cuFFT error when training with precision 16

[sau-arv-gul]

I am trying to train the model on custom dataset on my Windows 11 system & the code runs fine when precision is 32 but I am getting CUDA out of memory with precision 32. So due to limited GPU resource I set the precision to 16.
When I train with the precision 16, I am getting cuFFT error.

File "C:\Users\user\Desktop\lama_inpainting\lama\saicinpainting\training\modules\ffc.py", line 86, in forward
    ffted = torch.fft.rfftn(x, dim=fft_dim, norm=self.fft_norm)
RuntimeError: cuFFT only supports dimensions whose sizes are powers of two when computing in half precision, but got a signal size of[528, 422]