|
| 1 | +name: CI (self-hosted) |
| 2 | + |
| 3 | +on: |
| 4 | + workflow_dispatch: # allows manual triggering |
| 5 | + push: |
| 6 | + branches: |
| 7 | + - master |
| 8 | + paths: [ |
| 9 | + '.github/workflows/build.yml', |
| 10 | + '**/CMakeLists.txt', |
| 11 | + '**/.cmake', |
| 12 | + '**/*.h', |
| 13 | + '**/*.hpp', |
| 14 | + '**/*.c', |
| 15 | + '**/*.cpp', |
| 16 | + '**/*.cu', |
| 17 | + '**/*.cuh', |
| 18 | + '**/*.swift', |
| 19 | + '**/*.m', |
| 20 | + '**/*.metal', |
| 21 | + '**/*.comp', |
| 22 | + '**/*.glsl', |
| 23 | + '**/*.wgsl' |
| 24 | + ] |
| 25 | + |
| 26 | + pull_request: |
| 27 | + types: [opened, synchronize, reopened] |
| 28 | + paths: [ |
| 29 | + '.github/workflows/build-self-hosted.yml', |
| 30 | + '**/CMakeLists.txt', |
| 31 | + '**/.cmake', |
| 32 | + '**/*.h', |
| 33 | + '**/*.hpp', |
| 34 | + '**/*.c', |
| 35 | + '**/*.cpp', |
| 36 | + '**/*.cu', |
| 37 | + '**/*.cuh', |
| 38 | + '**/*.swift', |
| 39 | + '**/*.m', |
| 40 | + '**/*.metal', |
| 41 | + '**/*.comp', |
| 42 | + '**/*.glsl', |
| 43 | + '**/*.wgsl' |
| 44 | + ] |
| 45 | + |
| 46 | +concurrency: |
| 47 | + group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }} |
| 48 | + cancel-in-progress: true |
| 49 | + |
| 50 | +env: |
| 51 | + GGML_NLOOP: 3 |
| 52 | + GGML_N_THREADS: 1 |
| 53 | + LLAMA_LOG_COLORS: 1 |
| 54 | + LLAMA_LOG_PREFIX: 1 |
| 55 | + LLAMA_LOG_TIMESTAMPS: 1 |
| 56 | + |
| 57 | +jobs: |
| 58 | + ggml-ci-nvidia-cuda: |
| 59 | + runs-on: [self-hosted, Linux, NVIDIA] |
| 60 | + |
| 61 | + steps: |
| 62 | + - name: Clone |
| 63 | + id: checkout |
| 64 | + uses: actions/checkout@v6 |
| 65 | + |
| 66 | + - name: Test |
| 67 | + id: ggml-ci |
| 68 | + run: | |
| 69 | + nvidia-smi |
| 70 | + GG_BUILD_CUDA=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp |
| 71 | +
|
| 72 | + ggml-ci-nvidia-vulkan-cm: |
| 73 | + runs-on: [self-hosted, Linux, NVIDIA] |
| 74 | + |
| 75 | + steps: |
| 76 | + - name: Clone |
| 77 | + id: checkout |
| 78 | + uses: actions/checkout@v6 |
| 79 | + |
| 80 | + - name: Test |
| 81 | + id: ggml-ci |
| 82 | + run: | |
| 83 | + vulkaninfo --summary |
| 84 | + GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp |
| 85 | +
|
| 86 | + ggml-ci-nvidia-vulkan-cm2: |
| 87 | + runs-on: [self-hosted, Linux, NVIDIA, COOPMAT2] |
| 88 | + |
| 89 | + steps: |
| 90 | + - name: Clone |
| 91 | + id: checkout |
| 92 | + uses: actions/checkout@v6 |
| 93 | + |
| 94 | + - name: Test |
| 95 | + id: ggml-ci |
| 96 | + run: | |
| 97 | + vulkaninfo --summary |
| 98 | + GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp |
| 99 | +
|
| 100 | + ggml-ci-cpu-amx: |
| 101 | + runs-on: [self-hosted, Linux, CPU, AMX] |
| 102 | + |
| 103 | + steps: |
| 104 | + - name: Clone |
| 105 | + id: checkout |
| 106 | + uses: actions/checkout@v6 |
| 107 | + |
| 108 | + - name: Test |
| 109 | + id: ggml-ci |
| 110 | + run: | |
| 111 | + bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp |
| 112 | +
|
| 113 | + # ggml-ci-amd-vulkan: |
| 114 | + # runs-on: [self-hosted, Linux, AMD] |
| 115 | + |
| 116 | + # steps: |
| 117 | + # - name: Clone |
| 118 | + # id: checkout |
| 119 | + # uses: actions/checkout@v6 |
| 120 | + |
| 121 | + # - name: Test |
| 122 | + # id: ggml-ci |
| 123 | + # run: | |
| 124 | + # vulkaninfo --summary |
| 125 | + # GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp |
| 126 | + |
| 127 | + # ggml-ci-amd-rocm: |
| 128 | + # runs-on: [self-hosted, Linux, AMD] |
| 129 | + |
| 130 | + # steps: |
| 131 | + # - name: Clone |
| 132 | + # id: checkout |
| 133 | + # uses: actions/checkout@v6 |
| 134 | + |
| 135 | + # - name: Test |
| 136 | + # id: ggml-ci |
| 137 | + # run: | |
| 138 | + # amd-smi static |
| 139 | + # GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp |
| 140 | + |
| 141 | + ggml-ci-mac-metal: |
| 142 | + runs-on: [self-hosted, macOS, ARM64] |
| 143 | + |
| 144 | + steps: |
| 145 | + - name: Clone |
| 146 | + id: checkout |
| 147 | + uses: actions/checkout@v6 |
| 148 | + |
| 149 | + - name: Test |
| 150 | + id: ggml-ci |
| 151 | + run: | |
| 152 | + GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp |
| 153 | +
|
| 154 | + ggml-ci-mac-webgpu: |
| 155 | + runs-on: [self-hosted, macOS, ARM64] |
| 156 | + |
| 157 | + steps: |
| 158 | + - name: Clone |
| 159 | + id: checkout |
| 160 | + uses: actions/checkout@v6 |
| 161 | + |
| 162 | + - name: Dawn Dependency |
| 163 | + id: dawn-depends |
| 164 | + run: | |
| 165 | + DAWN_VERSION="v2.0.0" |
| 166 | + DAWN_OWNER="reeselevine" |
| 167 | + DAWN_REPO="dawn" |
| 168 | + DAWN_ASSET_NAME="Dawn-5e9a4865b1635796ccc77dd30057f2b4002a1355-macos-latest-Release" |
| 169 | + echo "Fetching release asset from https://github.com/${DAWN_OWNER}/${DAWN_REPO}/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.zip" |
| 170 | + curl -L -o artifact.zip \ |
| 171 | + "https://github.com/${DAWN_OWNER}/${DAWN_REPO}/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.zip" |
| 172 | + mkdir dawn |
| 173 | + unzip artifact.zip |
| 174 | + tar -xvf ${DAWN_ASSET_NAME}.tar.gz -C dawn --strip-components=1 |
| 175 | +
|
| 176 | + - name: Test |
| 177 | + id: ggml-ci |
| 178 | + run: | |
| 179 | + GG_BUILD_WEBGPU=1 GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \ |
| 180 | + bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp |
| 181 | +
|
| 182 | + ggml-ci-mac-vulkan: |
| 183 | + runs-on: [self-hosted, macOS, ARM64] |
| 184 | + |
| 185 | + steps: |
| 186 | + - name: Clone |
| 187 | + id: checkout |
| 188 | + uses: actions/checkout@v6 |
| 189 | + |
| 190 | + - name: Test |
| 191 | + id: ggml-ci |
| 192 | + run: | |
| 193 | + vulkaninfo --summary |
| 194 | + GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp |
| 195 | +
|
| 196 | + ggml-ci-linux-intel-vulkan: |
| 197 | + runs-on: [self-hosted, Linux, Intel] |
| 198 | + |
| 199 | + steps: |
| 200 | + - name: Clone |
| 201 | + id: checkout |
| 202 | + uses: actions/checkout@v6 |
| 203 | + with: |
| 204 | + persist-credentials: false |
| 205 | + |
| 206 | + - name: Test |
| 207 | + id: ggml-ci |
| 208 | + run: | |
| 209 | + vulkaninfo --summary |
| 210 | + GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp |
| 211 | +
|
| 212 | + ggml-ci-intel-openvino-gpu-low-perf: |
| 213 | + runs-on: [self-hosted, Linux, Intel, OpenVINO] |
| 214 | + |
| 215 | + env: |
| 216 | + # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile |
| 217 | + OPENVINO_VERSION_MAJOR: "2026.0" |
| 218 | + OPENVINO_VERSION_FULL: "2026.0.0.20965.c6d6a13a886" |
| 219 | + |
| 220 | + steps: |
| 221 | + - name: Clone |
| 222 | + id: checkout |
| 223 | + uses: actions/checkout@v6 |
| 224 | + |
| 225 | + - name: Use OpenVINO Toolkit Cache |
| 226 | + uses: actions/cache@v5 |
| 227 | + id: cache-openvino |
| 228 | + with: |
| 229 | + path: ./openvino_toolkit |
| 230 | + key: openvino-toolkit-v${{ env.OPENVINO_VERSION_FULL }}-${{ runner.os }} |
| 231 | + |
| 232 | + - name: Setup OpenVINO Toolkit |
| 233 | + if: steps.cache-openvino.outputs.cache-hit != 'true' |
| 234 | + uses: ./.github/actions/linux-setup-openvino |
| 235 | + with: |
| 236 | + path: ./openvino_toolkit |
| 237 | + version_major: ${{ env.OPENVINO_VERSION_MAJOR }} |
| 238 | + version_full: ${{ env.OPENVINO_VERSION_FULL }} |
| 239 | + |
| 240 | + - name: Install OpenVINO dependencies |
| 241 | + run: | |
| 242 | + cd ./openvino_toolkit |
| 243 | + chmod +x ./install_dependencies/install_openvino_dependencies.sh |
| 244 | + echo "Y" | sudo -E ./install_dependencies/install_openvino_dependencies.sh |
| 245 | +
|
| 246 | + - name: Test |
| 247 | + id: ggml-ci |
| 248 | + run: | |
| 249 | + source ./openvino_toolkit/setupvars.sh |
| 250 | + GG_BUILD_OPENVINO=1 GGML_OPENVINO_DEVICE=GPU GG_BUILD_LOW_PERF=1 bash ./ci/run.sh ./tmp/results ./tmp/mnt |
0 commit comments