diff --git a/docs/Coding_Conventions_and_Standards.md b/docs/Coding_Conventions_and_Standards.md
index 9484b3837f..b09408c2df 100644
--- a/docs/Coding_Conventions_and_Standards.md
+++ b/docs/Coding_Conventions_and_Standards.md
@@ -26,7 +26,7 @@ Other
 * Don't use else after return. see: [https://llvm.org/docs/CodingStandards.html#don-t-use-else-after-a-return](https://llvm.org/docs/CodingStandards.html#don-t-use-else-after-a-return)
 * Don't overuse std::shared\_ptr. Use std::shared\_ptr only if it's not clear when and where the object will be deallocated. See also: [https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#Rf-shared_ptr](https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#Rf-shared_ptr)
 * Avoid using the 'long' type, which could be either 32 bits or 64 bits.
-* If there is a legitimate need to allocate objects on the heap, prefer using std::make_unique(). References for the reasoning:
+* If there is a legitimate need to allocate objects on the heap, prefer using onnxruntime::make_unique(). References for the reasoning:
    * https://github.com/isocpp/CppCoreGuidelines/blob/master/CppCoreGuidelines.md#Rh-make_unique
    * https://herbsutter.com/2013/05/29/gotw-89-solution-smart-pointers/
    * https://abseil.io/tips/126
diff --git a/onnxruntime/core/session/onnxruntime_c_api.cc b/onnxruntime/core/session/onnxruntime_c_api.cc
index 5ef0478067..b2625b3e5d 100644
--- a/onnxruntime/core/session/onnxruntime_c_api.cc
+++ b/onnxruntime/core/session/onnxruntime_c_api.cc
@@ -989,7 +989,7 @@ static OrtStatus* OrtCreateValueImplSeqHelperTensor(const Tensor& tensor,
 
 static OrtStatus* OrtCreateValueImplSeqHelper(const OrtValue* const* in, size_t num_values,
                                               OrtValue** out) {
-  auto seq_ptr = std::make_unique<TensorSeq>();
+  auto seq_ptr = onnxruntime::make_unique<TensorSeq>();
   seq_ptr->tensors.resize(num_values);
 
   // use the data type of the first tensor as the data type of the seq
diff --git a/onnxruntime/test/providers/provider_test_utils.h b/onnxruntime/test/providers/provider_test_utils.h
index 6803806730..27c84fcffc 100644
--- a/onnxruntime/test/providers/provider_test_utils.h
+++ b/onnxruntime/test/providers/provider_test_utils.h
@@ -181,7 +181,12 @@ struct SequenceTensorTypeProto : ONNX_NAMESPACE::TypeProto {
 };
 
 template <typename ElemType>
-const SequenceTensorTypeProto<ElemType> s_sequence_tensor_type_proto;
+struct SequenceTensorType {
+  static const SequenceTensorTypeProto<ElemType> s_sequence_tensor_type_proto;
+};
+
+template <typename ElemType>
+const SequenceTensorTypeProto<ElemType> SequenceTensorType<ElemType>::s_sequence_tensor_type_proto;
 
 // To use OpTester:
 //  1. Create one with the op name
@@ -248,7 +253,7 @@ class OpTester {
   void AddSeqInput(const char* name, const SeqTensors<T>& seq_tensors) {
     auto mltype = DataTypeImpl::GetType<TensorSeq>();
     ORT_ENFORCE(mltype != nullptr, "TensorSeq must be a registered cpp type");
-    auto ptr = std::make_unique<TensorSeq>();
+    auto ptr = onnxruntime::make_unique<TensorSeq>();
     auto num_tensors = seq_tensors.tensors.size();
     ptr->tensors.resize(num_tensors);
     for (int i = 0; i < num_tensors; ++i) {
@@ -273,7 +278,8 @@ class OpTester {
     OrtValue value;
     value.Init(ptr.get(), mltype, mltype->GetDeleteFunc());
     ptr.release();
-    input_data_.push_back(Data(NodeArg(name, &s_sequence_tensor_type_proto<T>), std::move(value), optional<float>(), optional<float>()));
+    input_data_.push_back(Data(NodeArg(name, &SequenceTensorType<T>::s_sequence_tensor_type_proto), std::move(value),
+                               optional<float>(), optional<float>()));
   }
 
   template <typename TKey, typename TVal>