diff --git a/.circleci/scripts/binary_populate_env.sh b/.circleci/scripts/binary_populate_env.sh
index 7e663a64b..f9a542b53 100755
--- a/.circleci/scripts/binary_populate_env.sh
+++ b/.circleci/scripts/binary_populate_env.sh
@@ -48,9 +48,9 @@ if [[ -z "$DOCKER_IMAGE" ]]; then
   if [[ "$PACKAGE_TYPE" == conda ]]; then
     export DOCKER_IMAGE="pytorch/conda-cuda"
   elif [[ "$DESIRED_CUDA" == cpu ]]; then
-    export DOCKER_IMAGE="pytorch/manylinux-cpu"
+    export DOCKER_IMAGE="pytorch/manylinux-builder:cpu"
   else
-    export DOCKER_IMAGE="pytorch/manylinux-cuda${DESIRED_CUDA:2}"
+    export DOCKER_IMAGE="pytorch/manylinux-builer:${DESIRED_CUDA:2}"
   fi
 fi
 
@@ -121,6 +121,7 @@ export DESIRED_PYTHON="$DESIRED_PYTHON"
 export DESIRED_CUDA="$DESIRED_CUDA"
 export LIBTORCH_VARIANT="${LIBTORCH_VARIANT:-}"
 export BUILD_PYTHONLESS="${BUILD_PYTHONLESS:-}"
+export USE_SPLIT_BUILD="${USE_SPLIT_BUILD:-}"
 export DESIRED_DEVTOOLSET="$DESIRED_DEVTOOLSET"
 if [[ "${BUILD_FOR_SYSTEM:-}" == "windows" ]]; then
   export LIBTORCH_CONFIG="${LIBTORCH_CONFIG:-}"
diff --git a/.github/workflows/validate-binaries.yml b/.github/workflows/validate-binaries.yml
index f11d6c87a..f40c4f474 100644
--- a/.github/workflows/validate-binaries.yml
+++ b/.github/workflows/validate-binaries.yml
@@ -47,6 +47,13 @@ on:
         default: false
         required: false
         type: boolean
+      use_split_build:
+        description: |
+          [Experimental] Build a libtorch only wheel and build pytorch such that
+          are built from the libtorch wheel.
+        required: false
+        type: boolean
+        default: false
   workflow_dispatch:
     inputs:
       os:
@@ -105,6 +112,13 @@ on:
         default: false
         required: false
         type: boolean
+      use_split_build:
+        description: |
+          [Experimental] Build a libtorch only wheel and build pytorch such that
+          are built from the libtorch wheel.
+        required: false
+        type: boolean
+        default: false
 
 
 jobs:
@@ -141,6 +155,7 @@ jobs:
       include-test-ops: ${{ inputs.include-test-ops }}
       use-only-dl-pytorch-org: ${{ inputs.use-only-dl-pytorch-org }}
       use-force-reinstall: ${{ inputs.use-force-reinstall }}
+      use_split_build: ${{ inputs.use_split_build }}
 
   linux-aarch64:
     if:  inputs.os == 'linux-aarch64' || inputs.os == 'all'
diff --git a/CUDA_UPGRADE_GUIDE.MD b/CUDA_UPGRADE_GUIDE.MD
index be7bb5ff7..58897bee2 100644
--- a/CUDA_UPGRADE_GUIDE.MD
+++ b/CUDA_UPGRADE_GUIDE.MD
@@ -61,7 +61,7 @@ There are three types of Docker containers we maintain in order to build Linux b
 Add setup for our Docker `libtorch` and `manywheel`:
 1. Follow this PR [PR 1003](https://github.com/pytorch/builder/pull/1003) for all steps in this section
 2. For `libtorch`, the code changes are usually copy-paste. For `manywheel`, you should manually verify the versions of the shared libraries with the CUDA you downloaded before.
-3. This is Manual Step: Create a ticket for PyTorch Dev Infra team to Create a new repo to host manylinux-cuda images in docker hub, for example, https://hub.docker.com/r/pytorch/manylinux-cuda115. This repo should have public visibility and read & write access for bots. This step can be removed once the following [issue](https://github.com/pytorch/builder/issues/901) is addressed.
+3. This is Manual Step: Create a ticket for PyTorch Dev Infra team to Create a new repo to host manylinux-cuda images in docker hub, for example, https://hub.docker.com/r/pytorch/manylinux-builder:cuda115. This repo should have public visibility and read & write access for bots. This step can be removed once the following [issue](https://github.com/pytorch/builder/issues/901) is addressed.
 4. Push the images to Docker Hub. This step should be automated with the help with GitHub Actions in the `pytorch/builder` repo. Make sure to update the `cuda_version` to the version you're adding in respective YAMLs, such as `.github/workflows/build-manywheel-images.yml`, `.github/workflows/build-conda-images.yml`, `.github/workflows/build-libtorch-images.yml`.
 5. Verify that each of the workflows that push the images succeed by selecting and verifying them in the [Actions page](https://github.com/pytorch/builder/actions/workflows/build-libtorch-images.yml) of pytorch/builder. Furthermore, check [https://hub.docker.com/r/pytorch/manylinux-builder/tags](https://hub.docker.com/r/pytorch/manylinux-builder/tags), [https://hub.docker.com/r/pytorch/libtorch-cxx11-builder/tags](https://hub.docker.com/r/pytorch/libtorch-cxx11-builder/tags) to verify that the right tags exist for manylinux and libtorch types of images.
 6. Finally before enabling nightly binaries and CI builds we should make sure we post following PRs in [PR 1015](https://github.com/pytorch/builder/pull/1015) [PR 1017](https://github.com/pytorch/builder/pull/1017) and [this commit](https://github.com/pytorch/builder/commit/7d5e98f1336c7cb84c772604c5e0d1acb59f2d72) to enable the new CUDA build in wheels and conda.
diff --git a/aarch64_linux/build_aarch64_wheel.py b/aarch64_linux/build_aarch64_wheel.py
index 3956f0463..6338ae60b 100755
--- a/aarch64_linux/build_aarch64_wheel.py
+++ b/aarch64_linux/build_aarch64_wheel.py
@@ -53,7 +53,7 @@ def ec2_instances_by_id(instance_id):
     return rc[0] if len(rc) > 0 else None
 
 
-def start_instance(key_name, ami=ubuntu18_04_ami, instance_type='t4g.2xlarge'):
+def start_instance(key_name, ami=ubuntu18_04_ami, instance_type='t4g.2xlarge', ebs_size: int = 50):
     inst = ec2.create_instances(ImageId=ami,
                                 InstanceType=instance_type,
                                 SecurityGroups=['ssh-allworld'],
@@ -65,7 +65,7 @@ def start_instance(key_name, ami=ubuntu18_04_ami, instance_type='t4g.2xlarge'):
                                         'DeviceName': '/dev/sda1',
                                         'Ebs': {
                                             'DeleteOnTermination': True,
-                                            'VolumeSize': 50,
+                                            'VolumeSize': ebs_size,
                                             'VolumeType': 'standard'
                                         }
                                     }
@@ -714,6 +714,7 @@ def parse_arguments():
     parser.add_argument("--keep-running", action="store_true")
     parser.add_argument("--terminate-instances", action="store_true")
     parser.add_argument("--instance-type", type=str, default="t4g.2xlarge")
+    parser.add_argument("--ebs-size", type=int, default=50)
     parser.add_argument("--branch", type=str, default="master")
     parser.add_argument("--use-docker", action="store_true")
     parser.add_argument("--compiler", type=str, choices=['gcc-7', 'gcc-8', 'gcc-9', 'clang'], default="gcc-8")
@@ -746,7 +747,7 @@ def parse_arguments():
             check `~/.ssh/` folder or manually set SSH_KEY_PATH environment variable.""")
 
     # Starting the instance
-    inst = start_instance(key_name, ami=ami, instance_type=args.instance_type)
+    inst = start_instance(key_name, ami=ami, instance_type=args.instance_type, ebs_size=args.ebs_size)
     instance_name = f'{args.key_name}-{args.os}'
     if args.python_version is not None:
         instance_name += f'-py{args.python_version}'
diff --git a/common/install_cpython.sh b/common/install_cpython.sh
index 44ad3a4b7..4e2360653 100755
--- a/common/install_cpython.sh
+++ b/common/install_cpython.sh
@@ -6,7 +6,7 @@ PYTHON_DOWNLOAD_GITHUB_BRANCH=https://github.com/python/cpython/archive/refs/hea
 GET_PIP_URL=https://bootstrap.pypa.io/get-pip.py
 
 # Python versions to be installed in /opt/$VERSION_NO
-CPYTHON_VERSIONS=${CPYTHON_VERSIONS:-"3.7.5 3.8.1 3.9.0 3.10.1 3.11.0 3.12.0 3.13.0"}
+CPYTHON_VERSIONS=${CPYTHON_VERSIONS:-"3.8.1 3.9.0 3.10.1 3.11.0 3.12.0 3.13.0"}
 
 function check_var {
     if [ -z "$1" ]; then
diff --git a/conda/build_pytorch.sh b/conda/build_pytorch.sh
index 9f0649c50..9132b4058 100755
--- a/conda/build_pytorch.sh
+++ b/conda/build_pytorch.sh
@@ -413,7 +413,13 @@ for py_ver in "${DESIRED_PYTHON[@]}"; do
         else
           local_channel="$(pwd)/$output_folder"
         fi
-        conda install -y -c "file://$local_channel" pytorch==$PYTORCH_BUILD_VERSION -c pytorch -c numba/label/dev -c pytorch-nightly -c nvidia
+
+        CONDA_CHANNEL="pytorch-test"
+        if [[ -n "$OVERRIDE_PACKAGE_VERSION" && "$OVERRIDE_PACKAGE_VERSION" =~ .*dev.* ]]; then
+            CONDA_CHANNEL="pytorch-nightly"
+        fi
+
+        conda install -y -c "file://$local_channel" pytorch==$PYTORCH_BUILD_VERSION -c pytorch -c numba/label/dev -c $CONDA_CHANNEL -c nvidia
 
         echo "$(date) :: Running tests"
         pushd "$pytorch_rootdir"
diff --git a/ffmpeg/win/Dockerfile b/ffmpeg/win/Dockerfile
index 3f6015524..d9ab1f127 100644
--- a/ffmpeg/win/Dockerfile
+++ b/ffmpeg/win/Dockerfile
@@ -1,5 +1,5 @@
 # Base docker image for cross-compilling FFmpeg (LGPL) for Windows
-FROM pytorch/manylinux-cuda101
+FROM pytorch/manylinux-builder:cuda101
 COPY . /ffmpeg-build-src
 WORKDIR /ffmpeg-build-src
 
diff --git a/manywheel/Dockerfile b/manywheel/Dockerfile
index 2c805baa3..152cb802b 100644
--- a/manywheel/Dockerfile
+++ b/manywheel/Dockerfile
@@ -117,7 +117,7 @@ ENV SSL_CERT_FILE=/opt/_internal/certs.pem
 COPY --from=openssl            /opt/openssl                          /opt/openssl
 COPY --from=python             /opt/python                           /opt/python
 COPY --from=python             /opt/_internal                        /opt/_internal
-COPY --from=python             /opt/python/cp37-cp37m/bin/auditwheel /usr/local/bin/auditwheel
+COPY --from=python             /opt/python/cp39-cp39/bin/auditwheel /usr/local/bin/auditwheel
 COPY --from=intel              /opt/intel                            /opt/intel
 COPY --from=patchelf           /usr/local/bin/patchelf               /usr/local/bin/patchelf
 COPY --from=jni                /usr/local/include/jni.h              /usr/local/include/jni.h
diff --git a/manywheel/Dockerfile_s390x b/manywheel/Dockerfile_s390x
index caa5d00bd..5125e3830 100644
--- a/manywheel/Dockerfile_s390x
+++ b/manywheel/Dockerfile_s390x
@@ -69,5 +69,5 @@ RUN bash build_scripts/build.sh && rm -r build_scripts
 FROM openssl as final
 COPY --from=python             /opt/python                           /opt/python
 COPY --from=python             /opt/_internal                        /opt/_internal
-COPY --from=python             /opt/python/cp37-cp37m/bin/auditwheel /usr/local/bin/auditwheel
+COPY --from=python             /opt/python/cp39-cp39/bin/auditwheel /usr/local/bin/auditwheel
 COPY --from=patchelf           /usr/local/bin/patchelf               /usr/local/bin/patchelf
diff --git a/manywheel/build_common.sh b/manywheel/build_common.sh
index d68d9a323..e0a514402 100644
--- a/manywheel/build_common.sh
+++ b/manywheel/build_common.sh
@@ -45,8 +45,14 @@ fi
 if [[ -z "$TORCH_PACKAGE_NAME" ]]; then
     TORCH_PACKAGE_NAME='torch'
 fi
+
+if [[ -z "$TORCH_NO_PYTHON_PACKAGE_NAME" ]]; then
+    TORCH_NO_PYTHON_PACKAGE_NAME='torch_no_python'
+fi
+
 TORCH_PACKAGE_NAME="$(echo $TORCH_PACKAGE_NAME | tr '-' '_')"
-echo "Expecting the built wheels to all be called '$TORCH_PACKAGE_NAME'"
+TORCH_NO_PYTHON_PACKAGE_NAME="$(echo $TORCH_NO_PYTHON_PACKAGE_NAME | tr '-' '_')"
+echo "Expecting the built wheels to all be called '$TORCH_PACKAGE_NAME' or '$TORCH_NO_PYTHON_PACKAGE_NAME'"
 
 # Version: setup.py uses $PYTORCH_BUILD_VERSION.post$PYTORCH_BUILD_NUMBER if
 # PYTORCH_BUILD_NUMBER > 1
@@ -160,11 +166,29 @@ else
 fi
 
 echo "Calling setup.py bdist at $(date)"
-time CMAKE_ARGS=${CMAKE_ARGS[@]} \
-    EXTRA_CAFFE2_CMAKE_FLAGS=${EXTRA_CAFFE2_CMAKE_FLAGS[@]} \
+
+if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
+    echo "Calling setup.py bdist_wheel for split build (BUILD_LIBTORCH_WHL)"
+    time EXTRA_CAFFE2_CMAKE_FLAGS=${EXTRA_CAFFE2_CMAKE_FLAGS[@]} \
+    BUILD_LIBTORCH_WHL=1 BUILD_PYTHON_ONLY=0 \
     BUILD_LIBTORCH_CPU_WITH_DEBUG=$BUILD_DEBUG_INFO \
     USE_NCCL=${USE_NCCL} USE_RCCL=${USE_RCCL} USE_KINETO=${USE_KINETO} \
     python setup.py bdist_wheel -d /tmp/$WHEELHOUSE_DIR
+    echo "Finished setup.py bdist_wheel for split build (BUILD_LIBTORCH_WHL)"
+    echo "Calling setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
+    time EXTRA_CAFFE2_CMAKE_FLAGS=${EXTRA_CAFFE2_CMAKE_FLAGS[@]} \
+    BUILD_LIBTORCH_WHL=0 BUILD_PYTHON_ONLY=1 \
+    BUILD_LIBTORCH_CPU_WITH_DEBUG=$BUILD_DEBUG_INFO \
+    USE_NCCL=${USE_NCCL} USE_RCCL=${USE_RCCL} USE_KINETO=${USE_KINETO} \
+    python setup.py bdist_wheel -d /tmp/$WHEELHOUSE_DIR --cmake
+    echo "Finished setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
+else
+    time CMAKE_ARGS=${CMAKE_ARGS[@]} \
+        EXTRA_CAFFE2_CMAKE_FLAGS=${EXTRA_CAFFE2_CMAKE_FLAGS[@]} \
+        BUILD_LIBTORCH_CPU_WITH_DEBUG=$BUILD_DEBUG_INFO \
+        USE_NCCL=${USE_NCCL} USE_RCCL=${USE_RCCL} USE_KINETO=${USE_KINETO} \
+        python setup.py bdist_wheel -d /tmp/$WHEELHOUSE_DIR
+fi
 echo "Finished setup.py bdist at $(date)"
 
 # Build libtorch packages
@@ -282,6 +306,8 @@ echo 'Built this wheel:'
 ls /tmp/$WHEELHOUSE_DIR
 mkdir -p "/$WHEELHOUSE_DIR"
 mv /tmp/$WHEELHOUSE_DIR/torch*linux*.whl /$WHEELHOUSE_DIR/
+mv /tmp/$WHEELHOUSE_DIR/torch_no_python*.whl /$WHEELHOUSE_DIR/ || true
+
 if [[ -n "$BUILD_PYTHONLESS" ]]; then
     mkdir -p /$LIBTORCH_HOUSE_DIR
     mv /tmp/$LIBTORCH_HOUSE_DIR/*.zip /$LIBTORCH_HOUSE_DIR
@@ -292,7 +318,7 @@ rm -rf /tmp_dir
 mkdir /tmp_dir
 pushd /tmp_dir
 
-for pkg in /$WHEELHOUSE_DIR/torch*linux*.whl /$LIBTORCH_HOUSE_DIR/libtorch*.zip; do
+for pkg in /$WHEELHOUSE_DIR/torch*linux*.whl /$WHEELHOUSE_DIR/torch_no_python*.whl /$LIBTORCH_HOUSE_DIR/libtorch*.zip; do
 
     # if the glob didn't match anything
     if [[ ! -e $pkg ]]; then
@@ -408,6 +434,14 @@ for pkg in /$WHEELHOUSE_DIR/torch*linux*.whl /$LIBTORCH_HOUSE_DIR/libtorch*.zip;
         popd
     fi
 
+    # @sahanp todo: Remove this line
+    echo "current files in directory"
+    ls -l
+
+    echo "removing extraneous .so and .a files"
+    # todo @PaliC: Remove these .so and .a files before hand in the split build
+    rm *.so *.a || true
+
     # zip up the wheel back
     zip -rq $(basename $pkg) $PREIX*
 
@@ -441,7 +475,9 @@ if [[ -z "$BUILD_PYTHONLESS" ]]; then
   pushd $PYTORCH_ROOT/test
 
   # Install the wheel for this Python version
+  pip uninstall -y "$TORCH_NO_PYTHON_PACKAGE_NAME" || true
   pip uninstall -y "$TORCH_PACKAGE_NAME"
+  pip install "$TORCH_NO_PYTHON_PACKAGE_NAME" --no-index -f /$WHEELHOUSE_DIR --no-dependencies -v || true
   pip install "$TORCH_PACKAGE_NAME" --no-index -f /$WHEELHOUSE_DIR --no-dependencies -v
 
   # Print info on the libraries installed in this wheel
diff --git a/manywheel/build_docker.sh b/manywheel/build_docker.sh
index 819a4a003..58d74e578 100755
--- a/manywheel/build_docker.sh
+++ b/manywheel/build_docker.sh
@@ -16,14 +16,12 @@ case ${GPU_ARCH_TYPE} in
     cpu)
         TARGET=cpu_final
         DOCKER_TAG=cpu
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux-cpu
         GPU_IMAGE=centos:7
         DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=9"
         ;;
     cpu-manylinux_2_28)
         TARGET=cpu_final
         DOCKER_TAG=cpu
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux_2_28-cpu
         GPU_IMAGE=amd64/almalinux:8
         DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=11"
         MANY_LINUX_VERSION="2_28"
@@ -31,7 +29,6 @@ case ${GPU_ARCH_TYPE} in
     cpu-aarch64)
         TARGET=final
         DOCKER_TAG=cpu-aarch64
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux-cpu-aarch64
         GPU_IMAGE=arm64v8/centos:7
         DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=10"
         MANY_LINUX_VERSION="aarch64"
@@ -39,7 +36,6 @@ case ${GPU_ARCH_TYPE} in
     cpu-aarch64-2_28)
         TARGET=final
         DOCKER_TAG=cpu-aarch64
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux_2_28-cpu-aarch64
         GPU_IMAGE=arm64v8/almalinux:8
         DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=11"
         MANY_LINUX_VERSION="2_28_aarch64"
@@ -47,7 +43,6 @@ case ${GPU_ARCH_TYPE} in
     cpu-cxx11-abi)
         TARGET=final
         DOCKER_TAG=cpu-cxx11-abi
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux-cpu-cxx11-abi
         GPU_IMAGE=""
         DOCKER_GPU_BUILD_ARG=" --build-arg DEVTOOLSET_VERSION=9"
         MANY_LINUX_VERSION="cxx11-abi"
@@ -55,7 +50,6 @@ case ${GPU_ARCH_TYPE} in
     cpu-s390x)
         TARGET=final
         DOCKER_TAG=cpu-s390x
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux-cpu-s390x
         GPU_IMAGE=redhat/ubi9
         DOCKER_GPU_BUILD_ARG=""
         MANY_LINUX_VERSION="s390x"
@@ -63,7 +57,6 @@ case ${GPU_ARCH_TYPE} in
     cuda)
         TARGET=cuda_final
         DOCKER_TAG=cuda${GPU_ARCH_VERSION}
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux-cuda${GPU_ARCH_VERSION//./}
         # Keep this up to date with the minimum version of CUDA we currently support
         GPU_IMAGE=centos:7
         DOCKER_GPU_BUILD_ARG="--build-arg BASE_CUDA_VERSION=${GPU_ARCH_VERSION} --build-arg DEVTOOLSET_VERSION=9"
@@ -71,7 +64,6 @@ case ${GPU_ARCH_TYPE} in
     cuda-manylinux_2_28)
         TARGET=cuda_final
         DOCKER_TAG=cuda${GPU_ARCH_VERSION}
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux_2_28-cuda${GPU_ARCH_VERSION//./}
         GPU_IMAGE=amd64/almalinux:8
         DOCKER_GPU_BUILD_ARG="--build-arg BASE_CUDA_VERSION=${GPU_ARCH_VERSION} --build-arg DEVTOOLSET_VERSION=11"
         MANY_LINUX_VERSION="2_28"
@@ -79,7 +71,6 @@ case ${GPU_ARCH_TYPE} in
     cuda-aarch64)
         TARGET=cuda_final
         DOCKER_TAG=cuda${GPU_ARCH_VERSION}
-        LEGACY_DOCKER_IMAGE=''
         GPU_IMAGE=arm64v8/centos:7
         DOCKER_GPU_BUILD_ARG="--build-arg BASE_CUDA_VERSION=${GPU_ARCH_VERSION} --build-arg DEVTOOLSET_VERSION=11"
         MANY_LINUX_VERSION="aarch64"
@@ -88,7 +79,6 @@ case ${GPU_ARCH_TYPE} in
     rocm)
         TARGET=rocm_final
         DOCKER_TAG=rocm${GPU_ARCH_VERSION}
-        LEGACY_DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/manylinux-rocm:${GPU_ARCH_VERSION}
         GPU_IMAGE=rocm/dev-centos-7:${GPU_ARCH_VERSION}-complete
         PYTORCH_ROCM_ARCH="gfx900;gfx906;gfx908;gfx90a;gfx1030;gfx1100"
         ROCM_REGEX="([0-9]+)\.([0-9]+)[\.]?([0-9]*)"
@@ -114,7 +104,6 @@ DOCKER_NAME=manylinux${MANY_LINUX_VERSION}
 DOCKER_IMAGE=${DOCKER_REGISTRY}/pytorch/${DOCKER_NAME}-builder:${DOCKER_TAG}
 if [[ -n ${MANY_LINUX_VERSION} && -z ${DOCKERFILE_SUFFIX} ]]; then
     DOCKERFILE_SUFFIX=_${MANY_LINUX_VERSION}
-    LEGACY_DOCKER_IMAGE=''
 fi
 (
     set -x
@@ -135,9 +124,6 @@ DOCKER_IMAGE_SHA_TAG=${DOCKER_IMAGE}-${GIT_COMMIT_SHA}
 
 (
     set -x
-    if [[ -n ${LEGACY_DOCKER_IMAGE} ]]; then
-        docker tag ${DOCKER_IMAGE} ${LEGACY_DOCKER_IMAGE}
-    fi
     if [[ -n ${GITHUB_REF} ]]; then
         docker tag ${DOCKER_IMAGE} ${DOCKER_IMAGE_BRANCH_TAG}
         docker tag ${DOCKER_IMAGE} ${DOCKER_IMAGE_SHA_TAG}
@@ -148,9 +134,6 @@ if [[ "${WITH_PUSH}" == true ]]; then
     (
         set -x
         docker push "${DOCKER_IMAGE}"
-        if [[ -n ${LEGACY_DOCKER_IMAGE} ]]; then
-            docker push "${LEGACY_DOCKER_IMAGE}"
-        fi
         if [[ -n ${GITHUB_REF} ]]; then
             docker push "${DOCKER_IMAGE_BRANCH_TAG}"
             docker push "${DOCKER_IMAGE_SHA_TAG}"
diff --git a/manywheel/build_libtorch.sh b/manywheel/build_libtorch.sh
index ea11f0c51..c3c0391a8 100644
--- a/manywheel/build_libtorch.sh
+++ b/manywheel/build_libtorch.sh
@@ -69,12 +69,8 @@ fi
 # ever pass one python version, so we assume that DESIRED_PYTHON is not a list
 # in this case
 if [[ -n "$DESIRED_PYTHON" && "$DESIRED_PYTHON" != cp* ]]; then
-    if [[ "$DESIRED_PYTHON" == '3.7' ]]; then
-      DESIRED_PYTHON='cp37-cp37m'
-    else
-      python_nodot="$(echo $DESIRED_PYTHON | tr -d m.u)"
-      DESIRED_PYTHON="cp${python_nodot}-cp${python_nodot}"
-    fi
+    python_nodot="$(echo $DESIRED_PYTHON | tr -d m.u)"
+    DESIRED_PYTHON="cp${python_nodot}-cp${python_nodot}"
 fi
 pydir="/opt/python/$DESIRED_PYTHON"
 export PATH="$pydir/bin:$PATH"
@@ -97,9 +93,7 @@ fi
 pushd "$PYTORCH_ROOT"
 python setup.py clean
 retry pip install -qr requirements.txt
-if [[ "$DESIRED_PYTHON"  == "cp37-cp37m" ]]; then
-    retry pip install -q numpy==1.15
-elif [[ "$DESIRED_PYTHON"  == "cp38-cp38" ]]; then
+if [[ "$DESIRED_PYTHON"  == "cp38-cp38" ]]; then
     retry pip install -q numpy==1.15
 else
     retry pip install -q numpy==1.11
diff --git a/manywheel/build_scripts/build.sh b/manywheel/build_scripts/build.sh
index c545ca967..d7369f800 100644
--- a/manywheel/build_scripts/build.sh
+++ b/manywheel/build_scripts/build.sh
@@ -58,14 +58,14 @@ autoconf --version
 build_openssl $OPENSSL_ROOT $OPENSSL_HASH
 /build_scripts/install_cpython.sh
 
-PY37_BIN=/opt/python/cp37-cp37m/bin
+PY39_BIN=/opt/python/cp39-cp39/bin
 
 # Our openssl doesn't know how to find the system CA trust store
 #   (https://github.com/pypa/manylinux/issues/53)
 # And it's not clear how up-to-date that is anyway
 # So let's just use the same one pip and everyone uses
-$PY37_BIN/pip install certifi
-ln -s $($PY37_BIN/python -c 'import certifi; print(certifi.where())') \
+$PY39_BIN/pip install certifi
+ln -s $($PY39_BIN/python -c 'import certifi; print(certifi.where())') \
       /opt/_internal/certs.pem
 # If you modify this line you also have to modify the versions in the
 # Dockerfiles:
@@ -86,8 +86,8 @@ tar -xzf patchelf-0.10.tar.gz
 rm -rf patchelf-0.10.tar.gz patchelf-0.10
 
 # Install latest pypi release of auditwheel
-$PY37_BIN/pip install auditwheel
-ln -s $PY37_BIN/auditwheel /usr/local/bin/auditwheel
+$PY39_BIN/pip install auditwheel
+ln -s $PY39_BIN/auditwheel /usr/local/bin/auditwheel
 
 # Clean up development headers and other unnecessary stuff for
 # final image
diff --git a/test/smoke_test/smoke_test.py b/test/smoke_test/smoke_test.py
index 0b347a334..60574f434 100644
--- a/test/smoke_test/smoke_test.py
+++ b/test/smoke_test/smoke_test.py
@@ -162,11 +162,10 @@ def smoke_test_cuda(package: str, runtime_error_check: str, torch_compile_check:
                 version = imported_module._extension._check_cuda_version()
             print(f"{module['name']} CUDA: {version}")
 
-     # torch.compile is available on macos-arm64 and Linux for python 3.8-3.12
-    if torch_compile_check == "enabled" and sys.version_info < (3, 13, 0) and (
-        (target_os == "linux" and torch.cuda.is_available()) or
-        target_os in ["macos-arm64", "darwin"]):
-        smoke_test_compile()
+     # torch.compile is available on macos-arm64 and Linux for python 3.8-3.13
+    if (torch_compile_check == "enabled" and sys.version_info < (3, 13, 0)
+        and target_os in ["linux", "linux-aarch64", "macos-arm64", "darwin"]):
+        smoke_test_compile("cuda" if torch.cuda.is_available() else "cpu")
 
     if torch.cuda.is_available():
         if torch.version.cuda != gpu_arch_ver:
@@ -241,27 +240,26 @@ def test_linalg(device="cpu") -> None:
             torch.linalg.svd(A)
 
 
-def smoke_test_compile() -> None:
+def smoke_test_compile(device: str = "cpu") -> None:
     supported_dtypes = [torch.float16, torch.float32, torch.float64]
-    dv = "cuda" if target_os == "linux" else "cpu"
 
     def foo(x: torch.Tensor) -> torch.Tensor:
         return torch.sin(x) + torch.cos(x)
 
     for dtype in supported_dtypes:
-        print(f"Testing smoke_test_compile for {dtype}")
-        x = torch.rand(3, 3, device=dv).type(dtype)
+        print(f"Testing smoke_test_compile for {device} and {dtype}")
+        x = torch.rand(3, 3, device=device).type(dtype)
         x_eager = foo(x)
         x_pt2 = torch.compile(foo)(x)
-        print(torch.allclose(x_eager, x_pt2))
+        torch.testing.assert_close(x_eager, x_pt2)
 
     # Reset torch dynamo since we are changing mode
     torch._dynamo.reset()
     dtype = torch.float32
     torch.set_float32_matmul_precision('high')
     print(f"Testing smoke_test_compile with mode 'max-autotune' for {dtype}")
-    x = torch.rand(64, 1, 28, 28, device=dv).type(torch.float32)
-    model = Net().to(device=dv)
+    x = torch.rand(64, 1, 28, 28, device=device).type(torch.float32)
+    model = Net().to(device=device)
     x_pt2 = torch.compile(model, mode="max-autotune")(x)
 
 
diff --git a/wheel/build_wheel.sh b/wheel/build_wheel.sh
index 7320d6173..cb96d08db 100755
--- a/wheel/build_wheel.sh
+++ b/wheel/build_wheel.sh
@@ -194,7 +194,16 @@ export BUILD_TEST=OFF
 pushd "$pytorch_rootdir"
 echo "Calling setup.py bdist_wheel at $(date)"
 
-python setup.py bdist_wheel -d "$whl_tmp_dir"
+if [[ "$USE_SPLIT_BUILD" == "true" ]]; then
+    echo "Calling setup.py bdist_wheel for split build (BUILD_LIBTORCH_WHL)"
+    BUILD_LIBTORCH_WHL=1 BUILD_PYTHON_ONLY=0 python setup.py bdist_wheel -d "$whl_tmp_dir"
+    echo "Finished setup.py bdist_wheel for split build (BUILD_LIBTORCH_WHL)"
+    echo "Calling setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
+    BUILD_PYTHON_ONLY=1 BUILD_LIBTORCH_WHL=0 python setup.py bdist_wheel -d "$whl_tmp_dir" --cmake
+    echo "Finished setup.py bdist_wheel for split build (BUILD_PYTHON_ONLY)"
+else
+    python setup.py bdist_wheel -d "$whl_tmp_dir"
+fi
 
 echo "Finished setup.py bdist_wheel at $(date)"