diff --git a/lite/api/paddle_api.cc b/lite/api/paddle_api.cc
index f148096bb69a3a249521bcb847d5beae3f8297f9..131041012ae606f4581f0e6f0f4cae929931a10d 100644
--- a/lite/api/paddle_api.cc
+++ b/lite/api/paddle_api.cc
@@ -93,7 +93,7 @@ void Tensor::CopyFromCpu(const T *src_data) {
   }
 }
 template <typename T>
-void Tensor::CopyToCpu(T *data) {
+void Tensor::CopyToCpu(T *data) const {
   const T *src_data = tensor(raw_tensor_)->data<T>();
   int64_t num = tensor(raw_tensor_)->numel();
   CHECK(num > 0) << "You should call Resize interface first";
@@ -124,9 +124,9 @@ template void Tensor::CopyFromCpu<int, TargetType::kCUDA>(const int *);
 template void Tensor::CopyFromCpu<float, TargetType::kCUDA>(const float *);
 template void Tensor::CopyFromCpu<int8_t, TargetType::kCUDA>(const int8_t *);
 
-template void Tensor::CopyToCpu(int8_t *);
-template void Tensor::CopyToCpu(float *);
-template void Tensor::CopyToCpu(int *);
+template void Tensor::CopyToCpu(int8_t *) const;
+template void Tensor::CopyToCpu(float *) const;
+template void Tensor::CopyToCpu(int *) const;
 
 shape_t Tensor::shape() const {
   return ctensor(raw_tensor_)->dims().Vectorize();
diff --git a/lite/api/paddle_api.h b/lite/api/paddle_api.h
index 42b455da811fe1a21277d38f2e1237000276b1ff..c578769bd5159d27ad43e4e93de33f601223004b 100644
--- a/lite/api/paddle_api.h
+++ b/lite/api/paddle_api.h
@@ -49,7 +49,7 @@ struct LITE_API Tensor {
   void CopyFromCpu(const T* data);
 
   template <typename T>
-  void CopyToCpu(T* data);
+  void CopyToCpu(T* data) const;
   /// Shape of the tensor.
   shape_t shape() const;
   TargetType target() const;