diff --git a/mindspore/lite/src/delegate/npu/npu_delegate.cc b/mindspore/lite/src/delegate/npu/npu_delegate.cc
index 093e43b5bcc..34aaaa58b14 100644
--- a/mindspore/lite/src/delegate/npu/npu_delegate.cc
+++ b/mindspore/lite/src/delegate/npu/npu_delegate.cc
@@ -69,7 +69,7 @@ NPUDelegate::~NPUDelegate() {
 }
 
 int NPUDelegate::Init() {
-  npu_manager_ = new (std::nothrow) NPUManager();
+  npu_manager_ = new (std::nothrow) NPUManager(frequency_);
   if (npu_manager_ == nullptr) {
     MS_LOG(ERROR) << "New npu manager failed.";
     return RET_ERROR;
diff --git a/mindspore/lite/src/delegate/npu/npu_manager.h b/mindspore/lite/src/delegate/npu/npu_manager.h
index a90eaf50204..5ba2fa6213b 100644
--- a/mindspore/lite/src/delegate/npu/npu_manager.h
+++ b/mindspore/lite/src/delegate/npu/npu_manager.h
@@ -48,7 +48,7 @@ struct SubGraphModel {
 
 class NPUManager {
  public:
-  NPUManager() = default;
+  explicit NPUManager(int frequency) : frequency_(frequency) {}
 
   ~NPUManager() { Reset(); }
 
@@ -70,6 +70,8 @@ class NPUManager {
   int LoadModel(const std::shared_ptr<hiai::AiModelMngerClient> &client,
                 std::vector<std::shared_ptr<hiai::AiModelDescription>> desc_list);
 
+  int GetFrequency() { return frequency_; }
+
  private:
   bool IsKirinChip();
 
@@ -87,6 +89,7 @@ class NPUManager {
   bool is_support_ = false;
   std::unordered_map<std::string, std::shared_ptr<SubGraphModel>> models_;
   std::vector<std::shared_ptr<hiai::AiModelMngerClient>> clients_;
+  int frequency_ = 0;
 };
 
 }  // namespace mindspore
diff --git a/mindspore/lite/src/delegate/npu/npu_subgraph.cc b/mindspore/lite/src/delegate/npu/npu_subgraph.cc
index 3bbee48b46d..9d5eb674738 100644
--- a/mindspore/lite/src/delegate/npu/npu_subgraph.cc
+++ b/mindspore/lite/src/delegate/npu/npu_subgraph.cc
@@ -170,7 +170,6 @@ int NPUSubGraph::BuildNPUInputOp() {
       }
 
       bool is_weight_tensor = true;
-      // todo cpu in_ops can't be found
       auto pre_ops = FindPreOps(op);
       for (auto pre_op : pre_ops) {
         if (find(pre_op->outputs().begin(), pre_op->outputs().end(), in_tensor) != pre_op->outputs().end()) {
@@ -276,8 +275,7 @@ int NPUSubGraph::Init() {
   }
 
   MS_ASSERT(npu_manager_ != nullptr);
-  // todo y00520784, get frequency
-  npu_manager_->AddModel(model_buffer_data, GetOMModelName(), 3);
+  npu_manager_->AddModel(model_buffer_data, GetOMModelName(), npu_manager_->GetFrequency());
 
   executor_ = new (std::nothrow) NPUExecutor(GetOMModelName(), npu_manager_);
 
diff --git a/mindspore/lite/src/delegate/npu/pass/npu_pass_utils.h b/mindspore/lite/src/delegate/npu/pass/npu_pass_utils.h
index 842fb19daab..5f3c71aab07 100644
--- a/mindspore/lite/src/delegate/npu/pass/npu_pass_utils.h
+++ b/mindspore/lite/src/delegate/npu/pass/npu_pass_utils.h
@@ -57,7 +57,6 @@ class NPUPassUtils {
   static int MaskDataNHWC2NCHW(int mask);
 };
 
-// todo y00520784 : refactor the code of transform op
 class RuntimePass {
  public:
   RuntimePass(std::vector<NPUOp *> *ops, std::vector<tensor::MSTensor *> *tensors)