diff --git a/.github/actions/dependencies-action/action.yml b/.github/actions/dependencies-action/action.yml
deleted file mode 100644
index 051472c..0000000
--- a/.github/actions/dependencies-action/action.yml
+++ /dev/null
@@ -1,92 +0,0 @@
-# An action for installing dependencies. You can set
-# custom pip dependencies as following:
-#
-# Usage:
-#
-# - name: Install dependencies
-#   uses: ./.github/actions/dependencies-action
-#   with:
-#     pip_packages: |
-#       - torch
-#       - triton
-#     pip_requirements: |
-#       - /path/a/requirements.txt
-#       - /path/b/requirements.txt --no-deps
-
-name: Install dependencies
-description: Install OS and pip dependencies
-
-inputs:
-  pip_packages:
-    description: List of pip packages to install
-    required: false
-    default: ''
-  pip_requirements:
-    description: List of requirements files to install
-    required: false
-    default: ''
-
-runs:
-  using: composite
-  steps:
-    # TODO(shink): Remove this step after building devel images
-    - name: Install system dependencies
-      shell: bash
-      env:
-        DEBIAN_FRONTEND: noninteractive
-      run: |
-        echo "::group::Install system dependencies"
-        sed -i 's|ports.ubuntu.com|mirrors.tuna.tsinghua.edu.cn|g' /etc/apt/sources.list
-        apt update
-        apt install --no-install-recommends -y \
-            git \
-            gcc-10 \
-            g++-10 \
-            make \
-            cmake \
-            ninja-build
-        echo "::endgroup::"
-
-    - name: Create symlinks for gcc and g++
-      shell: bash
-      run: |
-        ln -s /usr/bin/gcc-10 /usr/bin/gcc
-        ln -s /usr/bin/g++-10 /usr/bin/g++
-
-    - name: Show versions
-      shell: bash
-      run: |
-        set -x
-        python --version
-        pip --version
-        gcc --version
-        g++ --version
-        make --version
-        cmake --version
-        ninja --version
-
-    - name: Install extra pip packages
-      if: ${{ inputs.pip_packages != '' }}
-      shell: bash
-      run: |
-        echo "${{ inputs.pip_packages }}" | while read -r package; do
-          package="${package#- }"
-          if [ -n "$(echo "$package" | xargs)" ]; then
-            echo "::group::Installing pip package: $package"
-            pip install $package -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
-            echo "::endgroup::"
-          fi
-        done
-
-    - name: Install extra pip requirements
-      if: ${{ inputs.pip_requirements != '' }}
-      shell: bash
-      run: |
-        echo "${{ inputs.pip_requirements }}" | while read -r requirement; do
-          requirement="${requirement#- }"
-          if [ -n "$(echo "$requirement" | xargs)" ]; then
-            echo "::group::Installing from requirements: $requirement"
-            pip install -r $requirement -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
-            echo "::endgroup::"
-          fi
-        done
diff --git a/.github/actions/fetch-and-rebase/action.yml b/.github/actions/fetch-and-rebase/action.yml
deleted file mode 100644
index fa72a6a..0000000
--- a/.github/actions/fetch-and-rebase/action.yml
+++ /dev/null
@@ -1,51 +0,0 @@
-name: Fetch and Rebase
-
-description: Fetch and rebase for repository
-
-inputs:
-  repo_path:
-    description: the path the repository locate on
-    required: true
-  remote_branch:
-    description: the branch rebased from
-    required: true
-  loop:
-    description: loop times when fails
-    required: true
-
-runs:
-  using: composite
-  steps:
-    - name: Fetch and rebase
-      env:
-        REPO_PATH: ${{ inputs.repo_path }}
-        REMOTE_BRANCH: ${{ inputs.remote_branch }}
-        LOOP: ${{ inputs.loop }}
-      shell: bash
-      run: |
-        set +e
-
-        COUNT=0
-
-        for i in $(seq 1 ${LOOP})
-        do
-          pushd ${REPO_PATH} &&
-          git fetch --all &&
-          git rebase ${REMOTE_BRANCH} &&
-          git submodule sync &&
-          git submodule update --init --recursive &&
-          git reset --hard HEAD &&
-          git clean -dfx &&
-          git submodule foreach git reset --hard HEAD &&
-          git submodule foreach git clean -dfx &&
-          popd
-
-          if [[ $? -ne 0 ]]
-          then
-            let COUNT++
-          else
-            break
-          fi
-        done
-
-        [[ ${COUNT} -lt ${LOOP} ]] && true || false
diff --git a/.github/workflows/_ascend_npu_build.yml b/.github/workflows/_ascend_npu_build.yml
new file mode 100644
index 0000000..dfcfc5e
--- /dev/null
+++ b/.github/workflows/_ascend_npu_build.yml
@@ -0,0 +1,98 @@
+name: '_ascend_npu_build'
+
+on:
+  workflow_call:
+    inputs:
+      runner:
+        required: true
+        type: string
+        description: 'The runner selected to run on'
+      image:
+        required: true
+        type: string
+        description: 'The docker image which will be used to build'
+    outputs:
+      artifact_name:
+        description: 'The torch_npu distribution artifact name'
+        value: ${{ jobs.build.outputs.dist_name }}
+
+# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
+# declared as "shell: bash -el {0}" on steps that need to be properly activated.
+# It's used to activate ascend-toolkit environment variables.
+defaults:
+  run:
+    shell: bash -el {0}
+
+jobs:
+  build:
+    name: build torch_npu
+    runs-on: ${{ inputs.runner }}
+    container:
+      image: ${{ inputs.image }}
+    outputs:
+      dist_name: ${{ steps.list-dist.outputs.dist_name }}
+    steps:
+      # TODO(shink): Should we add these dependencies to the image?
+      - name: Install system dependencies
+        run: |
+          sed -i 's|ports.ubuntu.com|mirrors.tuna.tsinghua.edu.cn|g' /etc/apt/sources.list
+          apt update
+          apt install --no-install-recommends -y git gcc g++ make cmake ninja-build
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Checkout torch_npu
+        uses: actions/checkout@v4
+        with:
+          # TODO(shink): Use Ascend/pytorch once this pr merged:
+          # https://gitee.com/ascend/pytorch/pulls/12854
+          # repository: Ascend/pytorch
+          repository: shink/torchnpu
+          ref: feat/autoload
+          submodules: recursive
+          path: torch_npu
+
+      - name: Install pip dependencies
+        working-directory: torch_npu
+        run: |
+          pip install wheel
+          pip install -r requirements.txt
+
+      - name: List Python version
+        id: list-py-version
+        working-directory: torch_npu
+        run: |
+          py_version=$(python --version | awk '{print $2}' | cut -d '.' -f 1,2)
+          echo "py_version=${py_version}" >> $GITHUB_OUTPUT
+
+      - name: Build torch_npu
+        working-directory: torch_npu
+        run: |
+          bash ci/build.sh --python=${{ steps.list-py-version.outputs.py_version }}
+
+      - name: List distribution package
+        id: list-dist
+        working-directory: torch_npu/dist
+        run: |
+          dist_name=$(ls torch_npu*.whl)
+          dist_path=$(pwd)/${dist_name}
+          echo "dist_name=${dist_name}" >> $GITHUB_OUTPUT
+          echo "dist_path=${dist_path}" >> $GITHUB_OUTPUT
+
+      - name: Upload distribution artifact
+        id: upload-dist
+        continue-on-error: true
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ steps.list-dist.outputs.dist_name }}
+          path: ${{ steps.list-dist.outputs.dist_path }}
+          if-no-files-found: error
+          retention-days: 1
+          overwrite: true
+
+      - name: Write to workflow job summary
+        if: ${{ steps.upload-dist.outputs.artifact-url }}
+        run: |
+          echo "## torch_npu built successfully! :rocket:" >> $GITHUB_STEP_SUMMARY
+          echo "You can download the distribution package [here](${{ steps.upload-dist.outputs.artifact-url }})." >> $GITHUB_STEP_SUMMARY
diff --git a/.github/workflows/_ascend_npu_test.yml b/.github/workflows/_ascend_npu_test.yml
new file mode 100644
index 0000000..37932ce
--- /dev/null
+++ b/.github/workflows/_ascend_npu_test.yml
@@ -0,0 +1,95 @@
+name: '_ascend_npu_test'
+
+on:
+  workflow_call:
+    inputs:
+      runner:
+        required: true
+        type: string
+        description: 'The runner selected to run on'
+      image:
+        required: true
+        type: string
+        description: 'The docker image which will be loaded'
+      device:
+        required: true
+        type: string
+        description: 'The device selected to run on'
+      artifact_name:
+        required: true
+        type: string
+        description: 'The torch_npu distribution artifact name'
+
+# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
+# declared as "shell: bash -el {0}" on steps that need to be properly activated.
+# It's used to activate ascend-toolkit environment variables.
+defaults:
+  run:
+    shell: bash -el {0}
+
+jobs:
+  test:
+    name: test torch_npu in ${{ inputs.image }} with ${{ inputs.device }}
+    runs-on: ${{ inputs.runner }}
+    container:
+      image: ${{ inputs.image }}
+      volumes:
+        - /usr/local/dcmi:/usr/local/dcmi
+        - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
+        - /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/
+        - /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info
+        - /etc/ascend_install.info:/etc/ascend_install.info
+      options: >-
+        --network host
+        --device ${{ inputs.device }}
+        --device /dev/davinci_manager
+        --device /dev/devmm_svm
+        --device /dev/hisi_hdc
+    steps:
+      - name: Show NPU info
+        run: |
+          npu-smi info
+
+      - name: Install system dependencies
+        run: |
+          sed -i 's|ports.ubuntu.com|mirrors.tuna.tsinghua.edu.cn|g' /etc/apt/sources.list
+          apt update
+          apt install --no-install-recommends -y git gcc g++ make cmake ninja-build
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Checkout torch_npu
+        uses: actions/checkout@v4
+        with:
+          # TODO(shink): Use Ascend/pytorch once this pr merged:
+          # https://gitee.com/ascend/pytorch/pulls/12854
+          # repository: Ascend/pytorch
+          repository: shink/torchnpu
+          ref: feat/autoload
+          submodules: recursive
+          path: torch_npu
+
+      - name: Download distribution artifact
+        uses: actions/download-artifact@v4
+        with:
+          name: ${{ inputs.artifact_name }}
+          path: torch_npu
+
+      - name: Install pip dependencies
+        working-directory: torch_npu
+        run: |
+          pip install wheel unittest-xml-reporting
+          pip install -r requirements.txt
+          pip install -r test/requirements.txt --no-deps
+
+      - name: Install torch_npu
+        working-directory: torch_npu
+        run: |
+          pip install ${{ inputs.artifact_name }}
+
+      # TODO(shink): Skip
+      - name: Do the test
+        continue-on-error: true
+        run: |
+          python torch_npu/ci/access_control_test.py
diff --git a/.github/workflows/_build-and-test.yml b/.github/workflows/_build-and-test.yml
deleted file mode 100644
index 3d71a13..0000000
--- a/.github/workflows/_build-and-test.yml
+++ /dev/null
@@ -1,153 +0,0 @@
-name: 'build-test'
-
-on:
-  workflow_call:
-    inputs:
-      runner:
-        required: true
-        type: string
-        description: The runner selected to run on
-      device:
-        required: true
-        type: string
-        description: The device selected to run on
-      image:
-        required: true
-        type: string
-        description: The docker image which will be loaded
-
-# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
-# declared as "shell: bash -el {0}" on steps that need to be properly activated.
-# It's used to activate ascend-toolkit environment variables.
-defaults:
-  run:
-    shell: bash -el {0}
-
-jobs:
-  build:
-    name: build torch_npu
-    runs-on: ${{ inputs.runner }}
-    container:
-      image: ${{ inputs.image }}
-      volumes:
-        - /home/runner/actions-runner/codes:/root/codes
-    outputs:
-      dist_name: ${{ steps.list-dist.outputs.dist_name }}
-    steps:
-      - name: Prepare the codes
-        run: |
-          cp -rf /root/codes /root/build
-
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Install dependencies
-        uses: ./.github/actions/dependencies-action
-        with:
-          pip_packages: |
-            - wheel
-          pip_requirements: |
-            - /root/build/npu/pytorch/requirements.txt
-
-      - name: Build torch_npu
-        working-directory: /root/build/npu/pytorch
-        run: |
-          py_version=$(python --version | awk '{print $2}' | cut -d '.' -f 1,2)
-          bash ci/build.sh --python=${py_version}
-
-      - name: List distribution package
-        id: list-dist
-        working-directory: /root/build/npu/pytorch/dist
-        run: |
-          dist_name=$(ls torch_npu*.whl)
-          dist_path=$(pwd)/${dist_name}
-          echo "dist_name=${dist_name}" >> $GITHUB_OUTPUT
-          echo "dist_path=${dist_path}" >> $GITHUB_OUTPUT
-
-      - name: Upload distribution artifact
-        id: upload-dist
-        continue-on-error: true
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ steps.list-dist.outputs.dist_name }}
-          path: ${{ steps.list-dist.outputs.dist_path }}
-          retention-days: 1
-
-      - name: Write to workflow job summary
-        if: ${{ steps.upload-dist.outputs.artifact-url }}
-        run: |
-          echo "## torch_npu built successfully! :rocket:" >> $GITHUB_STEP_SUMMARY
-          echo "You can download the distribution package [here](${{ steps.upload-dist.outputs.artifact-url }})." >> $GITHUB_STEP_SUMMARY
-
-  test:
-    name: test torch_npu in ${{ inputs.image }} with ${{ inputs.device }}
-    runs-on: ${{ inputs.runner }}
-    needs:
-      - build
-    container:
-      image: ${{ inputs.image }}
-      volumes:
-        - /usr/local/dcmi:/usr/local/dcmi
-        - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
-        - /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/
-        - /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info
-        - /etc/ascend_install.info:/etc/ascend_install.info
-        - /home/runner/actions-runner/codes:/root/codes
-      options: >-
-        --network host
-        --device ${{ inputs.device }}
-        --device /dev/davinci_manager
-        --device /dev/devmm_svm
-        --device /dev/hisi_hdc
-    steps:
-      - name: Show NPU info
-        run: |
-          npu-smi info
-
-      - name: Prepare the codes
-        run: |
-          cp -rf /root/codes /root/build
-
-      - name: Download distribution artifact
-        uses: actions/download-artifact@v4
-        with:
-          name: ${{ needs.build.outputs.dist_name }}
-          path: /root/build
-
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Install dependencies
-        uses: ./.github/actions/dependencies-action
-        with:
-          pip_packages: |
-            - wheel
-            - unittest-xml-reporting
-            - importlib-metadata
-          pip_requirements: |
-            - /root/build/npu/pytorch/requirements.txt
-            - /root/build/npu/pytorch/test/requirements.txt --no-deps
-
-      - name: Install torch_npu
-        working-directory: /root/build
-        run: |
-          pip install ${{ needs.build.outputs.dist_name }}
-
-      # TODO(shink): Skip
-      - name: Do the test
-        continue-on-error: true
-        working-directory: /root/build
-        run: |
-          python npu/pytorch/ci/access_control_test.py
-        env:
-          DISABLED_TESTS_FILE: /root/build/npu/pytorch/test/unsupported_test_cases/.pytorch-disabled-tests.json
-
-      - name: Train GPT2
-        working-directory: ./test
-        run: |
-          pip install -r requirements.txt
-          pip install accelerate -U
-          python gpt2_test.py
-        env:
-          IS_CI: true
-          HF_ENDPOINT: https://hf-mirror.com
diff --git a/.github/workflows/ascend_npu_test.yml b/.github/workflows/ascend_npu_test.yml
index 95711c7..f0e226b 100644
--- a/.github/workflows/ascend_npu_test.yml
+++ b/.github/workflows/ascend_npu_test.yml
@@ -1,6 +1,3 @@
-# Note:
-# Same runner only needs one job named like .*fetch-and-rebase
-
 name: Ascend NPU Test Suite
 
 on:
@@ -29,6 +26,16 @@ on:
           - npu-arm64
         default: 'self-hosted'
         description: 'The runner selected to run on'
+      image:
+        required: true
+        type: choice
+        options:
+          - ascendai/cann:7.1-openeuler2203sp2
+          - ascendai/cann:8.0.rc2.alpha003-910b-ubuntu22.04-py3.9
+          - ascendai/cann:8.0.rc3.alpha002-910b-ubuntu22.04-py3.9
+          - ascendai/cann:latest
+        default: 'ascendai/cann:latest'
+        description: 'The docker image which will be loaded'
       device:
         required: true
         type: choice
@@ -43,19 +50,10 @@ on:
           - /dev/davinci8
         default: '/dev/davinci6'
         description: 'The device selected to run on'
-      image:
-        required: true
-        type: choice
-        options:
-          - ascendai/cann:7.1-openeuler2203sp2
-          - ascendai/cann:8.0.rc2.alpha003-910b-ubuntu22.04-py3.9
-          - ascendai/cann:8.0.rc3.alpha002-910b-ubuntu22.04-py3.9
-        default: 'ascendai/cann:8.0.rc3.alpha002-910b-ubuntu22.04-py3.9'
-        description: 'The docker image which will be loaded'
 
 # Only cancel the previous runs when triggered by a pull request
 concurrency:
-  group: '${{ github.workflow }}-${{ github.event_name }}'
+  group: ${{ github.workflow }}-${{ github.event_name }}-${{ github.ref }}
   cancel-in-progress: ${{ github.event_name == 'pull_request' }}
 
 jobs:
@@ -64,8 +62,8 @@ jobs:
     runs-on: ubuntu-latest
     outputs:
       runner: ${{ steps.set-param.outputs.runner }}
-      device: ${{ steps.set-param.outputs.device }}
       image: ${{ steps.set-param.outputs.image }}
+      device: ${{ steps.set-param.outputs.device }}
     steps:
       - name: Set param
         id: set-param
@@ -73,28 +71,25 @@ jobs:
           set -e
           echo "runner=${{ github.event.inputs.runner || 'self-hosted' }}" >> $GITHUB_OUTPUT
           echo "device=${{ github.event.inputs.device || '/dev/davinci6' }}" >> $GITHUB_OUTPUT
-          echo "image=${{ github.event.inputs.image || 'ascendai/cann:8.0.rc3.alpha002-910b-ubuntu22.04-py3.9' }}" >> $GITHUB_OUTPUT
+          echo "image=${{ github.event.inputs.image || 'ascendai/cann:latest' }}" >> $GITHUB_OUTPUT
 
-  fetch-and-rebase:
-    name: Fetch and rebase
-    runs-on: ${{ needs.prepare.outputs.runner }}
+  build:
+    name: Build torch_npu
     needs:
       - prepare
-    steps:
-      - name: Pull latest codes for torch_npu
-        uses: ./.github/actions/fetch-and-rebase
-        with:
-          repo_path: /home/runner/actions-runner/codes/npu/pytorch
-          remote_branch: upstream/master
-          loop: 10
+    uses: ./.github/workflows/_ascend_npu_build.yml
+    with:
+      runner: ${{ needs.prepare.outputs.runner }}
+      image: ${{ needs.prepare.outputs.image }}
 
-  build-and-test:
-    name: Build and test
+  test:
+    name: Test torch_npu
     needs:
       - prepare
-      - fetch-and-rebase
-    uses: ./.github/workflows/_build-and-test.yml
+      - build
+    uses: ./.github/workflows/_ascend_npu_test.yml
     with:
       runner: ${{ needs.prepare.outputs.runner }}
-      device: ${{ needs.prepare.outputs.device }}
       image: ${{ needs.prepare.outputs.image }}
+      device: ${{ needs.prepare.outputs.device }}
+      artifact_name: ${{ needs.build.outputs.artifact_name }}