saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-06-24 02:47:54 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[WebNN] Don't skip scalar tensor registration (#22688)

Browse source 
ORT will optimize same scalar initializers into one, we should not skip
such scalar registration as a WebNN Constant.
This commit is contained in:
Wanming Lin 2024-11-04 13:33:34 +08:00 committed by GitHub
parent 777fe7922c
commit c64459fdd7
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 7 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
onnxruntime/core/providers/webnn/builders/model_builder.cc
View file

@ -88,11 +88,15 @@ Status ModelBuilder::RegisterInitializers() {
for (const auto& pair : GetInitializerTensors()) {
const auto& tensor = *pair.second;
const auto& name = tensor.name();
// Optional tensors can be indicated by an empty name, just ignore it.
if (name.empty() || Contains(skipped_initializers_, name))
const auto& shape = tensor.dims();
// Ignore the following tensors:
// 1. Empty tensors: optional tensors can be indicated by an empty name.
// 2. Tensors in skipped_initializers_: These are tensors that are not used as WebNN Constants.
// Note: Scalar tensors are excluded because ONNX Runtime will optimize same scalar initializers into one.
if (name.empty() || (Contains(skipped_initializers_, name) && !shape.empty()))
continue;
const auto& shape = tensor.dims();
std::vector<int32_t> dims;
// When the shape is empty, it is scalar initializer that dims = {};
std::transform(shape.cbegin(), shape.cend(),