Relax WeightBiasQuantization constraint for larger QDQ node group #25673
Conversation
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
Contributor
|
/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline,Windows x64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 5 pipeline(s). |
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR relaxes constraints in the WeightBiasQuantization transformer to enable quantization of weights in larger QDQ node groups that include type-preserving operations between the target node and QuantizeLinear nodes.
- Introduces a new function to check for valid paths to QuantizeLinear nodes that preserve type/shape without branching
- Replaces the strict single-consumer QuantizeLinear requirement with a more flexible path validation
- Fixes multiple spelling errors in function names across QNN builder components
Reviewed Changes
Copilot reviewed 7 out of 7 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| onnxruntime/core/optimizer/qdq_transformer/weight_bias_quantization.cc | Adds new path validation logic and replaces strict QuantizeLinear consumer check |
| onnxruntime/test/optimizer/qdq_transformer_test.cc | Adds test case for Conv with ReLU pattern and updates test comments |
| onnxruntime/core/providers/qnn/builder/opbuilder/base_op_builder.h | Fixes function name spelling from "ReArranagePads" to "ReArrangePads" |
| onnxruntime/core/providers/qnn/builder/opbuilder/conv_op_builder.cc | Updates function call and removes extra whitespace |
| onnxruntime/core/providers/qnn/builder/opbuilder/pool_op_builder.cc | Updates function call to use corrected spelling |
| onnxruntime/core/providers/qnn/builder/opbuilder/pad_op_builder.cc | Updates function call to use corrected spelling |
| onnxruntime/core/providers/qnn/builder/qnn_model_wrapper.cc | Fixes spelling error in comment |
onnxruntime/core/optimizer/qdq_transformer/weight_bias_quantization.cc
Outdated
Show resolved
Hide resolved
onnxruntime/core/optimizer/qdq_transformer/weight_bias_quantization.cc
Outdated
Show resolved
Hide resolved
qti-yuduo
force-pushed
the
dev/yuduo/fold-q
branch
from
f363878 to
34ed336
Compare
Contributor
|
/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline,Windows x64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 5 pipeline(s). |
HectorSVC
reviewed
Aug 13, 2025
qti-yuduo
force-pushed
the
dev/yuduo/fold-q
branch
from
34ed336 to
00bcce4
Compare
Contributor
|
/azp run Linux QNN CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Windows ARM64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline,Windows x64 QNN CI Pipeline |
|
Azure Pipelines successfully started running 5 pipeline(s). |
HectorSVC
approved these changes
Aug 14, 2025
snnn
pushed a commit
that referenced
this pull request
Aug 28, 2025
…5673) ### Description Relax WeightBiasQuantization constraint for larger QDQ node group ### Motivation and Context The transformer `WeightBiasQuantization` quantizes float weights on `Q -> DQ -> Conv/ConvTranspose/Gemm's Weights -> Q-> DQ` sequence; The check on `Weights -> Q` (`children_nodes.size() != 1 || children_nodes[0]->OpType() != QDQ::QOpName`) is an issue due to it would skip quantization for many common patterns such as unfused activations followed by `Conv` (`DQ - Conv -> ReLU -> Q`). It's actually unnecessary to check ending Q here (the fold can happen anyway without changing model semantics). However, in order to minimize the current behavior change, this PR simply extend the pattern to include single path (no branch), type-preserving path lead to `Q` to enable more quantization support.
snnn
added a commit
that referenced
this pull request
Aug 29, 2025
- **Relax WeightBiasQuantization constraint for larger QDQ node group (#25673)** - **Add cuda graph implementation for NV TRT RTX EP (#25787)** - **python GPU IO Bindings for NVIDIA (#25776)** - **Fixes for DynamicQuantizeMatMul and Attention3D tests (#25814)** - **Fix a long standing bug on file memory mapping on windows. (#25833)** - **Add API for precompiled model compatibility check using just the compat info (#25841)** - **Enable ABSL_FLAGS flag registration for onnxruntime_perf_test for mobile build (#25849)** - **Add default constructor to Ort::Status. (#25860)** - #25871 - #25878 - #25884 - #25886 - #25866
Member
|
The change is added to the release branch |
gedoensmax
pushed a commit
to gedoensmax/onnxruntime
that referenced
this pull request
Sep 2, 2025
…crosoft#25673) ### Description Relax WeightBiasQuantization constraint for larger QDQ node group ### Motivation and Context The transformer `WeightBiasQuantization` quantizes float weights on `Q -> DQ -> Conv/ConvTranspose/Gemm's Weights -> Q-> DQ` sequence; The check on `Weights -> Q` (`children_nodes.size() != 1 || children_nodes[0]->OpType() != QDQ::QOpName`) is an issue due to it would skip quantization for many common patterns such as unfused activations followed by `Conv` (`DQ - Conv -> ReLU -> Q`). It's actually unnecessary to check ending Q here (the fold can happen anyway without changing model semantics). However, in order to minimize the current behavior change, this PR simply extend the pattern to include single path (no branch), type-preserving path lead to `Q` to enable more quantization support.
qti-yuduo
pushed a commit
to CodeLinaro/onnxruntime
that referenced
this pull request
Sep 24, 2025
- pick microsoft#25584 - pick microsoft#25635 - pick microsoft#25673 - pick microsoft#25702 - pick microsoft#25738
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
Description
Relax WeightBiasQuantization constraint for larger QDQ node group
Motivation and Context
The transformer
WeightBiasQuantizationquantizes float weights onQ -> DQ -> Conv/ConvTranspose/Gemm's Weights -> Q-> DQsequence; The check onWeights -> Q(children_nodes.size() != 1 || children_nodes[0]->OpType() != QDQ::QOpName) is an issue due to it would skip quantization for many common patterns such as unfused activations followed byConv(DQ - Conv -> ReLU -> Q).It's actually unnecessary to check ending Q here (the fold can happen anyway without changing model semantics). However, in order to minimize the current behavior change, this PR simply extend the pattern to include single path (no branch), type-preserving path lead to
Qto enable more quantization support.