Problem with conversion when input tensor is not (518, 518)

[pzoltowski]

Hi thanks for this repo. I tried to change input tensor height to smaller (in a hope to speedup inference), however even if height is multiple of 14 the script failed to convert:

I tried both:

input_tensor = torch.randn(1, 3, 210, 518).to(device)

input_tensor = torch.randn(1, 3, 518, 210).to(device)

In both cases getting:

NotImplementedError: The operator 'aten::upsample_bicubic2d.out' is not currently implemented for the MPS device.

I guess patching somehow needs to be modified? Would be thankful for any tips regarding that.