only use plain tensors in kernel tests #7230
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
pmeier
commented
Feb 13, 2023
| @@ -237,13 +237,6 @@ class TensorLoader: | |||
| def load(self, device): | |||
| return self.fn(self.shape, self.dtype, device) | |||
|
|
|||
| def unwrap(self): | |||
There was a problem hiding this comment.
This just undoes the change in #7228 since we no longer need this.
Comment on lines
-671
to
-672
| yield ArgsKwargs(bounding_box_loader, new_format=new_format) | ||
| yield ArgsKwargs(bounding_box_loader.unwrap(), old_format=bounding_box_loader.format, new_format=new_format) |
There was a problem hiding this comment.
We no longer need to pass two things here, since the datapoint will be unwrapped by the test automatically.
| @@ -155,14 +156,12 @@ def _unbatch(self, batch, *, data_dims): | |||
| if batched_tensor.ndim == data_dims: | |||
| return batch | |||
|
|
|||
| unbatcheds = [] | |||
There was a problem hiding this comment.
Again, just undoing what #7227 did, since we no longer care for datapoints at this stage.
NicolasHug
approved these changes
Feb 13, 2023
| @@ -668,8 +669,7 @@ def sample_inputs_affine_video(): | |||
| def sample_inputs_convert_format_bounding_box(): | |||
| formats = list(datapoints.BoundingBoxFormat) | |||
| for bounding_box_loader, new_format in itertools.product(make_bounding_box_loaders(formats=formats), formats): | |||
| yield ArgsKwargs(bounding_box_loader, new_format=new_format) | |||
| yield ArgsKwargs(bounding_box_loader.unwrap(), old_format=bounding_box_loader.format, new_format=new_format) | |||
| yield ArgsKwargs(bounding_box_loader, old_format=bounding_box_loader.format, new_format=new_format) | |||
There was a problem hiding this comment.
Don't we still want to test both pure-tensors and BoundingBoxes, at least for those "hybrid" kernels ? (perhaps that's still covered?)
There was a problem hiding this comment.
It wasn't until 99dd341, but now it is 😇 So the kernel tests actually only test plain tensors, but the dispatcher tests take over the other part.
facebook-github-bot
pushed a commit
that referenced
this pull request
Mar 28, 2023
Reviewed By: vmoens Differential Revision: D44416278 fbshipit-source-id: 7ac022e18aec970d7b4c5091bd9840ea2d7e0ed6
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
With #7227 and #7228 we have a new paradigm of kernel dispatcher hybrids. So far we just simply passed the datapoints in the kernel tests directly, since it made no difference. With these recent changes, we can no longer do that without some limitations of the test suite.
Since some of our kernel test depend on the datapoint type as metadata, e.g.
vision/test/test_prototype_transforms_functional.py
Lines 172 to 188 in acabaf8
I've opted to unwrap inside the tests.
The largest part of the diff in this PR comes from our correctness tests for bounding boxes that we haven't ported to the proper framework yet. They happily mixed plain tensors and datapoints.
cc @vfdev-5 @bjuncek