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

name: Build wheels and upload to GitHub Releases

on:
  push:
    tags:
      - "v*"

jobs:
  create_releases:
    name: Create Releases
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Create Release
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          gh release create "${{ github.ref_name }}" \
            --title "${{ github.ref_name }}" \
            --notes "TBD"

  create_matrix:
    name: Create Matrix
    runs-on: ubuntu-latest
    outputs:
      matrix: ${{ steps.create_matrix.outputs.matrix }}
    steps:
      - uses: actions/checkout@v4
      - name: Create Matrix
        id: create_matrix
        run: |
          python create_matrix.py | tee /tmp/matrix.json
          echo "matrix=$(cat /tmp/matrix.json)" >> $GITHUB_OUTPUT

  # #########################################################
  # Linux
  # #########################################################

  # ################ GitHub-hosted runner ################
  build_wheels_linux:
    name: Build Linux x86_64
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).linux }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_linux.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
    secrets: inherit

  build_wheels_linux_arm64:
    name: Build Linux ARM64
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64 }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_linux.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
      runner: "ubuntu-22.04-arm"
    secrets: inherit

  # ################ Self-hosted runner ################
  build_wheels_linux_self_hosted:
    name: Build Linux (self-hosted)
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_self_hosted }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_self_hosted.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_self_hosted.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_self_hosted.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_self_hosted.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_linux_self_host.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
      runner: '["self-hosted", "x64"]'
      container-image: "quay.io/pypa/manylinux_2_28_x86_64"
    secrets: inherit

  build_wheels_linux_arm64_self_hosted:
    name: Build Linux ARM64 (self-hosted)
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64_self_hosted }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64_self_hosted.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64_self_hosted.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64_self_hosted.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).linux_arm64_self_hosted.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_linux_arm_self_host.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
      runner: '["self-hosted-arm", "arm64"]'
      container-image: "quay.io/pypa/manylinux_2_28_aarch64"
    secrets: inherit

  # #########################################################
  # Windows
  # #########################################################

  # ################ GitHub-hosted runner ################
  build_wheels_windows:
    name: Build Windows
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).windows }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_windows.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
    secrets: inherit

  build_wheels_windows_code_build:
    name: Build Windows (AWS CodeBuild)
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_code_build }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_code_build.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_code_build.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_code_build.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_code_build.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_windows_code_build.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
    secrets: inherit

  # ################ Self-hosted runner ################
  build_wheels_windows_self_hosted:
    name: Build Windows (self-hosted)
    needs: [create_releases, create_matrix]
    if: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_self_hosted }}
    strategy:
      fail-fast: false
      matrix:
        flash-attn-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_self_hosted.flash-attn-version }}
        python-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_self_hosted.python-version }}
        torch-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_self_hosted.torch-version }}
        cuda-version: ${{ fromjson(needs.create_matrix.outputs.matrix).windows_self_hosted.cuda-version }}
        exclude: ${{ fromjson(needs.create_matrix.outputs.matrix).exclude }}
    uses: ./.github/workflows/_build_windows_self_host.yml
    with:
      flash-attn-version: ${{ matrix.flash-attn-version }}
      python-version: ${{ matrix.python-version }}
      torch-version: ${{ matrix.torch-version }}
      cuda-version: ${{ matrix.cuda-version }}
    secrets: inherit

  # #########################################################
  # Post-processing: update release notes and docs
  # #########################################################
  update_release_notes:
    name: Update Release Notes
    needs:
      - build_wheels_linux
      - build_wheels_linux_arm64
      - build_wheels_linux_self_hosted
      - build_wheels_linux_arm64_self_hosted
      - build_wheels_windows
      - build_wheels_windows_code_build
      - build_wheels_windows_self_hosted
    if: always()
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-python@v5
        with:
          python-version: 3.14

      - name: Install dependencies
        run: pip install pandas

      - name: Generate Release Notes
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          gh release view "${{ github.ref_name }}" --json assets > /tmp/assets.json
          python create_release_note.py /tmp/assets.json > /tmp/release_notes.md
          gh release edit "${{ github.ref_name }}" --notes-file /tmp/release_notes.md

  update_docs:
    name: Update Docs
    needs:
      - build_wheels_linux
      - build_wheels_linux_arm64
      - build_wheels_linux_self_hosted
      - build_wheels_linux_arm64_self_hosted
      - build_wheels_windows
      - build_wheels_windows_code_build
      - build_wheels_windows_self_hosted
    permissions:
      contents: write
    if: always()
    runs-on: ubuntu-latest
    env:
      GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
      DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
    steps:
      - uses: actions/checkout@v4
        with:
          ref: ${{ github.event.repository.default_branch }}

      - uses: actions/setup-python@v5
        with:
          python-version: 3.14

      - name: Install dependencies
        run: pip install pandas

      - name: Update docs
        run: |
          gh release view "${{ github.ref_name }}" --json assets > /tmp/assets.json
          python create_release_history.py \
            --assets /tmp/assets.json \
            --tag "${{ github.ref_name }}" \
            --repo "${{ github.repository }}" \
            --output docs/release_history.md
          python create_packages.py --assets /tmp/assets.json --output docs/packages.md

      - name: Commit and push docs updates
        run: |
          git config --global user.name "github-actions[bot]"
          git config --global user.email "41898282+github-actions[bot]@users.noreply.github.com"
          git pull
          if git diff --quiet; then
            echo "No docs updates to commit."
            exit 0
          fi
          git commit -am "docs: update docs for ${{ github.ref_name }}"
          git push origin HEAD:"${DEFAULT_BRANCH}"