What was the grid-resolution of dataset that was used for shape-net car and Ahmed body CFD example in this paper (https://arxiv.org/pdf/2307.15034). I am wokring with torch==2.0.1+cu117' and it looks like cuFFT supports dimensions whose sizes are power of two when computing FFT in half-precision. Does the neuraloperator library uses torch.fft for doing FFT for half-precision input or something else?

Example to reproduce the error:

import torch
x = torch.rand(1,1,20,20,10, dtype=torch.float16).cuda()
xf = torch.fft.rfftn(x, dim=[-3,-2,-1])

Error:

Traceback (most recent call last):
  File "<string>", line 3, in <module>
RuntimeError: cuFFT only supports dimensions whose sizes are powers of two when computing in half precision, but got a signal size of[20, 20, 10]