flash-attention-prebuild-wheels-rocm/.github/workflows/build_linux.yml

# #########################################################
# Build wheels with GitHub hosted runner
# #########################################################

name: "[Linux x86_64] Build wheels and upload to GitHub Releases"

on:
  workflow_call:
    inputs:
      flash-attn-version:
        description: "Flash-Attention version"
        required: true
        type: string
      python-version:
        description: "Python version"
        required: true
        type: string
      torch-version:
        description: "PyTorch version"
        required: true
        type: string
      cuda-version:
        description: "CUDA version"
        required: true
        type: string

jobs:
  build_wheels:
    name: Build wheels and Upload (Linux x86_64)
    runs-on: ubuntu-22.04
    env:
      DEBIAN_FRONTEND: noninteractive
      TERM: xterm-256color
    steps:
      - uses: actions/checkout@v4

      - name: Maximize build space
        run: |
          df -h
          echo "-----------------------------"
          sudo rm -rf /usr/share/dotnet
          sudo rm -rf /usr/local/lib/android
          sudo rm -rf /opt/ghc
          sudo rm -rf /opt/hostedtoolcache/CodeQL
          df -h

      - name: Set Swap Space
        uses: pierotofy/set-swap-space@master
        with:
          swap-size-gb: 48

      - uses: actions/setup-python@v5
        with:
          python-version: ${{ inputs.python-version }}

      - uses: Jimver/cuda-toolkit@master
        with:
          cuda: ${{ inputs.cuda-version }}
          sub-packages: '["nvcc", "toolkit"]'
          method: "network"
          use-github-cache: false
          use-local-cache: false

      - name: Install build dependencies
        run: |
          sudo apt install -y ninja-build clang
          pip install -U pip setuptools==75.8.0 wheel setuptools packaging psutil

      - name: Set environment variables
        run: |
          export PATH=/usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
          export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH

      - name: Build wheels
        id: build_wheels
        env:
          MAX_JOBS: 2
          NVCC_THREADS: 2
        run: |
          chmod +x build_linux.sh
          ./build_linux.sh ${{ inputs.flash-attn-version }} ${{ inputs.python-version }} ${{ inputs.torch-version }} ${{ inputs.cuda-version }}
          wheel_name=$(basename $(ls flash-attention/dist/*.whl | head -n 1))
          echo "WHEEL_NAME=$wheel_name" >> $GITHUB_OUTPUT

      - name: Install Test
        run: |
          pip install --no-cache-dir flash-attention/dist/${{ steps.build_wheels.outputs.WHEEL_NAME }}
          python -c "import flash_attn; print(flash_attn.__version__)"

      - name: Upload Release Asset
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          tag_name=${{ github.ref_name }}
          wheel_path="flash-attention/dist/${{ steps.build_wheels.outputs.WHEEL_NAME }}"

          # Check if the file exists
          if [ ! -f "$wheel_path" ]; then
            echo "Error: Wheel file not found at $wheel_path"
            exit 1
          fi

          # Upload the release asset using GitHub CLI
          gh release upload "$tag_name" "$wheel_path" --clobber