[Relay, TOPI] Complete rewrite of where op to support broadcasting #6759
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
|
Nice work @masahi! |
kevinthesun
reviewed
Oct 26, 2020
masahi
force-pushed
the
where-broadcast
branch
from
0fe545e
to
5423c70
Compare
t-vi
reviewed
Oct 26, 2020
|
@t-vi In d8c5076, I ported c++ Now, invalid shapes are catched at runtime and result in an error. I added one test that intentionally tries to run a compiled model with invalid combination of shapes, and verifies that invaild broadcast are properly caught. Please have a look! |
t-vi
reviewed
Oct 27, 2020
|
Looks awesome to me. Thank you @masahi! |
|
@kevinthesun I've updated |
|
Thanks @masahi @junrushao1994 @t-vi |
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Oct 29, 2020
…pache#6759) * where type rel with broadcast * add tests for where with broadcast * clean up tests * uncomment other tests * add more tests * update doc * CHECK -> ICHECK * add where any test * fix format * remove useless detections for one * set manual seed * ported shape broadcast helper func to hybridscript * remove shape function helper from cpp Co-authored-by: masa <[email protected]>
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Dec 2, 2020
…pache#6759) * where type rel with broadcast * add tests for where with broadcast * clean up tests * uncomment other tests * add more tests * update doc * CHECK -> ICHECK * add where any test * fix format * remove useless detections for one * set manual seed * ported shape broadcast helper func to hybridscript * remove shape function helper from cpp Co-authored-by: masa <[email protected]>
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Dec 4, 2020
…pache#6759) * where type rel with broadcast * add tests for where with broadcast * clean up tests * uncomment other tests * add more tests * update doc * CHECK -> ICHECK * add where any test * fix format * remove useless detections for one * set manual seed * ported shape broadcast helper func to hybridscript * remove shape function helper from cpp Co-authored-by: masa <[email protected]>
trevor-m
pushed a commit
to neo-ai/tvm
that referenced
this pull request
Dec 4, 2020
…pache#6759) * where type rel with broadcast * add tests for where with broadcast * clean up tests * uncomment other tests * add more tests * update doc * CHECK -> ICHECK * add where any test * fix format * remove useless detections for one * set manual seed * ported shape broadcast helper func to hybridscript * remove shape function helper from cpp Co-authored-by: masa <[email protected]>
trevor-m
pushed a commit
to neo-ai/tvm
that referenced
this pull request
Mar 31, 2021
…pache#6759) * where type rel with broadcast * add tests for where with broadcast * clean up tests * uncomment other tests * add more tests * update doc * CHECK -> ICHECK * add where any test * fix format * remove useless detections for one * set manual seed * ported shape broadcast helper func to hybridscript * remove shape function helper from cpp Co-authored-by: masa <[email protected]>
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.
This is a follow up to the discussion in #6383.
Recently I've hit a model (
gpt2from transformers) that has awhereop with following shapes:It is not possible to support such combinations of shapes by yet another ad hoc band-aid, so I decided to roll a complete support for numpy style broadcasting. This turned out to be easy thanks to existing broadcasting utilities in relay/topi. I think this is desirable feature to align with other frameworks that follow numpy style where op (pytorch, onnx etc).
I added tests that replicate examples in numpy where op doc https://numpy.org/doc/stable/reference/generated/numpy.where.html. I think my implementation is correct, but since this is my first attempt to implement an op with broadcasting semantics, reviews by people who are familiar with broadcast issues are highly appreciated.
With this PR, I can load
gpt2model from transformers using PyTorch frontend, compile and get correct outputs compared to PyTorch.please review @zhiics @junrushao1994 @t-vi @kevinthesun @jwfromm @tqchen