diff --git a/src/framework/tensor.h b/src/framework/tensor.h
index a8f808519130140e76aab1ced34bbc4885314574..a221a26aa1435000646cf7d58321df28f3322834 100644
--- a/src/framework/tensor.h
+++ b/src/framework/tensor.h
@@ -85,6 +85,12 @@ class Tensor {
     }
   }
 
+  Tensor(const Tensor &inTensor) {
+    this->dims_ = inTensor.dims_;
+    this->holder_ = inTensor.holder_;
+    this->offset_ = inTensor.offset_;
+  }
+
   /*! Return a pointer to mutable memory block. */
   template <typename T>
   inline T *data() {
@@ -169,7 +175,9 @@ class Tensor {
   /*! The internal of two tensors share the same memory block. */
   inline Tensor &ShareDataWith(const Tensor &src) {
     src.check_memory_size();
-    *this = src;
+    if (holder_.get() != src.holder_.get()) {
+      *this = src;
+    }
     return *this;
   }
 
@@ -198,7 +206,6 @@ class Tensor {
       size_t base = numel() / dims_[0];
       Tensor dst;
       dst.holder_ = holder_;
-      dst.set_layout(layout_);
       DDim dst_dims = dims_;
       dst_dims[0] = end_idx - begin_idx;
       dst.Resize(dst_dims);
@@ -227,10 +234,6 @@ class Tensor {
                           "Tensor's dims_ is out of bound. ");
   }
 
-  inline DataLayout layout() const { return layout_; }
-
-  inline void set_layout(const DataLayout layout) { layout_ = layout; }
-
  private:
   /**
    * @note    Placeholder hides type T, so it doesn't appear as a
@@ -288,21 +291,6 @@ class Tensor {
 
   DDim dims_;
 
-  /**
-   * @brief the layout of memory block, default is NHWC.
-   *
-   * @note the memory allocation order, describe how weight/data is
-   * stored
-   *       For example, in 4-D Tensor(rank=4), there are three
-   * commonly
-   *       used layout. They are
-   *            NCHW, NHWC, CHWN.
-   *       N,C,H,W for respectively the batch size, the number of
-   *       feature maps, the height, the width.
-   */
-
-  DataLayout layout_ = DataLayout::kNHWC;
-
   /**
    * @brief   A PlaceHolder may be shared by more than one tensor.
    *
diff --git a/src/framework/tensor_util.cpp b/src/framework/tensor_util.cpp
index 465502cb19173e26361905752e76e75c15229893..6722ec3e37b8219eee9e1b9913799b08d8f902bc 100644
--- a/src/framework/tensor_util.cpp
+++ b/src/framework/tensor_util.cpp
@@ -20,7 +20,6 @@ namespace framework {
 void TensorCopy(const Tensor &src, Tensor *dst) {
   src.check_memory_size();
   dst->Resize(src.dims());
-  dst->set_layout(src.layout());
   auto src_ptr = src.data<void>();
   auto dst_ptr = dst->mutable_data(src.type());
   auto size = src.numel() * SizeOfType(src.type());
diff --git a/src/io/io.cpp b/src/io/io.cpp
index e5ca89f9f6fd495c6adaba9f982f0576a2a1af30..7931432bd1d4528ab9b0cda7ab05ab13c14dfcfe 100644
--- a/src/io/io.cpp
+++ b/src/io/io.cpp
@@ -477,7 +477,7 @@ std::shared_ptr<framework::Tensor> Executor<Dtype, P>::Predict(
   printf("====================[---------]======================\n");
 #endif
 
-  return std::shared_ptr<framework::Tensor>(output_tensor);
+  return std::make_shared<framework::Tensor>(framework::Tensor(*output_tensor));
 }
 template <typename Dtype, Precision P>
 std::shared_ptr<framework::Tensor> Executor<Dtype, P>::Predict(
diff --git a/test/net/test_googlenet.cpp b/test/net/test_googlenet.cpp
index e1bcd5212de214abbf2df5822ebc977619f8caa3..1695995a8d60d20e0d6c5f8911c39a948426a82a 100644
--- a/test/net/test_googlenet.cpp
+++ b/test/net/test_googlenet.cpp
@@ -30,7 +30,11 @@ int main() {
   std::vector<int64_t> dims{1, 3, 224, 224};
   GetInput<float>(g_test_image_1x3x224x224, &input, dims);
   auto time3 = time();
-  executor.Predict(input, dims);
+
+  for (int i = 0; i < 10; ++i) {
+    executor.Predict(input, dims);
+  }
+
   auto time4 = time();
   DLOG << "predict cost :" << time_diff(time3, time4) << "ms\n";
   return 0;