Arm backend: Make ConstantFoldingPass preserve Learnable Parameters#20372
Open
gggekov wants to merge 1 commit into
Open
Arm backend: Make ConstantFoldingPass preserve Learnable Parameters#20372gggekov wants to merge 1 commit into
gggekov wants to merge 1 commit into
Conversation
Avoid folding expressions that depend on trainable parameters during QAT preparation. Folding these values can turn learnable tensors into fixed constants, leaving the prepared model without the parameters needed for QAT training. Raised by pytorch#20148 Signed-off-by: George Gekov <george.gekov@arm.com> Change-Id: Idc9c089c87e7291567e73a0b48298c8bcea04081
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/20372
Note: Links to docs will display an error until the docs builds have been completed. ❌ 7 New Failures, 2 Unrelated FailuresAs of commit ce52f9f with merge base a6305cb ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
gggekov
added
the
partner: arm
This PR needs a
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates the Arm backend’s constant-folding pass to avoid folding subgraphs that depend on trainable parameters, preventing QAT preparation from accidentally turning learnable tensors into fixed constants (addressing the missing-parameters failure reported in #20148).
Changes:
- Teach
ConstantFoldingPassto preserveget_attrparameters withrequires_grad=True(and their dependent nodes) duringconstant_fold. - Add/extend Arm backend QAT training-loop tests, including a regression case for preserving a learnable parameter through QAT prepare.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 7 comments.
| File | Description |
|---|---|
backends/arm/_passes/constant_folding_pass.py |
Adds preservation logic to prevent constant folding of trainable-parameter-dependent subgraphs during QAT prep. |
backends/arm/test/misc/test_qat_training_loop.py |
Adds QAT training loop regression tests (learnable-param preservation + additional QAT smoke tests). |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+64
to
+68
| if node.op == "get_attr": | ||
| try: | ||
| attr = getattr(graph_module, node.target, None) | ||
| except AttributeError: | ||
| continue |
Comment on lines
+56
to
+60
| # Returns a set of the graph nodes with learnable | ||
| # parameters and any nodes depending on the learnable | ||
| # parameter. Needed for the case when doing | ||
| # QAT and we should not do constatnt folding | ||
| # for the learnable parameters and their dependends. |
Comment on lines
+134
to
+136
| def train_step( | ||
| model: torch.nn.Module, optimizer: torch.optim.Optimizer, input_output, loss_fn | ||
| ) -> None: |
Comment on lines
+167
to
+169
| optimizer = torch.optim.SGD(prepared_model.parameters(), lr=1.0, momentum=0.3) | ||
| for _ in range(3): | ||
| train_step(model, optimizer, _make_batch, torch.nn.functional.mse_loss) |
Comment on lines
+228
to
+230
| loss = train_step( | ||
| model, optimizer, _make_image_batch, torch.nn.functional.cross_entropy | ||
| ) |
Comment on lines
+271
to
+273
| quantizer = TOSAQuantizer(TosaSpecification.create_from_string("TOSA-1.0+INT")) | ||
| prepared_model = prepare_qat_pt2e(exported_model.module(), quantizer) | ||
| prepared_model = move_exported_model_to_train(prepared_model) |
| prepared_model, | ||
| optimizer, | ||
| _make_mha_batch, | ||
| torch.nn.functional.cross_entropy, |
zingo
approved these changes
Jun 18, 2026
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
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.
Avoid folding expressions that depend on trainable parameters during QAT preparation. Folding these values can turn learnable tensors into fixed constants, leaving the prepared model without the parameters needed for QAT training.
Raised by #20148
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani