Add DirectML support

[Looong01]

You can create a new branch to save this.

[Looong01]

@microsoft-github-policy-service agree

[ezioruan]

Awesome. Really need this

[ezioruan]

you can refractor these duplicate lines into a function

def get_touch_dtype(device):
    if isinstance(device, str):
            torch_dtype = torch.float16 if 'cuda' in device else torch.float32
        else:
            torch_dtype = torch.float32
        return torch_dtype

[ezioruan]

Thank you for adding this amazing feature. There is some high code repetition inside, it would be even better if it could be addressed.

[Looong01]

Thank you for adding this amazing feature. There is some high code repetition inside, it would be even better if it could be addressed.

Revised.

[Looong01]

Thank you for adding this amazing feature. There is some high code repetition inside, it would be even better if it could be addressed.

Hey, would you like to merge these codes? I have revised my codes under your criterion.

[ezioruan]

Thank you for adding this amazing feature. There is some high code repetition inside, it would be even better if it could be addressed.

Hey, would you like to merge these codes? I have revised my codes under your criterion.

I don't have the permission to do that, maybe ask the project owner

[Looong01]

@shengming-yin @BinayakJha @chenfei-wu Could you please merge this branch? This can help the users run it on Windows with AMD & Intel GPUs.

[Looong01]

@Wang-Xiaodong1899 @CODINNLG @rupeshs @bojieli @heysujal @WeizhenQ Could you please merge this branch? This can help the users run it on Windows with AMD & Intel GPUs.

[heysujal]

@Wang-Xiaodong1899 @CODINNLG @rupeshs @bojieli @heysujal @WeizhenQ Could you please merge this branch? This can help the users run it on Windows with AMD & Intel GPUs.

Hi , sorry but I am not the maintainer of the project