Skip to main content

pytest plugin for a better developer experience when working with the PyTorch test suite

Project description


license repo status isort black tests status

What is it?

pytest-pytorch is a lightweight pytest-plugin that enhances the developer experience when working with the PyTorch test suite if you come from a pytest background.

Why do I need it?

Some testcases in the PyTorch test suite are automatically generated when a module is loaded in order to parametrize them. Trying to collect them with their names as written, e.g. pytest or pytest, is unfortunately not possible. If you are used to this syntax or your IDE relies on it (PyCharm, VSCode), you can install pytest-pytorch to make it work.

How do I install it?

You can install pytest-pytorch with pip

$ pip install pytest-pytorch

or with conda:

$ conda install -c conda-forge pytest-pytorch

How do I use it?

With pytest-pytorch installed you can select test cases and tests as if the instantiation for different devices was performed by @pytest.mark.parametrize:

Use case Command
Run a test case against all devices pytest
Run a test case against one device pytest -k "$DEVICE"
Run a test against all devices pytest
Run a test against one device pytest -k "$DEVICE"

Can I have a little more background?

PyTorch uses its own method for generating tests that is for the most part compatible with unittest and pytest. Its custom test generation allows test templates to be written and instantiated for different device types, data types, and operators. Consider the following module

from torch.testing._internal.common_utils import TestCase
from torch.testing._internal.common_device_type import instantiate_device_type_tests

class TestFoo(TestCase):
    def test_bar(self, device):
    def test_baz(self, device):

instantiate_device_type_tests(TestFoo, globals())

Assuming we "cpu" and "cuda" are available as devices, we can collect four tests:

  3., and

From a pytest perspective this is similar to decorating TestFoo with @pytest.mark.parametrize("device", ("cpu", "cuda"))) which would result in

  3.[cpu], and

Since the PyTorch test framework renames testcases and tests, naively running pytest or pytest fails, because it can't find anything matching these names. Of course you can get around it by using the regular expression matching (-k command line flag) that pytest offers.

pytest-pytorch performs this matching so you can keep your familiar workflow and your IDE is happy out of the box.

How do I contribute?

First and foremost: Thank you for your interest in development of pytest-pytorch's! We appreciate all contributions be it code or something else. Check out our contribution guide lines for details.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pytest_pytorch-0.2.1.tar.gz (7.5 kB view hashes)

Uploaded source

Built Distribution

pytest_pytorch-0.2.1-py3-none-any.whl (6.1 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page