Fix: Prevent UnionTransformer type ambiguity in combination with PyTorchTypeTransformer #2726
Conversation
This file contains 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
…rchTypeTransformer Signed-off-by: Fabio Grätz <[email protected]>
fg91
requested review from
wild-endeavor,
kumare3,
eapolinario,
pingsutw,
cosmicBboy,
samhita-alla,
thomasjpfan and
Future-Outlier
as code owners
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #2726 +/- ##
===========================================
+ Coverage 49.76% 71.89% +22.13%
===========================================
Files 193 193
Lines 19659 19683 +24
Branches 4097 4101 +4
===========================================
+ Hits 9783 14152 +4369
+ Misses 9345 4841 -4504
- Partials 531 690 +159 ☔ View full report in Codecov by Sentry. |
thomasjpfan
reviewed
Sep 9, 2024
There was a problem hiding this comment.
Can we include one more unit test based on your example in #2726 (comment) to make sure this continues to work?
Signed-off-by: Fabio Grätz <[email protected]>
Absolutely 99df4fe |
|
@thomasjpfan are you ok with merging this PR? |
thomasjpfan
approved these changes
Sep 14, 2024
thomasjpfan
deleted the
fg91/fix/torch-type-transformer-ambiguous-union
branch
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Outdated suggestions cannot be applied.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.
Why are the changes needed?
This example fails with:
The type transformer choice should, however, not be ambiguous.
The problem arises as follows:
Here in the
UnionTransformerwe test all union variants and raise the observed error if more than one of the union variants work.In this case, the two union working variants are:
ListTransformerfor a list of tensors and then thePyTorchTensorTransformerfor each element in the list. This is the expected behaviour and does work.PyTorchTensorTransformerfor the list of tensors. This, however, works as well astorch.save([some_tensor], ...)works. This is not intended and hence causes the ambiguity.What changes were proposed in this pull request?
We need to prevent the pytorch type transformers from serializing lists of tensors or modules since the type engine (i.e. the
UnionTransformerexpects these cases to be handled by theListTransformer).How was this patch tested?
Tested with the snippet above and added a unit test that would have caught this.