build-vllm.yml

name: Build vLLM wheel

jobs:
  build:
    runs-on: ubuntu-latest

    steps:
      - name: Checkout
        uses: actions/checkout@v4
        with:
          ref: ${{ github.event.inputs.ref }}
          repository: ${{ github.event.inputs.repository }}

      - name: Apply patches
        run: |
          set -e

          # List of patches
          patches=(
            "https://raw.githubusercontent.com/${{ github.repository }}/${{ github.sha }}/patches/vllm/0000-enable-support-for-pascal-gpus.patch"
            "https://raw.githubusercontent.com/${{ github.repository }}/${{ github.sha }}/patches/vllm/1000-set-torch-cuda-arch-list.patch"
            "https://raw.githubusercontent.com/${{ github.repository }}/${{ github.sha }}/patches/vllm/9000-add-vllm-command-that-launches-api-server.patch"
          )

          # Apply patches
          for patch in ${patches[@]}; do
            echo "Applying patch $patch"
            curl "$patch" | patch -p1
          done

      - name: Build wheels
        uses: pypa/cibuildwheel@v2.18.0
        env:
          CIBW_BUILD: cp311-manylinux_x86_64
          CIBW_MANYLINUX_PYPY_X86_64_IMAGE: ghcr.io/sasha0552/manylinux2014_x86_64-cuda
          CIBW_MANYLINUX_X86_64_IMAGE: ghcr.io/sasha0552/manylinux2014_x86_64-cuda
          CIBW_REPAIR_WHEEL_COMMAND: ~

      - name: Create release
        uses: softprops/action-gh-release@v2
        with:
          files: wheelhouse/*.whl
          tag_name: ${{ github.event.inputs.tag_name }}

on:
  workflow_dispatch:
    inputs:
      repository:
        default: vllm-project/vllm
        description: Source repository
        required: true
        type: string

      ref:
        default: main
        description: Source ref
        required: true
        type: string

      tag_name:
        description: Target tag
        required: true
        type: string

permissions:
  contents: write