Felipe97/llama-cpp-compiled

1name: CI (CUDA, ubuntu)2 3on:4  workflow_dispatch: # allows manual triggering5  push:6    branches:7      - master8    paths: [9      '.github/workflows/build-cuda-ubuntu.yml',10      '**/CMakeLists.txt',11      '**/.cmake',12      '**/*.h',13      '**/*.hpp',14      '**/*.c',15      '**/*.cpp',16      '**/*.cu',17      '**/*.cuh'18    ]19 20  pull_request:21    types: [opened, synchronize, reopened]22    paths: [23      '.github/workflows/build-cuda-ubuntu.yml',24      'ggml/src/ggml-cuda/**'25    ]26 27concurrency:28  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}29  cancel-in-progress: true30 31env:32  GGML_NLOOP: 333  GGML_N_THREADS: 134  LLAMA_ARG_LOG_COLORS: 135  LLAMA_ARG_LOG_PREFIX: 136  LLAMA_ARG_LOG_TIMESTAMPS: 137 38jobs:39  cuda:40    runs-on: ubuntu-24.0441    container: nvidia/cuda:12.6.2-devel-ubuntu24.0442 43    steps:44      - name: Clone45        id: checkout46        uses: actions/checkout@v647 48      - name: Install dependencies49        env:50          DEBIAN_FRONTEND: noninteractive51        run: |52          apt update53          apt install -y cmake build-essential ninja-build libgomp1 git libssl-dev jq python3 python3-venv python3-pip54 55      - name: ccache56        uses: ggml-org/ccache-action@v1.2.2457        with:58          key: cuda-ubuntu-24.04-cuda59          save: false60 61      - name: ccache-buckets-restore62        uses: ./.github/actions/ccache-buckets63        env:64          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}65        with:66          key: cuda-ubuntu-24.04-cuda67          folder: llama.cpp68          hf_bucket: ggml-org/cache69 70      - name: Build with CMake71        # TODO: Remove GGML_CUDA_CUB_3DOT2 flag once CCCL 3.2 is bundled within CTK and that CTK version is used in this project72        run: |73          cmake -S . -B build -G Ninja \74            -DLLAMA_FATAL_WARNINGS=ON \75            -DCMAKE_BUILD_TYPE=Release \76            -DCMAKE_CUDA_ARCHITECTURES=89-real \77            -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \78            -DGGML_NATIVE=OFF \79            -DGGML_CUDA=ON \80            -DGGML_CUDA_CUB_3DOT2=ON81          cmake --build build82 83      - name: ccache-buckets-save84        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}85        uses: ./.github/actions/ccache-buckets86        env:87          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}88        with:89          key: cuda-ubuntu-24.04-cuda90          folder: llama.cpp91          evict-old-files: 1d92          hf_bucket: ggml-org/cache93          save: true94 95  hip:96    runs-on: ubuntu-22.0497    container: rocm/dev-ubuntu-22.04:6.1.298 99    steps:100      - name: Clone101        id: checkout102        uses: actions/checkout@v6103 104      - name: Dependencies105        id: depends106        run: |107          sudo apt-get update108          sudo apt-get install -y build-essential git cmake rocblas-dev hipblas-dev libssl-dev rocwmma-dev jq python3-venv109 110      - name: ccache111        uses: ggml-org/ccache-action@v1.2.24112        with:113          key: cuda-ubuntu-22.04-hip114          save: false115 116      - name: ccache-buckets-restore117        uses: ./.github/actions/ccache-buckets118        env:119          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}120        with:121          key: cuda-ubuntu-22.04-hip122          folder: llama.cpp123          hf_bucket: ggml-org/cache124 125      - name: Build with native CMake HIP support126        id: cmake_build127        run: |128          cmake -B build -S . \129            -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \130            -DGPU_TARGETS="gfx1030" \131            -DGGML_HIP=ON132          cmake --build build --config Release -j $(nproc)133 134      - name: ccache-buckets-save135        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}136        uses: ./.github/actions/ccache-buckets137        env:138          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}139        with:140          key: cuda-ubuntu-22.04-hip141          folder: llama.cpp142          evict-old-files: 1d143          hf_bucket: ggml-org/cache144          save: true145 146  musa:147    runs-on: ubuntu-22.04148    container: mthreads/musa:rc4.3.0-devel-ubuntu22.04-amd64149 150    steps:151      - name: Clone152        id: checkout153        uses: actions/checkout@v6154 155      - name: Dependencies156        id: depends157        run: |158          apt-get update159          apt-get install -y build-essential git cmake libssl-dev jq160 161      - name: ccache162        uses: ggml-org/ccache-action@v1.2.24163        with:164          key: cuda-ubuntu-22.04-musa165          save: false166 167      - name: ccache-buckets-restore168        uses: ./.github/actions/ccache-buckets169        env:170          HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}171        with:172          key: cuda-ubuntu-22.04-musa173          folder: llama.cpp174          hf_bucket: ggml-org/cache175 176      - name: Build with native CMake MUSA support177        id: cmake_build178        run: |179          cmake -B build -S . \180            -DGGML_MUSA=ON \181            -DMUSA_ARCHITECTURES=21182          time cmake --build build --config Release -j $(nproc)183 184      - name: ccache-buckets-save185        if: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}186        uses: ./.github/actions/ccache-buckets187        env:188          HF_TOKEN: ${{ secrets.HF_TOKEN_CACHE_OUTPUT }}189        with:190          key: cuda-ubuntu-22.04-musa191          folder: llama.cpp192          evict-old-files: 1d193          hf_bucket: ggml-org/cache194          save: true195