RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

[ani11452]

Getting this issue when I follow the dependency installation verbatim. Running the model on a NVIDIA L4 with nvcc --version 11.7.

This was solved by downloading torch with cu118:
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118

I am concerned that 11.7 is deprecated for torch.fft usage.

[lukegood]

I had the same problem. Do I need to reinstall cuda11.8 to replace 11.7 when I install pytorch 2.0.0？

[ani11452]

I believe so. I have somewhat forgotten exactly what helped with the fix, but I think there is just a versioning misalignment.
Downloading torch with cu118 (pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118) seemed to solve the problem for me. However, depending on your nvcc version and hardware, it might be different.

[lukegood]

I believe so. I have somewhat forgotten exactly what helped with the fix, but I think there is just a versioning misalignment. Downloading torch with cu118 (pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118) seemed to solve the problem for me. However, depending on your nvcc version and hardware, it might be different.

Thanks. I have solved this problem according to your help.