Apply alpha scale factor in torch add/sub converters#2680
Open
john-rocky wants to merge 1 commit intoapple:mainfrom
Open
Apply alpha scale factor in torch add/sub converters#2680john-rocky wants to merge 1 commit intoapple:mainfrom
john-rocky wants to merge 1 commit intoapple:mainfrom
Conversation
Collaborator
|
This pull request looks good. Once we get a clean CI run, I will merge it. However we recently updated our CI. @john-rocky - In order to get a CI run, I need you to rebase this pull request on top of latest |
`aten::add(self, other, alpha)` and `aten::sub(self, other, alpha)` compute `self ± alpha * other`, but the converters previously raised on positional `alpha != 1` and silently ignored kwarg `alpha` from the `torch.export` path. As a result, `torch.sub(x, y, alpha=5)` produced `x - y` instead of `x - 5*y` (issue apple#2573). Look up `alpha` from positional inputs (TorchScript) and kwinputs (torch.export), and apply `y = y * alpha` before the add/sub when `alpha != 1`, mirroring the existing `addmm` handler. The alpha=1 fast path is unchanged. Fixes apple#2573.
john-rocky
force-pushed
the
fix-add-sub-alpha
branch
from
35cbab6 to
765aee6
Compare
Contributor
Author
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
aten::add(self, other, alpha)andaten::sub(self, other, alpha)computeself ± alpha * other, but the converters previously raised on positionalalpha != 1and silently ignored the kwarg form fromtorch.export. As a result,torch.sub(x, y, alpha=5)producedx - yinstead ofx - 5*y(issue CoreML ignores alpha in torch.add/torch.sub and gives incorrect output #2573).alphafrom positional inputs (TorchScript) and kwinputs (torch.export), and applyy = y * alphabefore the add/sub whenalpha != 1. Mirrors the pattern already used by theaddmmhandler.alpha == 1fast path is unchanged.Test plan
pytest test_torch_ops.py::TestAddSubAlpha— 24 cases covering int/float alpha across TorchScript / torch.export / ExecuTorch frontends.pytest test_torch_ops.py::TestBoolOps test_torch_ops.py::TestAddmm— no regressions (104 cases).torch.sub(x, y, alpha=5)now returns-5, was3before.Fixes #2573.