Fix tests when PyTorch built with MPS support #1188
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Starting with Lightning 2.0,
Trainer
now defaults toaccelerator="auto"
, meaning it searches to see if any GPU/MPS device is available. While this works fine with CUDA, the MPS backend is still missing implementations of many operators, so the tests will fail. We should force the tests to use the CPU to ensure that they always pass, even on macOS. We don't see this issue in CI because the GitHub runner is an old x86_64 Mac and doesn't support MPS.