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

166 lines
6.3 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/*"
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 fabric through a symlink
- "requirements/fabric/**"
- "src/lightning/__about__.py"
- "src/lightning/__init__.py"
- "src/lightning/__main__.py"
- "src/lightning/__setup__.py"
- "src/lightning/__version__.py"
- "src/lightning/fabric/**"
- "src/lightning_fabric/*"
- "tests/tests_fabric/**"
- "pyproject.toml" # 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])' )
FREEZE_REQUIREMENTS: "1"
PIP_CACHE_DIR: "/var/tmp/pip"
container:
image: $(image)
# default shm size is 64m. Increase it to avoid:
# 'Error while creating shared memory: unhandled system error, NCCL version 2.7.8'
options: "--gpus=all --shm-size=2gb -v /var/tmp:/var/tmp"
strategy:
matrix:
"Fabric | latest":
image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.0"
PACKAGE_NAME: "fabric"
"Lightning | latest":
image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.0"
PACKAGE_NAME: "lightning"
workspace:
clean: all
steps:
- bash: |
echo "##vso[task.setvariable variable=CUDA_VISIBLE_DEVICES]$(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${cuda_ver}/torch_stable.html"
scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(fabric="lightning_fabric").get(n, n))')
echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
displayName: "set env. vars"
- bash: |
echo $(DEVICES)
echo $CUDA_VISIBLE_DEVICES
echo $CUDA_VERSION_MM
echo $TORCH_URL
echo $COVERAGE_SOURCE
whereis nvidia
nvidia-smi
which python && which pip
python --version
pip --version
pip list
displayName: "Image info & NVIDIA"
- bash: |
PYTORCH_VERSION=$(python -c "import torch; print(torch.__version__.split('+')[0])")
pip install -q wget packaging
python -m wget https://raw.githubusercontent.com/Lightning-AI/utilities/main/scripts/adjust-torch-versions.py
for fpath in `ls requirements/**/*.txt`; do \
python ./adjust-torch-versions.py $fpath ${PYTORCH_VERSION}; \
done
displayName: "Adjust dependencies"
- bash: |
extra=$(python -c "print({'lightning': 'fabric-'}.get('$(PACKAGE_NAME)', ''))")
pip install -e ".[${extra}dev]" pytest-timeout -U --find-links ${TORCH_URL}
displayName: "Install package & dependencies"
- bash: |
set -e
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 pytest lightning_fabric
workingDirectory: src
# without succeeded this could run even if the job has already failed
condition: and(succeeded(), eq(variables['PACKAGE_NAME'], 'fabric'))
displayName: "Testing: Fabric doctests"
- bash: |
pip install -q -r .actions/requirements.txt
python .actions/assistant.py copy_replace_imports --source_dir="./tests/tests_fabric" \
--source_import="lightning.fabric" \
--target_import="lightning_fabric"
python .actions/assistant.py copy_replace_imports --source_dir="./examples/fabric" \
--source_import="lightning.fabric" \
--target_import="lightning_fabric"
# without succeeded this could run even if the job has already failed
condition: and(succeeded(), eq(variables['PACKAGE_NAME'], 'fabric'))
displayName: "Adjust tests & examples"
- bash: python -m coverage run --source ${COVERAGE_SOURCE} -m pytest -v --durations=50
workingDirectory: tests/tests_fabric
env:
PL_RUN_CUDA_TESTS: "1"
displayName: "Testing: fabric standard"
timeoutInMinutes: "10"
- bash: bash run_standalone_tests.sh
workingDirectory: tests/tests_fabric
env:
PL_RUN_CUDA_TESTS: "1"
PL_STANDALONE_TESTS_SOURCE: $(COVERAGE_SOURCE)
displayName: "Testing: fabric standalone tests"
timeoutInMinutes: "10"
- bash: |
python -m coverage report
python -m coverage xml
python -m coverage html
# https://docs.codecov.com/docs/codecov-uploader
curl -Os https://uploader.codecov.io/latest/linux/codecov
chmod +x codecov
./codecov --token=$(CODECOV_TOKEN) --commit=$(Build.SourceVersion) \
--flags=gpu,pytest,${COVERAGE_SOURCE} --name="GPU-coverage" --env=linux,azure
ls -l
workingDirectory: tests/tests_fabric
displayName: "Statistics"
- script: |
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: fabric examples"