diff --git a/paddle/fluid/op_use_default_grad_op_maker.spec b/paddle/fluid/op_use_default_grad_op_maker.spec
index 6ac5f994222ec56775b38f8d077579b825419f22..6fed54d5367abd180255d09b62ba0dcbe26a51ee 100644
--- a/paddle/fluid/op_use_default_grad_op_maker.spec
+++ b/paddle/fluid/op_use_default_grad_op_maker.spec
@@ -1,19 +1,10 @@
-attention_lstm
 conv_shift
 cos_sim
 dequantize
 fc
 flatten
 fsp
-fused_embedding_fc_lstm
 fused_embedding_seq_pool
-fusion_gru
-fusion_lstm
-fusion_repeated_fc_relu
-fusion_seqconv_eltadd_relu
-fusion_seqexpand_concat_fc
-fusion_seqpool_concat
-fusion_squared_mat_sub
 gru
 lrn
 lstm_unit
diff --git a/paddle/fluid/operators/attention_lstm_op.cc b/paddle/fluid/operators/attention_lstm_op.cc
index f991bef96529b0a86ccf96822b95fb4e4c274d6f..275855cbb6e4c7323a0c57a8e00ed0b7fb7f8f9c 100644
--- a/paddle/fluid/operators/attention_lstm_op.cc
+++ b/paddle/fluid/operators/attention_lstm_op.cc
@@ -419,8 +419,7 @@ class AttentionLSTMKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(attention_lstm, ops::AttentionLSTMOp,
-                  ops::AttentionLSTMOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::AttentionLSTMOpMaker);
 
 REGISTER_OP_CPU_KERNEL(attention_lstm, ops::AttentionLSTMKernel<float>,
                        ops::AttentionLSTMKernel<double>);
diff --git a/paddle/fluid/operators/distributed_ops/split_ids_op.cc b/paddle/fluid/operators/distributed_ops/split_ids_op.cc
index 191ca1efe8ca5798ddbd38968eafde349af8a7d1..d46b57e7e15807756efd85fde765454260ea9d7b 100644
--- a/paddle/fluid/operators/distributed_ops/split_ids_op.cc
+++ b/paddle/fluid/operators/distributed_ops/split_ids_op.cc
@@ -81,27 +81,12 @@ class SplitIdsOpInferVarType : public framework::VarTypeInference {
   }
 };
 
-class SplitIdsOpGradMaker : public framework::SingleGradOpDescMaker {
- public:
-  using framework::SingleGradOpDescMaker::SingleGradOpDescMaker;
-
- protected:
-  std::unique_ptr<framework::OpDesc> Apply() const override {
-    auto grad = new framework::OpDesc();
-    grad->SetType("concat");
-    grad->SetInput("X", OutputGrad("Out"));
-    grad->SetOutput("Out", InputGrad("Ids"));
-    grad->SetAttr("axis", 0);
-    return std::unique_ptr<framework::OpDesc>(grad);
-  }
-};
-
 }  // namespace operators
 }  // namespace paddle
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(split_ids, ops::SplitIdsOp, ops::SplitIdsOpMaker,
-                  ops::SplitIdsOpGradMaker, ops::SplitIdsOpInferVarType);
+                  ops::SplitIdsOpInferVarType);
 
 REGISTER_OP_CPU_KERNEL(
     split_ids, ops::SplitIdsOpKernel<paddle::platform::CPUPlace, int64_t>,
diff --git a/paddle/fluid/operators/fused/fused_embedding_fc_lstm_op.cc b/paddle/fluid/operators/fused/fused_embedding_fc_lstm_op.cc
index 35a30854f22062efa594d02fecbbe6571fd75f97..2a2c583043a26ea69745253f099eb24ccc85bb58 100644
--- a/paddle/fluid/operators/fused/fused_embedding_fc_lstm_op.cc
+++ b/paddle/fluid/operators/fused/fused_embedding_fc_lstm_op.cc
@@ -589,8 +589,7 @@ class FusedEmbeddingFCLSTMKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(fused_embedding_fc_lstm, ops::FusedEmbeddingFCLSTMOp,
-                  ops::FusedEmbeddingFCLSTMOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::FusedEmbeddingFCLSTMOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fused_embedding_fc_lstm,
                        ops::FusedEmbeddingFCLSTMKernel<float>,
diff --git a/paddle/fluid/operators/fused/fusion_gru_op.cc b/paddle/fluid/operators/fused/fusion_gru_op.cc
index 56c41ef2a8ee096e31ca98b51556e0d0dbc237f6..e67c073b5be5e2e6d8fe20a45f91e8f623dc5d02 100644
--- a/paddle/fluid/operators/fused/fusion_gru_op.cc
+++ b/paddle/fluid/operators/fused/fusion_gru_op.cc
@@ -396,7 +396,7 @@ class FusionGRUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OPERATOR(fusion_gru, ops::FusionGRUOp, ops::FusionGRUOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+REGISTER_OPERATOR(fusion_gru, ops::FusionGRUOp, ops::FusionGRUOpMaker);
+
 REGISTER_OP_CPU_KERNEL(fusion_gru, ops::FusionGRUKernel<float>,
                        ops::FusionGRUKernel<double>);
diff --git a/paddle/fluid/operators/fused/fusion_lstm_op.cc b/paddle/fluid/operators/fused/fusion_lstm_op.cc
index 1a31fc7826512a3efda32eb3f5640e78844cfc99..f04aa017e3fc7527054c1bb90f8427638ccc9582 100644
--- a/paddle/fluid/operators/fused/fusion_lstm_op.cc
+++ b/paddle/fluid/operators/fused/fusion_lstm_op.cc
@@ -474,8 +474,7 @@ class FuisonLSTMKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-REGISTER_OPERATOR(fusion_lstm, ops::FusionLSTMOp, ops::FusionLSTMOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+REGISTER_OPERATOR(fusion_lstm, ops::FusionLSTMOp, ops::FusionLSTMOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fusion_lstm, ops::FuisonLSTMKernel<float>,
                        ops::FuisonLSTMKernel<double>);
diff --git a/paddle/fluid/operators/fused/fusion_repeated_fc_relu_op.cc b/paddle/fluid/operators/fused/fusion_repeated_fc_relu_op.cc
index 6be35de65f48525b2da7d5c9ef260b2d0798b67b..4c11482f5077eeeb2d446dc0cbe9c08f890f390f 100644
--- a/paddle/fluid/operators/fused/fusion_repeated_fc_relu_op.cc
+++ b/paddle/fluid/operators/fused/fusion_repeated_fc_relu_op.cc
@@ -144,8 +144,7 @@ class FusionRepeatedFCReluKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(fusion_repeated_fc_relu, ops::FusionRepeatedFCReluOp,
-                  ops::FusionRepeatedFCReluOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::FusionRepeatedFCReluOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fusion_repeated_fc_relu,
                        ops::FusionRepeatedFCReluKernel<float>,
diff --git a/paddle/fluid/operators/fused/fusion_seqconv_eltadd_relu_op.cc b/paddle/fluid/operators/fused/fusion_seqconv_eltadd_relu_op.cc
index b05329cfd072b767750f08dd73ae493880f11137..4a45177201af27709165bfc8bc881151575337b1 100644
--- a/paddle/fluid/operators/fused/fusion_seqconv_eltadd_relu_op.cc
+++ b/paddle/fluid/operators/fused/fusion_seqconv_eltadd_relu_op.cc
@@ -220,8 +220,7 @@ class FusionSeqConvEltAddReluKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(fusion_seqconv_eltadd_relu, ops::FusionSeqConvEltAddReluOp,
-                  ops::FusionSeqConvEltAddReluOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::FusionSeqConvEltAddReluOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fusion_seqconv_eltadd_relu,
                        ops::FusionSeqConvEltAddReluKernel<float>,
diff --git a/paddle/fluid/operators/fused/fusion_seqexpand_concat_fc_op.cc b/paddle/fluid/operators/fused/fusion_seqexpand_concat_fc_op.cc
index d091da5aa8a7e7ec30798d68021bfd2b9b87b32f..46632c1e9a4999a6e417e850874354f6f8817ba0 100644
--- a/paddle/fluid/operators/fused/fusion_seqexpand_concat_fc_op.cc
+++ b/paddle/fluid/operators/fused/fusion_seqexpand_concat_fc_op.cc
@@ -197,8 +197,7 @@ class FusionSeqExpandConcatFCOpKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(fusion_seqexpand_concat_fc, ops::FusionSeqExpandConcatFCOp,
-                  ops::FusionSeqExpandConcatFCOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::FusionSeqExpandConcatFCOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fusion_seqexpand_concat_fc,
                        ops::FusionSeqExpandConcatFCOpKernel<float>,
diff --git a/paddle/fluid/operators/fused/fusion_seqpool_concat_op.cc b/paddle/fluid/operators/fused/fusion_seqpool_concat_op.cc
index 25916768c08e7222ba95bd6e1999400a923b21a3..b14ee88aa53b64791fa09c848e23d4f01826e339 100644
--- a/paddle/fluid/operators/fused/fusion_seqpool_concat_op.cc
+++ b/paddle/fluid/operators/fused/fusion_seqpool_concat_op.cc
@@ -126,8 +126,7 @@ class FusionSeqPoolConcatKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(fusion_seqpool_concat, ops::FusionSeqPoolConcatOp,
-                  ops::FusionSeqPoolConcatOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::FusionSeqPoolConcatOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fusion_seqpool_concat,
                        ops::FusionSeqPoolConcatKernel<float>,
diff --git a/paddle/fluid/operators/fused/fusion_squared_mat_sub_op.cc b/paddle/fluid/operators/fused/fusion_squared_mat_sub_op.cc
index 53679ebddee1ceec102b5861c54b398aa4da4cde..2d10056044efa851898c8cf597fa14e495305fce 100644
--- a/paddle/fluid/operators/fused/fusion_squared_mat_sub_op.cc
+++ b/paddle/fluid/operators/fused/fusion_squared_mat_sub_op.cc
@@ -136,8 +136,7 @@ class FusionSquaredMatSubKernel : public framework::OpKernel<T> {
 
 namespace ops = paddle::operators;
 REGISTER_OPERATOR(fusion_squared_mat_sub, ops::FusionSquaredMatSubOp,
-                  ops::FusionSquaredMatSubOpMaker,
-                  paddle::framework::DefaultGradOpDescMaker<true>);
+                  ops::FusionSquaredMatSubOpMaker);
 
 REGISTER_OP_CPU_KERNEL(fusion_squared_mat_sub,
                        ops::FusionSquaredMatSubKernel<float>,