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.
Some testcases in the PyTorch test suite are only used as templates and will be instantiated at runtime. Unfortunately, PyTorch's naming scheme for parametrizations differs from pytest's. As a consequence, these tests cannot be selected by their names as written and one has to remember PyTorch's naming scheme. This can be especially disrupting to your workflow if your IDE (PyCharm, VSCode) relies on pytest's default selection syntax.
If this has ever been a source of frustration for you, worry no longer. pytest-pytorch was made for you.
You can install pytest-pytorch with pip
$ pip install pytest-pytorchor with conda:
$ conda install -c conda-forge pytest-pytorchWith pytest-pytorch installed you can select test cases and tests by their names as written:
| Use case | Command |
|---|---|
| Run a test case against all devices | pytest test_foo.py::TestBar |
| Run a test against all devices | pytest test_foo.py::TestBar::test_baz |
Similar to a parametrization by @pytest.mark.parametrize you can use the -k flag to select a specific set of parameters:
| Use case | Command |
|---|---|
| Run a test case against one device | pytest test_foo.py::TestBar -k "$DEVICE" |
| Run a test against one device | pytest test_foo.py::TestBar::test_baz -k "$DEVICE" |
Sure, we have written a blog post about pytest-pytorch that goes into details.
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.