Skip to content

Fix WebGPU inference crash in embedding and multi-modal feature allocation #2163

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
Baiju Meswani (baijumeswani) merged 1 commit into from
May 18, 2026
Merged

Fix WebGPU inference crash in embedding and multi-modal feature allocation #2163

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
4 changes: 2 additions & 2 deletions src/models/embeddings.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@ Embeddings::Embeddings(State& state, Embeddings::Mode mode, const std::string& n
// So only create the transient input and reuse that ortvalue for previous
// steps in the pipeline.
if (mode == Embeddings::Mode::Input) {
embeddings_ = OrtValue::CreateTensor(model_.p_device_->GetAllocator(), shape_, type_);
embeddings_ = OrtValue::CreateTensor(model_.p_device_inputs_->GetAllocator(), shape_, type_);
}
}

Expand Down Expand Up @@ -49,7 +49,7 @@ void Embeddings::UpdateSequenceLength(size_t new_length) {
shape_[1] = new_length;

if (mode_ == Embeddings::Mode::Input) {
embeddings_ = OrtValue::CreateTensor(model_.p_device_->GetAllocator(), shape_, type_);
embeddings_ = OrtValue::CreateTensor(model_.p_device_inputs_->GetAllocator(), shape_, type_);
state_.inputs_[index_] = embeddings_.get();
}
}
Expand Down
4 changes: 2 additions & 2 deletions src/models/multi_modal_features.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,7 @@ void MultiModalFeatures::Update(bool is_prompt) {
// num_feature_tokens will be 0 when no image is provided
if (!is_prompt && shape_[shape_.size() - 2] > 0) { // if num_image_tokens > 0
shape_[shape_.size() - 2] = 0;
features_ = OrtValue::CreateTensor(model_.p_device_->GetAllocator(), shape_, type_);
features_ = OrtValue::CreateTensor(model_.p_device_inputs_->GetAllocator(), shape_, type_);
state_.inputs_[index_] = features_.get();
}
}
Expand All @@ -82,7 +82,7 @@ void MultiModalFeatures::AllocateEmptyFeatures() {
// Skip if already allocated (avoids redundant allocation when called from
// both EmbeddingState::SetExtraInputs and the pipeline prompt path)
if (features_ && state_.inputs_[index_] == features_.get()) return;
features_ = OrtValue::CreateTensor(model_.p_device_->GetAllocator(), shape_, type_);
features_ = OrtValue::CreateTensor(model_.p_device_inputs_->GetAllocator(), shape_, type_);
state_.inputs_[index_] = features_.get();
}

Expand Down
Loading