build-llamacpp-binaries.yml

name: Build llama.cpp runtime binaries

# Cross-platform prebuilt binaries for the FunASR llama.cpp / GGUF runtime,
# like whisper.cpp's whisper-bin-*. Push a `runtime-llamacpp-v*` tag to publish
# a GitHub Release with the binaries; or run manually (workflow_dispatch) to test
# the build without creating a release.
on:
  workflow_dispatch:
  push:
    tags:
      - 'runtime-llamacpp-v*'

permissions:
  contents: write

jobs:
  build:
    name: build-${{ matrix.name }}
    strategy:
      fail-fast: false
      matrix:
        include:
          - os: ubuntu-latest
            name: linux-x64
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
          - os: ubuntu-latest
            name: linux-x64-avx2
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=ON
              -DGGML_AVX2=ON
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=ON
              -DGGML_F16C=ON
              -DGGML_BMI2=ON
          - os: ubuntu-latest
            name: linux-x64-vulkan
            vulkan: true
            build_target: llama-funasr-sensevoice
            timeout_minutes: 90
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_VULKAN=ON
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
          - os: ubuntu-24.04-arm
            name: linux-arm64
            cmake_flags: -DGGML_NATIVE=OFF
          - os: macos-latest
            name: macos-arm64
            cmake_flags: -DGGML_NATIVE=OFF
          - os: windows-latest
            name: windows-x64
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
          - os: windows-latest
            name: windows-x64-avx2
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=ON
              -DGGML_AVX2=ON
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=ON
              -DGGML_F16C=ON
              -DGGML_BMI2=ON
          - os: windows-latest
            name: windows-x64-vulkan
            vulkan: true
            windows_vulkan: true
            build_target: llama-funasr-sensevoice
            timeout_minutes: 90
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DCMAKE_MSVC_RUNTIME_LIBRARY=MultiThreaded
              -DGGML_VULKAN=ON
              -DGGML_OPENMP=OFF
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
          - os: windows-latest
            name: windows-x64-cuda
            cuda: true
            cuda_version: '13.2.0'
            cuda_architectures: '86'
            build_target: llama-funasr-sensevoice
            timeout_minutes: 90
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_CUDA=ON
              -DGGML_CUDA_FORCE_CUBLAS=ON
              -DGGML_CUDA_FA=OFF
              -DGGML_CUDA_NCCL=OFF
              -DGGML_CUDA_GRAPHS=OFF
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
    runs-on: ${{ matrix.os }}
    timeout-minutes: ${{ matrix.timeout_minutes || 30 }}
    defaults:
      run:
        shell: bash
        working-directory: runtime/llama.cpp
    steps:
      - uses: actions/checkout@v4
      - name: Install CUDA Toolkit
        if: matrix.cuda
        uses: Jimver/cuda-toolkit@v0.2.35
        with:
          cuda: ${{ matrix.cuda_version }}
          method: network
          log-file-suffix: ${{ matrix.name }}.txt
      - name: Show CUDA compiler
        if: matrix.cuda
        run: nvcc --version
      - name: Install Vulkan SDK packages
        if: matrix.vulkan && runner.os == 'Linux'
        run: |
          sudo apt-get update
          sudo apt-get install -y libvulkan-dev glslc spirv-headers vulkan-tools
          vulkaninfo --summary || true
      - name: Install Vulkan SDK
        if: matrix.windows_vulkan
        uses: humbletim/install-vulkan-sdk@30ba978f977e81b72d091fc8888feb1fb26f9aff
        with:
          version: '1.4.309.0'
          cache: true
      - name: Show Vulkan shader compiler
        if: matrix.windows_vulkan
        run: glslc --version
      - name: Install SPIR-V Headers CMake package
        if: matrix.windows_vulkan
        run: |
          runner_temp="$(cygpath -u "$RUNNER_TEMP")"
          headers_src="$runner_temp/SPIRV-Headers"
          headers_build="$runner_temp/SPIRV-Headers-build"
          headers_prefix="$runner_temp/SPIRV-Headers-install"
          git init -q "$headers_src"
          git -C "$headers_src" remote add origin https://github.com/KhronosGroup/SPIRV-Headers.git
          git -C "$headers_src" fetch --depth 1 origin 09913f088a1197aba4aefd300a876b2ebbaa3391
          git -C "$headers_src" checkout --detach FETCH_HEAD
          cmake -S "$headers_src" -B "$headers_build" \
            -DSPIRV_HEADERS_ENABLE_INSTALL=ON \
            -DSPIRV_HEADERS_ENABLE_TESTS=OFF \
            -DCMAKE_INSTALL_PREFIX="$headers_prefix"
          cmake --install "$headers_build" --config Release
          cmake_prefix_path="$(cygpath -m "$headers_prefix")"
          echo "CMAKE_PREFIX_PATH=$cmake_prefix_path" >> "$GITHUB_ENV"
      - name: Configure and build
        run: |
          # Release binaries must be portable across user machines, not tuned to
          # the ephemeral CI runner CPU. A native ggml build may emit AVX512 or
          # AVX-VNNI instructions and then crash with SIGILL on ordinary CPUs.
          # Keep the default x64 package conservative, and publish explicit
          # x64-avx2 assets for machines that support AVX2/FMA/F16C/BMI2.
          # The GPU release assets are SenseVoiceSmall-specific for now; building
          # only that target keeps the bounded jobs aligned with the backend
          # currently exposed by `--backend cuda` / `--backend vulkan`.
          cmake_args=(
            -B build
            -DCMAKE_BUILD_TYPE=Release
            ${{ matrix.cmake_flags }}
          )
          if [ -n "${{ matrix.cuda_architectures }}" ]; then
            cmake_args+=("-DCMAKE_CUDA_ARCHITECTURES=${{ matrix.cuda_architectures }}")
          fi
          cmake "${cmake_args[@]}"
          if [ -n "${{ matrix.build_target }}" ]; then
            cmake --build build --config Release -j 2 --target "${{ matrix.build_target }}"
          else
            cmake --build build --config Release -j 2
          fi
      - name: Package
        run: |
          mkdir -p pkg
          cp build/bin/llama-funasr-* pkg/ 2>/dev/null || true
          cp build/bin/Release/llama-funasr-* pkg/ 2>/dev/null || true
          cp README.md download-funasr-model.sh pkg/ 2>/dev/null || true
          echo "--- packaged binaries ---"; ls -la pkg
          if [ "${{ runner.os }}" = "Windows" ]; then
            7z a "funasr-llamacpp-${{ matrix.name }}.zip" ./pkg/*
          else
            tar czf "funasr-llamacpp-${{ matrix.name }}.tar.gz" -C pkg .
          fi
      - uses: actions/upload-artifact@v4
        with:
          name: funasr-llamacpp-${{ matrix.name }}
          path: runtime/llama.cpp/funasr-llamacpp-${{ matrix.name }}.*
          if-no-files-found: error

  release:
    needs: build
    if: startsWith(github.ref, 'refs/tags/')
    runs-on: ubuntu-latest
    steps:
      - uses: actions/download-artifact@v4
        with:
          path: dist
          merge-multiple: true
      - name: Create GitHub Release
        env:
          GH_TOKEN: ${{ github.token }}
        run: |
          tag="${{ github.ref_name }}"
          version="${tag#runtime-llamacpp-}"
          docs="https://github.com/${{ github.repository }}/blob/$tag/runtime/llama.cpp/README.md"
          gh release create "$tag" dist/* \
            --repo "${{ github.repository }}" \
            --title "FunASR llama.cpp runtime $version" \
            --notes "Prebuilt self-contained binaries for the FunASR llama.cpp / GGUF runtime: SenseVoice, Paraformer and Fun-ASR-Nano with built-in FSMN-VAD. Download the default quantized model with \`bash download-funasr-model.sh <sensevoice|paraformer|nano>\` (the helper requires the Hugging Face CLI: \`pip install -U huggingface_hub\`), then run \`llama-funasr-cli\` / \`llama-funasr-sensevoice\` / \`llama-funasr-paraformer\`. Use the default x64 asset for maximum CPU compatibility; use the x64-avx2 asset on CPUs with AVX2/FMA/F16C/BMI2 for higher throughput. The Vulkan assets are \`linux-x64-vulkan\` and \`windows-x64-vulkan\`; they require a working Vulkan driver/ICD and enable SenseVoiceSmall graph execution with \`llama-funasr-sensevoice ... --backend vulkan\`. Build from source with \`-DGGML_VULKAN=ON\` to validate platform-specific GPU stacks. The Windows CUDA asset is \`windows-x64-cuda\`; it requires an NVIDIA driver compatible with the CUDA Toolkit version configured by the release workflow, targets CUDA architecture 86, and enables SenseVoiceSmall graph execution with \`llama-funasr-sensevoice ... --backend cuda\`. Build from source for other GPU architectures. No Python ASR runtime or local build is required. Docs: $docs"