diff --git a/paddle/fluid/operators/while_op.cc b/paddle/fluid/operators/while_op.cc
index 16eac1ec2406c147fa765bc014038ae03a1416b2..3c8a01b6e47459760b05b5ca7fa4fa5e1d37d112 100644
--- a/paddle/fluid/operators/while_op.cc
+++ b/paddle/fluid/operators/while_op.cc
@@ -224,10 +224,12 @@ class WhileGradOp : public framework::OperatorBase {
         if (cur_scope_iter == step_scopes->rbegin()) {
           auto *var = (*cur_scope_iter)->FindVar(inside_grad_name);
           PADDLE_ENFORCE_NOT_NULL(var, "Can not find var %s", inside_grad_name);
-          PADDLE_ENFORCE(var->IsType<framework::LoDTensorArray>() ||
-                             var->IsType<LoDTensor>(),
-                         "Currently the type of var only can be LoDTensorArray "
-                         "or LoDTensor.");
+          PADDLE_ENFORCE(
+              var->IsType<framework::LoDTensorArray>() ||
+                  var->IsType<LoDTensor>(),
+              "Currently the type of var only can be LoDTensorArray, "
+              "or LoDTensor, but the received var[%s] is %s.",
+              inside_grad_name, var->Type().name());
 
           if (var->IsType<LoDTensor>()) {
             auto &inside_tensor = var->Get<framework::LoDTensor>();
diff --git a/python/paddle/fluid/layers/control_flow.py b/python/paddle/fluid/layers/control_flow.py
index c6250ff6ce5df8d8b0c78d538d736b77801f98f8..4af97e8632a47fbd981362dc8249a3f6b7269ecd 100644
--- a/python/paddle/fluid/layers/control_flow.py
+++ b/python/paddle/fluid/layers/control_flow.py
@@ -1570,6 +1570,10 @@ class DynamicRNN(object):
 
     The dynamic RNN can mark multiple variables as its output. Use `drnn()` to
     get the output sequence.
+    
+    NOTES:
+        Currently it is not supported that setting is_sparse to True of any 
+        layers within DynamicRNN.
     """
     BEFORE_RNN = 0
     IN_RNN = 1