 #include <cmath>
-#include <vector>
 
 
 namespace at::native {
@@ -304,6 +304,17 @@ static const Tensor& _exec_fft(Tensor& out, const Tensor& self, IntArrayRef out_
   CUFFT_CHECK(cufftSetWorkArea(plan, workspace.mutable_data_ptr()));
 
   // execute transform plan
+#if !defined(USE_ROCM)
+  CUcontext pctx = nullptr;
+  at::globalContext().getNVRTC().cuCtxGetCurrent(&pctx);
+  if (C10_UNLIKELY(!pctx)) {
+    // workaround for corner case where a primary context exists but is not
+    // the current context
+    TORCH_WARN_ONCE("Attempting to run cuFFT, but there was no current CUDA context! Attempting to set the primary context...");
+    at::globalContext().getNVRTC().cuDevicePrimaryCtxRetain(&pctx, 0);
+    at::globalContext().getNVRTC().cuCtxSetCurrent(pctx);
+  }
+#endif /* !defined(USE_ROCM) */
   exec_cufft_plan(*config, input.data_ptr(), out.data_ptr(), forward);
 
   // Inplace reshaping to original batch shape and inverting the dimension permutation

aten/src/ATen/native/mkldnn/Conv.cpp

@@ -727,7 +727,7 @@ Tensor _mkldnn_convolution_transpose(
 
   if (bias.defined()) {
     const ideep::tensor b = itensor_from_tensor(bias);
-    ideep::convolution_transpose_forward::compute(
+    ideep::convolution_transpose_forward::compute_v3(
         x,
         w,
         b,
@@ -738,9 +738,10 @@ Tensor _mkldnn_convolution_transpose(
         padding_r(padding_expanded, output_padding_expanded),
         dilation.vec(),
         groups,
+        use_channels_last,
         op_attr);
   } else {
-    ideep::convolution_transpose_forward::compute(
+    ideep::convolution_transpose_forward::compute_v3(
         x,
         w,
         output_sizes,
@@ -750,6 +751,7 @@ Tensor _mkldnn_convolution_transpose(
         padding_r(padding_expanded, output_padding_expanded),
         dilation.vec(),
         groups,
+        use_channels_last,
         op_attr);
   }
   if (input.is_mkldnn()) {
@@ -988,7 +990,7 @@ Tensor mkldnn_convolution_transpose_backward_input(
     grad_input.resize_(input_size, memory_format);
     grad_x = itensor_from_tensor(grad_input);
   }
-  ideep::convolution_transpose_backward_data::compute(
+  ideep::convolution_transpose_backward_data::compute_v3(
       grad_y,
       w,
       input_size.vec(),
@@ -997,7 +999,8 @@ Tensor mkldnn_convolution_transpose_backward_input(
       padding.vec(),
       padding_r(padding, output_padding),
       dilation.vec(),
-      groups);
+      groups,
+      is_channels_last);
 
   if (grad_output.is_mkldnn()) {
     return MKLDNNTensor(grad_x, grad_output.options());
@@ -1024,7 +1027,7 @@ std::tuple<Tensor,Tensor> mkldnn_convolution_transpose_backward_weights(
 
   ideep::tensor grad_w, grad_b;
   if (bias_defined) {
-    ideep::convolution_transpose_backward_weights::compute(
+    ideep::convolution_transpose_backward_weights::compute_v3(
         x,
         grad_y,
         weight_size.vec(),
@@ -1034,9 +1037,10 @@ std::tuple<Tensor,Tensor> mkldnn_convolution_transpose_backward_weights(
         padding.vec(),
         padding_r(padding, output_padding),
         dilation.vec(),
-        groups);
+        groups,
+        is_channels_last);
   } else {
-    ideep::convolution_transpose_backward_weights::compute(
+    ideep::convolution_transpose_backward_weights::compute_v3(
         x,
         grad_y,
         weight_size.vec(),
@@ -1045,7 +1049,8 @@ std::tuple<Tensor,Tensor> mkldnn_convolution_transpose_backward_weights(
         padding.vec(),
         padding_r(padding, output_padding),
         dilation.vec(),
-        groups);
+        groups,
+        is_channels_last);
   }
 
   if (!is_channels_last) {
@@ -1061,18 +1066,21 @@ std::tuple<Tensor,Tensor> mkldnn_convolution_transpose_backward_weights(
 }
 
 std::tuple<Tensor, Tensor, Tensor> mkldnn_convolution_transpose_backward(
-    const Tensor& input, const Tensor& grad_output_t, const Tensor& weight,
+    const Tensor& input_t, const Tensor& grad_output_t, const Tensor& weight_t,
     IntArrayRef padding, IntArrayRef output_padding, IntArrayRef stride, IntArrayRef dilation, int64_t groups,
     std::array<bool,3> output_mask)
 {
-  bool is_channels_last = mkldnn_conv_use_channels_last(input, weight);
-  auto memory_format = mkldnn_convolution_memory_format(input.ndimension(), is_channels_last);
+  bool is_channels_last = mkldnn_conv_use_channels_last(input_t, weight_t);
+  auto memory_format = mkldnn_convolution_memory_format(input_t.ndimension(), is_channels_last);
   Tensor grad_output = grad_output_t.is_mkldnn() ? grad_output_t : grad_output_t.contiguous(memory_format);
+  auto input = input_t.is_mkldnn() ? input_t : input_t.contiguous(memory_format);
+  auto weight = weight_t.is_mkldnn() ? weight_t : weight_t.contiguous(memory_format);
   int64_t dim = input.ndimension() - 2;
   const auto padding_expanded = expand_param_if_needed(padding, "padding", dim);
   const auto stride_expanded = expand_param_if_needed(stride, "stride", dim);
   const auto dilation_expanded = expand_param_if_needed(dilation, "dilation", dim);
   const auto output_padding_expanded = expand_param_if_needed(output_padding, "output_padding", dim);
+
   Tensor grad_input, grad_weight, grad_bias;
   if (output_mask[0]) {
     grad_input = mkldnn_convolution_transpose_backward_input(

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

@@ -72,7 +72,7 @@ static bool fill_mps_tensor_(Tensor& self, uint8_t value) {
   if (self.is_contiguous()) {
     MPSStream* stream = getCurrentMPSStream();
     auto storage_byte_offset = self.storage_offset() * self.itemsize();
-    stream->fill(mps::getMTLBufferStorage(self), 0, self.storage().nbytes(), storage_byte_offset);
+    stream->fill(mps::getMTLBufferStorage(self), value, self.nbytes(), storage_byte_offset);
     return true;
   }
   return false;

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

@@ -445,7 +445,7 @@ static Tensor mps_convolution_backward_weights(IntArrayRef weight_size,
     string key = "mps_convolution_backward_weights:" + to_string(stride[0]) + ":" + to_string(stride[1]) + ":" +
         to_string(dilation[0]) + ":" + to_string(dilation[1]) + ":" + to_string(padding[0]) + ":" +
         to_string(padding[1]) + ":" + to_string(groups) + ":" + mem_format_key +
-        getTensorsStringKey({grad_output_t, input_t}) + ":" + string([ns_shape_key UTF8String]);
+        getTensorsStringKey({grad_output_t, input_t, grad_weight_t}) + ":" + string([ns_shape_key UTF8String]);
 
     auto cachedGraph = LookUpOrCreateCachedGraph<CachedGraph>(key, [&](auto mpsGraph, auto newCachedGraph) {
       MPSGraphConvolution2DOpDescriptor* conv2dDescriptor_ = [[MPSGraphConvolution2DOpDescriptor new] autorelease];

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

@@ -100,7 +100,7 @@ static at::Tensor& copy_from_mps_(at::Tensor& dst_, const at::Tensor& src_, bool
     MTLResourceOptions options = MTLResourceCPUCacheModeDefaultCache | MTLResourceStorageModeShared;
     NSUInteger alignedLength = 0;
 
-    const void* host_dst = dst.storage().data();
+    const void* host_dst = static_cast<const char*>(dst.storage().data()) + dst.storage_offset() * dst.itemsize();
     void* alignedPtr = pageAlignedBlockPtr(host_dst, (NSUInteger)dst_tensor_nbytes, &alignedLength);
     NSUInteger destOffset = (uintptr_t(host_dst) - uintptr_t(alignedPtr));
     // 4 bytes alignment required on macos for blits.
@@ -293,7 +293,8 @@ at::Tensor& mps_copy_(at::Tensor& dst, const at::Tensor& src, bool non_blocking)
     dst.resize_as_(src);
   }
 
-  TORCH_CHECK(dst.dim() >= src.dim());
+  TORCH_CHECK(
+      dst.dim() >= src.dim(), "Destination ", dst.sym_sizes(), " doesn't match the broadcast shape ", src.sym_sizes());
   if (dst.dim() > src.dim()) {
     needs_broadcasting = true;
   } else {

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

@@ -16,15 +16,14 @@ namespace at::native {
 Scalar _local_scalar_dense_mps(const Tensor& self) {
   Scalar r;
 
+  auto output = at::empty_like(self, TensorOptions(kCPU));
+  mps::mps_copy_(output, self, false);
   AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND3(at::ScalarType::Half,
                                          at::ScalarType::Bool,
                                          at::ScalarType::BFloat16,
                                          self.scalar_type(),
                                          "_local_scalar_dense_mps",
                                          [&] {
-                                           Tensor output = at::empty({1}, TensorOptions(at::CPU(self.scalar_type())));
-
-                                           mps::mps_copy_(output, self, false);
                                            scalar_t value = *output.data_ptr<scalar_t>();
                                            r = Scalar(value);
                                          });

aten/src/ATen/native/transformers/attention.cpp

@@ -622,8 +622,13 @@ at::Tensor preprocess_mask(
     return pad_bias<mem_eff_alignment>(attn_mask);
   }
   // Check and make the tensor contiguous if needed
-  if (attn_mask.sym_stride(0) % 16 != 0 || attn_mask.sym_stride(1) % 16 != 0 ||
-      attn_mask.sym_stride(2) % 16 != 0) {
+  auto needs_contig = [](const c10::SymInt& stride) {
+    return (stride % 16 != 0) || (stride == 0);
+  };
+  if (needs_contig(attn_mask.sym_stride(0)) ||
+      needs_contig(attn_mask.sym_stride(1)) ||
+      needs_contig(attn_mask.sym_stride(2)) ||
+      needs_contig(attn_mask.sym_stride(3))) {
     return attn_mask.contiguous();
   }
 

aten/src/ATen/native/transformers/cuda/mem_eff_attention/kernel_forward.h

@@ -298,9 +298,11 @@ struct AttentionKernel {
       // 15/16th of tensor core compute In that case :
       //  - we only launch kernels for head_id % kQueriesPerBlock == 0
       //  - we iterate over heads instead of queries (strideM = strideH)
-      if (num_queries == 1 && k_strideH == 0 && v_strideH == 0) {
-        if (head_id % kQueriesPerBlock != 0)
+      if (num_queries == 1 && k_strideH == 0 && v_strideH == 0 &&
+          logsumexp_ptr == nullptr) {
+        if (head_id % kQueriesPerBlock != 0) {
           return false;
+        }
         q_strideM = q_strideH;
         num_queries = num_heads;
         num_heads = 1; // unused but here for intent

c10/cuda/driver_api.cpp

@@ -10,7 +10,7 @@ namespace {
 DriverAPI create_driver_api() {
 #define OPEN_LIBRARIES(name, n)               \
   void* handle_##n = dlopen(name, RTLD_LAZY); \
-  TORCH_INTERNAL_ASSERT(handle_##n);
+  TORCH_INTERNAL_ASSERT(handle_##n, "Can't open ", #name, ": ", dlerror());
 
   C10_FORALL_DRIVER_LIBRARIES(OPEN_LIBRARIES)
 #undef OPEN_LIBRARIES
@@ -18,7 +18,7 @@ DriverAPI create_driver_api() {
 
 #define LOOKUP_ENTRY(name, n)                              \
   r.name##_ = ((decltype(&name))dlsym(handle_##n, #name)); \
-  TORCH_INTERNAL_ASSERT(r.name##_)
+  TORCH_INTERNAL_ASSERT(r.name##_, "Can't find ", #name, ": ", dlerror())
   C10_FORALL_DRIVER_API(LOOKUP_ENTRY)
 #undef LOOKUP_ENTRY
   return r;

c10/cuda/driver_api.h

@@ -19,7 +19,7 @@
   } while (0)
 
 #define C10_FORALL_DRIVER_LIBRARIES(_) \
-  _("libcuda.so", 0)                   \
+  _("libcuda.so.1", 0)                 \
   _("libnvidia-ml.so.1", 1)
 
 #define C10_FORALL_DRIVER_API(_)         \

caffe2/utils/threadpool/ThreadPool.cc

@@ -41,8 +41,13 @@ namespace {
 }
 
 size_t getDefaultNumThreads() {
-  CAFFE_ENFORCE(cpuinfo_initialize(), "cpuinfo initialization failed");
-  int numThreads = cpuinfo_get_processors_count();
+  auto numThreads = 1U;
+  if (cpuinfo_initialize()) {
+    numThreads = std::max(cpuinfo_get_processors_count(), 1U);
+  } else {
+    LOG(WARNING) << "cpuinfo initialization failed";
+    numThreads = std::max(std::thread::hardware_concurrency(), 1U);
+  }
 
   bool applyCap = false;
 #if defined(C10_ANDROID)
@@ -109,7 +114,7 @@ size_t getDefaultNumThreads() {
    * detect if we are running under tsan, for now capping the default
    * threadcount to the tsan limit unconditionally.
    */
-  int tsanThreadLimit = 63;
+  auto tsanThreadLimit = 63U;
   numThreads = std::min(numThreads, tsanThreadLimit);
 
   return numThreads;

cmake/iOS.cmake

@@ -53,9 +53,9 @@ set(CMAKE_RANLIB ranlib CACHE FILEPATH "" FORCE)
 set(PKG_CONFIG_EXECUTABLE pkg-config CACHE FILEPATH "" FORCE)
 
 # Setup iOS platform unless specified manually with IOS_PLATFORM
-if(NOT DEFINED IOS_PLATFORM)
+if(NOT IOS_PLATFORM)
     set(IOS_PLATFORM "OS")
-endif(NOT DEFINED IOS_PLATFORM)
+endif(NOT IOS_PLATFORM)
 set(IOS_PLATFORM ${IOS_PLATFORM} CACHE STRING "Type of iOS Platform")
 
 # Check the platform selection and setup for developer root
@@ -118,9 +118,9 @@ set(CMAKE_FIND_LIBRARY_SUFFIXES ".dylib" ".so" ".a")
 # (where install_name_tool was hardcoded) and where CMAKE_INSTALL_NAME_TOOL isn't in the cache
 # and still cmake didn't fail in CMakeFindBinUtils.cmake (because it isn't rerun)
 # hardcode CMAKE_INSTALL_NAME_TOOL here to install_name_tool, so it behaves as it did before, Alex
-if(NOT DEFINED CMAKE_INSTALL_NAME_TOOL)
+if(NOT CMAKE_INSTALL_NAME_TOOL)
     find_program(CMAKE_INSTALL_NAME_TOOL install_name_tool)
-endif(NOT DEFINED CMAKE_INSTALL_NAME_TOOL)
+endif(NOT CMAKE_INSTALL_NAME_TOOL)
 
 # Setup iOS deployment target
 set(IOS_DEPLOYMENT_TARGET ${IOS_DEPLOYMENT_TARGET} CACHE STRING "Minimum iOS version")
@@ -130,17 +130,17 @@ set(IOS_DEPLOYMENT_TARGET ${IOS_DEPLOYMENT_TARGET} CACHE STRING "Minimum iOS ver
 exec_program(/usr/bin/xcode-select ARGS -print-path OUTPUT_VARIABLE CMAKE_XCODE_DEVELOPER_DIR)
 set(XCODE_POST_43_ROOT "${CMAKE_XCODE_DEVELOPER_DIR}/Platforms/${IOS_PLATFORM_LOCATION}/Developer")
 set(XCODE_PRE_43_ROOT "/Developer/Platforms/${IOS_PLATFORM_LOCATION}/Developer")
-if(NOT DEFINED CMAKE_IOS_DEVELOPER_ROOT)
+if(NOT CMAKE_IOS_DEVELOPER_ROOT)
     if(EXISTS ${XCODE_POST_43_ROOT})
         set(CMAKE_IOS_DEVELOPER_ROOT ${XCODE_POST_43_ROOT})
     elseif(EXISTS ${XCODE_PRE_43_ROOT})
         set(CMAKE_IOS_DEVELOPER_ROOT ${XCODE_PRE_43_ROOT})
     endif(EXISTS ${XCODE_POST_43_ROOT})
-endif(NOT DEFINED CMAKE_IOS_DEVELOPER_ROOT)
+endif(NOT CMAKE_IOS_DEVELOPER_ROOT)
 set(CMAKE_IOS_DEVELOPER_ROOT ${CMAKE_IOS_DEVELOPER_ROOT} CACHE PATH "Location of iOS Platform")
 
 # Find and use the most recent iOS sdk unless specified manually with CMAKE_IOS_SDK_ROOT
-if(NOT DEFINED CMAKE_IOS_SDK_ROOT)
+if(NOT CMAKE_IOS_SDK_ROOT)
     file(GLOB _CMAKE_IOS_SDKS "${CMAKE_IOS_DEVELOPER_ROOT}/SDKs/*")
     if(_CMAKE_IOS_SDKS)
         list(SORT _CMAKE_IOS_SDKS)
@@ -150,7 +150,7 @@ if(NOT DEFINED CMAKE_IOS_SDK_ROOT)
         message(FATAL_ERROR "No iOS SDK's found in default search path ${CMAKE_IOS_DEVELOPER_ROOT}. Manually set CMAKE_IOS_SDK_ROOT or install the iOS SDK.")
     endif(_CMAKE_IOS_SDKS)
     message(STATUS "Toolchain using default iOS SDK: ${CMAKE_IOS_SDK_ROOT}")
-endif(NOT DEFINED CMAKE_IOS_SDK_ROOT)
+endif(NOT CMAKE_IOS_SDK_ROOT)
 set(CMAKE_IOS_SDK_ROOT ${CMAKE_IOS_SDK_ROOT} CACHE PATH "Location of the selected iOS SDK")
 
 # Set the sysroot default to the most recent SDK

docker.Makefile

@@ -8,8 +8,9 @@ $(warning WARNING: No docker user found using results from whoami)
 DOCKER_ORG                = $(shell whoami)
 endif
 
-CUDA_VERSION              = 11.8.0
-CUDNN_VERSION             = 8
+CUDA_VERSION_SHORT       ?= 12.1
+CUDA_VERSION             ?= 12.1.1
+CUDNN_VERSION            ?= 8
 BASE_RUNTIME              = ubuntu:20.04
 BASE_DEVEL                = nvidia/cuda:$(CUDA_VERSION)-cudnn$(CUDNN_VERSION)-devel-ubuntu20.04
 CMAKE_VARS               ?=
@@ -71,25 +72,25 @@ all: devel-image
 
 .PHONY: devel-image
 devel-image: BASE_IMAGE := $(BASE_DEVEL)
-devel-image: DOCKER_TAG := $(PYTORCH_VERSION)-devel
+devel-image: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-devel
 devel-image:
 	$(DOCKER_BUILD)
 
 .PHONY: devel-push
 devel-push: BASE_IMAGE := $(BASE_DEVEL)
-devel-push: DOCKER_TAG := $(PYTORCH_VERSION)-devel
+devel-push: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-devel
 devel-push:
 	$(DOCKER_PUSH)
 
 .PHONY: runtime-image
 runtime-image: BASE_IMAGE := $(BASE_RUNTIME)
-runtime-image: DOCKER_TAG := $(PYTORCH_VERSION)-runtime
+runtime-image: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-runtime
 runtime-image:
 	$(DOCKER_BUILD)
 
 .PHONY: runtime-push
 runtime-push: BASE_IMAGE := $(BASE_RUNTIME)
-runtime-push: DOCKER_TAG := $(PYTORCH_VERSION)-runtime
+runtime-push: DOCKER_TAG := $(PYTORCH_VERSION)-cuda$(CUDA_VERSION_SHORT)-cudnn$(CUDNN_VERSION)-runtime
 runtime-push:
 	$(DOCKER_PUSH)
 

docs/Makefile

@@ -18,8 +18,8 @@ figures:
 	@$(PYCMD) source/scripts/build_quantization_configs.py
 
 onnx:
-	@$(PYCMD) source/scripts/onnx/build_onnx_supported_aten_op_csv_table.py
-	@$(PYCMD) source/scripts/onnx/build_onnx_diagnostics_rules_md.py $(SOURCEDIR)/generated/onnx_diagnostics_rules
+	@$(PYCMD) source/scripts/onnx/build_onnx_torchscript_supported_aten_op_csv_table.py
+	@$(PYCMD) source/scripts/onnx/build_onnx_dynamo_diagnostics_rules_md.py $(SOURCEDIR)/generated/onnx_dynamo_diagnostics_rules
 
 opset:
 	@$(PYCMD) source/scripts/build_opsets.py

docs/source/autograd.rst

@@ -179,6 +179,7 @@ Tensor autograd functions
    torch.Tensor.detach
    torch.Tensor.detach_
    torch.Tensor.register_hook
+   torch.Tensor.register_post_accumulate_grad_hook
    torch.Tensor.retain_grad
 
 :hidden:`Function`