diff --git a/paddle/fluid/operators/cast_op.cc b/paddle/fluid/operators/cast_op.cc
index 52ad19c385e97dd528aef251590a69c46337dd3a..ec7c820a1440e7d07a73428bd0d7da216bc2705f 100644
--- a/paddle/fluid/operators/cast_op.cc
+++ b/paddle/fluid/operators/cast_op.cc
@@ -72,15 +72,18 @@ class CastCompositeGradOpMaker : public prim::CompositeGradOpMakerBase {
   using prim::CompositeGradOpMakerBase::CompositeGradOpMakerBase;
 
   void Apply() override {
-    paddle::Tensor out_grad = paddle::Tensor(
-        std::make_shared<prim::DescTensor>(this->SingleOutputGrad("Out")));
-    paddle::Tensor x_grad = paddle::Tensor(
-        std::make_shared<prim::DescTensor>(this->SingleInputGrad("X")));
-    auto dx_ptr = this->GetOutputPtr(&x_grad);
-    std::string dx_name = this->GetOutputName(x_grad);
-    auto dtype = phi::TransToPhiDataType((this->Attr<int>("in_dtype")));
-    prim::cast_grad<prim::DescTensor>(out_grad, dtype, dx_ptr);
-    this->RecoverOutputName(x_grad, dx_name);
+    paddle::Tensor x = this->GetSingleForwardInput("X");
+    paddle::Tensor out_grad = this->GetSingleOutputGrad("Out");
+
+    // get outputs
+    paddle::Tensor x_grad_t = this->GetSingleInputGrad("X");
+    paddle::Tensor *x_grad = this->GetOutputPtr(&x_grad_t);
+    std::string x_grad_name = this->GetOutputName(x_grad_t);
+
+    VLOG(6) << "Runing cast_grad composite func";
+    prim::cast_grad<prim::DescTensor>(x, out_grad, x_grad);
+
+    this->RecoverOutputName(x_grad_t, x_grad_name);
   }
 };
 
diff --git a/paddle/fluid/prim/api/composite_backward/composite_backward_api.h b/paddle/fluid/prim/api/composite_backward/composite_backward_api.h
index 8ad54079ca087b5feb839f6efca7246682ec3f40..979be562db93a519d93275bfde43d08a8b7c6bc1 100644
--- a/paddle/fluid/prim/api/composite_backward/composite_backward_api.h
+++ b/paddle/fluid/prim/api/composite_backward/composite_backward_api.h
@@ -110,9 +110,9 @@ void softmax_grad(const Tensor& out,
 }
 
 template <typename T>
-void cast_grad(const Tensor& out_grad, DataType dtype, Tensor* x_grad) {
+void cast_grad(const Tensor& x, const Tensor& out_grad, Tensor* x_grad) {
   if (x_grad) {
-    auto res = cast<T>(out_grad, dtype);
+    auto res = cast<T>(out_grad, x.dtype());
     set_output<T>(res, x_grad);
   }
 }