|
| 1 | +name: Build Wheels (CUDA) |
| 2 | + |
| 3 | +on: workflow_dispatch |
| 4 | + |
| 5 | +permissions: |
| 6 | + contents: write |
| 7 | + |
| 8 | +jobs: |
| 9 | + define_matrix: |
| 10 | + name: Define Build Matrix |
| 11 | + runs-on: ubuntu-latest |
| 12 | + outputs: |
| 13 | + matrix: ${{ steps.set-matrix.outputs.matrix }} |
| 14 | + defaults: |
| 15 | + run: |
| 16 | + shell: pwsh |
| 17 | + |
| 18 | + steps: |
| 19 | + - name: Define Job Output |
| 20 | + id: set-matrix |
| 21 | + run: | |
| 22 | + $matrix = @{ |
| 23 | + 'os' = @('ubuntu-latest') |
| 24 | + 'pyver' = @("3.12") |
| 25 | + 'cuda' = @("12.6.1") |
| 26 | + 'releasetag' = @("basic") |
| 27 | + } |
| 28 | +
|
| 29 | + $matrixOut = ConvertTo-Json $matrix -Compress |
| 30 | + Write-Output ('matrix=' + $matrixOut) >> $env:GITHUB_OUTPUT |
| 31 | +
|
| 32 | + build_wheels: |
| 33 | + name: Build Wheel ${{ matrix.os }} ${{ matrix.pyver }} ${{ matrix.cuda }} ${{ matrix.releasetag == 'wheels' && 'AVX2' || matrix.releasetag }} |
| 34 | + needs: define_matrix |
| 35 | + runs-on: ${{ matrix.os }} |
| 36 | + strategy: |
| 37 | + matrix: ${{ fromJSON(needs.define_matrix.outputs.matrix) }} |
| 38 | + defaults: |
| 39 | + run: |
| 40 | + shell: pwsh |
| 41 | + env: |
| 42 | + CUDAVER: ${{ matrix.cuda }} |
| 43 | + AVXVER: ${{ matrix.releasetag }} |
| 44 | + |
| 45 | + steps: |
| 46 | + - name: Add MSBuild to PATH |
| 47 | + if: runner.os == 'Windows' |
| 48 | + uses: microsoft/setup-msbuild@v2 |
| 49 | + with: |
| 50 | + vs-version: '[16.11,16.12)' |
| 51 | + |
| 52 | + - uses: actions/checkout@v4 |
| 53 | + with: |
| 54 | + submodules: "recursive" |
| 55 | + |
| 56 | + - uses: actions/setup-python@v5 |
| 57 | + with: |
| 58 | + python-version: ${{ matrix.pyver }} |
| 59 | + cache: 'pip' |
| 60 | + |
| 61 | + - name: Setup Mamba |
| 62 | + uses: conda-incubator/setup-miniconda@v3.1.0 |
| 63 | + with: |
| 64 | + activate-environment: "llamacpp" |
| 65 | + python-version: ${{ matrix.pyver }} |
| 66 | + miniforge-version: latest |
| 67 | + add-pip-as-python-dependency: true |
| 68 | + auto-activate-base: false |
| 69 | + |
| 70 | + - name: VS Integration Cache |
| 71 | + id: vs-integration-cache |
| 72 | + if: runner.os == 'Windows' |
| 73 | + uses: actions/cache@v4 |
| 74 | + with: |
| 75 | + path: ./MSBuildExtensions |
| 76 | + key: cuda-${{ matrix.cuda }}-vs-integration |
| 77 | + |
| 78 | + - name: Get Visual Studio Integration |
| 79 | + if: runner.os == 'Windows' && steps.vs-integration-cache.outputs.cache-hit != 'true' |
| 80 | + run: | |
| 81 | + if ($env:CUDAVER -eq '12.1.1') {$x = '12.1.0'} else {$x = $env:CUDAVER} |
| 82 | + $links = (Invoke-RestMethod 'https://raw.githubusercontent.com/Jimver/cuda-toolkit/master/src/links/windows-links.ts').Trim().split().where({$_ -ne ''}) |
| 83 | + for ($i=$q=0;$i -lt $links.count -and $q -lt 2;$i++) {if ($links[$i] -eq "'$x',") {$q++}} |
| 84 | + Invoke-RestMethod $links[$i].Trim("'") -OutFile 'cudainstaller.zip' |
| 85 | + & 'C:\Program Files\7-Zip\7z.exe' e cudainstaller.zip -oMSBuildExtensions -r *\MSBuildExtensions\* > $null |
| 86 | + Remove-Item 'cudainstaller.zip' |
| 87 | +
|
| 88 | + - name: Install Visual Studio Integration |
| 89 | + if: runner.os == 'Windows' |
| 90 | + run: | |
| 91 | + $y = (gi '.\MSBuildExtensions').fullname + '\*' |
| 92 | + (gi 'C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\MSBuild\Microsoft\VC\*\BuildCustomizations').fullname.foreach({cp $y $_}) |
| 93 | + $cupath = 'CUDA_PATH_V' + $env:CUDAVER.Remove($env:CUDAVER.LastIndexOf('.')).Replace('.','_') |
| 94 | + echo "$cupath=$env:CONDA_PREFIX" >> $env:GITHUB_ENV |
| 95 | +
|
| 96 | + - name: Install Dependencies |
| 97 | + env: |
| 98 | + MAMBA_DOWNLOAD_FAILFAST: "0" |
| 99 | + MAMBA_NO_LOW_SPEED_LIMIT: "1" |
| 100 | + run: | |
| 101 | + $cudaVersion = $env:CUDAVER |
| 102 | + mamba install -y 'cuda' -c nvidia/label/cuda-$cudaVersion |
| 103 | + python -m pip install build wheel |
| 104 | +
|
| 105 | + - name: Build Wheel |
| 106 | + run: | |
| 107 | + $cudaVersion = $env:CUDAVER.Remove($env:CUDAVER.LastIndexOf('.')).Replace('.','') |
| 108 | + $env:CUDA_PATH = $env:CONDA_PREFIX |
| 109 | + $env:CUDA_HOME = $env:CONDA_PREFIX |
| 110 | + $env:CUDA_TOOLKIT_ROOT_DIR = $env:CONDA_PREFIX |
| 111 | + if ($IsLinux) { |
| 112 | + $env:LD_LIBRARY_PATH = $env:CONDA_PREFIX + '/lib:' + $env:LD_LIBRARY_PATH |
| 113 | + } |
| 114 | + $env:VERBOSE = '1' |
| 115 | + $env:CMAKE_ARGS = '-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=90' |
| 116 | + $env:CMAKE_ARGS = "-DGGML_CUDA_FORCE_MMQ=ON $env:CMAKE_ARGS" |
| 117 | + # if ($env:AVXVER -eq 'AVX') { |
| 118 | + $env:CMAKE_ARGS = $env:CMAKE_ARGS + ' -DGGML_AVX2=off -DGGML_FMA=off -DGGML_F16C=off' |
| 119 | + # } |
| 120 | + # if ($env:AVXVER -eq 'AVX512') { |
| 121 | + # $env:CMAKE_ARGS = $env:CMAKE_ARGS + ' -DGGML_AVX512=on' |
| 122 | + # } |
| 123 | + # if ($env:AVXVER -eq 'basic') { |
| 124 | + # $env:CMAKE_ARGS = $env:CMAKE_ARGS + ' -DGGML_AVX=off -DGGML_AVX2=off -DGGML_FMA=off -DGGML_F16C=off' |
| 125 | + # } |
| 126 | + python -m build --wheel |
| 127 | + # write the build tag to the output |
| 128 | + Write-Output "CUDA_VERSION=$cudaVersion" >> $env:GITHUB_ENV |
| 129 | +
|
| 130 | + - uses: softprops/action-gh-release@v2 |
| 131 | + with: |
| 132 | + files: dist/* |
| 133 | + # Set tag_name to <tag>-cu<cuda_version> |
| 134 | + tag_name: ${{ github.ref_name }}-cu${{ env.CUDA_VERSION }} |
| 135 | + env: |
| 136 | + GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} |
0 commit comments