diff --git a/paddle/fluid/operators/batch_norm_op.cc b/paddle/fluid/operators/batch_norm_op.cc
index 3467658e894d547d7a0106d6414624b10351d6f3..bc5bd118dbec4791e79790d304e5aa8729304976 100644
--- a/paddle/fluid/operators/batch_norm_op.cc
+++ b/paddle/fluid/operators/batch_norm_op.cc
@@ -257,13 +257,16 @@ void BatchNormOpMaker::Make() {
   AddOutput("ReserveSpace",
             "Reserve GPU space for triggering the new semi-persistent "
             "NHWC kernel")
-      .AsDispensable();
+      .AsDispensable()
+      .AsExtra();
   AddAttr<bool>("use_mkldnn",
                 "(bool, default false) Only used in mkldnn kernel")
-      .SetDefault(false);
+      .SetDefault(false)
+      .AsExtra();
   AddAttr<bool>("fuse_with_relu",
                 "(bool, default false) Only used in mkldnn kernel")
-      .SetDefault(false);
+      .SetDefault(false)
+      .AsExtra();
   AddAttr<bool>("use_global_stats",
                 "(bool, default false) Whether to use global mean and "
                 "variance. In inference or test mode, set use_global_stats "
diff --git a/paddle/fluid/operators/instance_norm_op.cc b/paddle/fluid/operators/instance_norm_op.cc
index 0a850400686c4949f6cda83f0e386d3c51d323f9..cfdaacf8cb6ee7dc958769322c3c03b44b921662 100644
--- a/paddle/fluid/operators/instance_norm_op.cc
+++ b/paddle/fluid/operators/instance_norm_op.cc
@@ -149,11 +149,13 @@ void InstanceNormOpMaker::Make() {
   AddOutput("SavedMean",
             "Mean of the current mini batch, "
             "will apply to output when training")
-      .AsIntermediate();
+      .AsIntermediate()
+      .AsExtra();
   AddOutput("SavedVariance",
             "Variance of the current mini batch, "
             "will apply to output when training")
-      .AsIntermediate();
+      .AsIntermediate()
+      .AsExtra();
   AddComment(R"DOC(
 Instance Normalization.
 
diff --git a/paddle/fluid/operators/mkldnn/batch_norm_mkldnn_op.cc b/paddle/fluid/operators/mkldnn/batch_norm_mkldnn_op.cc
index eb241b9157fecd8615797e276ab53f9f6812e21d..bd910e30672a47bab1452d59cef458d80cfcf56d 100644
--- a/paddle/fluid/operators/mkldnn/batch_norm_mkldnn_op.cc
+++ b/paddle/fluid/operators/mkldnn/batch_norm_mkldnn_op.cc
@@ -47,7 +47,9 @@ class BatchNormMKLDNNHandler : public platform::MKLDNNHandlerNoCachingT<
                                           mkldnn::batch_normalization_backward>(
             mkldnn_engine, ctx.GetPlace()) {
     const float epsilon = ctx.Attr<float>("epsilon");
-    const bool fuse_with_relu = ctx.Attr<bool>("fuse_with_relu");
+    const bool fuse_with_relu = ctx.HasAttr("fuse_with_relu")
+                                    ? ctx.Attr<bool>("fuse_with_relu")
+                                    : false;
 
     std::vector<std::string> DataLayout_error_msg = {"kNHWC", "kNCHW",
                                                      "kAnyLayout", "kMKLDNN"};
diff --git a/paddle/fluid/operators/squeeze_op.cc b/paddle/fluid/operators/squeeze_op.cc
index e13245797cff3e348666c21b89954a8eb61a28c4..cf24faf25db70fe931e110be451196fa81120e98 100644
--- a/paddle/fluid/operators/squeeze_op.cc
+++ b/paddle/fluid/operators/squeeze_op.cc
@@ -225,7 +225,7 @@ class Squeeze2Op : public framework::OperatorWithKernel {
       ctx->ShareLoD("X", "Out");
     }
 
-    OP_INOUT_CHECK(ctx->HasOutput("XShape"), "Output", "XShape", "Squeeze2");
+    if (!ctx->HasOutput("XShape")) return;
 
     std::vector<int64_t> xshape_dims(x_dims.size() + 1);
     xshape_dims[0] = 0;
@@ -323,7 +323,8 @@ class Squeeze2OpMaker : public SqueezeOpMaker {
     AddOutput("XShape",
               "XShape is just used to store the shape and lod of X, which will "
               "be used in SqueezeGradOp.")
-        .AsIntermediate();
+        .AsIntermediate()
+        .AsExtra();
   }
 };