Try creating LearningModelDevice

Samples/WinMLSamplesGallery/WinMLSamplesGalleryNative/AdapterList.cpp

@@ -3,6 +3,10 @@
 #include "AdapterList.g.cpp"
 #include <initguid.h>
 #include <dxcore.h>
+#include "d3d12.h"
+#include "winrt/Microsoft.AI.MachineLearning.h"
+#include "Microsoft.AI.MachineLearning.Native.h"
+//#include "Windows.AI.MachineLearning.Native.h"
 
 namespace winrt::WinMLSamplesGalleryNative::implementation
 {
@@ -70,4 +74,51 @@ namespace winrt::WinMLSamplesGalleryNative::implementation
 
         return found_adapter;
     }
+
+    winrt::Microsoft::AI::MachineLearning::LearningModelDevice CreateLearningModelDeviceFromAdapter(winrt::hstring description) {
+        winrt::com_ptr<IDXCoreAdapterFactory> adapterFactory;
+        winrt::check_hresult(::DXCoreCreateAdapterFactory(adapterFactory.put()));
+        winrt::com_ptr<IDXCoreAdapterList> d3D12CoreComputeAdapters;
+        GUID attributes[]{ DXCORE_ADAPTER_ATTRIBUTE_D3D12_CORE_COMPUTE };
+        winrt::check_hresult(
+            adapterFactory->CreateAdapterList(_countof(attributes),
+                attributes,
+                d3D12CoreComputeAdapters.put()));
+
+        const uint32_t count{ d3D12CoreComputeAdapters->GetAdapterCount() };
+        int found_driver = 0;
+        for (uint32_t i = 0; i < count; ++i)
+        {
+            winrt::com_ptr<IDXCoreAdapter> candidateAdapter;
+            winrt::check_hresult(
+                d3D12CoreComputeAdapters->GetAdapter(i, candidateAdapter.put()));
+            CHAR driver_description[128];
+            candidateAdapter->GetProperty(DXCoreAdapterProperty::DriverDescription, sizeof(driver_description), driver_description);
+            hstring hstr_driver_description = to_hstring(driver_description);
+
+            if (description == hstr_driver_description) {
+                found_driver = 1;
+
+                // create D3D12Device
+                com_ptr<IUnknown> spIUnknownAdapter;
+                candidateAdapter->QueryInterface(IID_IUnknown, spIUnknownAdapter.put_void());
+                com_ptr<ID3D12Device> spD3D12Device;
+                D3D12CreateDevice(spIUnknownAdapter.get(), D3D_FEATURE_LEVEL_11_0, _uuidof(ID3D12Device), spD3D12Device.put_void());
+
+                // create D3D12 command queue from device
+                D3D12_COMMAND_QUEUE_DESC queueDesc = {};
+                queueDesc.Flags = D3D12_COMMAND_QUEUE_FLAG_NONE;
+                queueDesc.Type = D3D12_COMMAND_LIST_TYPE_DIRECT;
+                com_ptr<ID3D12CommandQueue> spCommandQueue;
+                spD3D12Device->CreateCommandQueue(&queueDesc, IID_PPV_ARGS(spCommandQueue.put()));
+
+                // create LearningModelDevice from command queue	
+                com_ptr<ILearningModelDeviceFactoryNative> dFactory =
+                    get_activation_factory<winrt::Microsoft::AI::MachineLearning::LearningModelDevice, ILearningModelDeviceFactoryNative>();
+                com_ptr<::IUnknown> spLearningDevice;
+                dFactory->CreateFromD3D12CommandQueue(spCommandQueue.get(), spLearningDevice.put());
+                return spLearningDevice.as<winrt::Microsoft::AI::MachineLearning::LearningModelDevice>();
+            }
+        }
+    }
 }

Samples/WinMLSamplesGallery/WinMLSamplesGalleryNative/AdapterList.h

@@ -8,6 +8,7 @@ namespace winrt::WinMLSamplesGalleryNative::implementation
         AdapterList() = default;
         static winrt::com_array<hstring> GetAdapters();
         static winrt::hstring GetAdapterByDriverDescription(winrt::hstring description);
+        static winrt::Microsoft::AI::MachineLearning::LearningModelDevice CreateLearningModelDeviceFromAdapter(winrt::hstring description);
     };
 }
 namespace winrt::WinMLSamplesGalleryNative::factory_implementation

Samples/WinMLSamplesGallery/WinMLSamplesGalleryNative/WinMLSamplesGalleryNative.idl

@@ -22,5 +22,6 @@ namespace WinMLSamplesGalleryNative
     {
         static String[] GetAdapters();
         static String GetAdapterByDriverDescription(String description);
+        static Microsoft.AI.MachineLearning.LearningModelDevice CreateLearningModelDeviceFromAdapter(String description);
     }
 }