Skip to content

Normalize to positive axis for comparison on GEMM #26152

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
edgchen1 merged 6 commits into from
Sep 30, 2025
Merged

Normalize to positive axis for comparison on GEMM #26152

Changes from 5 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
245dafe
Normalize to positive axis for comparison on GEMM
qti-yuduo Sep 24, 2025
f3495d1
Address feedback
qti-yuduo Sep 25, 2025
a4a82e2
Update onnxruntime/core/optimizer/qdq_transformer/weight_bias_quantiz…
qti-yuduo Sep 25, 2025
08376eb
Address feedback
qti-yuduo Sep 25, 2025
0111231
Address feedback -- skip when unknown shape
qti-yuduo Sep 25, 2025
4441bfc
Add unit test
qti-yuduo Sep 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions onnxruntime/core/optimizer/qdq_transformer/weight_bias_quantization.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@
#include "core/optimizer/qdq_transformer/weight_bias_quantization.h"

#include "core/common/common.h"
#include "core/providers/common.h"
#include "core/util/qmath.h"
#include "core/graph/graph_utils.h"
#include "core/graph/graph_viewer.h"
Expand Down Expand Up @@ -128,6 +129,14 @@ Status WeightBiasQuantization::ApplyImpl(Graph& graph, bool& modified, int graph
int64_t axis = 1;
if (auto axis_iter = dq_attrs.find("axis"); axis_iter != dq_attrs.end()) {
axis = axis_iter->second.i();
const ONNX_NAMESPACE::TensorShapeProto* weight_shape = weight_arg->Shape();
if (!weight_shape && dq_1 && dq_1->InputDefs()[0]) {
Comment thread
edgchen1 marked this conversation as resolved.
Outdated
weight_shape = dq_1->InputDefs()[0]->Shape();
}
if (axis < 0 && !weight_shape) {
continue;
}
axis = HandleNegativeAxis(axis, weight_shape->dim_size());
Comment thread
qti-yuduo marked this conversation as resolved.
}

int64_t expected_axis = 0;
Expand Down