How to use upscaling models on GPU when using ONNX + Olive? Upscaling uses CPU instead

[FauxPrada]

Hello! Thank you Ishqqytiger and contributors for this implementation, it is the fastest and most stable on my card!!

I am using ONNX pipeline with OLive optimisations to generate images very fast (32 it/s on 7900 XTX).

But, when I use any upscaling in the "Extra" tab it uses the CPU instead of GPU.

This make sense because when using ONNX we don't use DirectML... but is it possible to use ONNX for txt2image, but DirectML for 'Extras' like upscaling and Face Restoration?

Or can the upscaling methods like DAT and LDSR be ran using ONNX as well?

Thank you for any help!