periodic.yml

name: periodic

on:
  schedule:
    - cron: '0,6,12,18 0 * * *'  # Runs at midnight UTC and every 6 hours
  push:
    tags:
      - ciflow/periodic/*
  workflow_dispatch:

jobs:
  gather-models:
    runs-on: ubuntu-22.04
    outputs:
      models: ${{ steps.gather-models.outputs.models }}
    steps:
      - uses: actions/checkout@v3
        with:
          submodules: 'false'
      - uses: actions/setup-python@v4
        with:
          python-version: '3.11'
      - name: Extract the list of models to test
        id: gather-models
        run: |
          set -eux
          PYTHONPATH="${PWD}" python .ci/scripts/gather_test_models.py --event "periodic"
  test-cpu:
    name: test-cpu (${{ matrix.platform }}, ${{ matrix.repo_name }})
    needs: gather-models
    strategy:
      matrix: ${{ fromJSON(needs.gather-models.outputs.models) }}
      fail-fast: false
    runs-on: ${{ matrix.runner }}
    env:
      TORCHCHAT_ROOT: ${{ github.workspace }}
      REPO_NAME: ${{ matrix.repo_name }}
      ENABKE_ET_PYBIND: ${{ matrix.runner == 'macos-14' && 'false' || 'true' }}
    steps:
      - name: Checkout repo
        uses: actions/checkout@v3
      - name: Setup Python
        uses: actions/setup-python@v4
        with:
          python-version: '3.11'
      - name: Print machine info
        run: |
          echo "$(uname -a)"
      - name: Install dependencies
        run: |
          bash ${TORCHCHAT_ROOT}/scripts/install_et.sh $ENABKE_ET_PYBIND
      - name: Download checkpoints
        run: |
          bash ${TORCHCHAT_ROOT}/.ci/scripts/wget_checkpoint.sh ${{ matrix.repo_name }} "${{ matrix.resources }}"
      - name: Run validation
        run: |
          pushd ${TORCHCHAT_ROOT}
          export CHECKPOINT_PATH=./checkpoints/${REPO_NAME}/model.pth
          bash .ci/scripts/convert_checkpoint.sh ${REPO_NAME}
          bash .ci/scripts/validate.sh ${CHECKPOINT_PATH}
  test-cuda:
    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
    name: test-cuda (linux, ${{ matrix.repo_name }})
    needs: gather-models
    strategy:
      matrix: ${{ fromJSON(needs.gather-models.outputs.models) }}
      fail-fast: false
    with:
      runner: linux.g5.4xlarge.nvidia.gpu
      gpu-arch-type: cuda
      gpu-arch-version: "12.1"
      script: |
        echo "::group::Print machine info"
        nvidia-smi
        echo "::endgroup::"

        echo "::group::Install required packages"
        pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121
        pip install -r ./requirements.txt
        pip list
        echo "::endgroup::"

        echo "::group::Download checkpoint"
        export REPO_NAME=${{ matrix.repo_name }}
        bash .ci/scripts/wget_checkpoint.sh ${REPO_NAME} ${{ matrix.resources }}
        echo "::endgroup::"

        echo "::group::Convert checkpoint"
        export CHECKPOINT_PATH=./checkpoints/${REPO_NAME}/model.pth
        bash .ci/scripts/convert_checkpoint.sh ${REPO_NAME}
        echo "::endgroup::"

        echo "::group::Run inference"
        bash .ci/scripts/validate.sh ${CHECKPOINT_PATH} cuda
        echo "::endgroup::"