lightning/.azure/gpu-tests-pytorch.yml

# Python package
# Create and test a Python package on multiple Python versions.
# Add steps that analyze code, save the dist with the build record, publish to a PyPI-compatible index, and more:
# https://docs.microsoft.com/azure/devops/pipelines/languages/python

trigger:
  tags:
    include:
      - '*'
  branches:
    include:
      - "master"
      - "release/*"
      - "refs/tags/*"

pr:
  branches:
    include:
      - "master"
      - "release/*"
  paths:
    include:
      - ".actions/**"
      - ".azure/gpu-tests-pytorch.yml"
      - "examples/run_pl_examples.sh"
      - "examples/pl_basics/backbone_image_classifier.py"
      - "examples/pl_basics/autoencoder.py"
      - "examples/pl_fault_tolerant/automatic.py"
      - "requirements/pytorch/**"
      - "src/pytorch_lightning/**"
      - "tests/tests_pytorch/**"
      - "setup.cfg"  # includes pytest config
      - "requirements/fabric/**"
      - "src/lightning_fabric/**"
    exclude:
      - "requirements/*/docs.txt"
      - "*.md"
      - "**/*.md"

jobs:
  - job: testing
    strategy:
      matrix:
        'PyTorch & strategies':  # this uses torch 1.12 as not all strategies support 1.13 yet
          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.9-torch1.12-cuda11.6.1"
          scope: "strategies"
        'PyTorch - latest':
          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.9-torch1.13-cuda11.7.1"
          scope: ""
    # how long to run the job before automatically cancelling
    timeoutInMinutes: "80"
    # how much time to give 'run always even if cancelled tasks' before stopping them
    cancelTimeoutInMinutes: "2"
    pool: lit-rtx-3090
    variables:
      DEVICES: $( python -c 'print("$(Agent.Name)".split("_")[-1])' )
    container:
      image: $(image)
      # default shm size is 64m. Increase it to avoid:
      # 'Error while creating shared memory: unhandled system error, NCCL version 2.7.8'
      options: "--gpus=all --shm-size=2gb"
    workspace:
      clean: all

    steps:

    - bash: |
        echo "##vso[task.setvariable variable=CUDA_VISIBLE_DEVICES]$(DEVICES)"
        cuda_ver=$(python -c "import torch ; print(''.join(map(str, torch.version.cuda.split('.')[:2])))")
        echo "##vso[task.setvariable variable=CUDA_VERSION_MM]$cuda_ver"
        echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/cu${cuda_ver}/torch_stable.html"        
      displayName: 'set env. vars'

    - bash: |
        echo $CUDA_VISIBLE_DEVICES
        echo $CUDA_VERSION_MM
        echo $TORCH_URL
        lspci | egrep 'VGA|3D'
        whereis nvidia
        nvidia-smi
        which python && which pip
        python --version
        pip --version
        pip list        
      displayName: 'Image info & NVIDIA'

    - bash: |
        PYTORCH_VERSION=$(python -c "import torch; print(torch.__version__.split('+')[0])")
        for fpath in `ls requirements/**/*.txt`; do \
          python ./requirements/pytorch/adjust-versions.py $fpath ${PYTORCH_VERSION}; \
        done        
      displayName: 'Adjust dependencies'

    - bash: pip install -e .[extra,test,examples] --find-links ${TORCH_URL}
      env:
        PACKAGE_NAME: "pytorch"
        FREEZE_REQUIREMENTS: "1"
      displayName: 'Install package & extras'

    - bash: pip uninstall -y -r requirements/pytorch/strategies.txt
      condition: eq(variables['scope'], '')
      displayName: 'UnInstall strategies'

    - bash: |
        set -e
        CUDA_VERSION_BAGUA=$(python -c "print([ver for ver in [116,113,111,102] if $CUDA_VERSION_MM >= ver][0])")
        pip install "bagua-cuda$CUDA_VERSION_BAGUA"

        PYTORCH_VERSION_COLOSSALAI=$(python -c "import torch; print(torch.__version__.split('+')[0][:4])")
        CUDA_VERSION_MM_COLOSSALAI=$(python -c "import torch ; print(''.join(map(str, torch.version.cuda)))")
        CUDA_VERSION_COLOSSALAI=$(python -c "print([ver for ver in [11.3, 11.1] if $CUDA_VERSION_MM_COLOSSALAI >= ver][0])")
        pip install "colossalai==0.1.10+torch${PYTORCH_VERSION_COLOSSALAI}cu${CUDA_VERSION_COLOSSALAI}" --find-links https://release.colossalai.org

        pip install -r requirements/pytorch/strategies.txt --find-links ${TORCH_URL}

        python requirements/pytorch/check-avail-strategies.py        
      condition: eq(variables['scope'], 'strategies')
      displayName: 'Install strategies'

    - bash: |
        set -e
        pip list
        python requirements/collect_env_details.py
        python -c "import torch ; mgpu = torch.cuda.device_count() ; assert mgpu == 2, f'GPU: {mgpu}'"
        python requirements/pytorch/check-avail-extras.py        
      displayName: 'Env details'

    - bash: bash .actions/pull_legacy_checkpoints.sh
      displayName: 'Get legacy checkpoints'

    - bash: python -m pytest pytorch_lightning
      workingDirectory: src
      displayName: 'Testing: PyTorch doctests'

    - bash: python -m coverage run --source pytorch_lightning -m pytest --ignore benchmarks -v --junitxml=$(Build.StagingDirectory)/test-results.xml --durations=50
      env:
        PL_RUN_CUDA_TESTS: "1"
      workingDirectory: tests/tests_pytorch
      displayName: 'Testing: PyTorch standard'
      timeoutInMinutes: "35"

    - bash: bash run_standalone_tests.sh
      workingDirectory: tests/tests_pytorch
      env:
        PL_USE_MOCKED_MNIST: "1"
        PL_RUN_CUDA_TESTS: "1"
        PL_STANDALONE_TESTS_SOURCE: "pytorch_lightning"
      displayName: 'Testing: PyTorch standalone tests'
      timeoutInMinutes: "35"

    - bash: bash run_standalone_tasks.sh
      workingDirectory: tests/tests_pytorch
      env:
        PL_USE_MOCKED_MNIST: "1"
        PL_RUN_CUDA_TESTS: "1"
      displayName: 'Testing: PyTorch standalone tasks'
      timeoutInMinutes: "10"

    - bash: |
        python -m coverage report
        python -m coverage xml
        python -m coverage html
        python -m codecov --token=$(CODECOV_TOKEN) --commit=$(Build.SourceVersion) --flags=gpu,pytest --name="GPU-coverage" --env=linux,azure
        ls -l        
      workingDirectory: tests/tests_pytorch
      displayName: 'Statistics'

    - task: PublishTestResults@2
      displayName: 'Publish test results'
      inputs:
        testResultsFiles: '$(Build.StagingDirectory)/test-results.xml'
        testRunTitle: '$(Agent.OS) - $(Build.DefinitionName) - Python $(python.version)'
      condition: succeededOrFailed()

    - script: |
        set -e
        bash run_pl_examples.sh --trainer.accelerator=gpu --trainer.devices=1
        bash run_pl_examples.sh --trainer.accelerator=gpu --trainer.devices=2 --trainer.strategy=ddp
        bash run_pl_examples.sh --trainer.accelerator=gpu --trainer.devices=2 --trainer.strategy=ddp --trainer.precision=16        
      workingDirectory: examples
      env:
        PL_USE_MOCKED_MNIST: "1"
      displayName: 'Testing: PyTorch examples'

    - bash: python -m pytest benchmarks -v --maxfail=2 --durations=0
      workingDirectory: tests/tests_pytorch
      env:
        PL_RUN_CUDA_TESTS: "1"
      displayName: 'Testing: PyTorch benchmarks'