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

136 lines
4.7 KiB
YAML

# 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/*"
- "lite/debug"
paths:
include:
- ".actions/**"
- ".azure/gpu-tests-fabric.yml"
- "examples/fabric/**"
- "examples/run_fabric_examples.sh"
- "tests/tests_fabric/run_standalone_*.sh"
- "tests/tests_pytorch/run_standalone_tests.sh" # used by Lite through a symlink
- "requirements/fabric/**"
- "src/lightning_fabric/**"
- "tests/tests_fabric/**"
- "setup.cfg" # includes pytest config
exclude:
- "requirements/*/docs.txt"
- "*.md"
- "**/*.md"
jobs:
- job: testing
# how long to run the job before automatically cancelling
timeoutInMinutes: "20"
# 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: "pytorchlightning/pytorch_lightning:base-cuda-py3.9-torch1.13-cuda11.6.1"
# 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"
# TODO: package parametrization
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=TORCH_URL]https://download.pytorch.org/whl/cu${cuda_ver}/torch_stable.html"
displayName: 'set env. vars'
- bash: |
echo $CUDA_VISIBLE_DEVICES
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 .[dev,strategies,examples] --find-links ${TORCH_URL}
env:
PACKAGE_NAME: "fabric"
FREEZE_REQUIREMENTS: "1"
displayName: 'Install package & dependencies'
- bash: |
set -e
pip list
python requirements/collect_env_details.py
python -c "import torch ; mgpu = torch.cuda.device_count() ; assert mgpu == 2, f'GPU: {mgpu}'"
displayName: 'Env details'
- bash: python -m coverage run --source lightning_fabric -m pytest --ignore benchmarks -v --junitxml=$(Build.StagingDirectory)/test-results.xml --durations=50
env:
PL_RUN_CUDA_TESTS: "1"
workingDirectory: tests/tests_fabric
displayName: 'Testing: Lite standard'
timeoutInMinutes: "10"
- bash: bash run_standalone_tests.sh
workingDirectory: tests/tests_fabric
env:
PL_RUN_CUDA_TESTS: "1"
PL_STANDALONE_TESTS_SOURCE: "lightning_fabric"
displayName: 'Testing: Lite standalone tests'
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_fabric
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: |
# In order to run the examples, we need to substitute the meta package imports with the standalone package
python ../.actions/assistant.py copy_replace_imports --source_dir="./fabric" --source_import="lightning.fabric" --target_import="lightning_fabric.fabric"
set -e
bash run_fabric_examples.sh --accelerator=cuda --devices=1
bash run_fabric_examples.sh --accelerator=cuda --devices=2 --strategy ddp
workingDirectory: examples
displayName: 'Testing: Lite examples'