Pytorch bilinear interpolate wrong result

[NicerWang]

Problem

in_cpu = torch.rand((1, 32, 512, 512), device='cpu', dtype=torch.float32)
in_dml = torch.tensor(in_cpu, device=torch_directml.device())

out_cpu = F.interpolate(in_cpu, scale_factor=2, mode='bilinear')
out_dml = F.interpolate(in_dml, scale_factor=2, mode='bilinear')

out_cpu and out_dml is not the same, and have a great difference:

The problem may relate to operator upsample_bilinear2d.

Versions

torch 2.0.0
torch-directml 0.2.0.dev230426
torchvision 0.15.1

[lyleaf-81]

I have the same problem

[Adele101]

Hello, thank you for submitting this issue. While I can't provide a timeline for resolution as the moment, please know that your feedback is valuable to us. We will follow up once we can review this issue.

[AlexAwesome88]

Any updates on this? I'm also trying to get GPFPGAN running on directML and this issue is corrupting the output :(