Arm backend: Minor composable_quantizer improvements#19330
Open
AdrianLundell wants to merge 1 commit intopytorch:mainfrom
Open
Arm backend: Minor composable_quantizer improvements#19330AdrianLundell wants to merge 1 commit intopytorch:mainfrom
AdrianLundell wants to merge 1 commit intopytorch:mainfrom
Conversation
- Make sure the root node of a shared cluster is the topological first node of the model to avoid crash in torchao. - Do not report get_attr as non-annotated as these are never quantized, leading to unnecessarily long logs. - Make SharedQuantization logging more straight to the point. Do not warn for multiple users as this is resported in the quantizer report and use one single simple log message when nodes are left unquantized. - Make the pre-transform for annotation report more minimal and only print it when relevant + update the notebook example to this fact. Signed-off-by: Adrian Lundell <adrian.lundell@arm.com> Change-Id: I412918235799c99ab4b1b1e1f6412de4c906766f
AdrianLundell
added
partner: arm
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19330
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (10 Unrelated Failures)As of commit e712c31 with merge base 1debeb6 ( 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
Contributor
There was a problem hiding this comment.
Pull request overview
This PR refines Arm backend composable quantizer reporting and SharedQspecQuantizer behavior to reduce noisy logs and avoid a torchao crash by stabilizing shared-cluster root selection.
Changes:
- Update pre-transform reporting to be more minimal and only emitted when relevant; update the Arm quantizer tutorial accordingly.
- Exclude
get_attrnodes from “non-annotated” reporting to avoid misleading/verbose reports. - Ensure SharedQspecQuantizer uses the topologically-first node as the shared-cluster root and simplify its accept/reject reporting.
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.
| File | Description |
|---|---|
| examples/arm/quantizer_tutorial.ipynb | Updates documentation to explain the two reports (pre-transform vs final) and aligns example output formatting. |
| backends/cortex_m/quantizer_reporter.py | Filters get_attr out of unannotated-node reporting to reduce unnecessary noise. |
| backends/arm/quantizer/arm_quantizer.py | Replaces the pre-transform quantizer reporter output with a minimal debug log of nodes not marked for decomposition. |
| backends/arm/quantizer/arm_quantizer_utils.py | Stabilizes shared-cluster root selection and simplifies SharedQuantization reject handling/logging. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+1119
to
+1127
| msg = """ | ||
| ---------------------------------------------------------------------------------------------------- | ||
| PRE-TRANSFORM FOR ANNOTATION QUANTIZATION REPORT | ||
| ---------------------------------------------------------------------------------------------------- | ||
| The following nodes are not marked for quantization and will not be decomposed in the transform for annotation pipeline:\n""" | ||
| for node in non_quantized_nodes: | ||
| msg += f" {node.name}\n" | ||
|
|
||
| logger.debug(msg) |
Comment on lines
+625
to
+628
| self.report_reject( | ||
| ordered_nodes, | ||
| "All inputs and outputs to these nodes are non-quantized.", | ||
| ) |
Comment on lines
+595
to
+596
| # Ensure the root node is the first one in the graph. | ||
| root_node = ordered_nodes[0] |
| "decomposition must not happen if it should be kept in float.\n", | ||
| "\n", | ||
| "**This is important to be aware of when doing mixed quantization since this means that for an operator to be fully quantized,\n", | ||
| "both the original operator and the decomposition needs to be targeted.**\n", |
zingo
approved these changes
May 6, 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.
Make sure the root node of a shared cluster is the topological first node of the model to avoid crash in torchao.
Do not report get_attr as non-annotated as these are never quantized, leading to unnecessarily long logs.
Make SharedQuantization logging more straight to the point. Do not warn for multiple users as this is resported in the quantizer report and use one single simple log message when nodes are left unquantized.
Make the pre-transform for annotation report more minimal and only print it when relevant + update the notebook example to this fact.
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell