How to run NPU on OpenVINO EP?

[fan2tamo]

When I want to run OpenVINO EP on a GPU by C#, I specify:

SessionOptions so = new SessionOptions();
so.AppendExecutionProvider_OpenVINO("GPU_FP16"); // or GPU_FP32

But how should I specify it for an NPU?
Given the following excerpt from the onnxruntime_c_api.h, it is unclear whether NPU is supported:

/** \brief OpenVINO Provider Options
 *
 * \see OrtApi::SessionOptionsAppendExecutionProvider_OpenVINO
 */
typedef struct OrtOpenVINOProviderOptions {
#ifdef __cplusplus
  OrtOpenVINOProviderOptions() : device_type{},
                                 enable_npu_fast_compile{},
                                 device_id{},
                                 num_of_threads{},
                                 cache_dir{},
                                 context{},
                                 enable_opencl_throttling{},
                                 enable_dynamic_shapes{} {}
#endif
  /** \brief Device type string
   *
   * Valid settings are one of: "CPU_FP32", "CPU_FP16", "GPU_FP32", "GPU_FP16"
   */
  const char* device_type;
  unsigned char enable_npu_fast_compile;  ///< 0 = disabled, nonzero = enabled
  const char* device_id;
  size_t num_of_threads;  ///< 0 = Use default number of threads
  const char* cache_dir;  // path is set to empty by default
  void* context;
  unsigned char enable_opencl_throttling;  ///< 0 = disabled, nonzero = enabled
  unsigned char enable_dynamic_shapes;     ///< 0 = disabled, nonzero = enabled
} OrtOpenVINOProviderOptions;

[hariharans29]

Tagging @jywu-msft and @sfatimar for the OpenVINO question

[fan2tamo]

With the latest OnnxRuntime (v1.18.0), I have confirmed that OpenVINO EP (C#) can be used to reason on Intel NPU.