Materialize zero bias for conv replacement (#23354) - #23354
Merged
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/23354
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 1 PendingAs of commit 590bcaf with merge base 305638c ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
Contributor
|
@pcwu0329 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D110672307. |
This PR needs a
|
pcwu0329
added a commit
to pcwu0329/executorch
that referenced
this pull request
Oct 2, 2026
Summary: Pull Request resolved: pytorch#23354 When replacing biasless `aten.convolution` with Cadence convolution, materialize an explicit zero bias tensor so strict Edge validation sees a schema-compatible Cadence conv. This keeps strict validation enabled for QAT compile paths that leave some backbone convolutions FP32. Differential Revision: D110672307
meta-codesync
Bot
changed the title
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 07:52
2d35485 to
ae663e1
Compare
pcwu0329
requested review from
DrJessop,
JacobSzwejbka and
zonglinpeng
and removed request for
DrJessop
October 2, 2026 07:53
pcwu0329
added a commit
to pcwu0329/executorch
that referenced
this pull request
Oct 2, 2026
Summary: When replacing biasless `aten.convolution` with Cadence convolution, materialize an explicit zero bias tensor so strict Edge validation sees a schema-compatible Cadence conv. This keeps strict validation enabled for QAT compile paths that leave some backbone convolutions FP32. Differential Revision: D110672307
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 07:59
ae663e1 to
ab5bd4b
Compare
pcwu0329
added a commit
to pcwu0329/executorch
that referenced
this pull request
Oct 2, 2026
Summary: Pull Request resolved: pytorch#23354 When replacing biasless `aten.convolution` with Cadence convolution, materialize an explicit zero bias tensor so strict Edge validation sees a schema-compatible Cadence conv. This keeps strict validation enabled for QAT compile paths that leave some backbone convolutions FP32. Differential Revision: D110672307
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 08:03
ab5bd4b to
254721d
Compare
pcwu0329
added a commit
to pcwu0329/executorch
that referenced
this pull request
Oct 2, 2026
Summary: When replacing biasless `aten.convolution` with Cadence convolution, materialize an explicit zero bias tensor so strict Edge validation sees a schema-compatible Cadence conv. This keeps strict validation enabled for QAT compile paths that leave some backbone convolutions FP32. Differential Revision: D110672307
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 08:18
254721d to
9d29ad9
Compare
pcwu0329
added a commit
to pcwu0329/executorch
that referenced
this pull request
Oct 2, 2026
Summary: Pull Request resolved: pytorch#23354 When replacing biasless `aten.convolution` with Cadence convolution, materialize an explicit zero bias tensor so strict Edge validation sees a schema-compatible Cadence conv. This keeps strict validation enabled for QAT compile paths that leave some backbone convolutions FP32. Differential Revision: D110672307
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 08:21
9d29ad9 to
b4a2fe6
Compare
DrJessop
approved these changes
Oct 2, 2026
pcwu0329
added a commit
to pcwu0329/executorch
that referenced
this pull request
Oct 2, 2026
Summary: `aten::convolution` declares `Tensor? bias`, so a conv built with `bias=False` lowers to a node whose bias argument is `None`. `cadence::conv1d/conv2d/conv3d` declare a non-optional `Tensor bias`, with no way to express absence. Replacing a biasless convolution therefore yields a node that strict Edge validation rejects for schema mismatch. Materialize an explicit zero bias via `edge.aten.full.default`, sized to the convolution's output channels and typed from the weight, so the replacement is schema-valid. Adding zeros is numerically a no-op. If the Cadence kernels gain optional-bias support, as discussed on the diff, this method can be deleted outright. Reviewed By: DrJessop Differential Revision: D110672307
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 21:12
b4a2fe6 to
1b65a46
Compare
Summary: Pull Request resolved: pytorch#23354 `aten::convolution` declares `Tensor? bias`, so a conv built with `bias=False` lowers to a node whose bias argument is `None`. `cadence::conv1d/conv2d/conv3d` declare a non-optional `Tensor bias`, with no way to express absence. Replacing a biasless convolution therefore yields a node that strict Edge validation rejects for schema mismatch. Materialize an explicit zero bias via `edge.aten.full.default`, sized to the convolution's output channels and typed from the weight, so the replacement is schema-valid. Adding zeros is numerically a no-op. If the Cadence kernels gain optional-bias support, as discussed on the diff, this method can be deleted outright. Reviewed By: DrJessop Differential Revision: D110672307
pcwu0329
force-pushed
the
export-D110672307
branch
from
October 2, 2026 21:15
1b65a46 to
590bcaf
Compare
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::convolutiondeclaresTensor? bias, so a conv built withbias=Falselowers to a node whose bias argument isNone.cadence::conv1d/conv2d/conv3ddeclare a non-optionalTensor bias, with no way to express absence. Replacing a biasless convolution therefore yields a node that strict Edge validation rejects for schema mismatch.Materialize an explicit zero bias via
edge.aten.full.default, sized to the convolution's output channels and typed from the weight, so the replacement is schema-valid. Adding zeros is numerically a no-op.If the Cadence kernels gain optional-bias support, as discussed on the diff, this method can be deleted outright.
Reviewed By: DrJessop
Differential Revision: D110672307