[ONNX] Fix Expand operator to properly handle target shapes #17779
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
This fixes issue apache#17746 where the ONNX Expand operator was not correctly expanding tensors to higher dimensions. The issue manifested when a downstream ArgMin operation received a tensor with fewer dimensions than expected, causing an 'axis out of bounds' error. Specifically: 1. The Expand op was incorrectly skipping the broadcast when input and target shapes had the same values but different ranks 2. This caused a tensor with shape [5,60] to remain [5,60] when it should have been expanded to [1,1,5,60] 3. The subsequent ArgMin op with axis=2 then failed as the tensor only had 2 dimensions instead of the expected 4 The fix ensures that Expand always broadcasts to the target shape, preserving the rank specified in the ONNX model. This allows downstream operations to work with the correct tensor dimensions. Fixes apache#17746
singh20anurag
force-pushed
the
fix_onnx_expand_op
branch
from
d4e0e43 to
80f1ce9
Compare
Member
|
it would be good to have a test case |
Contributor
|
hi, I add the test case in pr 17900 |
Member
|
Thanks, @xinxilwl for the fix! Please feel free to chime in and re-open this issue if it persists |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.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.
This fixes issue #17746 where the ONNX Expand operator was not correctly expanding tensors to higher dimensions. The issue manifested when a downstream ArgMin operation received a tensor with fewer dimensions than expected, causing an 'axis out of bounds' error.
Specifically:
The fix ensures that Expand always broadcasts to the target shape, preserving the rank specified in the ONNX model. This allows downstream operations to work with the correct tensor dimensions.
Fixes #17746