Skip to content

Commit

Permalink
[GPU] Fix SD3 hybrid static failure. (#28430)
Browse files Browse the repository at this point in the history
Fix functional issue when checks data type during
DynamicQuantizeFullyConnected callback. Target model has two output
node, it caused the error.


### Tickets:
 - *160518*

Signed-off-by: hyunback <hyunback.kim@intel.com>
  • Loading branch information
hyunback authored Jan 15, 2025
1 parent 5166440 commit 3ee2339
Showing 1 changed file with 5 additions and 3 deletions.
8 changes: 5 additions & 3 deletions src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -1055,9 +1055,11 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
if (device_info.supports_immad) {
auto dynamic_quantization_group_size = config.get_property(ov::hint::dynamic_quantization_group_size);
pass_config->set_callback<ov::intel_gpu::DynamicQuantizeFullyConnected>([=](const_node_ptr& root) -> bool {
if (root->get_input_node_shared_ptr(0)->get_element_type() == ov::element::Type_t::f32) {
GPU_DEBUG_TRACE << root->get_friendly_name() << " dyn_quan is turned off: input type is not supported" << std::endl;
return true;
for (size_t i = 0 ; i < root->get_input_node_shared_ptr(0)->get_output_size(); ++i) {
if (root->get_input_node_shared_ptr(0)->get_output_element_type(i) == ov::element::Type_t::f32) {
GPU_DEBUG_TRACE << root->get_friendly_name() << " dyn_quan is turned off: input type is not supported" << std::endl;
return true;
}
}

auto weight_shape = root->get_input_partial_shape(1);
Expand Down

0 comments on commit 3ee2339

Please sign in to comment.