ci.yml

name: ci

permissions:
  contents: read
  pull-requests: read
  actions: read

env:
  VERSION: "${{ github.ref_name }}"

on:
  workflow_dispatch: { }
  push:
    tags:
      - "v*.*.*"
    branches:
      - main
    paths-ignore:
      - "docs/**"
      - "**.md"
      - "**.mdx"
      - "**.png"
      - "**.jpg"
  pull_request:
    branches:
      - main
    paths-ignore:
      - "docs/**"
      - "**.md"
      - "**.mdx"
      - "**.png"
      - "**.jpg"

concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
  cancel-in-progress: true

jobs:
  darwin-metal:
    if: ${{ false }}
    strategy:
      fail-fast: false
      matrix:
        # see https://github.com/actions/runner-images?tab=readme-ov-file#available-images.
        os: [ macos-13, macos-14 ]
    runs-on: ${{ matrix.os }}
    steps:
      - name: Setup XCode
        if: ${{ matrix.os == 'macos-13' }}
        uses: maxim-lobanov/setup-xcode@v1
        with:
          xcode-version: '15.2'
      - name: Checkout
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          submodules: 'recursive'
      - name: Deps
        continue-on-error: true
        run: |
          brew update
      - name: Build
        run: |
          echo "===== BUILD ====="
          make -j LLAMA_METAL=1
          
          echo "===== RESULT ====="
          file ./.dist/llama-box
      - name: Release
        if: ${{ startsWith(github.ref, 'refs/tags/') }}
        uses: actions/upload-artifact@v4
        with:
          path: ./.dist/llama-box
          name: llama-box-darwin-${{ endsWith(matrix.os, '-13') && 'amd64' || 'arm64' }}-metal

  linux-hip:
    if: ${{ false }}
    strategy:
      fail-fast: false
      matrix:
        arch: [ amd64 ]
        version: [ '6.0.2' ]
    runs-on: ubuntu-22.04
    steps:
      - name: Maximize Space
        # see https://github.com/easimon/maximize-build-space/blob/master/action.yml.
        run: |
          sudo rm -rf /usr/share/dotnet
          sudo rm -rf /usr/local/lib/android
          sudo rm -rf /opt/ghc
          sudo rm -rf /opt/hostedtoolcache/CodeQL
          sudo docker image prune --all --force
      - name: Clone
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          submodules: 'recursive'
      - name: Setup QEMU
        uses: docker/setup-qemu-action@v3
        with:
          image: tonistiigi/binfmt:qemu-v7.0.0
          platforms: "arm64"
      - name: Build
        # disable OpenMP to support static linking,
        # see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
        #     https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
        run: |
          echo "===== SCRIPT ====="
          cat <<EOF > /tmp/entrypoint.sh
          #!/bin/bash
          apt-get update && apt-get install -y build-essential git rocblas-dev hipblas-dev libgomp1
          make -j LLAMA_HIPBLAS=1 LLAMA_NO_OPENMP=1
          EOF
          chmod +x /tmp/entrypoint.sh
          cat /tmp/entrypoint.sh
          
          echo "===== BUILD ====="
          docker run \
            --rm \
            --privileged \
            --platform linux/${{ matrix.arch }} \
            --volume $(pwd):/workspace \
            --volume /tmp/entrypoint.sh:/entrypoint.sh \
            --entrypoint /entrypoint.sh \
            --workdir /workspace \
            --env CC=/opt/rocm/llvm/bin/clang \
            --env CXX=/opt/rocm/llvm/bin/clang++ \
            --env GPU_TARGETS="gfx803 gfx900 gfx906 gfx908 gfx90a gfx1010 gfx1030 gfx1100 gfx1101 gfx1102" \
            rocm/dev-ubuntu-22.04:${{ matrix.version }}
          
          echo "===== RESULT ====="
          file ./.dist/llama-box
      - name: Release
        if: ${{ startsWith(github.ref, 'refs/tags/') }}
        uses: actions/upload-artifact@v4
        with:
          path: ./.dist/llama-box
          name: llama-box-linux-${{ matrix.arch }}-hip

  linux-cuda:
    strategy:
      fail-fast: false
      matrix:
        arch: [ amd64, arm64 ]
        version: [ '11.7.1' ]
    runs-on: ubuntu-22.04
    steps:
      - name: Maximize Space
        # see https://github.com/easimon/maximize-build-space/blob/master/action.yml.
        run: |
          sudo rm -rf /usr/share/dotnet
          sudo rm -rf /usr/local/lib/android
          sudo rm -rf /opt/ghc
          sudo rm -rf /opt/hostedtoolcache/CodeQL
          sudo docker image prune --all --force
      - name: Clone
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
          submodules: 'recursive'
      - name: Setup QEMU
        uses: docker/setup-qemu-action@v3
        with:
          image: tonistiigi/binfmt:qemu-v7.0.0
          platforms: "arm64"
      - name: Build
        # disable OpenMP to support static linking,
        # see https://github.com/ggerganov/llama.cpp/issues/7743#issuecomment-2148342691,
        #     https://github.com/ggerganov/llama.cpp/issues/7719#issuecomment-2147631216.
        run: |
          echo "===== SCRIPT ====="
          cat <<EOF > /tmp/entrypoint.sh
          #!/bin/bash
          apt-get update && apt-get install -y build-essential git libgomp1
          make -j LLAMA_CUDA=1 LLAMA_NO_OPENMP=1
          EOF
          chmod +x /tmp/entrypoint.sh
          cat /tmp/entrypoint.sh
          
          echo "===== BUILD ====="
          docker run \
            --rm \
            --privileged \
            --platform linux/${{ matrix.arch }} \
            --volume $(pwd):/workspace \
            --workdir /workspace \
            --volume /tmp/entrypoint.sh:/entrypoint.sh \
            --entrypoint /entrypoint.sh \
            --env CUDA_DOCKER_ARCH=all \
            nvidia/cuda:${{ matrix.version }}-devel-ubuntu22.04
          
          echo "===== RESULT ====="
          file ./.dist/llama-box
      - name: Release
        if: ${{ startsWith(github.ref, 'refs/tags/') }}
        uses: actions/upload-artifact@v4
        with:
          path: ./.dist/llama-box
          name: llama-box-linux-${{ matrix.arch }}-cuda