diff --git a/paddle/fluid/framework/operator.cc b/paddle/fluid/framework/operator.cc
index 040f669f2b..1723a9a78a 100644
--- a/paddle/fluid/framework/operator.cc
+++ b/paddle/fluid/framework/operator.cc
@@ -884,6 +884,8 @@ void OperatorWithKernel::RunImpl(const Scope& scope,
   // result of HasAttr.
   if (!enable_cache_runtime_context && HasAttr(kEnableCacheRuntimeContext))
     enable_cache_runtime_context = true;
+  if (!enable_cache_expected_kernel && HasAttr(kEnableCacheExpectedKernel))
+    enable_cache_expected_kernel = true;
   if (!all_kernels_must_compute_runtime_shape &&
       HasAttr(kAllKernelsMustComputeRuntimeShape))
     all_kernels_must_compute_runtime_shape = true;
@@ -906,7 +908,7 @@ void OperatorWithKernel::RunImpl(const Scope& scope,
   platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
   auto* dev_ctx = pool.Get(place);
 
-  if (!HasAttr(kEnableCacheExpectedKernel) || !kernel_type_) {
+  if (!enable_cache_expected_kernel || !kernel_type_) {
     ChooseKernel(*runtime_ctx, scope, place);
   }
 
diff --git a/paddle/fluid/framework/operator.h b/paddle/fluid/framework/operator.h
index 2e733badb8..489b660996 100644
--- a/paddle/fluid/framework/operator.h
+++ b/paddle/fluid/framework/operator.h
@@ -507,6 +507,7 @@ class OperatorWithKernel : public OperatorBase {
   mutable std::unique_ptr<RuntimeContext> runtime_ctx_;
   mutable const Scope* pre_scope_ = nullptr;
   mutable bool enable_cache_runtime_context = false;
+  mutable bool enable_cache_expected_kernel = false;
   mutable bool all_kernels_must_compute_runtime_shape = false;
 };