diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 0a9872771ff65a795d397511326446990c676cdb..50a630686aba02591371d0cd26d563e768ef0a3d 100755
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -24,7 +24,7 @@ repos:
         files: (?!.*third_party)^.*$ | (?!.*book)^.*$
     -   id: end-of-file-fixer
     -   id: sort-simple-yaml
-        files: (api|backward|api_[a-z_]+)\.yaml$
+        files: (op|backward|op_[a-z_]+)\.yaml$
     -   id: trailing-whitespace
         files: (.*\.(py|bzl|md|rst|c|cc|cxx|cpp|cu|h|hpp|hxx|xpu|kps|cmake)|BUILD|.*\.BUILD|WORKSPACE|CMakeLists.txt)$
 -   repo: local
diff --git a/paddle/fluid/operators/minus_op.cc b/paddle/fluid/operators/minus_op.cc
index 310d28738fc756ad252305e018021d1ef38d5c57..398a254f45cbcc3f5efb54412e1775f61de75c28 100644
--- a/paddle/fluid/operators/minus_op.cc
+++ b/paddle/fluid/operators/minus_op.cc
@@ -130,6 +130,7 @@ class MinusGradMaker : public imperative::GradOpBaseMakerBase {
       op.SetInput("X", this->OutputGrad("Out"));
       op.SetOutput("Out", x_g);
       op.SetAttr("scale", 1.0f);
+      op.SetDefaultAttrsMap(DefaultAttrsMap());
     }
 
     if (!y_g.empty()) {
@@ -138,6 +139,7 @@ class MinusGradMaker : public imperative::GradOpBaseMakerBase {
       op.SetInput("X", this->OutputGrad("Out"));
       op.SetOutput("Out", y_g);
       op.SetAttr("scale", -1.0f);
+      op.SetDefaultAttrsMap(DefaultAttrsMap());
     }
 
     return node;
diff --git a/paddle/fluid/operators/scale_op.cc b/paddle/fluid/operators/scale_op.cc
index 3c2b5363d81ce2609bb7f974d65c84db97c5df36..cab04e43e8681f3f0d674f0d0e7633b07c3971c5 100644
--- a/paddle/fluid/operators/scale_op.cc
+++ b/paddle/fluid/operators/scale_op.cc
@@ -75,10 +75,6 @@ $$Out = scale*(X + bias)$$
         "Apply bias addition after or before scaling. It is useful for "
         "numeric stability in some circumstances.")
         .SetDefault(true);
-    AddAttr<bool>("use_mkldnn",
-                  "(bool, default false) Only used in mkldnn kernel")
-        .SetDefault(false)
-        .AsExtra();
   }
 };
 
@@ -108,11 +104,6 @@ class ScaleGradMaker : public framework::SingleGradOpMaker<T> {
     VLOG(6) << "Finish Set Attr bias";
     grad_op->SetAttr("bias_after_scale", true);
     VLOG(6) << "Finish Set Attr bias_after_scale";
-    if (grad_op->HasAttr("use_mkldnn")) {
-      VLOG(6) << "Finish Check Attr use_mkldnn";
-      grad_op->SetAttr("use_mkldnn", this->GetAttr("use_mkldnn"));
-      VLOG(6) << "Finish Set Attr use_mkldnn";
-    }
     VLOG(6) << "Finish Apply";
   }
 };
diff --git a/paddle/fluid/operators/sum_op.cc b/paddle/fluid/operators/sum_op.cc
index ca851b8ee75b1ebd80dcde5c7cf73df81526db6e..fcedf2d24bc0fed08f8a12a14bf01e4a123208e2 100644
--- a/paddle/fluid/operators/sum_op.cc
+++ b/paddle/fluid/operators/sum_op.cc
@@ -334,6 +334,7 @@ class SumGradOpBaseMaker : public imperative::GradOpBaseMakerBase {
         op.SetInput("X", og);
         op.SetOutput("Out", InputGradsType{x_grad});
         op.SetAttr("scale", 1.0f);
+        op.SetDefaultAttrsMap(DefaultAttrsMap());
       }
       return node;
     } else {
diff --git a/paddle/phi/api/yaml/op_compat.yaml b/paddle/phi/api/yaml/op_compat.yaml
index ec268a529adb353e109c6e197224493c059b45ee..fb2723fb25b266004dafa92012b41507130c8b29 100644
--- a/paddle/phi/api/yaml/op_compat.yaml
+++ b/paddle/phi/api/yaml/op_compat.yaml
@@ -3,17 +3,17 @@
   extra :
     attrs : [bool use_cudnn = false, bool use_mkldnn = false]
 
+- op : acosh
+  backward : acosh_grad
+  extra :
+    attrs : [bool use_mkldnn = false, bool use_cudnn = false]
+
 - op : add (elementwise_add)
   backward : add_grad (elementwise_add_grad)
   extra :
     attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
              bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
 
-- op : acosh
-  backward : acosh_grad
-  extra :
-    attrs : [bool use_mkldnn = false, bool use_cudnn = false]
-
 - op : addmm
   backward : addmm_grad
   extra :
@@ -266,6 +266,11 @@
   inputs: {x: X}
   outputs: {out: Out}
 
+- op : floor
+  backward : floor_grad
+  extra :
+    attrs : [bool use_mkldnn = false, bool use_cudnn = false]
+
 - op : floor_divide (elementwise_floordiv)
   extra :
     attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
@@ -283,11 +288,6 @@
     attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
              bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
 
-- op : floor
-  backward : floor_grad
-  extra :
-    attrs : [bool use_mkldnn = false, bool use_cudnn = false]
-
 - op : frobenius_norm
   backward : frobenius_norm_grad
   extra :
@@ -313,17 +313,17 @@
   extra :
     attrs : [bool is_test = false]
 
+- op : hard_swish
+  backward : hard_swish_grad
+  extra :
+    attrs : [bool use_mkldnn = false]
+
 - op : heaviside (elementwise_heaviside)
   backward : heaviside_grad (elementwise_heaviside_grad)
   extra :
     attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
              bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
 
-- op : hard_swish
-  backward : hard_swish_grad
-  extra :
-    attrs : [bool use_mkldnn = false]
-
 - op : inplace_abn
   backward : inplace_abn_grad
   extra :
@@ -404,17 +404,17 @@
     attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
              bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
 
+- op : mish
+  backward : mish_grad
+  extra :
+    attrs : [bool use_mkldnn = false]
+
 - op : multiply (elementwise_mul)
   backward : multiply_grad (elementwise_mul_grad)
   extra :
     attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
              bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
 
-- op : mish
-  backward : mish_grad
-  extra :
-    attrs : [bool use_mkldnn = false]
-
 - op : mv
   inputs :
     {x : X, vec : Vec}
@@ -500,11 +500,6 @@
   extra :
     attrs : [bool use_mkldnn = false]
 
-- op : remainder (elementwise_mod)
-  extra :
-    attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
-             bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
-
 - op : relu
   backward : relu_grad
   extra :
@@ -515,6 +510,11 @@
   extra :
     attrs : [bool use_mkldnn = false]
 
+- op : remainder (elementwise_mod)
+  extra :
+    attrs : [bool use_mkldnn = false, str x_data_format = "", str y_data_format = "", str mkldnn_data_type = "float32",
+             bool use_quantizer = false, float Scale_x = 1.0f, float Scale_y = 1.0f, float Scale_out = 1.0f]
+
 - op : renorm
   backward : renorm_grad
   extra :
@@ -535,6 +535,10 @@
   extra :
     attrs : [bool use_mkldnn = false, bool use_cudnn = false]
 
+- op : scale
+  extra :
+    attrs : [bool use_mkldnn = false]
+
 - op : seed
   extra :
     attrs : [bool deterministic = false, str rng_name = "", bool force_cpu = false]