diff --git a/.github/actions/setup/action.yml b/.github/actions/setup/action.yml
index bb67d27d1..6ccc65428 100644
--- a/.github/actions/setup/action.yml
+++ b/.github/actions/setup/action.yml
@@ -40,31 +40,3 @@ runs:
       run: |
         bash ./.github/workflows/cuda/${{ runner.os }}.sh ${{ inputs.cuda-version }}
       shell: bash
-
-    - name: Install PyTorch ${{ inputs.torch-version }}+${{ inputs.cuda-version }}
-      run: |
-        pip install torch==${{ inputs.torch-version }} --extra-index-url https://download.pytorch.org/whl/${{ inputs.cuda-version }}
-        python -c "import torch; print('PyTorch:', torch.__version__)"
-        python -c "import torch; print('CUDA:', torch.version.cuda)"
-        python -c "import torch; print('CXX11 ABI:', torch.compiled_with_cxx11_abi())"
-      shell: bash
-
-    - name: Disable CUDNN
-      if: ${{ (inputs.cuda-version != 'cpu') && ((inputs.torch-version == '1.12.0') || (inputs.torch-version == '1.13.0')) }}
-      run: |
-        Torch_DIR=`python -c 'import torch;print(torch.utils.cmake_prefix_path)'`
-        sed -i '95,100d' ${Torch_DIR}/Caffe2/Caffe2Config.cmake
-        sed -i 's/;caffe2::cudnn-public//g' ${Torch_DIR}/Caffe2/Caffe2Targets.cmake
-      shell: bash
-
-    - name: Downgrade GLIBC
-      if: ${{ runner.os == 'Linux' }}
-      run: |
-        sed -i '1s/^/#if defined(__linux__) \&\& defined(__x86_64__)\n__asm__(".symver log,log@GLIBC_2.2.5");\n#endif\n/' third_party/METIS/GKlib/gk_proto.h
-        sed -i '1s/^/#if defined(__linux__) \&\& defined(__x86_64__)\n__asm__(".symver pow,pow@GLIBC_2.2.5");\n#endif\n/' third_party/METIS/libmetis/metislib.h
-      shell: bash
-
-    - name: Install additional dependencies
-      run: |
-        pip install setuptools ninja wheel
-      shell: bash
diff --git a/.github/workflows/building.yml b/.github/workflows/building.yml
index 96d8084a4..ba75d1006 100644
--- a/.github/workflows/building.yml
+++ b/.github/workflows/building.yml
@@ -1,6 +1,14 @@
 name: Building Wheels
 
-on: [workflow_dispatch]  # yamllint disable-line rule:truthy
+on:  # yamllint disable-line rule:truthy
+  workflow_dispatch:
+  push:
+    branches:
+      - aki/manylinux
+
+concurrency:
+  group: aki_hobby
+  cancel-in-progress: true
 
 jobs:
 
@@ -10,113 +18,9 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-22.04, macos-14, windows-2019]
-        python-version: ['3.9', '3.10', '3.11', '3.12', '3.13']
-        # torch-version: [1.13.0, 2.0.0, 2.1.0, 2.2.0, 2.3.0, 2.4.0, 2.5.0, 2.6.0, 2.7.0]
-        torch-version: [2.7.0]
-        cuda-version: ['cpu', 'cu117', 'cu118', 'cu121', 'cu124', 'cu126', 'cu128']
-        exclude:
-          - torch-version: 1.13.0
-            python-version: '3.13'
-          - torch-version: 2.0.0
-            python-version: '3.13'
-          - torch-version: 2.1.0
-            python-version: '3.13'
-          - torch-version: 2.2.0
-            python-version: '3.13'
-          - torch-version: 2.3.0
-            python-version: '3.13'
-          - torch-version: 2.4.0
-            python-version: '3.13'
-          - torch-version: 2.5.0
-            python-version: '3.13'
-          - torch-version: 1.13.0
-            python-version: '3.12'
-          - torch-version: 2.0.0
-            python-version: '3.12'
-          - torch-version: 2.1.0
-            python-version: '3.12'
-          - torch-version: 1.13.0
-            python-version: '3.11'
-          - torch-version: 1.13.0
-            cuda-version: 'cu118'
-          - torch-version: 1.13.0
-            cuda-version: 'cu121'
-          - torch-version: 1.13.0
-            cuda-version: 'cu124'
-          - torch-version: 1.13.0
-            cuda-version: 'cu126'
-          - torch-version: 1.13.0
-            cuda-version: 'cu128'
-          - torch-version: 2.0.0
-            cuda-version: 'cu124'
-          - torch-version: 2.0.0
-            cuda-version: 'cu126'
-          - torch-version: 2.0.0
-            cuda-version: 'cu128'
-          - torch-version: 2.1.0
-            cuda-version: 'cu117'
-          - torch-version: 2.1.0
-            cuda-version: 'cu124'
-          - torch-version: 2.1.0
-            cuda-version: 'cu126'
-          - torch-version: 2.1.0
-            cuda-version: 'cu128'
-          - torch-version: 2.2.0
-            cuda-version: 'cu117'
-          - torch-version: 2.2.0
-            cuda-version: 'cu124'
-          - torch-version: 2.2.0
-            cuda-version: 'cu126'
-          - torch-version: 2.2.0
-            cuda-version: 'cu128'
-          - torch-version: 2.3.0
-            cuda-version: 'cu117'
-          - torch-version: 2.3.0
-            cuda-version: 'cu124'
-          - torch-version: 2.3.0
-            cuda-version: 'cu126'
-          - torch-version: 2.3.0
-            cuda-version: 'cu128'
-          - torch-version: 2.4.0
-            cuda-version: 'cu117'
-          - torch-version: 2.4.0
-            cuda-version: 'cu126'
-          - torch-version: 2.4.0
-            cuda-version: 'cu128'
-          - torch-version: 2.5.0
-            cuda-version: 'cu117'
-          - torch-version: 2.5.0
-            cuda-version: 'cu126'
-          - torch-version: 2.5.0
-            cuda-version: 'cu128'
-          - torch-version: 2.6.0
-            cuda-version: 'cu117'
-          - torch-version: 2.6.0
-            cuda-version: 'cu121'
-          - torch-version: 2.6.0
-            cuda-version: 'cu128'
-          - torch-version: 2.7.0
-            cuda-version: 'cu117'
-          - torch-version: 2.7.0
-            cuda-version: 'cu121'
-          - torch-version: 2.7.0
-            cuda-version: 'cu124'
-          - os: macos-14
-            cuda-version: 'cu117'
-          - os: macos-14
-            cuda-version: 'cu118'
-          - os: macos-14
-            cuda-version: 'cu121'
-          - os: macos-14
-            cuda-version: 'cu124'
-          - os: macos-14
-            cuda-version: 'cu126'
-          - os: macos-14
-            cuda-version: 'cu128'
-          - os: windows-2019
-            torch-version: 2.0.0
-            cuda-version: 'cu121'
+        include:
+          - {os: ubuntu-22.04, torch-version: '2.6.0', python-version: '3.10', cuda-version: 'cpu'}
+          - {os: ubuntu-22.04, torch-version: '2.6.0', python-version: '3.10', cuda-version: 'cu118'}
 
     steps:
       - name: Checkout repository
@@ -124,12 +28,10 @@ jobs:
         with:
           submodules: recursive
 
-      - name: Install dependencies
-        uses: ./.github/actions/setup
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-          torch-version: ${{ matrix.torch-version }}
-          cuda-version: ${{ matrix.cuda-version }}
 
       - name: Set version on non-macOS
         if: ${{ runner.os != 'macOS' }}
@@ -155,27 +57,34 @@ jobs:
       - name: Build wheel
         run: |
           source ./.github/workflows/cuda/${{ runner.os }}-env.sh ${{ matrix.cuda-version }}
-          pip install build
-          python -m build --wheel --no-isolation --outdir dist
+          pip install cibuildwheel -U
+          python -m cibuildwheel --output-dir dist
         shell: bash
+        env:
+          CIBW_BUILD: cp311-manylinux_x86_64
+          CIBW_BUILD_FRONTEND: 'build; args: --no-isolation'
+          CIBW_BEFORE_BUILD: 'dnf install -y python3.11-devel && bash ./tools/prepare_for_build.sh 2.6.0 cpu'
+          CIBW_MANYLINUX_X86_64_IMAGE: manylinux_2_28
 
       - name: Test wheel
+        working-directory: dist
         run: |
-          cd dist
           ls -lah
+          pip install auditwheel
+          auditwheel repair *.whl
+
           pip install *.whl
           python -c "import pyg_lib; print('pyg-lib:', pyg_lib.__version__)"
           python -c "import pyg_lib; print('CUDA:', pyg_lib.cuda_version())"
-          cd ..
         shell: bash
 
-      - name: Configure AWS
-        uses: aws-actions/configure-aws-credentials@v1
-        with:
-          aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
-          aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
-          aws-region: us-west-1
+      # - name: Configure AWS
+      #   uses: aws-actions/configure-aws-credentials@v1
+      #   with:
+      #     aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
+      #     aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+      #     aws-region: us-west-1
 
-      - name: Upload wheel
-        run: |
-          aws s3 sync dist s3://data.pyg.org/whl/torch-${{ matrix.torch-version }}+${{ matrix.cuda-version }} --grants read=uri=http://acs.amazonaws.com/groups/global/AllUsers
+      # - name: Upload wheel
+      #   run: |
+      #     aws s3 sync dist s3://data.pyg.org/whl/torch-${{ matrix.torch-version }}+${{ matrix.cuda-version }} --grants read=uri=http://acs.amazonaws.com/groups/global/AllUsers
diff --git a/.github/workflows/cuda/Linux-env.sh b/.github/workflows/cuda/Linux-env.sh
index a6f8c3759..128fb04e7 100755
--- a/.github/workflows/cuda/Linux-env.sh
+++ b/.github/workflows/cuda/Linux-env.sh
@@ -1,4 +1,5 @@
 #!/bin/bash
+set -ex
 
 case ${1} in
   cu128)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index d821637b0..b8006f90c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -73,6 +73,10 @@ if(WITH_CUDA)
     set(CMAKE_CUDA_ARCHITECTURES "60;70;75")
   endif()
 
+  enable_language(CUDA)
+  add_definitions(-DWITH_CUDA)
+  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr -allow-unsupported-compiler")
+
   if (NOT "$ENV{EXTERNAL_CUTLASS_INCLUDE_DIR}" STREQUAL "")
     include_directories($ENV{EXTERNAL_CUTLASS_INCLUDE_DIR})
   else()
diff --git a/tools/prepare_for_build.sh b/tools/prepare_for_build.sh
new file mode 100644
index 000000000..10d1ec2c6
--- /dev/null
+++ b/tools/prepare_for_build.sh
@@ -0,0 +1,58 @@
+#!/bin/bash
+
+# Usage:
+# ./prepare_for_build.sh <torch_version> <cuda_version>
+set -ex
+
+# print system info:
+uname -a
+
+TORCH_VERSION="${1:?Specify torch version, e.g. 2.6.0}"
+CUDA_VERSION="${2:?Specify cuda version, e.g. cu118}"
+
+dnf install -y python3.11-devel python3.11-libs
+
+pip install --progress-bar off -q setuptools ninja wheel build
+pip install --progress-bar off -q torch=="${TORCH_VERSION}" --extra-index-url "https://download.pytorch.org/whl/${CUDA_VERSION}"
+python -c "import torch; print(f'PyTorch: {torch.__version__}\nCUDA: {torch.version.cuda}\nCXX11 ABI: {torch.compiled_with_cxx11_abi()}')"
+
+sed -i '1s/^/#if defined(__linux__) \&\& defined(__x86_64__)\n__asm__(".symver log,log@GLIBC_2.2.5");\n#endif\n/' third_party/METIS/GKlib/gk_proto.h
+sed -i '1s/^/#if defined(__linux__) \&\& defined(__x86_64__)\n__asm__(".symver pow,pow@GLIBC_2.2.5");\n#endif\n/' third_party/METIS/libmetis/metislib.h
+
+######################
+
+# Install CUDA 12.8
+# aria2c -s 16 -x 16 https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda-repo-rhel8-12-8-local-12.8.0_570.86.10-1.x86_64.rpm
+# rpm -i cuda-repo-rhel8-12-8-local-12.8.0_570.86.10-1.x86_64.rpm
+# dnf clean all && dnf -y install cuda-toolkit-12-8
+
+# Install CUDA 11.8
+if [ ! -f cuda-repo-rhel8-11-8-local-11.8.0_520.61.05-1.x86_64.rpm ]; then
+  dnf install -y aria2
+  aria2c -s 16 -x 16 https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda-repo-rhel8-11-8-local-11.8.0_520.61.05-1.x86_64.rpm
+else
+  echo "cuda-repo-rhel8-11-8-local-11.8.0_520.61.05-1.x86_64.rpm already exists"
+fi
+rpm -i cuda-repo-rhel8-11-8-local-11.8.0_520.61.05-1.x86_64.rpm
+dnf clean all
+dnf -y module install nvidia-driver:latest-dkms
+dnf -y install cuda
+
+ls -al /usr/local/ | grep -i cuda
+
+export CUDA_HOME=/usr/local/cuda-11.8
+export PATH=$CUDA_HOME/bin:$PATH
+export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH
+
+echo "CUDA_HOME: $CUDA_HOME"
+echo "PATH: $PATH"
+echo "LD_LIBRARY_PATH: $LD_LIBRARY_PATH"
+
+nvcc --version
+
+# source .github/workflows/cuda/Linux-env.sh cu118
+
+echo "CUDA_HOME: $CUDA_HOME"
+echo "PATH: $PATH"
+echo "LD_LIBRARY_PATH: $LD_LIBRARY_PATH"
+echo "TORCH_CUDA_ARCH_LIST: $TORCH_CUDA_ARCH_LIST"