fix

Author: likholat

samples/cpp/text_generation/lora_greedy_causal_lm.cpp

@@ -19,6 +19,7 @@ int main(int argc, char* argv[]) try {
 
     // Resetting config to set greedy behaviour ignoring generation config from model directory.
     // It helps to compare two generations with and without LoRA adapter.
+    // LLMPipeline pipe(models_path, device); 
     ov::genai::GenerationConfig config;
     config.max_new_tokens = 100;
     pipe.set_generation_config(config);
@@ -30,7 +31,8 @@ int main(int argc, char* argv[]) try {
     std::cout << "\n-----------------------------";
     std::cout << "\nGenerate without LoRA adapter:" << std::endl;
     // // LLMPipeline pipe1(models_path, device);    // register all required adapters here
-    std::cout << pipe.generate(prompt, max_new_tokens(100), adapters()) << std::endl;
+    
+    // std::cout << pipe.generate(prompt, max_new_tokens(100)) << std::endl;
 
 } catch (const std::exception& error) {
     std::cerr << error.what() << '\n';

src/cpp/src/lora/adapter.cpp

@@ -477,7 +477,7 @@ struct LoRAStateGetterForConst : public BaseStateGetter {
         variable_ids(variable_ids) {}
 
     std::optional<LoRAConstantNode> operator() (NodePtr node) const {
-        std::cout << "LoRAStateGetterForConst operator()" << std::endl;
+        // std::cout << "LoRAStateGetterForConst operator()" << std::endl;
         std::string name = node->get_friendly_name();
         if (auto params = getter(name)) {
             // FIXME: Potential name conflict if LoRA is applied multiple times by using this infrastructure independently each time (not a recommended approach).
@@ -486,8 +486,6 @@ struct LoRAStateGetterForConst : public BaseStateGetter {
             LoRAConstantNode result;
             ov::op::util::VariableInfo variable_info;
 
-            std::cout << "111 " << params->tensor->get_output_element_type(0) << std::endl;
-
             // FIXME: No guarantees on ordering of state in InferRequest makes impossible using indices of variables later, forced to use variable_id instead
             variable_info = ov::op::util::VariableInfo{
                 params->tensor->get_output_shape(0),
@@ -619,9 +617,6 @@ class LoRAReplaceConstantTransformDynamic : public LoRAReplaceConstantTransform
     bool apply(NodePtr node, const LoRAConstantNode& lora_weight) override {
         auto consumers = node->get_output_target_inputs(0);
 
-        std::cout << lora_weight.tensor->get_element_type() << std::endl;
-        std::cout << node->get_element_type() << std::endl;
-
         const auto node_type = node->get_element_type();
 
         // Приводим tensor к типу node
@@ -653,7 +648,6 @@ class LoRAReplaceConstantTransformDynamic : public LoRAReplaceConstantTransform
 
         for (auto& consumer : consumers) {
             consumer.replace_source_output(if_node->output(0));
-            std::cout << consumer.get_node()->get_element_type() << std::endl;
         }
         return true;
     }
@@ -690,14 +684,12 @@ NodePtr tensors_multiplication(NodePtr input,
     const auto target_shape = target.get_partial_shape();
     const auto target_rank = target_shape.rank().get_length();
 
-    std::cout << 'tensors_multiplication' << std::endl;
+    std::cout << '!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!tensors_multiplication' << std::endl;
 
     for (size_t i = 0; i < multipliers.size(); ++i) {
         NodePtr normalized = multipliers[i];
-        std::cout <<"aaa "<<  normalized->get_output_element_type(0).get_type_name() << std::endl;
         if (normalized->get_output_element_type(0) != target_type) {
             normalized = std::make_shared<v0::Convert>(normalized, target_type);
-            std::cout <<"bbb "<<  normalized->get_output_element_type(0).get_type_name() << std::endl;
             if (std::dynamic_pointer_cast<v0::Constant>(normalized)) {
                 input->get_rt_info()["decompression"];
             }