Minor corrections

src/common/transformations/include/ov_ops/lora_subgraph.hpp

@@ -16,7 +16,8 @@ namespace internal {
  * @brief LoraSubgraph operation, which is used for LoRA subgraphs fusion.
  * It always has only 1 output, and the following inputs, whose order is fixed:
  * 1. main_flow_input: input from original model.
- * 2. LoRA_input: input to which the Low-Rank adaptation is applied. The adapted input is combined with `main_flow_input`.
+ * 2. LoRA_input: input to which the Low-Rank adaptation is applied.
+ *    The adapted input is combined with `main_flow_input`.
  * 3. LoRA_matrices: 3 Low-Rank adaptation matrices applied to `LoRA_input`.
  * The fused subgraph can be optimized in runtime based on LoRA semantic.
  * For instance, `main_flow_input` can be fast-forwarded to output in case of empty `LoRA_matrices`.

src/common/transformations/src/ov_ops/lora_subgraph.cpp

@@ -4,6 +4,8 @@
 
 #include "ov_ops/lora_subgraph.hpp"
 
+#include "itt.hpp"
+
 namespace ov {
 namespace op {
 namespace internal {
@@ -18,11 +20,13 @@ LoraSubgraph::LoraSubgraph(const OutputVector& args, const std::shared_ptr<ov::M
 }
 
 std::shared_ptr<Node> LoraSubgraph::clone_with_new_inputs(const OutputVector& new_args) const {
+    INTERNAL_OP_SCOPE(internal_LoraSubgraph_clone_with_new_inputs);
     check_new_args_count(this, new_args);
     return std::make_shared<LoraSubgraph>(new_args, get_function()->clone());
 }
 
 void LoraSubgraph::validate_and_infer_types() {
+    INTERNAL_OP_SCOPE(internal_LoraSubgraph_validate_and_infer_types);
     OPENVINO_ASSERT(get_input_size() == 5, "LoraSubgraph must have 5 inputs whereas it has ", get_input_size());
     OPENVINO_ASSERT(get_output_size() == 1, "LoraSubgraph must have 1 output whereas it has ", get_output_size());
     const auto& body = get_function();

src/common/transformations/src/transformations/common_optimizations/lora_subgraph_fusion.cpp

@@ -65,8 +65,8 @@ ov::pass::LoraSubgraphFusion::LoraSubgraphFusion() {
 
         // Note: internal_inputs/external_connections order corresponds to LoraSubgraph semantic
         const std::vector<ov::Input<ov::Node>> internal_inputs{
-            find_connected_input(add.get_node(), main_flow.get_node()),
             // For commutative eltwise ops, input idx may be any, so it must be computed
+            find_connected_input(add.get_node(), main_flow.get_node()),
             pattern_map.count(transpose1_m) ? pattern_map.at(transpose1_m).get_node()->input(0)
                                             : matmul1.get_node()->input(0),
             matmul1.get_node()->input(1),