|
| 1 | +name: Build Wheels & Release |
| 2 | + |
| 3 | +on: |
| 4 | + workflow_dispatch: |
| 5 | + inputs: |
| 6 | + release: |
| 7 | + description: 'Release? 1 = yes, 0 = no' |
| 8 | + default: '0' |
| 9 | + required: true |
| 10 | + type: string |
| 11 | + |
| 12 | +permissions: |
| 13 | + contents: write |
| 14 | + |
| 15 | +jobs: |
| 16 | + build_wheels: |
| 17 | + name: ${{ matrix.os }} P${{ matrix.pyver }} C${{ matrix.cuda }} R${{ matrix.rocm }} T${{ matrix.torch }} |
| 18 | + runs-on: ${{ matrix.os }} |
| 19 | + strategy: |
| 20 | + matrix: |
| 21 | + include: |
| 22 | + |
| 23 | + # Ubuntu 20.04 CUDA |
| 24 | + |
| 25 | + # Python 3.10 |
| 26 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 27 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 28 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 29 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 30 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 31 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 32 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 33 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 34 | + - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 35 | + |
| 36 | + # Python 3.11 |
| 37 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 38 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '12.1.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 39 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 40 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '12.1.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 41 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 42 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 43 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 44 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 45 | + - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 46 | + |
| 47 | + # Python 3.12 |
| 48 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 49 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '12.1.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 50 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 51 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '12.1.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 52 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 53 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 54 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 55 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 56 | + - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 57 | + |
| 58 | + # Python 3.13 |
| 59 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.13', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 60 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.13', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 61 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.13', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 62 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.13', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 63 | + - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.13', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 64 | + |
| 65 | + # Windows 2022 CUDA |
| 66 | + |
| 67 | + # Python 3.10 |
| 68 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 69 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 70 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 71 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 72 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 73 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 74 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 75 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 76 | + - { artname: 'wheel', os: windows-2022, pyver: '3.10', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 77 | + |
| 78 | + # Python 3.11 |
| 79 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 80 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '12.1.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 81 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 82 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '12.1.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 83 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 84 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 85 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 86 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 87 | + - { artname: 'wheel', os: windows-2022, pyver: '3.11', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 88 | + |
| 89 | + # Python 3.12 |
| 90 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 91 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '12.1.0', rocm: '', torch: '2.3.1', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 92 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 93 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '12.1.0', rocm: '', torch: '2.4.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 94 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 95 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '12.1.0', rocm: '', torch: '2.5.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 96 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 97 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 98 | + - { artname: 'wheel', os: windows-2022, pyver: '3.12', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 99 | + |
| 100 | + # Python 3.13 |
| 101 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.13', cuda: '11.8.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 102 | +# - { artname: 'wheel', os: windows-2022, pyver: '3.13', cuda: '12.4.0', rocm: '', torch: '2.6.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 103 | + - { artname: 'wheel', os: windows-2022, pyver: '3.13', cuda: '12.8.1', rocm: '', torch: '2.7.0', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX' } |
| 104 | + |
| 105 | + # Ubuntu 20.04 ROCm |
| 106 | + |
| 107 | + # ROCm 5.6 |
| 108 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '', rocm: '5.6', torch: '2.2.2', cudaarch: '' } |
| 109 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '', rocm: '5.6', torch: '2.2.2', cudaarch: '' } |
| 110 | + |
| 111 | + # ROCm 6.0 |
| 112 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '', rocm: '6.0', torch: '2.3.1', cudaarch: '' } |
| 113 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '', rocm: '6.0', torch: '2.3.1', cudaarch: '' } |
| 114 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '', rocm: '6.0', torch: '2.3.1', cudaarch: '' } |
| 115 | + |
| 116 | + # ROCm 6.1 |
| 117 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '', rocm: '6.1', torch: '2.4.0', cudaarch: '' } |
| 118 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.11', cuda: '', rocm: '6.1', torch: '2.4.0', cudaarch: '' } |
| 119 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.12', cuda: '', rocm: '6.1', torch: '2.4.0', cudaarch: '' } |
| 120 | + |
| 121 | + # sdist |
| 122 | +# - { artname: 'sdist', os: ubuntu-22.04, pyver: '3.11', cuda: '', rocm: '', torch: '2.3.1', cudaarch: '' } |
| 123 | + |
| 124 | + # Extra wheel for HF spaces |
| 125 | +# - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.10', cuda: '12.1.0', rocm: '', torch: '2.2.2', cudaarch: '6.0 6.1 7.0 7.5 8.0 8.6 8.9 9.0+PTX' } |
| 126 | + |
| 127 | + fail-fast: false |
| 128 | + |
| 129 | + defaults: |
| 130 | + run: |
| 131 | + shell: pwsh |
| 132 | + |
| 133 | + steps: |
| 134 | + # Free disk space |
| 135 | + |
| 136 | + - name: Free Disk Space |
| 137 | + uses: jlumbroso/free-disk-space@v1.3.1 |
| 138 | + if: runner.os == 'Linux' |
| 139 | + with: |
| 140 | + tool-cache: true |
| 141 | + android: true |
| 142 | + dotnet: true |
| 143 | + haskell: true |
| 144 | + large-packages: false |
| 145 | + swap-storage: true |
| 146 | + |
| 147 | + # Setup Python |
| 148 | + |
| 149 | + - uses: actions/checkout@v4 |
| 150 | + |
| 151 | + - uses: actions/setup-python@v5.4.0 |
| 152 | + with: |
| 153 | + python-version: ${{ matrix.pyver }} |
| 154 | + |
| 155 | + # Get version string from package |
| 156 | + |
| 157 | + - name: Get version string |
| 158 | + id: package_version |
| 159 | + run: | |
| 160 | + $versionString = Get-Content $(Join-Path 'exllamav2' 'version.py') -raw |
| 161 | + if ($versionString -match '__version__ = "(\d+\.(?:\d+\.?(?:dev\d+)?)*)"') |
| 162 | + { |
| 163 | + Write-Output $('::notice file=build-wheels-release.yml,line=200,title=Package Version::Detected package version is: {0}' -f $Matches[1]) |
| 164 | + Write-Output "PACKAGE_VERSION=$($Matches[1])" >> "$env:GITHUB_OUTPUT" |
| 165 | + } |
| 166 | + else |
| 167 | + { |
| 168 | + Write-Output '::error file=build-wheels-release.yml,line=203::Could not parse version from exllamav2/version.py! You must upload wheels manually!' |
| 169 | + Write-Output "PACKAGE_VERSION=None" >> "$env:GITHUB_OUTPUT" |
| 170 | + } |
| 171 | +
|
| 172 | + # Pin VS build tools to 17.9 so builds won't fail |
| 173 | + |
| 174 | + - name: Install VS2022 BuildTools 17.9.7 |
| 175 | + run: choco install -y visualstudio2022buildtools --version=117.9.7.0 --params "--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64 --installChannelUri https://aka.ms/vs/17/release/180911598_-255012421/channel" |
| 176 | + if: runner.os == 'Windows' |
| 177 | + |
| 178 | + # Install ROCm SDK, apparently needs to happen before setting up Python |
| 179 | + |
| 180 | + - name: Build for ROCm |
| 181 | + if: matrix.rocm != '' |
| 182 | + shell: bash |
| 183 | + run: | |
| 184 | + # --- Install ROCm SDK |
| 185 | +
|
| 186 | + export ROCM_VERSION=${{ matrix.rocm }} |
| 187 | + export TORCH_VERSION=${{ matrix.torch }} |
| 188 | +
|
| 189 | + [ ! -d /etc/apt/keyrings ] && sudo mkdir --parents --mode=0755 /etc/apt/keyrings |
| 190 | + wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null |
| 191 | + echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/$ROCM_VERSION focal main" | sudo tee --append /etc/apt/sources.list.d/rocm.list |
| 192 | + echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600 |
| 193 | + |
| 194 | + sudo apt update |
| 195 | + sudo apt install rocm-hip-sdk -y |
| 196 | + sudo apt clean -y |
| 197 | +
|
| 198 | + echo "/opt/rocm/bin" >> $GITHUB_PATH |
| 199 | + echo "ROCM_PATH=/opt/rocm" >> $GITHUB_ENV |
| 200 | + echo "ROCM_VERSION=$ROCM_VERSION" >> $GITHUB_ENV |
| 201 | + echo "USE_ROCM=1" >> $GITHUB_ENV |
| 202 | +
|
| 203 | + # --- Install dependencies |
| 204 | +
|
| 205 | + python3 -m ensurepip --upgrade |
| 206 | + pip3 install torch==${{ matrix.torch }} --index-url="https://download.pytorch.org/whl/rocm$ROCM_VERSION" |
| 207 | + pip3 install --upgrade setuptools==69.5.1 build wheel safetensors sentencepiece ninja numpy |
| 208 | + pip3 cache purge |
| 209 | +
|
| 210 | + # --- Build wheel |
| 211 | +
|
| 212 | + python3 -m build -n --wheel -C--build-option=egg_info "-C--build-option=--tag-build=+rocm${{ matrix.rocm }}-torch${{ matrix.torch }}" |
| 213 | +
|
| 214 | + # Build for CUDA |
| 215 | + |
| 216 | + - name: Setup Mamba |
| 217 | + if: matrix.cuda != '' |
| 218 | + uses: conda-incubator/setup-miniconda@v3.1.0 |
| 219 | + with: |
| 220 | + activate-environment: "exllama" |
| 221 | + python-version: ${{ matrix.pyver }} |
| 222 | +# miniforge-variant: Mambaforge |
| 223 | + miniforge-version: latest |
| 224 | +# use-mamba: true |
| 225 | + add-pip-as-python-dependency: true |
| 226 | + auto-activate-base: false |
| 227 | + |
| 228 | + - name: Build for CUDA |
| 229 | + if: matrix.cuda != '' |
| 230 | + run: | |
| 231 | + # --- Spawn the VS shell |
| 232 | + if ($IsWindows) { |
| 233 | + Import-Module 'C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\Common7\Tools\Microsoft.VisualStudio.DevShell.dll' |
| 234 | + Enter-VsDevShell -VsInstallPath 'C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools' -DevCmdArguments '-arch=x64 -host_arch=x64' |
| 235 | + $env:DISTUTILS_USE_SDK=1 |
| 236 | + } |
| 237 | + |
| 238 | + # --- Install CUDA using Conda |
| 239 | + $cudaVersion = '${{ matrix.cuda }}' |
| 240 | + $cudaVersionPytorch = '${{ matrix.cuda }}'.Remove('${{ matrix.cuda }}'.LastIndexOf('.')).Replace('.','') |
| 241 | + |
| 242 | + $env:MAMBA_NO_LOW_SPEED_LIMIT = 1 |
| 243 | + mamba install -y -c nvidia/label/cuda-$cudaVersion cuda-toolkit cuda-runtime |
| 244 | + |
| 245 | + if (!(mamba list cuda)[-1].contains('cuda')) {sleep -s 10; mamba install -y 'cuda' $cudaVersion} |
| 246 | + if (!(mamba list cuda)[-1].contains('cuda')) {throw 'CUDA Toolkit failed to install!'} |
| 247 | + |
| 248 | + $env:CUDA_PATH = $env:CONDA_PREFIX |
| 249 | + $env:CUDA_HOME = $env:CONDA_PREFIX |
| 250 | + if ($IsLinux) {$env:LD_LIBRARY_PATH = $env:CONDA_PREFIX + '/lib:' + $env:LD_LIBRARY_PATH} |
| 251 | + |
| 252 | + # --- Install dependencies |
| 253 | + |
| 254 | + python -m ensurepip --upgrade |
| 255 | + python -m pip install torch==${{ matrix.torch }} --index-url https://download.pytorch.org/whl/cu$cudaVersionPytorch |
| 256 | + python -m pip install --upgrade setuptools==69.5.1 build wheel safetensors sentencepiece ninja numpy |
| 257 | + |
| 258 | + # --- Build wheel |
| 259 | + |
| 260 | + $BUILDTAG = "+cu$cudaVersionPytorch-torch${{ matrix.torch }}" |
| 261 | + $env:TORCH_CUDA_ARCH_LIST = '${{ matrix.cudaarch }}' |
| 262 | + python -m build -n --wheel -C--build-option=egg_info "-C--build-option=--tag-build=$BUILDTAG" |
| 263 | + |
| 264 | + # Build sdist |
| 265 | + |
| 266 | + - name: Build sdist |
| 267 | + if: matrix.cuda == '' && matrix.rocm == '' |
| 268 | + run: | |
| 269 | + # --- Spawn the VS shell |
| 270 | + if ($IsWindows) { |
| 271 | + Import-Module 'C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\Common7\Tools\Microsoft.VisualStudio.DevShell.dll' |
| 272 | + Enter-VsDevShell -VsInstallPath 'C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools' -DevCmdArguments '-arch=x64 -host_arch=x64' |
| 273 | + $env:DISTUTILS_USE_SDK=1 |
| 274 | + } |
| 275 | +
|
| 276 | + # --- Install dependencies |
| 277 | + |
| 278 | + python -m pip install torch==${{ matrix.torch }} --index-url https://download.pytorch.org/whl/cpu |
| 279 | + python -m pip install build wheel ninja |
| 280 | +
|
| 281 | + # --- Build wheel |
| 282 | + |
| 283 | + $env:EXLLAMA_NOCOMPILE=1 |
| 284 | + python -m build -n |
| 285 | +
|
| 286 | + # Upload files |
| 287 | + |
| 288 | + - uses: actions/upload-artifact@v4 |
| 289 | + if: matrix.artname == 'wheel' |
| 290 | + with: |
| 291 | + name: wheel-${{ matrix.os }}-py${{ matrix.pyver }}-cuda${{ matrix.cuda }}-torch${{ matrix.torch }} |
| 292 | + path: ./dist/* |
| 293 | + |
| 294 | + - uses: actions/upload-artifact@v4 |
| 295 | + if: matrix.artname == 'sdist' |
| 296 | + with: |
| 297 | + name: 'sdist' |
| 298 | + path: ./dist/* |
| 299 | + |
| 300 | + - name: Upload files to GitHub release |
| 301 | + if: steps.package_version.outputs.PACKAGE_VERSION != 'None' && inputs.release == '1' |
| 302 | + uses: svenstaro/upload-release-action@2.6.1 |
| 303 | + with: |
| 304 | + file: ./dist/*.whl |
| 305 | + tag: ${{ format('v{0}', steps.package_version.outputs.PACKAGE_VERSION) }} |
| 306 | + file_glob: true |
| 307 | + overwrite: true |
| 308 | + release_name: ${{ steps.package_version.outputs.PACKAGE_VERSION }} |
0 commit comments