How to use Intel NPU force ?

[bonihaniboni]

I use onnx with directML to run session, but they choose CPU-DML Automatically.

Is there any way to use my intel NPU force? instead of using CPU or GPU?

import onnxruntime
import numpy as np

# 모델 경로 (로지스틱 회귀 모델)
model_path = "logreg_iris.onnx"

session = onnxruntime.InferenceSession(model_path, providers=['DmlExecutionProvider'])

# 로지스틱 회귀 모델에 맞는 입력 데이터
input_data = np.random.rand(1, 4).astype(np.float32)  # Iris 데이터셋 특성에 맞게 4개 특성

input_name = session.get_inputs()[0].name

input_feed = {input_name: input_data}

output = session.run(None, input_feed)

print(output)

# print(onnxruntime.get_device())

[RyanUnderhill]

Sounds like DML won't use the NPU (as it's not a Direct3D device), but OpenVINO should support Intel NPUs:

https://onnxruntime.ai/docs/execution-providers/OpenVINO-ExecutionProvider.html