diff --git a/cmake/winml_unittests.cmake b/cmake/winml_unittests.cmake
index 9f99adbf43..4cbbb792f3 100644
--- a/cmake/winml_unittests.cmake
+++ b/cmake/winml_unittests.cmake
@@ -152,6 +152,7 @@ file(GLOB winml_test_common_src CONFIGURE_DEPENDS
     "${WINML_TEST_SRC_DIR}/common/*.h"
     "${WINML_TEST_SRC_DIR}/common/*.cpp")
 add_library(winml_test_common STATIC ${winml_test_common_src})
+target_compile_options(winml_test_common PRIVATE /wd5205)  # workaround cppwinrt SDK bug https://github.com/microsoft/cppwinrt/issues/584
 add_dependencies(winml_test_common
   onnx
   winml_api
diff --git a/onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.cpp b/onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.cpp
index 9698bbcbf0..d21daa36c3 100644
--- a/onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.cpp
+++ b/onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.cpp
@@ -742,8 +742,8 @@ namespace OperatorHelper
         // Determine the number of output dimensions.
         ML_CHECK_VALID_ARGUMENT(inputDimensions.size() >= 1);
         ML_CHECK_VALID_ARGUMENT(indicesDimensions.size() >= 1);
-        ML_CHECK_VALID_ARGUMENT(inputDimensions.size() > batchCount);
-        ML_CHECK_VALID_ARGUMENT(indicesDimensions.size() > batchCount);
+        ML_CHECK_VALID_ARGUMENT(static_cast<int64_t>(inputDimensions.size()) > static_cast<int64_t>(batchCount));
+        ML_CHECK_VALID_ARGUMENT(static_cast<int64_t>(indicesDimensions.size()) > static_cast<int64_t>(batchCount));
         const uint32_t numberOfCoordinatesPerIndex = indicesDimensions.back();
         ML_CHECK_VALID_ARGUMENT(inputDimensions.size() >= batchCount + numberOfCoordinatesPerIndex);
         const uint32_t numberOfOutputDimensionsFromInput = static_cast<uint32_t>(inputDimensions.size()) - batchCount - numberOfCoordinatesPerIndex;
diff --git a/tools/ci_build/build.py b/tools/ci_build/build.py
index 735a2489db..077fb3fc6f 100755
--- a/tools/ci_build/build.py
+++ b/tools/ci_build/build.py
@@ -582,7 +582,7 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
             "OFF" if args.skip_winml_tests else "ON"),
         "-Donnxruntime_GENERATE_TEST_REPORTS=ON",
         "-Donnxruntime_DEV_MODE=" + (
-            "OFF" if args.use_acl or args.use_armnn or args.use_winml or
+            "OFF" if args.use_acl or args.use_armnn or
             (args.ios and is_macOS()) else "ON"),
         "-DPYTHON_EXECUTABLE=" + sys.executable,
         "-Donnxruntime_USE_CUDA=" + ("ON" if args.use_cuda else "OFF"),
diff --git a/winml/lib/Api.Image/NominalRangeConverter.cpp b/winml/lib/Api.Image/NominalRangeConverter.cpp
index f9c8d4bf9a..83cd705f97 100644
--- a/winml/lib/Api.Image/NominalRangeConverter.cpp
+++ b/winml/lib/Api.Image/NominalRangeConverter.cpp
@@ -27,12 +27,12 @@ namespace _winml {
   }
 
   DirectX::PackedVector::HALF NominalRangeConverter::Normalize(DirectX::PackedVector::HALF val) const {
-    return val / scale - shift;
+    return static_cast<DirectX::PackedVector::HALF>(val / scale - shift);
   }
 
 #if defined(_M_AMD64) || defined(_M_IX86)
   __m128 NominalRangeConverter::Normalize(__m128 sse_data) const {
-    __m128 sse_shift = _mm_set1_ps(shift);
+    __m128 sse_shift = _mm_set1_ps(static_cast<float>(shift));
     __m128 sse_scale = _mm_set1_ps(scale);
 
     auto sse_dived = _mm_div_ps(sse_data, sse_scale);
@@ -48,16 +48,16 @@ namespace _winml {
   }
 
   DirectX::PackedVector::HALF NominalRangeConverter::Denormalize(DirectX::PackedVector::HALF val) const {
-    return scale * (val + shift);
+    return static_cast<DirectX::PackedVector::HALF>(scale * (val + shift));
   }
 
 #if defined(_M_AMD64) || defined(_M_IX86)
   __m128 NominalRangeConverter::Denormalize(__m128 sse_data) const {
-    __m128 sse_shift = _mm_set1_ps(shift);
+    __m128 sse_shift = _mm_set1_ps(static_cast<float>(shift));
     __m128 sse_scale = _mm_set1_ps(scale);
 
     auto sse_added = _mm_add_ps(sse_data, sse_shift);
     return _mm_mul_ps(sse_added, sse_scale);
   }
 #endif
-} // namespace _winml
\ No newline at end of file
+} // namespace _winml
diff --git a/winml/test/adapter/AdapterSessionTest.cpp b/winml/test/adapter/AdapterSessionTest.cpp
index c81b1544ae..f76da0d51d 100644
--- a/winml/test/adapter/AdapterSessionTest.cpp
+++ b/winml/test/adapter/AdapterSessionTest.cpp
@@ -223,7 +223,7 @@ void CopyInputAcrossDevices() {
   constexpr size_t input_tensor_size = [&dimensions]() {
     size_t size = 1;
     for (auto dim : dimensions)
-      size *= dim;
+      size *= static_cast<size_t>(dim);
     return size;
   } ();
 
@@ -259,7 +259,7 @@ void CopyInputAcrossDevices_DML() {
   constexpr size_t input_tensor_size = [&dimensions]() {
     size_t size = 1;
     for (auto dim : dimensions)
-      size *= dim;
+      size *= static_cast<size_t>(dim);
     return size;
   } ();
 
diff --git a/winml/test/adapter/adapter_test.cpp b/winml/test/adapter/adapter_test.cpp
index d20aee70a7..81e1f53c4a 100644
--- a/winml/test/adapter/adapter_test.cpp
+++ b/winml/test/adapter/adapter_test.cpp
@@ -105,13 +105,13 @@ static void ModelGetVersion() {
 static void ModelGetInputCount() {
   size_t input_count;
   winml_adapter_api->ModelGetInputCount(squeezenet_model, &input_count);
-  WINML_EXPECT_EQUAL(input_count, 1);
+  WINML_EXPECT_EQUAL(input_count, 1u);
 }
 
 static void ModelGetOutputCount() {
   size_t output_count;
   winml_adapter_api->ModelGetOutputCount(squeezenet_model, &output_count);
-  WINML_EXPECT_EQUAL(output_count, 1);
+  WINML_EXPECT_EQUAL(output_count, 1u);
 }
 
 static void ModelGetInputName() {
@@ -163,7 +163,7 @@ static void ModelGetInputTypeInfo() {
 
   size_t dim_count;
   ort_api->GetDimensionsCount(tensor_info, &dim_count);
-  WINML_EXPECT_EQUAL(dim_count, 4);
+  WINML_EXPECT_EQUAL(dim_count, 4u);
 
   int64_t dim_values[4]; 
   ort_api->GetDimensions(tensor_info, dim_values, 4);
@@ -192,7 +192,7 @@ static void ModelGetOutputTypeInfo() {
 
   size_t dim_count;
   ort_api->GetDimensionsCount(tensor_info, &dim_count);
-  WINML_EXPECT_EQUAL(dim_count, 4);
+  WINML_EXPECT_EQUAL(dim_count, 4u);
 
   int64_t dim_values[4];
   ort_api->GetDimensions(tensor_info, dim_values, 4);
@@ -207,7 +207,7 @@ static void ModelGetOutputTypeInfo() {
 static void ModelGetMetadataCount() {
   size_t metadata_count;
   winml_adapter_api->ModelGetMetadataCount(metadata_model, &metadata_count);
-  WINML_EXPECT_EQUAL(metadata_count, 2);
+  WINML_EXPECT_EQUAL(metadata_count, 2u);
 }
 
 static void ModelGetMetadata() {
@@ -218,15 +218,15 @@ static void ModelGetMetadata() {
 
   winml_adapter_api->ModelGetMetadata(metadata_model, 0, &metadata_key, &metadata_key_len, &metadata_value, &metadata_value_len);
   WINML_EXPECT_EQUAL(std::string(metadata_key), "thisisalongkey");
-  WINML_EXPECT_EQUAL(metadata_key_len, 14);
+  WINML_EXPECT_EQUAL(metadata_key_len, 14u);
   WINML_EXPECT_EQUAL(std::string(metadata_value), "thisisalongvalue");
-  WINML_EXPECT_EQUAL(metadata_value_len, 16);
+  WINML_EXPECT_EQUAL(metadata_value_len, 16u);
 
   winml_adapter_api->ModelGetMetadata(metadata_model, 1, &metadata_key, &metadata_key_len, &metadata_value, &metadata_value_len);
   WINML_EXPECT_EQUAL(std::string(metadata_key), "key2");
-  WINML_EXPECT_EQUAL(metadata_key_len, 4);
+  WINML_EXPECT_EQUAL(metadata_key_len, 4u);
   WINML_EXPECT_EQUAL(std::string(metadata_value), "val2");
-  WINML_EXPECT_EQUAL(metadata_value_len, 4);
+  WINML_EXPECT_EQUAL(metadata_value_len, 4u);
 }
 
 static void ModelEnsureNoFloat16() {
@@ -328,4 +328,4 @@ const AdapterTestApi& getapi() {
           EnvConfigureCustomLoggerAndProfiler,
           };
   return api;
-}
\ No newline at end of file
+}
diff --git a/winml/test/api/LearningModelAPITest.cpp b/winml/test/api/LearningModelAPITest.cpp
index a9ddeb07bd..44e07974c0 100644
--- a/winml/test/api/LearningModelAPITest.cpp
+++ b/winml/test/api/LearningModelAPITest.cpp
@@ -24,7 +24,7 @@ static void CreateModelFromFilePath() {
 
 static void CreateModelFromUnicodeFilePath() {
   LearningModel learningModel = nullptr;
-  WINML_EXPECT_NO_THROW(APITest::LoadModel(L"UnicodePath\\こんにちは maçã\\foo.onnx", learningModel));
+  WINML_EXPECT_NO_THROW(APITest::LoadModel(L"UnicodePath\\\u3053\u3093\u306B\u3061\u306F maçã\\foo.onnx", learningModel));
 }
 
 static void CreateModelFileNotFound() {
@@ -45,7 +45,6 @@ static void CreateCorruptModel() {
     APITest::LoadModel(L"corrupt-model.onnx", learningModel),
     winrt::hresult_error,
     [](const winrt::hresult_error& e) -> bool {
-        auto f = __HRESULT_FROM_WIN32(e.code());
           return e.code() == __HRESULT_FROM_WIN32(ERROR_FILE_CORRUPT);
     });
 }
@@ -309,4 +308,4 @@ const LearningModelApiTestsApi& getapi() {
     api.CloseModelCheckEval = SkipTest;
   }
   return api;
-}
\ No newline at end of file
+}
diff --git a/winml/test/api/LearningModelSessionAPITest.cpp b/winml/test/api/LearningModelSessionAPITest.cpp
index b314fede2c..9c08476492 100644
--- a/winml/test/api/LearningModelSessionAPITest.cpp
+++ b/winml/test/api/LearningModelSessionAPITest.cpp
@@ -320,7 +320,8 @@ static void NamedDimensionOverride()
   WINML_EXPECT_NO_THROW(device = LearningModelDevice(LearningModelDeviceKind::Cpu));
 
   // the model input shape. the batch size, n, is overriden to 5
-  int64_t n = 5, c = 3, h = 720, w = 720;
+  uint32_t n = 5;
+  int64_t c = 3, h = 720, w = 720;
 
   LearningModelSessionOptions options;
   options.OverrideNamedDimension(L"None", n);
diff --git a/winml/test/api/RawApiTests.cpp b/winml/test/api/RawApiTests.cpp
index 2308641898..bb00b467ec 100644
--- a/winml/test/api/RawApiTests.cpp
+++ b/winml/test/api/RawApiTests.cpp
@@ -17,7 +17,7 @@ auto CreateModelAsBuffer(const wchar_t* model_path)
     std::streamsize size = input_stream.tellg();
     input_stream.seekg(0, std::ios::beg);
 
-    std::vector<char> buffer(size);
+    std::vector<char> buffer(static_cast<std::vector<char>::size_type>(size));
     input_stream.read(buffer.data(), size);
 
     return std::make_pair(buffer, size);
@@ -68,12 +68,10 @@ static void EvaluateNoInputCopy() {
 static void EvaluateFromModelFromBuffer() {
   std::wstring model_path = L"model.onnx";
 
-  size_t size;
-  std::vector<char> buffer;
-  std::tie(buffer, size) = CreateModelAsBuffer(model_path.c_str());
+  auto [buffer, size] = CreateModelAsBuffer(model_path.c_str());
 
   std::unique_ptr<ml::learning_model> model = nullptr;
-  WINML_EXPECT_NO_THROW(model = std::make_unique<ml::learning_model>(buffer.data(), size));
+  WINML_EXPECT_NO_THROW(model = std::make_unique<ml::learning_model>(buffer.data(), static_cast<size_t>(size)));
 
   std::unique_ptr<ml::learning_model_device> device = nullptr;
   WINML_EXPECT_NO_THROW(device = std::make_unique<ml::learning_model_device>());
diff --git a/winml/test/common/googleTestMacros.h b/winml/test/common/googleTestMacros.h
index 7b17249b49..79a7f51ce5 100644
--- a/winml/test/common/googleTestMacros.h
+++ b/winml/test/common/googleTestMacros.h
@@ -82,6 +82,9 @@
 #define WINML_EXPECT_HRESULT_FAILED(hresult_expression) EXPECT_HRESULT_FAILED(hresult_expression)
 #define WINML_EXPECT_THROW_SPECIFIC(statement, exception, condition) EXPECT_THROW_SPECIFIC(statement, exception, condition)
 
+#pragma warning(push)
+#pragma warning(disable:4505)  // unreferenced local function has been removed
+
 static bool RuntimeParameterExists(std::wstring param) 
 {
   std::string narrowParam = std::wstring_convert<std::codecvt_utf8<wchar_t>>().to_bytes(param);
@@ -89,7 +92,7 @@ static bool RuntimeParameterExists(std::wstring param)
   return no_gpu_tests != RuntimeParameters::Parameters.end() && no_gpu_tests->second != "0";
 }
 
-static bool SkipGpuTests() 
+static bool SkipGpuTests()
 {
 #ifndef USE_DML
   return true;
@@ -110,3 +113,5 @@ static bool SkipTestsImpactedByOpenMP() {
   return false;
 #endif
 }
+
+#pragma warning(pop)
diff --git a/winml/test/common/protobufHelpers.cpp b/winml/test/common/protobufHelpers.cpp
index f84c9042c7..1bc2817820 100644
--- a/winml/test/common/protobufHelpers.cpp
+++ b/winml/test/common/protobufHelpers.cpp
@@ -83,7 +83,10 @@ std::vector<int64_t> GetTypeSpecificDataFromTensorProto(
 template <>
 std::vector<uint8_t> GetTypeSpecificDataFromTensorProto(
     onnx::TensorProto tensorProto) {
+#pragma warning(push)
+#pragma warning(disable:4244)  // conversion with possible loss of data
   return std::vector<uint8_t>(std::begin(tensorProto.int32_data()), std::end(tensorProto.int32_data()));
+#pragma warning(pop)
 }
 template <>
 std::vector<double> GetTypeSpecificDataFromTensorProto(
diff --git a/winml/test/common/test.h b/winml/test/common/test.h
index 1a8380da2b..3e8e023c0b 100644
--- a/winml/test/common/test.h
+++ b/winml/test/common/test.h
@@ -9,6 +9,9 @@ using TeardownClass = VoidTest;
 using SetupTest = VoidTest;
 using TeardownTest = VoidTest;
 
+#pragma warning(push)
+#pragma warning(disable:4505)  // unreferenced local function has been removed
+
 constexpr bool alwaysTrue() {
     return true;
 }
@@ -27,3 +30,5 @@ constexpr bool alwaysFalse() {
 static void SkipTest() {
   WINML_SKIP_TEST("");
 }
+
+#pragma warning(pop)
diff --git a/winml/test/image/imageTestHelper.cpp b/winml/test/image/imageTestHelper.cpp
index fd06f3380a..1f769eeb46 100644
--- a/winml/test/image/imageTestHelper.cpp
+++ b/winml/test/image/imageTestHelper.cpp
@@ -201,8 +201,8 @@ namespace ImageTestHelper {
         // Copy from Cpu to GPU
         D3D12_SUBRESOURCE_DATA CPUData = {};
         CPUData.pData = reinterpret_cast<BYTE*>(pCPUTensor);
-        CPUData.RowPitch = bufferbytesize;
-        CPUData.SlicePitch = bufferbytesize;
+        CPUData.RowPitch = static_cast<LONG_PTR>(bufferbytesize);
+        CPUData.SlicePitch = static_cast<LONG_PTR>(bufferbytesize);
         UpdateSubresources(cmdList.get(), pGPUResource.get(), imageUploadHeap.get(), 0, 0, 1, &CPUData);
 
         // Close the command list and execute it to begin the initial GPU setup.