Build v0.3.2 Preview #72
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Build / Release | |
| run-name: ${{ github.event_name == 'push' && format('Build {0} Preview', github.ref_name) || format('Release {0}', inputs.tag_name) }} | |
| on: | |
| push: | |
| branches: | |
| - "v*" | |
| workflow_dispatch: | |
| inputs: | |
| source_ref: | |
| description: "Branch or tag name that requested the build" | |
| required: true | |
| type: string | |
| source_sha: | |
| description: "Exact commit SHA to build" | |
| required: true | |
| type: string | |
| tag_name: | |
| description: "Release/package version name, for example v0.3.0" | |
| required: true | |
| type: string | |
| publish_release: | |
| description: "Publish a stable release from its v* tag" | |
| required: true | |
| default: false | |
| type: boolean | |
| concurrency: | |
| group: ${{ github.workflow }}-${{ github.event_name == 'push' && github.ref || inputs.source_ref }} | |
| cancel-in-progress: ${{ github.event_name == 'push' }} | |
| permissions: | |
| actions: write | |
| contents: read | |
| env: | |
| CMAKE_ARGS: "-DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_TOOLS=ON -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON" | |
| jobs: | |
| release-meta: | |
| name: Publication metadata | |
| runs-on: ubuntu-24.04 | |
| outputs: | |
| source_ref: ${{ steps.meta.outputs.source_ref }} | |
| source_sha: ${{ steps.meta.outputs.source_sha }} | |
| tag_name: ${{ steps.meta.outputs.tag_name }} | |
| version_name: ${{ steps.meta.outputs.version_name }} | |
| release_title: ${{ steps.meta.outputs.release_title }} | |
| preview: ${{ steps.meta.outputs.preview }} | |
| stable_release: ${{ steps.meta.outputs.stable_release }} | |
| build_packages: ${{ steps.meta.outputs.build_packages }} | |
| publish_docker: ${{ steps.meta.outputs.publish_docker }} | |
| ccache_ref: ${{ steps.meta.outputs.ccache_ref }} | |
| image_repo: ${{ steps.meta.outputs.image_repo }} | |
| short_sha: ${{ steps.meta.outputs.short_sha }} | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| fetch-depth: 0 | |
| - name: Resolve release metadata | |
| id: meta | |
| shell: bash | |
| run: | | |
| set -euo pipefail | |
| build_packages="true" | |
| publish_docker="true" | |
| if [[ "${EVENT_NAME}" == "push" ]]; then | |
| if [[ "${GITHUB_REF}" != refs/heads/v* ]]; then | |
| echo "Unsupported preview ref ${GITHUB_REF}; preview builds require a v* branch." >&2 | |
| exit 1 | |
| fi | |
| source_ref="${GITHUB_REF_NAME}" | |
| source_sha="${GITHUB_SHA}" | |
| version_name="${source_ref}" | |
| tag_name="preview-${source_ref}" | |
| release_title="${source_ref} Preview" | |
| preview="true" | |
| stable_release="false" | |
| elif [[ "${EVENT_NAME}" == "workflow_dispatch" ]]; then | |
| if [[ "${INPUT_PUBLISH_RELEASE}" != "true" ]]; then | |
| echo "workflow_dispatch is reserved for main-owned stable releases." >&2 | |
| exit 1 | |
| fi | |
| if [[ "${GITHUB_REF}" != "refs/heads/main" ]]; then | |
| echo "Stable release workflow_dispatch must run on main, not ${GITHUB_REF}." >&2 | |
| exit 1 | |
| fi | |
| source_ref="${INPUT_SOURCE_REF}" | |
| source_sha="${INPUT_SOURCE_SHA}" | |
| tag_name="${INPUT_TAG_NAME}" | |
| version_name="${tag_name}" | |
| release_title="${tag_name}" | |
| preview="false" | |
| stable_release="true" | |
| else | |
| echo "Unsupported event ${EVENT_NAME}." >&2 | |
| exit 1 | |
| fi | |
| if [[ ! "${version_name}" =~ ^v[0-9][0-9A-Za-z._-]*$ ]]; then | |
| echo "Version name ${version_name} is not safe for release assets and Docker tags." >&2 | |
| exit 1 | |
| fi | |
| if [[ -z "${source_sha}" || ! "${source_sha}" =~ ^[0-9a-f]{40}$ ]]; then | |
| echo "source_sha must be a full 40-character commit SHA." >&2 | |
| exit 1 | |
| fi | |
| if ! git cat-file -e "${source_sha}^{commit}"; then | |
| git fetch origin +refs/heads/"${source_ref}":refs/remotes/origin/"${source_ref}" --no-tags || true | |
| git fetch origin +refs/tags/"${source_ref}":refs/tags/"${source_ref}" || true | |
| fi | |
| if ! git cat-file -e "${source_sha}^{commit}"; then | |
| git fetch origin "${source_sha}" --no-tags | |
| fi | |
| git cat-file -e "${source_sha}^{commit}" | |
| if [[ "${preview}" == "true" ]]; then | |
| if [[ "${source_ref}" != v* || "${tag_name}" != "preview-${source_ref}" ]]; then | |
| echo "Invalid preview metadata: source_ref=${source_ref}, tag_name=${tag_name}." >&2 | |
| exit 1 | |
| fi | |
| git fetch origin +refs/heads/"${source_ref}":refs/remotes/origin/"${source_ref}" --no-tags | |
| branch_commit="$(git rev-parse refs/remotes/origin/"${source_ref}")" | |
| if [[ "${branch_commit}" != "${source_sha}" ]]; then | |
| echo "Branch ${source_ref} points to ${branch_commit}, not requested source ${source_sha}." >&2 | |
| exit 1 | |
| fi | |
| else | |
| if [[ "${tag_name}" != v* ]]; then | |
| echo "Release publishing requires a v* tag_name." >&2 | |
| exit 1 | |
| fi | |
| if [[ "${source_ref}" != "${tag_name}" ]]; then | |
| echo "Stable release source_ref ${source_ref} must match tag_name ${tag_name}." >&2 | |
| exit 1 | |
| fi | |
| git fetch origin +refs/tags/"${tag_name}":refs/tags/"${tag_name}" | |
| tag_commit="$(git rev-list -n 1 "${tag_name}")" | |
| if [[ "${tag_commit}" != "${source_sha}" ]]; then | |
| echo "Tag ${tag_name} points to ${tag_commit}, not requested source ${source_sha}." >&2 | |
| exit 1 | |
| fi | |
| git fetch origin +refs/heads/main:refs/remotes/origin/main --no-tags | |
| if ! git merge-base --is-ancestor "${source_sha}" refs/remotes/origin/main; then | |
| echo "Tag ${tag_name} is not reachable from origin/main; refusing to publish." >&2 | |
| exit 1 | |
| fi | |
| fi | |
| if [[ "${preview}" == "true" ]]; then | |
| ccache_ref="${source_ref}" | |
| else | |
| ccache_ref="main" | |
| fi | |
| owner="$(printf '%s' "${GITHUB_REPOSITORY_OWNER}" | tr '[:upper:]' '[:lower:]')" | |
| repo="$(printf '%s' "${GITHUB_REPOSITORY#*/}" | tr '[:upper:]' '[:lower:]')" | |
| image_repo="ghcr.io/${owner}/${repo}" | |
| echo "source_ref=${source_ref}" >> "${GITHUB_OUTPUT}" | |
| echo "source_sha=${source_sha}" >> "${GITHUB_OUTPUT}" | |
| echo "tag_name=${tag_name}" >> "${GITHUB_OUTPUT}" | |
| echo "version_name=${version_name}" >> "${GITHUB_OUTPUT}" | |
| echo "release_title=${release_title}" >> "${GITHUB_OUTPUT}" | |
| echo "preview=${preview}" >> "${GITHUB_OUTPUT}" | |
| echo "stable_release=${stable_release}" >> "${GITHUB_OUTPUT}" | |
| echo "build_packages=${build_packages}" >> "${GITHUB_OUTPUT}" | |
| echo "publish_docker=${publish_docker}" >> "${GITHUB_OUTPUT}" | |
| echo "ccache_ref=${ccache_ref}" >> "${GITHUB_OUTPUT}" | |
| echo "image_repo=${image_repo}" >> "${GITHUB_OUTPUT}" | |
| short_sha="${source_sha::12}" | |
| echo "short_sha=${short_sha}" >> "${GITHUB_OUTPUT}" | |
| echo "Using source ref: ${source_ref}" | |
| echo "Using source SHA: ${source_sha}" | |
| echo "Using package version: ${tag_name}" | |
| echo "Release title: ${release_title}" | |
| echo "Preview: ${preview}" | |
| echo "Build packages: ${build_packages}" | |
| echo "Publish Docker images: ${publish_docker}" | |
| env: | |
| EVENT_NAME: ${{ github.event_name }} | |
| INPUT_SOURCE_REF: ${{ inputs.source_ref }} | |
| INPUT_SOURCE_SHA: ${{ inputs.source_sha }} | |
| INPUT_TAG_NAME: ${{ inputs.tag_name }} | |
| INPUT_PUBLISH_RELEASE: ${{ inputs.publish_release }} | |
| macos-arm64: | |
| name: macOS arm64 Metal | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: macos-14 | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-macos-arm64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-macos-arm64- | |
| release-macos-arm64- | |
| - name: Build | |
| run: | | |
| cmake -B build \ | |
| -DCMAKE_BUILD_TYPE=Release \ | |
| -DGGML_METAL_USE_BF16=ON \ | |
| -DGGML_METAL_EMBED_LIBRARY=ON \ | |
| -DCMAKE_INSTALL_RPATH='@loader_path' \ | |
| -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ | |
| -DLLAMA_FATAL_WARNINGS=ON \ | |
| -DLLAMA_BUILD_BORINGSSL=ON \ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \ | |
| ${{ env.CMAKE_ARGS }} | |
| cmake --build build --config Release -j "$(sysctl -n hw.logicalcpu)" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-macos-arm64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Smoke package binaries | |
| run: | | |
| ./build/bin/llama-server --version | |
| ./build/bin/llama-cli --version | |
| - name: Pack artifacts | |
| run: | | |
| cp LICENSE ./build/bin/ | |
| tar -czvf "beellama-${{ needs.release-meta.outputs.tag_name }}-bin-macos-arm64.tar.gz" \ | |
| -s ",^\.,beellama-${{ needs.release-meta.outputs.tag_name }}," \ | |
| -C ./build/bin . | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-${{ needs.release-meta.outputs.tag_name }}-bin-macos-arm64.tar.gz | |
| name: beellama-bin-macos-arm64.tar.gz | |
| ubuntu-cpu: | |
| name: Ubuntu ${{ matrix.build }} CPU | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: ${{ matrix.os }} | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| include: | |
| - build: x64 | |
| os: ubuntu-22.04 | |
| - build: arm64 | |
| os: ubuntu-24.04-arm | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-${{ matrix.os }}-cpu | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-${{ matrix.os }}-cpu- | |
| release-${{ matrix.os }}-cpu- | |
| - name: Dependencies | |
| run: | | |
| sudo apt-get update | |
| sudo apt-get install -y build-essential libssl-dev | |
| - name: Toolchain workaround | |
| if: ${{ contains(matrix.os, 'ubuntu-24.04') }} | |
| run: | | |
| sudo apt-get install -y gcc-14 g++-14 | |
| echo "CC=gcc-14" >> "${GITHUB_ENV}" | |
| echo "CXX=g++-14" >> "${GITHUB_ENV}" | |
| - name: Build | |
| run: | | |
| cmake -B build \ | |
| -DCMAKE_BUILD_TYPE=Release \ | |
| -DCMAKE_INSTALL_RPATH='$ORIGIN' \ | |
| -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ | |
| -DGGML_BACKEND_DL=ON \ | |
| -DGGML_NATIVE=OFF \ | |
| -DGGML_CPU_ALL_VARIANTS=ON \ | |
| -DLLAMA_FATAL_WARNINGS=ON \ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \ | |
| ${{ env.CMAKE_ARGS }} | |
| cmake --build build --config Release -j "$(nproc)" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-${{ matrix.os }}-cpu | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Smoke package binaries | |
| run: | | |
| file ./build/bin/llama-server | |
| ./build/bin/llama-server --version | |
| ./build/bin/llama-cli --version | |
| - name: Pack artifacts | |
| run: | | |
| cp LICENSE ./build/bin/ | |
| tar -czvf "beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-${{ matrix.build }}.tar.gz" \ | |
| --transform "s,^\.,beellama-${{ needs.release-meta.outputs.tag_name }}," \ | |
| -C ./build/bin . | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-${{ matrix.build }}.tar.gz | |
| name: beellama-bin-ubuntu-${{ matrix.build }}.tar.gz | |
| ubuntu-vulkan: | |
| name: Ubuntu x64 Vulkan | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: ubuntu-22.04 | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-vulkan-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-vulkan-x64- | |
| release-ubuntu-vulkan-x64- | |
| - name: Dependencies | |
| run: | | |
| wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo apt-key add - | |
| sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list | |
| sudo apt-get update -y | |
| sudo apt-get install -y build-essential mesa-vulkan-drivers vulkan-sdk libssl-dev | |
| - name: Build | |
| run: | | |
| cmake -B build \ | |
| -DCMAKE_BUILD_TYPE=Release \ | |
| -DCMAKE_INSTALL_RPATH='$ORIGIN' \ | |
| -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ | |
| -DGGML_BACKEND_DL=ON \ | |
| -DGGML_NATIVE=OFF \ | |
| -DGGML_CPU_ALL_VARIANTS=ON \ | |
| -DGGML_VULKAN=ON \ | |
| -DLLAMA_FATAL_WARNINGS=ON \ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \ | |
| ${{ env.CMAKE_ARGS }} | |
| cmake --build build --config Release -j "$(nproc)" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-vulkan-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Smoke package binaries | |
| run: | | |
| ./build/bin/llama-server --version | |
| ./build/bin/llama-cli --version | |
| - name: Pack artifacts | |
| run: | | |
| cp LICENSE ./build/bin/ | |
| tar -czvf "beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-vulkan-x64.tar.gz" \ | |
| --transform "s,^\.,beellama-${{ needs.release-meta.outputs.tag_name }}," \ | |
| -C ./build/bin . | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-vulkan-x64.tar.gz | |
| name: beellama-bin-ubuntu-vulkan-x64.tar.gz | |
| ubuntu-rocm: | |
| name: Ubuntu x64 ROCm | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: ubuntu-22.04 | |
| env: | |
| ROCM_VERSION: "7.2.1" | |
| GPU_TARGETS: "gfx908;gfx90a;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1151;gfx1150;gfx1200;gfx1201" | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: Free up disk space | |
| uses: ggml-org/free-disk-space@v1.3.1 | |
| with: | |
| tool-cache: true | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-rocm-${{ env.ROCM_VERSION }}-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-rocm-${{ env.ROCM_VERSION }}-x64- | |
| release-ubuntu-rocm-${{ env.ROCM_VERSION }}-x64- | |
| - name: Dependencies | |
| run: | | |
| sudo apt-get update | |
| sudo apt-get install -y build-essential git cmake wget libssl-dev | |
| - name: Setup ROCm | |
| run: | | |
| sudo mkdir --parents --mode=0755 /etc/apt/keyrings | |
| wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | \ | |
| gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null | |
| sudo tee /etc/apt/sources.list.d/rocm.list << EOF | |
| deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/${ROCM_VERSION} jammy main | |
| EOF | |
| sudo tee /etc/apt/preferences.d/rocm-pin-600 << EOF | |
| Package: * | |
| Pin: release o=repo.radeon.com | |
| Pin-Priority: 600 | |
| EOF | |
| sudo apt-get update | |
| sudo apt-get install -y rocm-hip-sdk | |
| - name: Build | |
| run: | | |
| cmake -B build -S . \ | |
| -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \ | |
| -DCMAKE_BUILD_TYPE=Release \ | |
| -DGGML_BACKEND_DL=ON \ | |
| -DGGML_NATIVE=OFF \ | |
| -DCMAKE_INSTALL_RPATH='$ORIGIN' \ | |
| -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ | |
| -DGGML_CPU_ALL_VARIANTS=ON \ | |
| -DGPU_TARGETS="${GPU_TARGETS}" \ | |
| -DGGML_HIP=ON \ | |
| -DGGML_CUDA_FA=ON \ | |
| -DHIP_PLATFORM=amd \ | |
| -DGGML_HIP_ROCWMMA_FATTN=ON \ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_HIP_COMPILER_LAUNCHER=ccache \ | |
| ${{ env.CMAKE_ARGS }} | |
| cmake --build build --config Release -j "$(nproc)" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-rocm-${{ env.ROCM_VERSION }}-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Smoke package binaries | |
| run: | | |
| ./build/bin/llama-server --version | |
| ./build/bin/llama-cli --version | |
| - name: Pack artifacts | |
| run: | | |
| rocm_short="$(echo "${ROCM_VERSION}" | cut -d '.' -f 1,2)" | |
| cp LICENSE ./build/bin/ | |
| tar -czvf "beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-rocm-${rocm_short}-x64.tar.gz" \ | |
| --transform "s,^\.,beellama-${{ needs.release-meta.outputs.tag_name }}," \ | |
| -C ./build/bin . | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-rocm-7.2-x64.tar.gz | |
| name: beellama-bin-ubuntu-rocm-7.2-x64.tar.gz | |
| ubuntu-sycl: | |
| name: Ubuntu x64 SYCL | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: ubuntu-24.04 | |
| env: | |
| ONEAPI_ROOT: /opt/intel/oneapi | |
| ONEAPI_INSTALLER_VERSION: "2025.3.3" | |
| ONEAPI_INSTALLER_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/56f7923a-adb8-43f3-8b02-2b60fcac8cab/intel-deep-learning-essentials-2025.3.3.16_offline.sh | |
| LEVEL_ZERO_VERSION: "1.28.2" | |
| LEVEL_ZERO_UBUNTU_VERSION: "u24.04" | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: Free up disk space | |
| uses: ggml-org/free-disk-space@v1.3.1 | |
| with: | |
| tool-cache: true | |
| - name: Prepare oneAPI cache path | |
| run: | | |
| sudo mkdir -p "${ONEAPI_ROOT}" | |
| sudo chown -R "${USER}:${USER}" /opt/intel | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64- | |
| release-ubuntu-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64- | |
| - name: Dependencies | |
| run: | | |
| sudo apt-get update | |
| sudo apt-get install -y build-essential git cmake ninja-build wget ca-certificates libssl-dev | |
| - name: Install oneAPI | |
| run: | | |
| installer="${RUNNER_TEMP}/intel-deep-learning-essentials-${ONEAPI_INSTALLER_VERSION}.sh" | |
| wget "${ONEAPI_INSTALLER_URL}" -O "${installer}" | |
| sudo bash "${installer}" -s -a --silent --eula accept | |
| sudo chown -R "${USER}:${USER}" "${ONEAPI_ROOT}" | |
| - name: Install Level Zero SDK | |
| run: | | |
| cd /tmp | |
| wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero.deb | |
| wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero-devel_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero-devel.deb | |
| sudo apt-get -o Dpkg::Options::="--force-overwrite" install -y ./level-zero.deb ./level-zero-devel.deb | |
| - name: Build | |
| run: | | |
| source "${ONEAPI_ROOT}/setvars.sh" | |
| cmake -B build -G "Ninja" \ | |
| -DCMAKE_BUILD_TYPE=Release \ | |
| -DCMAKE_INSTALL_RPATH='$ORIGIN' \ | |
| -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ | |
| -DGGML_SYCL=ON \ | |
| -DGGML_SYCL_F16=OFF \ | |
| -DCMAKE_C_COMPILER=icx \ | |
| -DCMAKE_CXX_COMPILER=icpx \ | |
| -DGGML_BACKEND_DL=ON \ | |
| -DGGML_NATIVE=OFF \ | |
| -DGGML_CPU_ALL_VARIANTS=ON \ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \ | |
| ${{ env.CMAKE_ARGS }} | |
| cmake --build build --config Release -j "$(nproc)" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Smoke package binaries | |
| run: | | |
| source "${ONEAPI_ROOT}/setvars.sh" | |
| test -s ./build/bin/libggml-sycl.so | |
| ldd ./build/bin/libggml-sycl.so | |
| mv ./build/bin/libggml-sycl.so ./build/bin/libggml-sycl.so.smoke-disabled | |
| trap 'mv ./build/bin/libggml-sycl.so.smoke-disabled ./build/bin/libggml-sycl.so' EXIT | |
| ./build/bin/llama-server --version | |
| ./build/bin/llama-cli --version | |
| - name: Pack artifacts | |
| run: | | |
| cp LICENSE ./build/bin/ | |
| tar -czvf "beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-sycl-x64.tar.gz" \ | |
| --transform "s,^\.,beellama-${{ needs.release-meta.outputs.tag_name }}," \ | |
| -C ./build/bin . | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-sycl-x64.tar.gz | |
| name: beellama-bin-ubuntu-sycl-x64.tar.gz | |
| ubuntu-cuda: | |
| name: Ubuntu x64 CUDA ${{ matrix.cuda }} | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: ubuntu-24.04 | |
| container: nvidia/cuda:${{ matrix.cuda_container }} | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| include: | |
| - cuda: "12.4" | |
| cuda_container: "12.4.1-devel-ubuntu22.04" | |
| - cuda: "13.1" | |
| cuda_container: "13.1.1-devel-ubuntu24.04" | |
| steps: | |
| - name: Dependencies | |
| run: | | |
| apt-get update | |
| apt-get install -y --no-install-recommends build-essential ca-certificates cmake ninja-build ccache git file libssl-dev libgomp1 | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-cuda-${{ matrix.cuda }}-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-cuda-${{ matrix.cuda }}-x64- | |
| release-ubuntu-cuda-${{ matrix.cuda }}-x64- | |
| max-size: 5G | |
| sloppiness: time_macros | |
| - name: Build | |
| run: | | |
| cmake -S . -B build -G Ninja \ | |
| -DCMAKE_BUILD_TYPE=Release \ | |
| -DCMAKE_INSTALL_RPATH='$ORIGIN' \ | |
| -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ | |
| -DGGML_NATIVE=OFF \ | |
| -DGGML_CUDA=ON \ | |
| -DGGML_CUDA_FA=ON \ | |
| -DGGML_BACKEND_DL=ON \ | |
| -DGGML_CPU_ALL_VARIANTS=ON \ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \ | |
| -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache \ | |
| ${{ env.CMAKE_ARGS }} \ | |
| -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined | |
| cmake --build build --config Release -j "$(nproc)" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-ubuntu-cuda-${{ matrix.cuda }}-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Smoke package binaries | |
| run: | | |
| file ./build/bin/llama-server | |
| ./build/bin/llama-server --version | |
| ./build/bin/llama-cli --version | |
| - name: Pack artifacts | |
| run: | | |
| cp LICENSE ./build/bin/ | |
| tar -czvf "beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-cuda-${{ matrix.cuda }}-x64.tar.gz" \ | |
| --transform "s,^\.,beellama-${{ needs.release-meta.outputs.tag_name }}," \ | |
| -C ./build/bin . | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-${{ needs.release-meta.outputs.tag_name }}-bin-ubuntu-cuda-${{ matrix.cuda }}-x64.tar.gz | |
| name: beellama-bin-ubuntu-cuda-${{ matrix.cuda }}-x64.tar.gz | |
| windows-cpu: | |
| name: Windows x64 CPU | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: windows-2025 | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-cpu-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-windows-cpu-x64- | |
| release-windows-cpu-x64- | |
| - name: Install Ninja | |
| run: choco install ninja -y | |
| - name: Build | |
| shell: cmd | |
| run: | | |
| call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64 | |
| cmake -S . -B build -G "Ninja Multi-Config" ^ | |
| -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake ^ | |
| -DLLAMA_BUILD_BORINGSSL=ON ^ | |
| -DGGML_NATIVE=OFF ^ | |
| -DGGML_BACKEND_DL=ON ^ | |
| -DGGML_CPU_ALL_VARIANTS=ON ^ | |
| -DGGML_OPENMP=ON ^ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache ^ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ^ | |
| %CMAKE_ARGS% | |
| cmake --build build --config Release -j %NUMBER_OF_PROCESSORS% | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-cpu-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Pack artifacts | |
| shell: pwsh | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| $bin = ".\build\bin\Release" | |
| Copy-Item ".\LICENSE" $bin | |
| $redistRoot = "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Redist\MSVC" | |
| $omp = Get-ChildItem $redistRoot -Recurse -Filter "libomp140.x86_64.dll" | | |
| Where-Object { $_.FullName -like "*\debug_nonredist\x64\Microsoft.VC143.OpenMP.LLVM\libomp140.x86_64.dll" } | | |
| Sort-Object FullName -Descending | | |
| Select-Object -First 1 | |
| if (-not $omp) { | |
| throw "VC143 x64 libomp140.x86_64.dll not found under $redistRoot" | |
| } | |
| Copy-Item $omp.FullName $bin -Force | |
| python scripts\verify-windows-package.py $bin | |
| & "$bin\llama-server.exe" --version | |
| & "$bin\llama-cli.exe" --version | |
| 7z a -snl "beellama-bin-win-cpu-x64.zip" "$bin\*" | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-bin-win-cpu-x64.zip | |
| name: beellama-bin-win-cpu-x64.zip | |
| windows-sycl: | |
| name: Windows x64 SYCL | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: windows-2022 | |
| env: | |
| WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/b60765d1-2b85-4e85-86b6-cb0e9563a699/intel-deep-learning-essentials-2025.3.3.18_offline.exe | |
| WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel:intel.oneapi.win.dnnl:intel.oneapi.win.tbb.devel | |
| LEVEL_ZERO_SDK_URL: https://github.com/oneapi-src/level-zero/releases/download/v1.28.2/level-zero-win-sdk-1.28.2.zip | |
| ONEAPI_ROOT: 'C:\Program Files (x86)\Intel\oneAPI' | |
| ONEAPI_INSTALLER_VERSION: "2025.3.3" | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Install oneAPI | |
| shell: cmd | |
| run: scripts\install-oneapi.bat "%WINDOWS_BASEKIT_URL%" "%WINDOWS_DPCPP_MKL%" | |
| - name: Install Level Zero SDK | |
| shell: pwsh | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| $archive = Join-Path $env:RUNNER_TEMP "level-zero-win-sdk.zip" | |
| $extractDir = Join-Path $env:RUNNER_TEMP "level-zero-sdk" | |
| Invoke-WebRequest -Uri "${{ env.LEVEL_ZERO_SDK_URL }}" -OutFile $archive | |
| Expand-Archive -Path $archive -DestinationPath $extractDir -Force | |
| $sdkRoot = Get-ChildItem -Path $extractDir -Directory | Select-Object -First 1 | |
| if (-not $sdkRoot) { | |
| $sdkRoot = Get-Item $extractDir | |
| } | |
| "LEVEL_ZERO_V1_SDK_PATH=$($sdkRoot.FullName)" | Out-File -FilePath $env:GITHUB_ENV -Append | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-windows-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64- | |
| release-windows-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64- | |
| - name: Install Ninja | |
| run: choco install ninja -y | |
| - name: Build | |
| shell: cmd | |
| run: | | |
| call "%ONEAPI_ROOT%\setvars.bat" intel64 --force | |
| icx --version | |
| cmake -G "Ninja" -B build ^ | |
| -DCMAKE_C_COMPILER=cl ^ | |
| -DCMAKE_CXX_COMPILER=icx ^ | |
| -DCMAKE_BUILD_TYPE=Release ^ | |
| -DGGML_BACKEND_DL=ON ^ | |
| -DBUILD_SHARED_LIBS=ON ^ | |
| -DGGML_CPU=OFF ^ | |
| -DGGML_SYCL=ON ^ | |
| -DGGML_SYCL_F16=OFF ^ | |
| -DLLAMA_BUILD_EXAMPLES=OFF ^ | |
| -DLLAMA_BUILD_TESTS=OFF ^ | |
| -DLLAMA_BUILD_TOOLS=OFF ^ | |
| -DLLAMA_BUILD_SERVER=OFF ^ | |
| -DLLAMA_BUILD_BORINGSSL=ON ^ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache ^ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache | |
| cmake --build build --config Release -j %NUMBER_OF_PROCESSORS% --target ggml-sycl | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-sycl-${{ env.ONEAPI_INSTALLER_VERSION }}-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Pack SYCL backend | |
| shell: pwsh | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| $bin = ".\build\bin" | |
| Copy-Item ".\LICENSE" $bin -Force | |
| $runtimeFiles = @( | |
| "$env:ONEAPI_ROOT\mkl\latest\bin\mkl_sycl_blas.5.dll", | |
| "$env:ONEAPI_ROOT\mkl\latest\bin\mkl_core.2.dll", | |
| "$env:ONEAPI_ROOT\mkl\latest\bin\mkl_tbb_thread.2.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\ur_adapter_level_zero.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\ur_adapter_level_zero_v2.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\ur_adapter_opencl.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\ur_loader.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\ur_win_proxy_loader.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\sycl8.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\svml_dispmd.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\libmmd.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\libiomp5md.dll", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\sycl-ls.exe", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\libsycl-fallback-bfloat16.spv", | |
| "$env:ONEAPI_ROOT\compiler\latest\bin\libsycl-native-bfloat16.spv", | |
| "$env:ONEAPI_ROOT\dnnl\latest\bin\dnnl.dll", | |
| "$env:ONEAPI_ROOT\tbb\latest\bin\tbb12.dll", | |
| "$env:ONEAPI_ROOT\tcm\latest\bin\tcm.dll", | |
| "$env:ONEAPI_ROOT\tcm\latest\bin\libhwloc-15.dll", | |
| "$env:ONEAPI_ROOT\umf\latest\bin\umf.dll" | |
| ) | |
| foreach ($file in $runtimeFiles) { | |
| Copy-Item -LiteralPath $file -Destination $bin -Force | |
| } | |
| $zeLoader = Get-ChildItem -Path @($env:ONEAPI_ROOT, $env:LEVEL_ZERO_V1_SDK_PATH) -Recurse -Filter "ze_loader.dll" -ErrorAction SilentlyContinue | Select-Object -First 1 | |
| if ($zeLoader) { | |
| Copy-Item -LiteralPath $zeLoader.FullName -Destination $bin -Force | |
| } else { | |
| Write-Warning "ze_loader.dll was not found in oneAPI or the Level Zero SDK; relying on the system driver runtime" | |
| } | |
| python scripts\verify-windows-package.py $bin | |
| 7z a -snl "beellama-bin-win-sycl-x64.zip" "$bin\*" | |
| - name: Upload SYCL backend | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-bin-win-sycl-x64.zip | |
| name: beellama-bin-win-sycl-x64.zip | |
| windows-vulkan: | |
| name: Windows x64 Vulkan | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: windows-2025 | |
| env: | |
| VULKAN_VERSION: 1.4.313.2 | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: Install Vulkan SDK | |
| shell: pwsh | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| $installer = Join-Path $env:RUNNER_TEMP "VulkanSDK-Installer.exe" | |
| curl.exe -o $installer -L "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/vulkansdk-windows-X64-${env:VULKAN_VERSION}.exe" | |
| & $installer --accept-licenses --default-answer --confirm-command install | |
| "VULKAN_SDK=C:\VulkanSDK\${env:VULKAN_VERSION}" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append | |
| "C:\VulkanSDK\${env:VULKAN_VERSION}\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-vulkan-${{ env.VULKAN_VERSION }}-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-windows-vulkan-${{ env.VULKAN_VERSION }}-x64- | |
| release-windows-vulkan-${{ env.VULKAN_VERSION }}-x64- | |
| - name: Install Ninja | |
| run: choco install ninja -y | |
| - name: Build | |
| shell: cmd | |
| run: | | |
| call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64 | |
| cmake -S . -B build -G "Ninja Multi-Config" ^ | |
| -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake ^ | |
| -DGGML_BACKEND_DL=ON ^ | |
| -DGGML_NATIVE=OFF ^ | |
| -DGGML_CPU=OFF ^ | |
| -DGGML_VULKAN=ON ^ | |
| -DLLAMA_BUILD_BORINGSSL=ON ^ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache ^ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache | |
| cmake --build build --config Release -j %NUMBER_OF_PROCESSORS% --target ggml-vulkan | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-vulkan-${{ env.VULKAN_VERSION }}-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Pack Vulkan backend | |
| run: | | |
| 7z a -snl "beellama-bin-win-vulkan-x64.zip" .\build\bin\Release\ggml-vulkan.dll | |
| - name: Upload Vulkan backend | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-bin-win-vulkan-x64.zip | |
| name: beellama-bin-win-vulkan-x64.zip | |
| windows-cuda: | |
| name: Windows x64 CUDA ${{ matrix.cuda }} | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: windows-2022 | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| cuda: ["12.4", "13.1"] | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-cuda-${{ matrix.cuda }} | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-windows-cuda-${{ matrix.cuda }}- | |
| release-windows-cuda-${{ matrix.cuda }}- | |
| max-size: 5G | |
| sloppiness: time_macros | |
| - name: Install CUDA Toolkit | |
| uses: ./.github/actions/windows-setup-cuda | |
| with: | |
| cuda_version: ${{ matrix.cuda }} | |
| - name: Configure CUDA environment | |
| shell: pwsh | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| $cudaPath = "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v${{ matrix.cuda }}" | |
| $nvcc = Join-Path $cudaPath "bin\nvcc.exe" | |
| if (-not (Test-Path $nvcc)) { | |
| throw "CUDA Toolkit not found at $cudaPath" | |
| } | |
| Join-Path $cudaPath "bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append | |
| $libNvvp = Join-Path $cudaPath "libnvvp" | |
| if (Test-Path $libNvvp) { | |
| $libNvvp | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append | |
| } | |
| "CUDA_PATH=$cudaPath" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append | |
| $versionEnv = "CUDA_PATH_V${{ matrix.cuda }}".Replace(".", "_") | |
| "$versionEnv=$cudaPath" | Out-File -FilePath $env:GITHUB_ENV -Encoding utf8 -Append | |
| - name: Install Ninja | |
| run: choco install ninja -y | |
| - name: Build | |
| shell: cmd | |
| run: | | |
| call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64 | |
| cmake -S . -B build -G "Ninja Multi-Config" ^ | |
| -DGGML_BACKEND_DL=ON ^ | |
| -DGGML_NATIVE=OFF ^ | |
| -DGGML_CPU=OFF ^ | |
| -DGGML_CUDA=ON ^ | |
| -DGGML_CUDA_FA=ON ^ | |
| -DLLAMA_BUILD_BORINGSSL=ON ^ | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache ^ | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ^ | |
| -DCMAKE_CUDA_COMPILER_LAUNCHER=ccache ^ | |
| -DGGML_CUDA_CUB_3DOT2=ON | |
| set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1 | |
| cmake --build build --config Release -j %NINJA_JOBS% --target ggml-cuda | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-cuda-${{ matrix.cuda }} | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Pack CUDA backend | |
| run: 7z a -snl "beellama-bin-win-cuda-${{ matrix.cuda }}-x64.zip" .\build\bin\Release\ggml-cuda.dll | |
| - name: Upload CUDA backend | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-bin-win-cuda-${{ matrix.cuda }}-x64.zip | |
| name: beellama-bin-win-cuda-${{ matrix.cuda }}-x64.zip | |
| - name: Pack CUDA runtime | |
| shell: pwsh | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| $dst = ".\build\bin\cudart" | |
| New-Item -ItemType Directory -Force -Path $dst | Out-Null | |
| $roots = @("${{ env.CUDA_PATH }}\bin", "${{ env.CUDA_PATH }}\lib", "${{ env.CUDA_PATH }}\bin\x64") | |
| foreach ($root in $roots) { | |
| if (Test-Path $root) { | |
| Get-ChildItem -Path (Join-Path $root "*") -File -ErrorAction SilentlyContinue | | |
| Where-Object { $_.Name -like "cudart64_*.dll" -or $_.Name -like "cublas64_*.dll" -or $_.Name -like "cublasLt64_*.dll" } | | |
| Copy-Item -Destination $dst -Force | |
| } | |
| } | |
| 7z a "cudart-beellama-bin-win-cuda-${{ matrix.cuda }}-x64.zip" "$dst\*" | |
| - name: Upload CUDA runtime | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: cudart-beellama-bin-win-cuda-${{ matrix.cuda }}-x64.zip | |
| name: cudart-beellama-bin-win-cuda-${{ matrix.cuda }}-x64.zip | |
| windows-hip: | |
| name: Windows x64 HIP/Radeon | |
| needs: release-meta | |
| if: ${{ needs.release-meta.outputs.build_packages == 'true' }} | |
| runs-on: windows-2022 | |
| env: | |
| HIPSDK_INSTALLER_VERSION: "26.Q1" | |
| GPU_TARGETS: "gfx1150;gfx1151;gfx1200;gfx1201;gfx1100;gfx1101;gfx1102;gfx1030;gfx1031;gfx1032" | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Setup Node.js | |
| uses: actions/setup-node@v6 | |
| with: | |
| node-version: "24" | |
| - name: Grab rocWMMA package | |
| run: | | |
| curl -o rocwmma.deb "https://repo.radeon.com/rocm/apt/7.2.1/pool/main/r/rocwmma-dev/rocwmma-dev_2.2.0.70201-81~24.04_amd64.deb" | |
| 7z x rocwmma.deb | |
| 7z x data.tar | |
| - name: ccache | |
| uses: ./.github/actions/setup-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-x64 | |
| restore-keys: | | |
| release-${{ needs.release-meta.outputs.ccache_ref }}-windows-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-x64- | |
| release-windows-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-x64- | |
| - name: Install ROCm | |
| run: | | |
| $ErrorActionPreference = "Stop" | |
| Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ env.HIPSDK_INSTALLER_VERSION }}-Win11-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe" | |
| $installTimeoutSeconds = 1800 | |
| $proc = Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -WindowStyle Hidden -PassThru | |
| $completed = $proc.WaitForExit($installTimeoutSeconds * 1000) | |
| if (-not $completed) { | |
| $proc.Kill() | |
| Write-Error "ROCm installation timed out after $($installTimeoutSeconds / 60) minutes" | |
| } | |
| if ($proc.ExitCode -ne 0) { | |
| Write-Error "ROCm installation failed with exit code $($proc.ExitCode)" | |
| } | |
| - name: Verify ROCm | |
| run: | | |
| $clangPath = Get-ChildItem 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | Select-Object -First 1 | |
| if (-not $clangPath) { | |
| Write-Error "ROCm installation not found" | |
| } | |
| & $clangPath.FullName --version | |
| - name: Build | |
| run: | | |
| $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | Split-Path | Split-Path) | |
| $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}" | |
| cmake -G "Unix Makefiles" -B build -S . ` | |
| -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" ` | |
| -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" ` | |
| -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/opt/rocm-7.2.1/include/ -Wno-ignored-attributes -Wno-nested-anon-types" ` | |
| -DCMAKE_BUILD_TYPE=Release ` | |
| -DGGML_BACKEND_DL=ON ` | |
| -DGGML_NATIVE=OFF ` | |
| -DGGML_CPU=OFF ` | |
| -DGPU_TARGETS="${env:GPU_TARGETS}" ` | |
| -DGGML_HIP_ROCWMMA_FATTN=ON ` | |
| -DGGML_HIP=ON ` | |
| -DGGML_CUDA_FA=ON ` | |
| -DCMAKE_C_COMPILER_LAUNCHER=ccache ` | |
| -DCMAKE_CXX_COMPILER_LAUNCHER=ccache ` | |
| -DCMAKE_HIP_COMPILER_LAUNCHER=ccache ` | |
| -DLLAMA_BUILD_BORINGSSL=ON | |
| cmake --build build --target ggml-hip -j ${env:NUMBER_OF_PROCESSORS} | |
| New-Item -ItemType Directory -Force -Path "build\bin\rocblas\library" | Out-Null | |
| New-Item -ItemType Directory -Force -Path "build\bin\hipblaslt\library" | Out-Null | |
| Copy-Item "${env:HIP_PATH}\bin\libhipblas.dll" "build\bin\" | |
| Copy-Item "${env:HIP_PATH}\bin\libhipblaslt.dll" "build\bin\" | |
| Copy-Item "${env:HIP_PATH}\bin\rocblas.dll" "build\bin\" | |
| Copy-Item "${env:HIP_PATH}\bin\rocblas\library\*" "build\bin\rocblas\library\" | |
| Copy-Item "${env:HIP_PATH}\bin\hipblaslt\library\*" "build\bin\hipblaslt\library\" | |
| - name: ccache stats | |
| if: always() | |
| run: ccache --show-stats | |
| - name: Save ccache | |
| if: always() | |
| uses: ./.github/actions/save-rolling-ccache | |
| with: | |
| key: release-${{ needs.release-meta.outputs.ccache_ref }}-windows-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-x64 | |
| ref: refs/heads/${{ needs.release-meta.outputs.ccache_ref }} | |
| evict-old-files: 1d | |
| - name: Pack artifacts | |
| run: | | |
| python scripts\verify-windows-package.py .\build\bin | |
| 7z a -snl "beellama-bin-win-hip-radeon-x64.zip" .\build\bin\* | |
| - name: Upload artifacts | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| path: beellama-bin-win-hip-radeon-x64.zip | |
| name: beellama-bin-win-hip-radeon-x64.zip | |
| package-assets: | |
| name: Assemble release assets | |
| needs: | |
| - release-meta | |
| - macos-arm64 | |
| - ubuntu-cpu | |
| - ubuntu-vulkan | |
| - ubuntu-rocm | |
| - ubuntu-sycl | |
| - ubuntu-cuda | |
| - windows-cpu | |
| - windows-sycl | |
| - windows-vulkan | |
| - windows-cuda | |
| - windows-hip | |
| runs-on: ubuntu-24.04 | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Download package artifacts | |
| uses: actions/download-artifact@v7 | |
| with: | |
| pattern: beellama-bin-* | |
| path: ./artifact | |
| merge-multiple: true | |
| - name: Download CUDA runtime artifacts | |
| uses: actions/download-artifact@v7 | |
| with: | |
| pattern: cudart-beellama-* | |
| path: ./artifact | |
| merge-multiple: true | |
| - name: Prepare release assets | |
| shell: bash | |
| env: | |
| TAG_NAME: ${{ needs.release-meta.outputs.tag_name }} | |
| run: | | |
| set -euo pipefail | |
| shopt -s nullglob | |
| mkdir -p release | |
| cpu_zip="artifact/beellama-bin-win-cpu-x64.zip" | |
| if [[ -f "${cpu_zip}" ]]; then | |
| temp_dir="$(mktemp -d)" | |
| unzip -q "${cpu_zip}" -d "${temp_dir}" | |
| for target_zip in artifact/beellama-bin-win-*-x64.zip; do | |
| if [[ "$(basename "${target_zip}")" == "beellama-bin-win-cpu-x64.zip" ]]; then | |
| continue | |
| fi | |
| realpath_target_zip="$(realpath "${target_zip}")" | |
| (cd "${temp_dir}" && zip -qr "${realpath_target_zip}" .) | |
| done | |
| rm -rf "${temp_dir}" | |
| fi | |
| for zip_file in artifact/beellama-bin-win-*.zip; do | |
| base_name="$(basename "${zip_file}" .zip)" | |
| mv "${zip_file}" "release/beellama-${TAG_NAME}-${base_name#beellama-}.zip" | |
| done | |
| for zip_file in artifact/cudart-beellama-bin-win-cuda-*.zip; do | |
| base_name="$(basename "${zip_file}" .zip)" | |
| suffix="${base_name#cudart-beellama-bin-win-}" | |
| mv "${zip_file}" "release/beellama-${TAG_NAME}-cudart-win-${suffix}.zip" | |
| done | |
| for tar_file in artifact/*.tar.gz; do | |
| mv "${tar_file}" release/ | |
| done | |
| python3 scripts/verify-windows-package.py release/beellama-${TAG_NAME}-bin-win-*.zip | |
| for cuda_version in 12.4 13.1; do | |
| package_zip="release/beellama-${TAG_NAME}-bin-win-cuda-${cuda_version}-x64.zip" | |
| runtime_zip="release/beellama-${TAG_NAME}-cudart-win-cuda-${cuda_version}-x64.zip" | |
| if [[ -f "${package_zip}" && -f "${runtime_zip}" ]]; then | |
| temp_dir="$(mktemp -d)" | |
| unzip -q "${package_zip}" -d "${temp_dir}" | |
| unzip -q "${runtime_zip}" -d "${temp_dir}" | |
| python3 scripts/verify-windows-package.py "${temp_dir}" | |
| rm -rf "${temp_dir}" | |
| fi | |
| done | |
| sha256sum release/* > release/SHA256SUMS.txt | |
| ls -lh release | |
| - name: Write release notes | |
| shell: bash | |
| env: | |
| TAG_NAME: ${{ needs.release-meta.outputs.tag_name }} | |
| VERSION_NAME: ${{ needs.release-meta.outputs.version_name }} | |
| PREVIEW: ${{ needs.release-meta.outputs.preview }} | |
| SOURCE_REF: ${{ needs.release-meta.outputs.source_ref }} | |
| SOURCE_SHA: ${{ needs.release-meta.outputs.source_sha }} | |
| SHORT_SHA: ${{ needs.release-meta.outputs.short_sha }} | |
| IMAGE_REPO: ${{ needs.release-meta.outputs.image_repo }} | |
| run: | | |
| set -euo pipefail | |
| release_url="${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/releases/download/${TAG_NAME}" | |
| changelog_url="${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/blob/${SOURCE_SHA}/CHANGELOG.md" | |
| repo_name="${GITHUB_REPOSITORY#*/}" | |
| repo_name="${repo_name,,}" | |
| package_url="${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/pkgs/container/${repo_name}" | |
| python3 scripts/extract-changelog.py \ | |
| --tag "${VERSION_NAME}" \ | |
| --changelog CHANGELOG.md \ | |
| --fallback-url "${changelog_url}" \ | |
| --output changelog-fragment.md | |
| python3 scripts/write-release-notes.py \ | |
| --tag "${TAG_NAME}" \ | |
| --release-url "${release_url}" \ | |
| --image-repo "${IMAGE_REPO}" \ | |
| --package-url "${package_url}" \ | |
| --changelog changelog-fragment.md \ | |
| --output release-notes.md | |
| if [[ "${PREVIEW}" == "true" ]]; then | |
| { | |
| echo "> [!WARNING]" | |
| echo "> Rolling preview build from [\`${SOURCE_REF}\`](${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/tree/${SOURCE_REF}) at [\`${SHORT_SHA}\`](${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/commit/${SOURCE_SHA}). Assets and moving Docker tags are replaced by newer successful branch builds. Last updated: $(date -u +'%d/%m/%Y %H:%M')." | |
| echo | |
| cat release-notes.md | |
| } > release-notes-preview.md | |
| mv release-notes-preview.md release-notes.md | |
| fi | |
| - name: Upload release assets | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| name: beellama-release-assets | |
| path: | | |
| release/* | |
| release-notes.md | |
| docker-build: | |
| name: Docker (${{ matrix.config.display }}) | |
| if: ${{ needs.release-meta.outputs.publish_docker == 'true' }} | |
| needs: | |
| - release-meta | |
| - ubuntu-cpu | |
| - ubuntu-vulkan | |
| - ubuntu-rocm | |
| - ubuntu-sycl | |
| - ubuntu-cuda | |
| runs-on: ${{ matrix.config.runs_on }} | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| config: | |
| - name: cpu-amd64 | |
| display: CPU amd64 | |
| group: cpu | |
| arch: amd64 | |
| platform: linux/amd64 | |
| artifact: beellama-bin-ubuntu-x64.tar.gz | |
| dockerfile: .devops/runtime-server.Dockerfile | |
| base_image: ubuntu:24.04 | |
| runtime_packages: ca-certificates curl libgomp1 | |
| runs_on: ubuntu-24.04 | |
| - name: cpu-arm64 | |
| display: CPU arm64 | |
| group: cpu | |
| arch: arm64 | |
| platform: linux/arm64 | |
| artifact: beellama-bin-ubuntu-arm64.tar.gz | |
| dockerfile: .devops/runtime-server.Dockerfile | |
| base_image: ubuntu:24.04 | |
| runtime_packages: ca-certificates curl libgomp1 | |
| runs_on: ubuntu-24.04-arm | |
| - name: cuda12 | |
| display: CUDA 12.4 | |
| group: cuda12 | |
| arch: amd64 | |
| platform: linux/amd64 | |
| artifact: beellama-bin-ubuntu-cuda-12.4-x64.tar.gz | |
| dockerfile: .devops/runtime-server.Dockerfile | |
| base_image: nvidia/cuda:12.4.1-runtime-ubuntu22.04 | |
| runtime_packages: ca-certificates curl libgomp1 | |
| runs_on: ubuntu-24.04 | |
| - name: cuda13 | |
| display: CUDA 13.1 | |
| group: cuda13 | |
| arch: amd64 | |
| platform: linux/amd64 | |
| artifact: beellama-bin-ubuntu-cuda-13.1-x64.tar.gz | |
| dockerfile: .devops/runtime-server.Dockerfile | |
| base_image: nvidia/cuda:13.1.1-runtime-ubuntu24.04 | |
| runtime_packages: ca-certificates curl libgomp1 | |
| runs_on: ubuntu-24.04 | |
| - name: rocm | |
| display: ROCm | |
| group: rocm | |
| arch: amd64 | |
| platform: linux/amd64 | |
| artifact: beellama-bin-ubuntu-rocm-7.2-x64.tar.gz | |
| dockerfile: .devops/runtime-server.Dockerfile | |
| base_image: rocm/dev-ubuntu-24.04:7.2.1-complete | |
| runtime_packages: ca-certificates curl libgomp1 | |
| runs_on: ubuntu-24.04 | |
| - name: vulkan | |
| display: Vulkan | |
| group: vulkan | |
| arch: amd64 | |
| platform: linux/amd64 | |
| artifact: beellama-bin-ubuntu-vulkan-x64.tar.gz | |
| dockerfile: .devops/runtime-server.Dockerfile | |
| base_image: ubuntu:24.04 | |
| runtime_packages: ca-certificates curl libegl1 libgl1 libgles2 libglvnd0 libglx0 libgomp1 libvulkan1 mesa-vulkan-drivers | |
| runs_on: ubuntu-24.04 | |
| - name: sycl | |
| display: SYCL | |
| group: sycl | |
| arch: amd64 | |
| platform: linux/amd64 | |
| artifact: beellama-bin-ubuntu-sycl-x64.tar.gz | |
| dockerfile: .devops/runtime-intel-server.Dockerfile | |
| base_image: unused | |
| runtime_packages: unused | |
| runs_on: ubuntu-24.04 | |
| permissions: | |
| contents: read | |
| packages: write | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Download package artifact | |
| uses: actions/download-artifact@v7 | |
| with: | |
| name: ${{ matrix.config.artifact }} | |
| path: ./package | |
| - name: Prepare Docker context | |
| shell: bash | |
| run: | | |
| set -euo pipefail | |
| mkdir -p docker-context/app | |
| tar_file="$(find package -type f -name '*.tar.gz' | head -n 1)" | |
| if [[ -z "${tar_file}" ]]; then | |
| echo "No tar.gz package found for ${{ matrix.config.artifact }}" >&2 | |
| exit 1 | |
| fi | |
| tar -xzf "${tar_file}" -C docker-context/app --strip-components=1 | |
| test -x docker-context/app/llama-server | |
| cp "${{ matrix.config.dockerfile }}" docker-context/Dockerfile | |
| ls -lh docker-context/app/llama-server | |
| - name: Set up Docker Buildx | |
| uses: docker/setup-buildx-action@4d04d5d9486b7bd6fa91e7baf45bbb4f8b9deedd # v4 | |
| - name: Log in to GHCR | |
| uses: docker/login-action@b45d80f862d83dbcd57f89517bcf500b2ab88fb2 # v4 | |
| with: | |
| registry: ghcr.io | |
| username: ${{ github.repository_owner }} | |
| password: ${{ secrets.GITHUB_TOKEN }} | |
| - name: Get build date | |
| id: build_date | |
| run: echo "date=$(date -u +"%Y-%m-%dT%H:%M:%SZ")" >> "${GITHUB_OUTPUT}" | |
| - name: Build and push image digest | |
| id: build | |
| uses: docker/build-push-action@d08e5c354a6adb9ed34480a06d141179aa583294 # v7 | |
| with: | |
| context: ./docker-context | |
| file: ./docker-context/Dockerfile | |
| platforms: ${{ matrix.config.platform }} | |
| outputs: type=image,name=${{ needs.release-meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true,oci-mediatypes=true | |
| provenance: false | |
| build-args: | | |
| BASE_IMAGE=${{ matrix.config.base_image }} | |
| RUNTIME_PACKAGES=${{ matrix.config.runtime_packages }} | |
| BUILD_DATE=${{ steps.build_date.outputs.date }} | |
| APP_VERSION=${{ needs.release-meta.outputs.tag_name }} | |
| APP_REVISION=${{ needs.release-meta.outputs.source_sha }} | |
| IMAGE_URL=${{ github.server_url }}/${{ github.repository }} | |
| IMAGE_SOURCE=${{ github.server_url }}/${{ github.repository }} | |
| annotations: | | |
| manifest:org.opencontainers.image.created=${{ steps.build_date.outputs.date }} | |
| manifest:org.opencontainers.image.version=${{ needs.release-meta.outputs.tag_name }} | |
| manifest:org.opencontainers.image.revision=${{ needs.release-meta.outputs.source_sha }} | |
| manifest:org.opencontainers.image.title=BeeLlama.cpp | |
| manifest:org.opencontainers.image.description=BeeLlama.cpp GGUF inference with DFlash, TurboQuant, and TCQ cache types | |
| manifest:org.opencontainers.image.url=${{ github.server_url }}/${{ github.repository }} | |
| manifest:org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }} | |
| cache-from: type=registry,ref=${{ needs.release-meta.outputs.image_repo }}:buildcache-runtime-${{ matrix.config.name }} | |
| cache-to: type=registry,ref=${{ needs.release-meta.outputs.image_repo }}:buildcache-runtime-${{ matrix.config.name }},mode=max | |
| - name: Record digest | |
| shell: bash | |
| run: | | |
| set -euo pipefail | |
| digest="${{ steps.build.outputs.digest }}" | |
| if [[ -z "${digest}" ]]; then | |
| echo "Docker build did not return a digest." >&2 | |
| exit 1 | |
| fi | |
| mkdir -p /tmp/docker-digests | |
| printf '%s\t%s\t%s\n' "${{ matrix.config.group }}" "${{ matrix.config.arch }}" "${digest}" > "/tmp/docker-digests/${{ matrix.config.name }}.tsv" | |
| - name: Upload digest metadata | |
| uses: actions/upload-artifact@v6 | |
| with: | |
| name: docker-digest-${{ matrix.config.name }} | |
| path: /tmp/docker-digests/${{ matrix.config.name }}.tsv | |
| if-no-files-found: error | |
| docker-merge: | |
| name: Docker tags (${{ matrix.config.display }}) | |
| if: ${{ needs.release-meta.outputs.publish_docker == 'true' }} | |
| needs: | |
| - release-meta | |
| - docker-build | |
| runs-on: ubuntu-24.04 | |
| permissions: | |
| contents: read | |
| packages: write | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| config: | |
| - group: cpu | |
| display: CPU | |
| platforms: linux/amd64 linux/arm64 | |
| tag_suffixes: none -cpu | |
| - group: cuda12 | |
| display: CUDA 12.4 | |
| platforms: linux/amd64 | |
| tag_suffixes: -cuda -cuda12 | |
| - group: cuda13 | |
| display: CUDA 13.1 | |
| platforms: linux/amd64 | |
| tag_suffixes: -cuda13 | |
| - group: rocm | |
| display: ROCm | |
| platforms: linux/amd64 | |
| tag_suffixes: -rocm | |
| - group: vulkan | |
| display: Vulkan | |
| platforms: linux/amd64 | |
| tag_suffixes: -vulkan | |
| - group: sycl | |
| display: SYCL | |
| platforms: linux/amd64 | |
| tag_suffixes: -sycl | |
| steps: | |
| - name: Download digest metadata | |
| uses: actions/download-artifact@v7 | |
| with: | |
| pattern: docker-digest-* | |
| path: /tmp/docker-digests | |
| merge-multiple: true | |
| - name: Set up Docker Buildx | |
| uses: docker/setup-buildx-action@4d04d5d9486b7bd6fa91e7baf45bbb4f8b9deedd # v4 | |
| - name: Log in to GHCR | |
| uses: docker/login-action@b45d80f862d83dbcd57f89517bcf500b2ab88fb2 # v4 | |
| with: | |
| registry: ghcr.io | |
| username: ${{ github.repository_owner }} | |
| password: ${{ secrets.GITHUB_TOKEN }} | |
| - name: Create Docker tags | |
| shell: bash | |
| env: | |
| IMAGE_REPO: ${{ needs.release-meta.outputs.image_repo }} | |
| TAG_NAME: ${{ needs.release-meta.outputs.tag_name }} | |
| STABLE_RELEASE: ${{ needs.release-meta.outputs.stable_release }} | |
| SOURCE_REF: ${{ needs.release-meta.outputs.source_ref }} | |
| SOURCE_SHA: ${{ needs.release-meta.outputs.source_sha }} | |
| SHORT_SHA: ${{ needs.release-meta.outputs.short_sha }} | |
| GROUP: ${{ matrix.config.group }} | |
| PLATFORMS: ${{ matrix.config.platforms }} | |
| TAG_SUFFIXES: ${{ matrix.config.tag_suffixes }} | |
| run: | | |
| set -euo pipefail | |
| if [[ "${STABLE_RELEASE}" != "true" ]]; then | |
| current_sha="$(git ls-remote "https://github.com/${GITHUB_REPOSITORY}.git" "refs/heads/${SOURCE_REF}" | awk '{ print $1 }')" | |
| if [[ -z "${current_sha}" || "${current_sha}" != "${SOURCE_SHA}" ]]; then | |
| echo "Branch ${SOURCE_REF} moved from ${SOURCE_SHA} to ${current_sha:-missing}; refusing to publish stale preview Docker tags." >&2 | |
| exit 1 | |
| fi | |
| fi | |
| digest_glob="/tmp/docker-digests/*.tsv" | |
| if ! ls ${digest_glob} >/dev/null 2>&1; then | |
| echo "No Docker digest metadata found." >&2 | |
| exit 1 | |
| fi | |
| sources=() | |
| for platform in ${PLATFORMS}; do | |
| arch="${platform#linux/}" | |
| digest="$(awk -F '\t' -v g="${GROUP}" -v a="${arch}" '$1 == g && $2 == a { print $3; exit }' ${digest_glob})" | |
| if [[ -z "${digest}" ]]; then | |
| echo "Missing digest for group=${GROUP} arch=${arch}" >&2 | |
| exit 1 | |
| fi | |
| sources+=("${IMAGE_REPO}@${digest}") | |
| done | |
| tag_args=() | |
| for suffix in ${TAG_SUFFIXES}; do | |
| if [[ "${suffix}" == "none" ]]; then | |
| suffix="" | |
| fi | |
| if [[ "${STABLE_RELEASE}" == "true" ]]; then | |
| tag_args+=("-t" "${IMAGE_REPO}:server${suffix}") | |
| tag_args+=("-t" "${IMAGE_REPO}:server${suffix}-${TAG_NAME}") | |
| else | |
| tag_args+=("-t" "${IMAGE_REPO}:server${suffix}-${TAG_NAME}") | |
| tag_args+=("-t" "${IMAGE_REPO}:server${suffix}-${TAG_NAME}-${SHORT_SHA}") | |
| fi | |
| done | |
| printf 'Creating Docker tags:\n' | |
| printf ' %s\n' "${tag_args[@]}" | |
| docker buildx imagetools create "${tag_args[@]}" "${sources[@]}" | |
| release: | |
| name: Publish GitHub release | |
| needs: | |
| - release-meta | |
| - package-assets | |
| - docker-merge | |
| runs-on: ubuntu-24.04 | |
| permissions: | |
| contents: write | |
| steps: | |
| - name: Clone | |
| uses: actions/checkout@v6 | |
| with: | |
| ref: ${{ needs.release-meta.outputs.source_sha }} | |
| fetch-depth: 0 | |
| - name: Prepare publication tag | |
| env: | |
| PREVIEW: ${{ needs.release-meta.outputs.preview }} | |
| SOURCE_REF: ${{ needs.release-meta.outputs.source_ref }} | |
| TAG_NAME: ${{ needs.release-meta.outputs.tag_name }} | |
| SOURCE_SHA: ${{ needs.release-meta.outputs.source_sha }} | |
| run: | | |
| set -euo pipefail | |
| if [[ "${PREVIEW}" == "true" ]]; then | |
| current_sha="$(git ls-remote origin "refs/heads/${SOURCE_REF}" | awk '{ print $1 }')" | |
| if [[ -z "${current_sha}" || "${current_sha}" != "${SOURCE_SHA}" ]]; then | |
| echo "Branch ${SOURCE_REF} moved from ${SOURCE_SHA} to ${current_sha:-missing}; refusing to publish stale preview assets." >&2 | |
| exit 1 | |
| fi | |
| git tag -f "${TAG_NAME}" "${SOURCE_SHA}" | |
| git push origin "refs/tags/${TAG_NAME}" --force | |
| else | |
| git fetch origin +refs/tags/"${TAG_NAME}":refs/tags/"${TAG_NAME}" | |
| tag_commit="$(git rev-list -n 1 "${TAG_NAME}")" | |
| if [[ "${tag_commit}" != "${SOURCE_SHA}" ]]; then | |
| echo "Tag ${TAG_NAME} moved from ${SOURCE_SHA} to ${tag_commit}; refusing to publish." >&2 | |
| exit 1 | |
| fi | |
| fi | |
| - name: Download release assets | |
| uses: actions/download-artifact@v7 | |
| with: | |
| name: beellama-release-assets | |
| path: ./release-bundle | |
| - name: Create or update release | |
| env: | |
| GH_TOKEN: ${{ secrets.GITHUB_TOKEN }} | |
| TAG_NAME: ${{ needs.release-meta.outputs.tag_name }} | |
| RELEASE_TITLE: ${{ needs.release-meta.outputs.release_title }} | |
| PREVIEW: ${{ needs.release-meta.outputs.preview }} | |
| SOURCE_SHA: ${{ needs.release-meta.outputs.source_sha }} | |
| run: | | |
| set -euo pipefail | |
| notes_file="$(find release-bundle -type f -name release-notes.md | head -n 1)" | |
| if [[ -z "${notes_file}" ]]; then | |
| echo "release-notes.md not found in release asset bundle" >&2 | |
| exit 1 | |
| fi | |
| mapfile -t release_files < <(find release-bundle -type f ! -name release-notes.md | sort) | |
| if [[ "${#release_files[@]}" -eq 0 ]]; then | |
| echo "No release files found in release asset bundle" >&2 | |
| exit 1 | |
| fi | |
| if gh release view "${TAG_NAME}" --repo "${GITHUB_REPOSITORY}" >/dev/null 2>&1; then | |
| gh release upload "${TAG_NAME}" "${release_files[@]}" --clobber --repo "${GITHUB_REPOSITORY}" | |
| if [[ "${PREVIEW}" == "true" ]]; then | |
| mapfile -t existing_assets < <(gh release view "${TAG_NAME}" --repo "${GITHUB_REPOSITORY}" --json assets --jq '.assets[].name') | |
| for existing_asset in "${existing_assets[@]}"; do | |
| keep_asset="false" | |
| for release_file in "${release_files[@]}"; do | |
| if [[ "$(basename "${release_file}")" == "${existing_asset}" ]]; then | |
| keep_asset="true" | |
| break | |
| fi | |
| done | |
| if [[ "${keep_asset}" != "true" ]]; then | |
| gh release delete-asset "${TAG_NAME}" "${existing_asset}" --yes --repo "${GITHUB_REPOSITORY}" | |
| fi | |
| done | |
| fi | |
| if [[ "${PREVIEW}" == "true" ]]; then | |
| gh release edit "${TAG_NAME}" --repo "${GITHUB_REPOSITORY}" --title "${RELEASE_TITLE}" --notes-file "${notes_file}" --target "${SOURCE_SHA}" --prerelease --latest=false --verify-tag | |
| else | |
| gh release edit "${TAG_NAME}" --repo "${GITHUB_REPOSITORY}" --title "${RELEASE_TITLE}" --notes-file "${notes_file}" --target "${SOURCE_SHA}" --prerelease=false --latest=true --verify-tag | |
| fi | |
| else | |
| if [[ "${PREVIEW}" == "true" ]]; then | |
| gh release create "${TAG_NAME}" "${release_files[@]}" --repo "${GITHUB_REPOSITORY}" --title "${RELEASE_TITLE}" --notes-file "${notes_file}" --target "${SOURCE_SHA}" --prerelease --latest=false --verify-tag | |
| else | |
| gh release create "${TAG_NAME}" "${release_files[@]}" --repo "${GITHUB_REPOSITORY}" --title "${RELEASE_TITLE}" --notes-file "${notes_file}" --target "${SOURCE_SHA}" --latest=true --verify-tag | |
| fi | |
| fi |