diff --git a/.github/workflows/integration-test-backup.yml b/.github/workflows/integration-test-backup.yml
new file mode 100644
index 000000000..ffc5fe3af
--- /dev/null
+++ b/.github/workflows/integration-test-backup.yml
@@ -0,0 +1,75 @@
+name: IntegrationTest
+
+on: workflow_dispatch
+
+jobs:
+  IntegrationTest:
+    runs-on: self-hosted
+    strategy:
+      matrix:
+        container-image: [ghcr.io/microsoft/mscclpp/mscclpp:base-cuda11.8, ghcr.io/microsoft/mscclpp/mscclpp:base-cuda12.1]
+
+    container:
+      image: ${{ matrix.container-image }}
+      options: --privileged --ipc=host --gpus=all --ulimit memlock=-1:-1
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v2
+
+      - name: Install CMake
+        run: |
+          curl -L https://github.com/Kitware/CMake/releases/download/v3.26.4/cmake-3.26.4-linux-x86_64.tar.gz -o /tmp/cmake-3.26.4-linux-x86_64.tar.gz
+          tar xzf /tmp/cmake-3.26.4-linux-x86_64.tar.gz -C /tmp
+
+      - name: Build
+        run: |
+          mkdir build && cd build
+          MPI_HOME=/usr/local/mpi /tmp/cmake-3.26.4-linux-x86_64/bin/cmake -DCMAKE_BUILD_TYPE=Release ..
+          make -j
+
+      - name: Lock GPU clock frequency
+        run: |
+          sudo nvidia-smi -pm 1
+          for i in $(seq 0 $(( $(nvidia-smi -L | wc -l) - 1 ))); do
+            sudo nvidia-smi -ac $(nvidia-smi --query-gpu=clocks.max.memory,clocks.max.sm --format=csv,noheader,nounits -i $i | sed 's/\ //') -i $i
+          done
+
+      - name: Run mscclpp AllGather test
+        run: |
+          set -e
+          export PATH=/usr/local/mpi/bin:$PATH
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 2 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 3 -o output.jsonl
+
+      - name: Run mscclpp SendRecv test
+        run: |
+          set -e
+          export PATH=/usr/local/mpi/bin:$PATH
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/sendrecv_test_perf -b 1K -e 1G -f 2 -o output.jsonl
+
+      - name: Run mscclpp AllReduce test
+        run: |
+          set -e
+          export PATH=/usr/local/mpi/bin:$PATH
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 2 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 3 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 4 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 12M -e 48M -i 3145728 2 -k 5 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 24K -e 768K -i 24576 -k 6 -w 100 -n 100 -o output.jsonl
+
+      - name: Run mscclpp AllToAll test
+        run: |
+          set -e
+          export PATH=/usr/local/mpi/bin:$PATH
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -o output.jsonl
+          mpirun -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl
+
+      - name: Check collective primitives performance
+        run: |
+          set -e
+          python3 test/mscclpp-test/check_perf_result.py --perf-file output.jsonl --baseline-file test/deploy/perf_ndmv4.jsonl
diff --git a/.github/workflows/lint.yml b/.github/workflows/lint.yml
index f06e1272d..9cdcf443d 100644
--- a/.github/workflows/lint.yml
+++ b/.github/workflows/lint.yml
@@ -38,12 +38,8 @@ jobs:
       - name: Install Python dependencies
         run: python3.8 -m pip install black
 
-      - name: Run linters
-        uses: wearerequired/lint-action@v2
-        with:
-          black: true
-          black_auto_fix: false
-          black_args: "--config pyproject.toml --check"
+      - name: Run black
+        run: python3.8 -m black --check --config pyproject.toml .
 
   spelling:
     runs-on: ubuntu-20.04
diff --git a/.github/workflows/ut-backup.yml b/.github/workflows/ut-backup.yml
new file mode 100644
index 000000000..736c800e7
--- /dev/null
+++ b/.github/workflows/ut-backup.yml
@@ -0,0 +1,63 @@
+name: UnitTest
+
+on: workflow_dispatch
+
+jobs:
+  UnitTest:
+    runs-on: self-hosted
+    timeout-minutes: 30
+    strategy:
+      matrix:
+        container-image: [ghcr.io/microsoft/mscclpp/mscclpp:base-cuda11.8, ghcr.io/microsoft/mscclpp/mscclpp:base-cuda12.1]
+
+    container:
+      image: ${{ matrix.container-image }}
+      options: --privileged --ipc=host --gpus=all --ulimit memlock=-1:-1
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v2
+
+      - name: Build
+        run: |
+          curl -L -C- https://github.com/Kitware/CMake/releases/download/v3.26.4/cmake-3.26.4-linux-x86_64.tar.gz -o /tmp/cmake-3.26.4-linux-x86_64.tar.gz
+          tar xzf /tmp/cmake-3.26.4-linux-x86_64.tar.gz -C /tmp
+          mkdir build && cd build
+          MPI_HOME=/usr/local/mpi /tmp/cmake-3.26.4-linux-x86_64/bin/cmake -DCMAKE_BUILD_TYPE=Release ..
+          make -j
+        working-directory: ${{ github.workspace }}
+
+      - name: LockGPUClock
+        run: |
+          sudo nvidia-smi -pm 1
+          for i in $(seq 0 $(( $(nvidia-smi -L | wc -l) - 1 ))); do
+            sudo nvidia-smi -ac $(nvidia-smi --query-gpu=clocks.max.memory,clocks.max.sm --format=csv,noheader,nounits -i $i | sed 's/\ //') -i $i
+          done
+        working-directory: ${{ github.workspace }}
+
+      - name: UnitTests
+        run: |
+          ./build/test/unit_tests
+        working-directory: ${{ github.workspace }}
+
+      - name: MpUnitTests
+        run: |
+          set -e
+          export PATH=/usr/local/mpi/bin:$PATH
+          mpirun -tag-output -np 2 ./build/test/mp_unit_tests
+          mpirun -tag-output -np 4 ./build/test/mp_unit_tests
+          mpirun -tag-output -np 8 ./build/test/mp_unit_tests
+        working-directory: ${{ github.workspace }}
+
+      - name: PyTests
+        run: |
+          set -e
+          export PATH=/usr/local/mpi/bin:$PATH
+          cd build && make pylib-copy
+          if [[ '${{ matrix.container-image }}' == *'cuda11'* ]]; then
+            pip3 install -r ../python/test/requirements_cu11.txt
+          else
+            pip3 install -r ../python/test/requirements_cu12.txt
+          fi
+          mpirun -tag-output -np 8 ~/.local/bin/pytest ../python/test/test_mscclpp.py -x
+        working-directory: ${{ github.workspace }}