RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

[408550969]

When I built the environment in conda as required, the training model reported an error:
====Training Epoch: 0====
Train epoch 0: 0%| | 0/25 [00:07<?, ?it/s]
Traceback (most recent call last):
File "main.py", line 296, in
train_and_test(config, data_loader_dict)
File "main.py", line 67, in train_and_test
model_trainer.train(data_loader_dict)
File "/home/chenlili/RhythmMamba-main/neural_methods/trainer/RhythmMambaTrainer.py", line 71, in train
pred_ppg = self.model(data)
File "/home/chenlili/anaconda3/envs/rppg-toolbox/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/chenlili/anaconda3/envs/rppg-toolbox/lib/python3.8/site-packages/torch/nn/parallel/data_parallel.py", line 169, in forward
return self.module(*inputs[0], **kwargs[0])
File "/home/chenlili/anaconda3/envs/rppg-toolbox/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/chenlili/RhythmMamba-main/neural_methods/model/RhythmMamba.py", line 332, in forward
x = blk(x)
File "/home/chenlili/anaconda3/envs/rppg-toolbox/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/chenlili/RhythmMamba-main/neural_methods/model/RhythmMamba.py", line 215, in forward
x = x + self.drop_path(self.mlp(self.norm2(x)))
File "/home/chenlili/anaconda3/envs/rppg-toolbox/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
return forward_call(*input, **kwargs)
File "/home/chenlili/RhythmMamba-main/neural_methods/model/RhythmMamba.py", line 120, in forward
x_fre = torch.fft.fft(x, dim=1, norm='ortho') # FFT on N dimension
RuntimeError: cuFFT error: CUFFT_INTERNAL_ERROR

[zizheng-guo]

The torch.fft library used in the code seems to temporarily not support RTX 4090. You can use other GPUs or other spectral transformation methods.

[408550969]

Thanks!

[408550969]

把CUDA切到11.8能解决问题，具体可以参考下面这个链接：
https://blog.csdn.net/weixin_44007713/article/details/136475398

注意cuda toolkit安装完毕后需要添加到环境中，
export CPATH=/usr/local/cuda-11.8/targets/x86_64-linux/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/targets/x86_64-linux/lib:$LD_LIBRARY_PATH
export PATH=/usr/local/cuda-11.8/bin:$PATH