CI: fix workflows #61
Workflow file for this run
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Python package | |
| on: | |
| push: | |
| branches: [ main ] | |
| pull_request: | |
| paths: | |
| - '.github/workflows/python-package.yml' | |
| - 'bitsandbytes/**' | |
| - 'csrc/**' | |
| - 'include/**' | |
| - 'tests/**' | |
| - 'CMakeLists.txt' | |
| - 'requirements*.txt' | |
| - 'setup.py' | |
| - 'pyproject.toml' | |
| - 'pytest.ini' | |
| - '**/*.md' | |
| release: | |
| branches: [ main ] | |
| types: [ published ] | |
| concurrency: | |
| group: cmake-${{ github.ref }} | |
| cancel-in-progress: true | |
| jobs: | |
| ## | |
| # This job matrix builds the non-CUDA versions of the libraries for all supported platforms. | |
| ## | |
| build-shared-libs: | |
| strategy: | |
| # Set fail-fast to false to ensure that feedback is delivered for all matrix combinations. Consider changing this to true when your workflow is stable. | |
| fail-fast: false | |
| matrix: | |
| os: [ubuntu-latest, macos-latest, windows-latest] | |
| arch: [x86_64, aarch64] | |
| build_type: [Release] | |
| exclude: | |
| - os: windows-latest # This probably requires arm64 Windows agents | |
| arch: aarch64 | |
| runs-on: ${{ matrix.os }} # One day, we could run them on native agents. Azure supports this now but it's planned only for Q3 2023 for hosted agents | |
| steps: | |
| # Check out code | |
| - uses: actions/checkout@v4 | |
| - name: Allow cross-compile on aarch64 | |
| if: ${{ matrix.os == 'ubuntu-latest' && matrix.arch == 'aarch64' }} | |
| run: | | |
| # Allow cross-compile on aarch64 | |
| sudo apt-get install -y gcc-aarch64-linux-gnu binutils-aarch64-linux-gnu | |
| - name: Setup cmake | |
| uses: jwlawson/[email protected] | |
| with: | |
| cmake-version: '3.26.x' | |
| - name: Set up MSVC | |
| if: startsWith(matrix.os, 'windows') | |
| uses: ilammy/[email protected] | |
| # Check out dependencies code | |
| - uses: actions/checkout@v4 | |
| name: Check out NVidia cub | |
| with: | |
| repository: nvidia/cub | |
| ref: 1.11.0 | |
| path: dependencies/cub | |
| - name: Set reusable strings | |
| # Turn repeated input strings (such as the build output directory) into step outputs. These step outputs can be used throughout the workflow file. | |
| id: strings | |
| shell: bash | |
| run: | | |
| echo "build-output-dir=${{ github.workspace }}/build" >> "$GITHUB_OUTPUT" | |
| - name: Prep Compilers | |
| shell: bash -el {0} | |
| run: | | |
| python3 -m pip install cmake==3.27.9 ninja | |
| if [ "${{ matrix.os }}" = "windows-latest" ]; then | |
| echo CXX_COMPILER=cl >> "$GITHUB_ENV" | |
| else | |
| echo CXX_COMPILER=g++ >> "$GITHUB_ENV" | |
| fi | |
| if [ "${{ matrix.os }}" = "macos-latest" -a "${{ matrix.arch }}" == "aarch64" ]; then | |
| echo DCMAKE_OSX_ARCHITECTURES=-DCMAKE_OSX_ARCHITECTURES=arm64 >> "$GITHUB_ENV" | |
| fi | |
| - name: Configure CPU | |
| run: > | |
| cmake -B ${{ steps.strings.outputs.build-output-dir }} | |
| -G Ninja ${{ env.DCMAKE_OSX_ARCHITECTURES }} | |
| -DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }} | |
| -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} | |
| -DCOMPUTE_BACKEND=cpu | |
| -S ${{ github.workspace }} | |
| - name: Build CPU | |
| run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }} | |
| - name: Copy libraries | |
| shell: bash | |
| run: | | |
| mkdir -p output/${{ matrix.os }}/${{ matrix.arch }} | |
| ( shopt -s nullglob && cp bitsandbytes/*.{so,dylib,dll} output/${{ matrix.os }}/${{ matrix.arch }}/ ) | |
| - name: Upload build artifact | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: shared_library_${{ matrix.os }}_${{ matrix.arch }} | |
| path: output/* | |
| ## | |
| # This job matrix builds the CUDA versions of the libraries for platforms that support CUDA (Windows) | |
| ## | |
| build-shared-libs-cuda: | |
| strategy: | |
| # Set fail-fast to false to ensure that feedback is delivered for all matrix combinations. Consider changing this to true when your workflow is stable. | |
| fail-fast: false | |
| matrix: | |
| os: [windows-latest] | |
| arch: [x86_64] | |
| cuda-version: ['11.8.0', '12.1.1'] | |
| build_type: [Release] | |
| runs-on: ${{ matrix.os }} # One day, we could run them on native agents. Azure supports this now but it's planned only for Q3 2023 for hosted agents | |
| steps: | |
| # Check out code | |
| - uses: actions/checkout@v4 | |
| # Linux: We use Docker to build cross platform Cuda (aarch64 is built in emulation) | |
| - name: Setup cmake | |
| uses: jwlawson/[email protected] | |
| with: | |
| cmake-version: '3.26.x' | |
| - name: Set up Python 3.10 | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: "3.10" | |
| - name: Set up MSVC | |
| if: startsWith(matrix.os, 'windows') | |
| uses: ilammy/[email protected] | |
| # Check out dependencies code | |
| - uses: actions/checkout@v4 | |
| name: Check out NVidia cub | |
| with: | |
| repository: nvidia/cub | |
| ref: 1.11.0 | |
| path: dependencies/cub | |
| - name: Setup Mambaforge | |
| uses: conda-incubator/[email protected] | |
| with: | |
| miniforge-variant: Mambaforge | |
| miniforge-version: latest | |
| activate-environment: bnb-env | |
| use-mamba: true | |
| - uses: conda-incubator/[email protected] | |
| with: | |
| auto-update-conda: true | |
| activate-environment: bnb-env | |
| environment-file: environment-bnb.yml | |
| use-only-tar-bz2: false | |
| auto-activate-base: true | |
| python-version: "3.10" | |
| mamba-version: "*" | |
| - name: Set reusable strings | |
| # Turn repeated input strings (such as the build output directory) into step outputs. These step outputs can be used throughout the workflow file. | |
| id: strings | |
| shell: bash | |
| run: | | |
| echo "build-output-dir=${{ github.workspace }}/build" >> "$GITHUB_OUTPUT" | |
| - name: Allow cross-compile on aarch64 | |
| if: ${{ matrix.os == 'ubuntu-latest' && matrix.arch == 'aarch64' }} | |
| run: | | |
| # Allow cross-compile on aarch64 | |
| sudo apt-get install -y gcc-aarch64-linux-gnu binutils-aarch64-linux-gnu | |
| - name: CUDA Toolkit | |
| shell: bash -el {0} | |
| run: | | |
| if [ "${{ matrix.os }}" = "ubuntu-latest" ]; then | |
| # to prepare space | |
| sudo rm -rf /usr/share/dotnet | |
| sudo rm -rf /opt/ghc | |
| sudo rm -rf /usr/local/share/boost | |
| fi | |
| addon="" | |
| cuda_version=${{ matrix.cuda-version }} | |
| [ "$cuda_version" = "12.1.1" ] && [ "${{ matrix.os }}" = "ubuntu-latest" ] && addon="cuda-cudart-static cuda-nvrtc" | |
| [ "$cuda_version" = "11.8.0" ] && [ "${{ matrix.os }}" = "windows-latest" ] && addon="cuda-nvrtc" | |
| [ "$cuda_version" = "11.8.0" ] && cuda__version="11.8" | |
| [ "$cuda_version" = "12.1.1" ] && cuda__version="12.1" | |
| conda install pytorch-cuda=$cuda__version -c pytorch # it's dependency not correctly resolved sometime | |
| conda install cuda-python=$cuda__version cuda-libraries-dev cuda-nvcc cuda-nvtx cuda-cupti cuda-cudart cuda-cudart-dev cuda-runtime cuda-libraries $addon -c "nvidia/label/cuda-$cuda_version" | |
| CUDA_HOME="${{ env.CONDA }}/envs/bnb-env" | |
| echo CUDA_HOME=$CUDA_HOME >> "$GITHUB_ENV" | |
| echo CUDA_PATH=$CUDA_HOME >> "$GITHUB_ENV" | |
| if [ "${{ matrix.os }}" = "windows-latest" ]; then | |
| echo CXX_COMPILER=cl >> "$GITHUB_ENV" | |
| # without -DCMAKE_CUDA_COMPILER=nvcc, cmake config always fail for cuda-11.8 | |
| echo DCMAKE_CUDA_COMPILER=-DCMAKE_CUDA_COMPILER=nvcc >> "$GITHUB_ENV" | |
| else | |
| echo CXX_COMPILER=g++ >> "$GITHUB_ENV" | |
| fi | |
| nvcc --version | |
| - name: Update environment | |
| run: mamba env update -n bnb-env -f environment-bnb.yml | |
| - name: Prep build | |
| run: python -m pip install cmake==3.27.9 ninja | |
| # TODO: the following steps (CUDA, NOBLASLT, CPU) could be moved to the matrix, so they're built in parallel | |
| - name: Configure CUDA | |
| run: > | |
| cmake -B ${{ steps.strings.outputs.build-output-dir }} | |
| -G Ninja ${{ env.DCMAKE_CUDA_COMPILER }} | |
| -DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }} | |
| -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} | |
| -DCOMPUTE_CAPABILITY="50;52;60;61;62;70;72;75;80;86;87;89;90" | |
| -DCOMPUTE_BACKEND=cuda | |
| -S ${{ github.workspace }} | |
| - name: Build CUDA | |
| run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }} | |
| - name: Configure NOBLASLT | |
| run: > | |
| cmake -B ${{ steps.strings.outputs.build-output-dir }} | |
| -G Ninja ${{ env.DCMAKE_CUDA_COMPILER }} | |
| -DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }} | |
| -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} | |
| -DCOMPUTE_CAPABILITY="50;52;60;61;62;70;72;75;80;86;87;89;90" | |
| -DCOMPUTE_BACKEND=cuda | |
| -DNO_CUBLASLT=ON | |
| -S ${{ github.workspace }} | |
| - name: Build NOBLASLT | |
| run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }} | |
| - name: Copy libraries | |
| shell: bash | |
| run: | | |
| mkdir -p output/${{ matrix.os }}/${{ matrix.arch }} | |
| ( shopt -s nullglob && cp bitsandbytes/*.{so,dylib,dll} output/${{ matrix.os }}/${{ matrix.arch }}/ ) | |
| - name: Upload build artifact | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: shared_library_cuda_${{ matrix.os }}_${{ matrix.arch }}_${{ matrix.cuda-version }} | |
| path: output/* | |
| ## | |
| # This job matrix builds the CUDA versions of the libraries for platforms that support CUDA (Linux x64/aarch64) | |
| ## | |
| build-shared-libs-cuda-docker: | |
| strategy: | |
| # Set fail-fast to false to ensure that feedback is delivered for all matrix combinations. Consider changing this to true when your workflow is stable. | |
| fail-fast: false | |
| matrix: | |
| os: [ubuntu-latest] | |
| arch: [x86_64, aarch64] | |
| cuda-version: ['11.8.0', '12.1.1'] | |
| build_type: [Release] | |
| exclude: | |
| - os: windows-latest # This probably requires arm64 Windows agents | |
| arch: aarch64 | |
| runs-on: ${{ matrix.os }} # One day, we could run them on native agents. Azure supports this now but it's planned only for Q3 2023 for hosted agents | |
| container: | |
| image: ${{ matrix.os == 'windows-latest' && 'null' || format('nvidia/cuda:{0}-{1}', matrix.cuda-version, 'devel-ubuntu22.04') }} | |
| volumes: | |
| - /home/runner/work:/home/runner/work | |
| steps: | |
| # Check out code | |
| - uses: actions/checkout@v4 | |
| # Linux: We use Docker to build cross platform Cuda (aarch64 is built in emulation) | |
| - name: Setup cmake | |
| uses: jwlawson/[email protected] | |
| with: | |
| cmake-version: '3.26.x' | |
| - name: Set up Python 3.10 | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: "3.10" | |
| - name: Setup Docker image | |
| if: startsWith(matrix.os, 'ubuntu') | |
| shell: bash | |
| run: | | |
| apt-get update | |
| DEBIAN_FRONTEND=noninteractive apt-get install -y sudo cmake | |
| - name: Set reusable strings | |
| # Turn repeated input strings (such as the build output directory) into step outputs. These step outputs can be used throughout the workflow file. | |
| id: strings | |
| shell: bash | |
| run: | | |
| echo "build-output-dir=${{ github.workspace }}/build" >> "$GITHUB_OUTPUT" | |
| - name: Allow cross-compile on aarch64 | |
| if: ${{ matrix.os == 'ubuntu-latest' && matrix.arch == 'aarch64' }} | |
| run: | | |
| # Allow cross-compile on aarch64 | |
| sudo apt-get install -y gcc-aarch64-linux-gnu binutils-aarch64-linux-gnu | |
| - name: CUDA Toolkit | |
| shell: bash -el {0} | |
| run: | | |
| cuda_version=${{ matrix.cuda-version }} | |
| [ "$cuda_version" = "11.8.0" ] && cuda__version="11.8" | |
| [ "$cuda_version" = "12.1.1" ] && cuda__version="12.1" | |
| CUDA_HOME="${{ env.CONDA }}/envs/bnb-env" | |
| echo CUDA_HOME=$CUDA_HOME >> "$GITHUB_ENV" | |
| echo CUDA_PATH=$CUDA_HOME >> "$GITHUB_ENV" | |
| echo CXX_COMPILER=g++ >> "$GITHUB_ENV" | |
| nvcc --version | |
| - name: Prep build | |
| run: python -m pip install cmake==3.27.9 ninja | |
| # TODO: the following steps (CUDA, NOBLASLT, CPU) could be moved to the matrix, so they're built in parallel | |
| - name: Configure CUDA | |
| run: > | |
| cmake -B ${{ steps.strings.outputs.build-output-dir }} | |
| -G Ninja ${{ env.DCMAKE_CUDA_COMPILER }} | |
| -DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }} | |
| -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} | |
| -DCOMPUTE_CAPABILITY="50;52;60;61;62;70;72;75;80;86;87;89;90" | |
| -DCOMPUTE_BACKEND=cuda | |
| -S ${{ github.workspace }} | |
| - name: Build CUDA | |
| run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }} | |
| - name: Configure NOBLASLT | |
| run: > | |
| cmake -B ${{ steps.strings.outputs.build-output-dir }} | |
| -G Ninja ${{ env.DCMAKE_CUDA_COMPILER }} | |
| -DCMAKE_CXX_COMPILER=${{ env.CXX_COMPILER }} | |
| -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} | |
| -DCOMPUTE_CAPABILITY="50;52;60;61;62;70;72;75;80;86;87;89;90" | |
| -DCOMPUTE_BACKEND=cuda | |
| -DNO_CUBLASLT=ON | |
| -S ${{ github.workspace }} | |
| - name: Build NOBLASLT | |
| run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }} | |
| - name: Copy libraries | |
| shell: bash | |
| run: | | |
| mkdir -p output/${{ matrix.os }}/${{ matrix.arch }} | |
| ( shopt -s nullglob && cp bitsandbytes/*.{so,dylib,dll} output/${{ matrix.os }}/${{ matrix.arch }}/ ) | |
| - name: Upload build artifact | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: shared_library_cuda_${{ matrix.os }}_${{ matrix.arch }}_${{ matrix.cuda-version }} | |
| path: output/* | |
| build-wheels: | |
| needs: | |
| - build-shared-libs | |
| - build-shared-libs-cuda | |
| - build-shared-libs-cuda-docker | |
| strategy: | |
| matrix: | |
| os: [ubuntu-latest, macos-latest, windows-latest] | |
| arch: [x86_64, aarch64] | |
| exclude: | |
| - os: windows-latest # This probably requires arm64 Windows agents | |
| arch: aarch64 | |
| runs-on: ${{ matrix.os }} | |
| steps: | |
| # Check out code | |
| - uses: actions/checkout@v4 | |
| # Download shared libraries | |
| - name: Download build artifact | |
| uses: actions/download-artifact@v4 | |
| with: | |
| merge-multiple: true | |
| pattern: "shared_library*_${{ matrix.os }}_${{ matrix.arch }}*" | |
| path: output/ | |
| - name: Copy correct platform shared library | |
| shell: bash | |
| run: | | |
| ls -lR output/ | |
| cp output/${{ matrix.os }}/${{ matrix.arch }}/* bitsandbytes/ | |
| # Set up the Python version needed | |
| - name: Set up Python 3.10 | |
| uses: actions/setup-python@v5 | |
| with: | |
| python-version: "3.10" | |
| cache: pip | |
| - name: Install build package | |
| shell: bash | |
| run: pip install build | |
| - name: Install Python test dependencies | |
| shell: bash | |
| run: pip install -r requirements-ci.txt | |
| # TODO: How to run CUDA tests on GitHub actions? | |
| #- name: Run unit tests | |
| # if: ${{ matrix.arch == 'x86_64' }} # Tests are too slow to run in emulation. Wait for real aarch64 agents | |
| # run: | | |
| # PYTHONPATH=. pytest --log-cli-level=DEBUG tests | |
| - name: Build wheel | |
| shell: bash | |
| run: python -m build . --wheel | |
| - name: Upload build artifact | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: bdist_wheel_${{ matrix.os }}-${{ matrix.arch }} | |
| path: ${{ github.workspace }}/dist/bitsandbytes-*.whl | |
| publish: | |
| needs: build-wheels | |
| runs-on: ubuntu-latest | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Download build artifact | |
| uses: actions/download-artifact@v4 | |
| with: | |
| path: dist/ | |
| merge-multiple: true | |
| pattern: "bdist_wheel_*" | |
| - run: | | |
| ls -lR dist/ | |
| - name: Publish to PyPi | |
| if: startsWith(github.ref, 'refs/tags') | |
| uses: pypa/gh-action-pypi-publish@release/v1 | |
| with: | |
| password: ${{ secrets.pypi }} |