Is it possible for python to use OrtSessionOptionsAppendExecutionProvider_DML function?

[JiaPai12138]

I can set it manually through graphic settings, but how to do it in program?

[skottmckay]

You would need a build that included DML. You can check the output from get_available_providers to make sure the DML execution provider is listed.

import onnxruntime as ort
# include all execution providers. alternatively you could use providers=['DmlExecutionProvider']
sess = ort.InferenceSession('model.onnx', providers=ort.get_available_providers())
``

[JiaPai12138]

You would need a build that included DML. You can check the output from get_available_providers to make sure the DML execution provider is listed.

import onnxruntime as ort
# include all execution providers. alternatively you could use providers=['DmlExecutionProvider']
sess = ort.InferenceSession('model.onnx', providers=ort.get_available_providers())
``

I know and I did. And i found by default it will run on my nvidia device. The problem is that i also need to use dxgi screenshot, which must be run on my Intel UHD graphic card. Seems like i cannot handle both at the same time

[skottmckay]

I believe DML has a device_id option. This may help:

providers = [
    ('DmlExecutionProvider', {
        'device_id': 0  # default is 0 so you may need to set this to 1 to run on the other graphics card
    }),
    'CPUExecutionProvider',
]

sess = ort.InferenceSession('model.onnx', providers=providers)

[JiaPai12138]

Thanks very much! That works well!