[Relay][Frontend][Onnx] Allow A to B broadcasting of batch_matmul and reverse strided slice#6681
Merged
masahi merged 6 commits intoapache:mainfrom Oct 15, 2020
Merged
[Relay][Frontend][Onnx] Allow A to B broadcasting of batch_matmul and reverse strided slice#6681masahi merged 6 commits intoapache:mainfrom
masahi merged 6 commits intoapache:mainfrom
Conversation
Contributor
Author
|
@mbrookhart @masahi can you guys take a look at this PR? |
Member
|
There is an error from auto schedular test. Not sure if it is related or flaky |
Contributor
Author
|
It seems very unlikely that these changes would cause that error. I'll try to rebuild. It looks like it passes locally as well. |
Member
|
hey @comaniac @merrymercy can you check if the error from evolutionary search test https://ci.tlcpack.ai/blue/organizations/jenkins/tvm/detail/PR-6681/1/pipeline/ can be flaky? |
Member
|
@masahi It seems to be an internal bug of the auto-scheduler rather than a flaky test. |
mbrookhart
reviewed
Oct 14, 2020
mbrookhart
approved these changes
Oct 14, 2020
masahi
approved these changes
Oct 15, 2020
Member
|
Thanks @jwfromm @mbrookhart |
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Oct 29, 2020
… reverse strided slice (apache#6681) * slice and batch_matmul fixes. * Bug fix in shape inference. * Test backwards strided slice. * Fix batch_matmul dynamic shape function. * formatting. * Fix edge case for implicit broadcast
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Dec 2, 2020
… reverse strided slice (apache#6681) * slice and batch_matmul fixes. * Bug fix in shape inference. * Test backwards strided slice. * Fix batch_matmul dynamic shape function. * formatting. * Fix edge case for implicit broadcast
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Dec 4, 2020
… reverse strided slice (apache#6681) * slice and batch_matmul fixes. * Bug fix in shape inference. * Test backwards strided slice. * Fix batch_matmul dynamic shape function. * formatting. * Fix edge case for implicit broadcast
trevor-m
pushed a commit
to neo-ai/tvm
that referenced
this pull request
Dec 4, 2020
… reverse strided slice (apache#6681) * slice and batch_matmul fixes. * Bug fix in shape inference. * Test backwards strided slice. * Fix batch_matmul dynamic shape function. * formatting. * Fix edge case for implicit broadcast
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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 PR contains two otherwise unrelated bug fixes / improvements that allow us to import a class of segmentation models using the Onnx frontend.
The first is the ability to broadcast the batch dimension of the left hand side argument of
batch_matmulto to the right hand side argument, where we previously only allowed rhs to lhs broadcasting. There are a few bug fixes in relay to enable this.The second is a fix to our strided slice shape inference to allow reverse slicing, specifically when the end is a negative number, indicating that we should iterate all the way to the first element. A test case in relay is added to catch this in the future.
Finally, I removed some unneeded type inference and shape checking in padding.