diff --git a/mindspore/lite/src/lite_session.cc b/mindspore/lite/src/lite_session.cc
index 8e01ac7ced..1504b14939 100644
--- a/mindspore/lite/src/lite_session.cc
+++ b/mindspore/lite/src/lite_session.cc
@@ -547,6 +547,7 @@ LiteSession::~LiteSession() {
   mindspore::lite::NPUPassManager::GetInstance()->Clear();
   mindspore::lite::NPUManager::GetInstance()->Reset();
 #endif
+  delete (model_);
   is_running_.store(false);
 }
 
@@ -723,7 +724,7 @@ session::LiteSession *session::LiteSession::CreateSession(const char *model_buf,
     return nullptr;
   }
   model->buf = nullptr;
-  delete (model);
+  (reinterpret_cast<lite::LiteSession *>(session))->set_model(model);
   return session;
 }
 }  // namespace mindspore
diff --git a/mindspore/lite/src/lite_session.h b/mindspore/lite/src/lite_session.h
index 91e47bf5eb..7cbbfadfc6 100644
--- a/mindspore/lite/src/lite_session.h
+++ b/mindspore/lite/src/lite_session.h
@@ -66,6 +66,8 @@ class LiteSession : public session::LiteSession {
   int Resize(const std::vector<mindspore::tensor::MSTensor *> &inputs,
              const std::vector<std::vector<int>> &dims) override;
 
+  void set_model(Model *model) { this->model_ = model; }
+
  protected:
   static void ConvertTensorsQuantParam(const schema::Tensor *src_tensor, lite::Tensor *dst_tensor);
 
@@ -123,6 +125,7 @@ class LiteSession : public session::LiteSession {
   // graph output tensor name -- output tensor
   std::unordered_map<std::string, mindspore::tensor::MSTensor *> output_tensor_map_;
   Executor *executor_ = nullptr;
+  Model *model_ = nullptr;
   std::atomic<bool> is_running_ = false;
 #if SUPPORT_GPU && !SUPPORT_TRAIN
   opencl::OpenCLRuntimeWrapper ocl_runtime_wrap_;