diff --git a/paddle/operators/transpose_op.cc b/paddle/operators/transpose_op.cc
index 9b7812c79d39cacdd1a7a3749615effbe091a530..ea6b2a9ec5ac2b749a6c07116c5f16cf2e15fb4e 100644
--- a/paddle/operators/transpose_op.cc
+++ b/paddle/operators/transpose_op.cc
@@ -28,7 +28,7 @@ class TransposeOp : public framework::OperatorWithKernel {
  protected:
   void InferShape(const framework::InferShapeContext &ctx) const override {
     auto in_dim = ctx.Input<Tensor>("X")->dims();
-    auto axis = ctx.GetAttr<std::vector<int>>("axis");
+    auto axis = ctx.Attr<std::vector<int>>("axis");
     size_t in_dim_size = in_dim.size();
     size_t axis_size = axis.size();
 
diff --git a/paddle/operators/transpose_op.cu b/paddle/operators/transpose_op.cu
index 853659e3c3ed6e1111d64a2c1fd2c1dbe5f994d2..24feeea4b2cd4ad311925744cae537ade0eac346 100644
--- a/paddle/operators/transpose_op.cu
+++ b/paddle/operators/transpose_op.cu
@@ -98,7 +98,7 @@ class TransposeCUDAKernel : public framework::OpKernel {
                    "It must use GPUPlace.");
     auto* in = context.Input<framework::Tensor>("X");
     auto* out = context.Output<framework::Tensor>("Out");
-    auto axis = context.GetAttr<std::vector<int>>("axis");
+    auto axis = context.Attr<std::vector<int>>("axis");
     TransposeCUDA<T>(context, *in, *out, axis);
   }
 };
@@ -111,7 +111,7 @@ class TransposeGradCUDAKernel : public framework::OpKernel {
                    "It must use GPUPlace.");
     auto* in = context.Input<framework::Tensor>(framework::GradVarName("Out"));
     auto* out = context.Output<framework::Tensor>(framework::GradVarName("X"));
-    auto axis_temp = context.GetAttr<std::vector<int>>("axis");
+    auto axis_temp = context.Attr<std::vector<int>>("axis");
 
     std::vector<int> axis(axis_temp);
 
diff --git a/paddle/operators/transpose_op.h b/paddle/operators/transpose_op.h
index ca64b5a636417ed11720669e0378818ee42d6284..57f63e60e9a9e0febbb9af23be2984ef4b217119 100644
--- a/paddle/operators/transpose_op.h
+++ b/paddle/operators/transpose_op.h
@@ -77,7 +77,7 @@ class TransposeKernel : public framework::OpKernel {
     auto* out = context.Output<framework::Tensor>("Out");
     out->mutable_data<T>(context.GetPlace());
 
-    auto axis = context.GetAttr<std::vector<int>>("axis");
+    auto axis = context.Attr<std::vector<int>>("axis");
     int ndims = axis.size();
     switch (ndims) {
       case 2:
@@ -107,7 +107,7 @@ class TransposeGradKernel : public framework::OpKernel {
     auto* out = context.Output<framework::Tensor>(framework::GradVarName("X"));
     out->mutable_data<T>(context.GetPlace());
 
-    auto axis_temp = context.GetAttr<std::vector<int>>("axis");
+    auto axis_temp = context.Attr<std::vector<int>>("axis");
     std::vector<int> axis(axis_temp);
 
     for (size_t i = 0; i < axis.size(); i++) {