name: CI (self-hosted Metal backend)

on:
  workflow_dispatch: # allows manual triggering
  push:
    branches:
      - master
    paths: [
      '.github/workflows/ci-self-hosted-metal.yml',
      'ci/run.sh',
      '**/CMakeLists.txt',
      '**/.cmake',
      '**/*.h',
      '**/*.hpp',
      '**/*.c',
      '**/*.cpp',
      '**/*.swift',
      '**/*.m',
      '**/*.metal'
    ]

  pull_request:
    types: [opened, synchronize, reopened]
    paths: [
      '.github/workflows/ci-self-hosted-metal.yml',
      'ci/run.sh',
      '**/CMakeLists.txt',
      '**/.cmake',
      'ggml/src/*',
      'ggml/src/ggml-cpu/**',
      'ggml/src/ggml-metal/**'
    ]

cache-mode: none
permissions:
  contents: read

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

env:
  # note: this is dud token to avoid rate limiting (https://github.com/ggml-org/llama.cpp/pull/25706#issuecomment-4979941302)
  HF_TOKEN: ${{ secrets.HF_TOKEN_CI }}
  GGML_NLOOP: 3
  GGML_N_THREADS: 1
  LLAMA_ARG_LOG_COLORS: 1
  LLAMA_ARG_LOG_PREFIX: 1
  LLAMA_ARG_LOG_TIMESTAMPS: 1

jobs:
  gpu-metal:
    runs-on: [self-hosted, macOS, ARM64]

    steps:
      - name: Clone
        id: checkout
        uses: actions/checkout@v6

      - name: Test
        id: ggml-ci
        run: |
          GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp/$RUNNER_NAME ~/mnt/llama.cpp