simply change __init__.py, add onnxruntime with dml support.

[FNsi]

switch Non-NVIDIA onnx version from CPU to directml version to support whatever GPU in windows,

which could handle CPU inference too.

[zelenooki87]

Just wanted to test directml performance versus CUDA.
Have Nvidia GPU. Unfortunately, for me, Directml is not listed. I have built chainner by following your commit and npm.

[FNsi]

Just wanted to test directml performance versus CUDA.

Have Nvidia GPU. Unfortunately, for me, Directml is not listed. I have built chainner by following your commit and npm.

Nice! The reason for dml is unlimited vram, throw a 4K picture without tiling in iGPU!

Update:
😅To clarify, try onnx-dml only needed with onnxruntime-directml installed because current script will automatically looking for "provider"

That means, you don't need to build from my source,

just using system python, simply run

pip install onnxruntime-directml

Could enable the ability to function dml.

[zelenooki87]

Script working okay. No need to modify. Directml working, this is super helpful for folks with AMD cards.

[FNsi]

Script working okay. No need to modify. Directml working, this is super helpful for folks with AMD cards.

That would be fine.

No much clue since my last nv burned years ago🤣