name: Self-hosted runner (scheduled)

# Note that each job's dependencies go into a corresponding docker file.
#
# For example for `run_torch_cuda_extensions_gpu` the docker image is
# `huggingface/transformers-pytorch-deepspeed-latest-gpu`, which can be found at
# `docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile`

on:
  workflow_call:
    inputs:
      job:
        required: true
        type: string
      slack_report_channel:
        required: true
        type: string
      runner:
        required: true
        type: string
      docker:
        required: true
        type: string
      ci_event:
        required: true
        type: string
      working-directory-prefix:
        default: ''
        required: false
        type: string

env:
  HF_HOME: /mnt/cache
  TRANSFORMERS_IS_CI: yes
  OMP_NUM_THREADS: 8
  MKL_NUM_THREADS: 8
  RUN_SLOW: yes
  # For gated repositories, we still need to agree to share information on the Hub repo. page in order to get access.
  # This token is created under the bot `hf-transformers-bot`.
  HF_HUB_READ_TOKEN: ${{ secrets.HF_HUB_READ_TOKEN }}
  SIGOPT_API_TOKEN: ${{ secrets.SIGOPT_API_TOKEN }}
  TF_FORCE_GPU_ALLOW_GROWTH: true
  CUDA_VISIBLE_DEVICES: 0,1
  NUM_SLICES: 2

jobs:
  setup:
    if: contains(fromJSON('["run_models_gpu", "run_trainer_and_fsdp_gpu", "run_quantization_torch_gpu"]'), inputs.job)
    name: Setup
    strategy:
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
    runs-on:
      group: '${{ matrix.machine_type }}'
    container:
      image: huggingface/transformers-all-latest-gpu
      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    outputs:
      folder_slices: ${{ steps.set-matrix.outputs.folder_slices }}
      slice_ids: ${{ steps.set-matrix.outputs.slice_ids }}
      quantization_matrix: ${{ steps.set-matrix-quantization.outputs.quantization_matrix }}
    steps:
      - name: Update clone
        working-directory: /transformers
        run: |
          git fetch && git checkout ${{ github.sha }}

      - name: Cleanup
        working-directory: /transformers
        run: |
          rm -rf tests/__pycache__
          rm -rf tests/models/__pycache__
          rm -rf reports

      - name: Show installed libraries and their versions
        working-directory: /transformers
        run: pip freeze

      - id: set-matrix
        if: contains(fromJSON('["run_models_gpu", "run_trainer_and_fsdp_gpu"]'), inputs.job)
        name: Identify models to test
        working-directory: /transformers/tests
        run: |
          if [ "${{ inputs.job }}" = "run_models_gpu" ]; then
            echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
            echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
          elif [ "${{ inputs.job }}" = "run_trainer_and_fsdp_gpu" ]; then
            echo "folder_slices=[['trainer'], ['fsdp']]" >> $GITHUB_OUTPUT
            echo "slice_ids=[0, 1]" >> $GITHUB_OUTPUT
          fi

      - id: set-matrix-quantization
        if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
        name: Identify quantization method to test
        working-directory: /transformers/tests
        run: |
          echo "quantization_matrix=$(python3 -c 'import os; tests = os.getcwd(); quantization_tests = os.listdir(os.path.join(tests, "quantization")); d = sorted(list(filter(os.path.isdir, [f"quantization/{x}" for x in quantization_tests]))) ;  print(d)')" >> $GITHUB_OUTPUT

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

  run_models_gpu:
    if: ${{ inputs.job == 'run_models_gpu' }}
    name: " "
    needs: setup
    strategy:
      fail-fast: false
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
        slice_id: ${{ fromJSON(needs.setup.outputs.slice_ids) }}
    uses: ./.github/workflows/model_jobs.yml
    with:
      folder_slices: ${{ needs.setup.outputs.folder_slices }}
      machine_type: ${{ matrix.machine_type }}
      slice_id: ${{ matrix.slice_id }}
      runner: ${{ inputs.runner }}
      docker: ${{ inputs.docker }}
    secrets: inherit

  run_trainer_and_fsdp_gpu:
    if: ${{ inputs.job == 'run_trainer_and_fsdp_gpu' }}
    name: " "
    needs: setup
    strategy:
      fail-fast: false
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
        slice_id: [0, 1]
    uses: ./.github/workflows/model_jobs.yml
    with:
      folder_slices: ${{ needs.setup.outputs.folder_slices }}
      machine_type: ${{ matrix.machine_type }}
      slice_id: ${{ matrix.slice_id }}
      runner: ${{ inputs.runner }}
      docker: ${{ inputs.docker }}
      report_name_prefix: run_trainer_and_fsdp_gpu
    secrets: inherit

  run_pipelines_torch_gpu:
    if: ${{ inputs.job == 'run_pipelines_torch_gpu' }}
    name: PyTorch pipelines
    strategy:
      fail-fast: false
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
    runs-on:
      group: '${{ matrix.machine_type }}'
    container:
      image: huggingface/transformers-pytorch-gpu
      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - name: Update clone
        working-directory: /transformers
        run: git fetch && git checkout ${{ github.sha }}

      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
        working-directory: /transformers
        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

      - name: Environment
        working-directory: /transformers
        run: |
          python3 utils/print_env.py

      - name: Show installed libraries and their versions
        working-directory: /transformers
        run: pip freeze

      - name: Set `machine_type` for report and artifact names
        working-directory: /transformers
        shell: bash
        run: |
          echo "${{ matrix.machine_type }}"

          if [ "${{ matrix.machine_type }}" = "aws-g4dn-4xlarge-cache" ]; then
            machine_type=single-gpu
          elif [ "${{ matrix.machine_type }}" = "aws-g4dn-12xlarge-cache" ]; then
            machine_type=multi-gpu
          else
            machine_type=${{ matrix.machine_type }}
          fi

          echo "$machine_type"
          echo "machine_type=$machine_type" >> $GITHUB_ENV

      - name: Run all pipeline tests on GPU
        working-directory: /transformers
        run: |
          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports tests/pipelines

      - name: Failure short reports
        if: ${{ failure() }}
        continue-on-error: true
        run: cat /transformers/reports/${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports/failures_short.txt

      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports"
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: ${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports
          path: /transformers/reports/${{ env.machine_type }}_run_pipelines_torch_gpu_test_reports

  run_pipelines_tf_gpu:
    if: ${{ inputs.job == 'run_pipelines_tf_gpu' }}
    name: TensorFlow pipelines
    strategy:
      fail-fast: false
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
    runs-on:
      group: '${{ matrix.machine_type }}'
    container:
      image: huggingface/transformers-tensorflow-gpu
      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - name: Update clone
        working-directory: /transformers
        run: |
          git fetch && git checkout ${{ github.sha }}

      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
        working-directory: /transformers
        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

      - name: Environment
        working-directory: /transformers
        run: |
          python3 utils/print_env.py

      - name: Show installed libraries and their versions
        working-directory: /transformers
        run: pip freeze

      - name: Set `machine_type` for report and artifact names
        working-directory: /transformers
        shell: bash
        run: |
          echo "${{ matrix.machine_type }}"

          if [ "${{ matrix.machine_type }}" = "aws-g4dn-4xlarge-cache" ]; then
            machine_type=single-gpu
          elif [ "${{ matrix.machine_type }}" = "aws-g4dn-12xlarge-cache" ]; then
            machine_type=multi-gpu
          else
            machine_type=${{ matrix.machine_type }}
          fi

          echo "$machine_type"
          echo "machine_type=$machine_type" >> $GITHUB_ENV

      - name: Run all pipeline tests on GPU
        working-directory: /transformers
        run: |
          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ env.machine_type }}_run_pipelines_tf_gpu_test_reports tests/pipelines

      - name: Failure short reports
        if: ${{ always() }}
        run: |
          cat /transformers/reports/${{ env.machine_type }}_run_pipelines_tf_gpu_test_reports/failures_short.txt

      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_pipelines_tf_gpu_test_reports"
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: ${{ env.machine_type }}_run_pipelines_tf_gpu_test_reports
          path: /transformers/reports/${{ env.machine_type }}_run_pipelines_tf_gpu_test_reports

  run_examples_gpu:
    if: ${{ inputs.job == 'run_examples_gpu' }}
    name: Examples directory
    strategy:
      fail-fast: false
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache]
    runs-on:
      group: '${{ matrix.machine_type }}'
    container:
      image: huggingface/transformers-all-latest-gpu
      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - name: Update clone
        working-directory: /transformers
        run: git fetch && git checkout ${{ github.sha }}

      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
        working-directory: /transformers
        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

      - name: Environment
        working-directory: /transformers
        run: |
          python3 utils/print_env.py

      - name: Show installed libraries and their versions
        working-directory: /transformers
        run: pip freeze

      - name: Set `machine_type` for report and artifact names
        working-directory: /transformers
        shell: bash
        run: |
          echo "${{ matrix.machine_type }}"

          if [ "${{ matrix.machine_type }}" = "aws-g4dn-4xlarge-cache" ]; then
            machine_type=single-gpu
          elif [ "${{ matrix.machine_type }}" = "aws-g4dn-12xlarge-cache" ]; then
            machine_type=multi-gpu
          else
            machine_type=${{ matrix.machine_type }}
          fi

          echo "$machine_type"
          echo "machine_type=$machine_type" >> $GITHUB_ENV

      - name: Run examples tests on GPU
        working-directory: /transformers
        run: |
          pip install -r examples/pytorch/_tests_requirements.txt
          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_examples_gpu_test_reports examples/pytorch

      - name: Failure short reports
        if: ${{ failure() }}
        continue-on-error: true
        run: cat /transformers/reports/${{ env.machine_type }}_run_examples_gpu_test_reports/failures_short.txt

      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_examples_gpu_test_reports"
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: ${{ env.machine_type }}_run_examples_gpu_test_reports
          path: /transformers/reports/${{ env.machine_type }}_run_examples_gpu_test_reports

  run_torch_cuda_extensions_gpu:
    if: ${{ inputs.job == 'run_torch_cuda_extensions_gpu' }}
    name: Torch CUDA extension tests
    strategy:
      fail-fast: false
      matrix:
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
    runs-on:
      group: '${{ matrix.machine_type }}'
    container:
      image: ${{ inputs.docker }}
      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - name: Update clone
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        run: git fetch && git checkout ${{ github.sha }}

      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .

      - name: Update / Install some packages (for Past CI)
        if: ${{ contains(inputs.docker, '-past-') && contains(inputs.docker, '-pytorch-') }}
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        run: |
          python3 -m pip install -U datasets
          python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate

      - name: Remove cached torch extensions
        run: rm -rf /github/home/.cache/torch_extensions/

      # To avoid unknown test failures
      - name: Pre build DeepSpeed *again* (for daily CI)
        if: ${{ contains(inputs.ci_event, 'Daily CI') }}
        working-directory: ${{ inputs.working-directory-prefix }}/
        run: |
          python3 -m pip uninstall -y deepspeed
          DS_DISABLE_NINJA=1 DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install deepspeed --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check

      # To avoid unknown test failures
      - name: Pre build DeepSpeed *again* (for nightly & Past CI)
        if: ${{ contains(inputs.ci_event, 'Nightly CI') || contains(inputs.ci_event, 'Past CI') }}
        working-directory: ${{ inputs.working-directory-prefix }}/
        run: |
          python3 -m pip uninstall -y deepspeed
          rm -rf DeepSpeed
          git clone https://github.com/deepspeedai/DeepSpeed && cd DeepSpeed && rm -rf build
          DS_BUILD_CPU_ADAM=1 DS_BUILD_FUSED_ADAM=1 python3 -m pip install . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

      - name: Environment
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        run: |
          python3 utils/print_env.py

      - name: Show installed libraries and their versions
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        run: pip freeze

      - name: Set `machine_type` for report and artifact names
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        shell: bash
        run: |
          echo "${{ matrix.machine_type }}"

          if [ "${{ matrix.machine_type }}" = "aws-g4dn-4xlarge-cache" ]; then
            machine_type=single-gpu
          elif [ "${{ matrix.machine_type }}" = "aws-g4dn-12xlarge-cache" ]; then
            machine_type=multi-gpu
          else
            machine_type=${{ matrix.machine_type }}
          fi

          echo "$machine_type"
          echo "machine_type=$machine_type" >> $GITHUB_ENV

      - name: Run all tests on GPU
        working-directory: ${{ inputs.working-directory-prefix }}/transformers
        run: |
          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports tests/deepspeed tests/extended

      - name: Failure short reports
        if: ${{ failure() }}
        continue-on-error: true
        run: cat ${{ inputs.working-directory-prefix }}/transformers/reports/${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports/failures_short.txt

      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports"
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: ${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports
          path: ${{ inputs.working-directory-prefix }}/transformers/reports/${{ env.machine_type }}_run_torch_cuda_extensions_gpu_test_reports

  run_quantization_torch_gpu:
    if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
    name: " "
    needs: setup
    strategy:
      max-parallel: 4
      fail-fast: false
      matrix:
        folders: ${{ fromJson(needs.setup.outputs.quantization_matrix) }}
        machine_type: [aws-g4dn-4xlarge-cache, aws-g4dn-12xlarge-cache]
    runs-on:
      group: '${{ matrix.machine_type }}'
    container:
      image: huggingface/transformers-quantization-latest-gpu
      options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
    steps:
      - name: Echo folder ${{ matrix.folders }}
        shell: bash
        run: |
          echo "${{ matrix.folders }}"
          matrix_folders=${{ matrix.folders }}
          matrix_folders=${matrix_folders/'quantization/'/'quantization_'}
          echo "$matrix_folders"
          echo "matrix_folders=$matrix_folders" >> $GITHUB_ENV

      - name: Update clone
        working-directory: /transformers
        run: git fetch && git checkout ${{ github.sha }}

      - name: Reinstall transformers in edit mode (remove the one installed during docker image build)
        working-directory: /transformers
        run: python3 -m pip uninstall -y transformers && python3 -m pip install -e .

      - name: NVIDIA-SMI
        run: |
          nvidia-smi

      - name: Environment
        working-directory: /transformers
        run: |
          python3 utils/print_env.py

      - name: Show installed libraries and their versions
        working-directory: /transformers
        run: pip freeze

      - name: Set `machine_type` for report and artifact names
        working-directory: /transformers
        shell: bash
        run: |
          echo "${{ matrix.machine_type }}"

          if [ "${{ matrix.machine_type }}" = "aws-g4dn-4xlarge-cache" ]; then
            machine_type=single-gpu
          elif [ "${{ matrix.machine_type }}" = "aws-g4dn-12xlarge-cache" ]; then
            machine_type=multi-gpu
          else
            machine_type=${{ matrix.machine_type }}
          fi

          echo "$machine_type"
          echo "machine_type=$machine_type" >> $GITHUB_ENV

      - name: Run quantization tests on GPU
        working-directory: /transformers
        run: |
          python3 -m pytest -v --make-reports=${{ env.machine_type }}_run_quantization_torch_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}

      - name: Failure short reports
        if: ${{ failure() }}
        continue-on-error: true
        run: cat /transformers/reports/${{ env.machine_type }}_run_quantization_torch_gpu_${{ matrix.folders }}_test_reports/failures_short.txt

      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_quantization_torch_gpu_${{ env.matrix_folders }}_test_reports"
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: ${{ env.machine_type }}_run_quantization_torch_gpu_${{ env.matrix_folders }}_test_reports
          path: /transformers/reports/${{ env.machine_type }}_run_quantization_torch_gpu_${{ matrix.folders }}_test_reports

  run_extract_warnings:
    # Let's only do this for the job `run_models_gpu` to simplify the (already complex) logic.
    if: ${{ always() && inputs.job == 'run_models_gpu' }}
    name: Extract warnings in CI artifacts
    runs-on: ubuntu-22.04
    needs: [setup, run_models_gpu]
    steps:
      - name: Checkout transformers
        uses: actions/checkout@v4
        with:
          fetch-depth: 2

      - name: Install transformers
        run: pip install transformers

      - name: Show installed libraries and their versions
        run: pip freeze

      - name: Create output directory
        run: mkdir warnings_in_ci

      - uses: actions/download-artifact@v4
        with:
          path: warnings_in_ci

      - name: Show artifacts
        run: echo "$(python3 -c 'import os; d = os.listdir(); print(d)')"
        working-directory: warnings_in_ci

      - name: Extract warnings in CI artifacts
        run: |
          python3 utils/extract_warnings.py --workflow_run_id ${{ github.run_id }} --output_dir warnings_in_ci --token ${{ secrets.ACCESS_REPO_INFO_TOKEN }} --from_gh
          echo "$(python3 -c 'import os; import json; fp = open("warnings_in_ci/selected_warnings.json"); d = json.load(fp); d = "\n".join(d) ;print(d)')"

      - name: Upload artifact
        if: ${{ always() }}
        uses: actions/upload-artifact@v4
        with:
          name: warnings_in_ci
          path: warnings_in_ci/selected_warnings.json

  send_results:
    name: Slack Report
    needs: [
      setup,
      run_models_gpu,
      run_trainer_and_fsdp_gpu,
      run_pipelines_torch_gpu,
      run_pipelines_tf_gpu,
      run_examples_gpu,
      run_torch_cuda_extensions_gpu,
      run_quantization_torch_gpu,
      run_extract_warnings
    ]
    if: ${{ always() }}
    uses: ./.github/workflows/slack-report.yml
    with:
      job: ${{ inputs.job }}
      # This would be `skipped` if `setup` is skipped.
      setup_status: ${{ needs.setup.result }}
      slack_report_channel: ${{ inputs.slack_report_channel }}
      # This would be an empty string if `setup` is skipped.
      folder_slices: ${{ needs.setup.outputs.folder_slices }}
      quantization_matrix: ${{ needs.setup.outputs.quantization_matrix }}
      ci_event: ${{ inputs.ci_event }}

    secrets: inherit

  check_new_model_failures:
    if: ${{ always() && inputs.ci_event == 'Daily CI' && inputs.job == 'run_models_gpu' && needs.send_results.result == 'success' }}
    name: Check new model failures
    needs: send_results
    uses: ./.github/workflows/check_failed_model_tests.yml
    with:
      docker: ${{ inputs.docker }}
      start_sha: ${{ github.sha }}
    secrets: inherit