Avoid modifying the default dtype on exception #18500
Conversation
carmocca
requested review from
awaelchli,
justusschock and
williamFalcon
as code owners
⚡ Required checks status: All passing 🟢Groups summary🟢 pytorch_lightning: Tests workflowThese checks are required after the changes to 🟢 pytorch_lightning: Azure GPU
These checks are required after the changes to 🟢 pytorch_lightning: Benchmarks
These checks are required after the changes to 🟢 fabric: Docs
These checks are required after the changes to 🟢 pytorch_lightning: Docs
These checks are required after the changes to 🟢 lightning_fabric: CPU workflowThese checks are required after the changes to 🟢 lightning_fabric: Azure GPU
These checks are required after the changes to 🟢 mypy
These checks are required after the changes to 🟢 installThese checks are required after the changes to Thank you for your contribution! 💜
|
|
@awaelchli Do you want a test for each class for both trainer and fabric? |
|
That's a possibility yes. Or a parameterized test for the class in a test_all.py file in the plugins folder if the assertion is the same for all. |
|
Done |
What does this PR do?
Since
torch.set_default_dtypemodifies the global state, we need to make sure that the default dtype is restored if an exception happens while we are modifying it.An example where this is a problem is tests, if one test fails, the default dtype will be different for others, possibly creating a chain of failures
cc @Borda @carmocca @justusschock @awaelchli