Stuck after 'Setting up PyTorch plugin "bias_act_plugin"'

[festinais]

Hi, I'm trying to train StyleGAN2-D2DCE-ADA.yaml . All of a sudden it got stuck at this step: 'Setting up PyTorch plugin "bias_act_plugin"', as you can see in the screenshot:

I left for almost an hour and it was stuck there. Any ideas why this might happen?

This is the command I'm using to train:
CUDA_VISIBLE_DEVICES=0,1 python3 src/main.py -cfg src/configs/CIFAR10/StyleGAN2-D2DCE-ADA.yaml -t -metrics fid -save stylegan_curated_overlap_d2d_ada -data data/CUSTOM -ref "test" -mpc

torch version: 1.13.0.dev20220825+cu116
cuda: 11.6
Also, pulled the latest updates.

[mingukkang]

Please execute following commands:

export LIBRARY_PATH=/opt/conda/lib/

conda deactivate

conda activate

if it does not work, I suggest removing PyTorch 1.13.0.dev20220825+cu116 and reinstalling PyTorch 1.12 with cu113.

Thanks.

[festinais]

Thank you for your quick answer! Which python version do you recommend also?

[festinais]

It works now. I have CUDA 11.6, python 3.9, and torch 1.12+cu116

For CUDA 11.6 there is only torch with version 1.12. This is how I installed torch: pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116
You can take a look more at this link https://download.pytorch.org/whl/cu116.

Also, I had many problems with upfirdn2d_plugin and bias_act_plugin plugins, till I made it work. What helped me was:

1. Create a new venv with conda
2. Make sure to clean everything with: pip freeze | xargs pip uninstall -y
3. Also, make sure to clean pip cache with: pip cache purge
4. Check inside this link https://download.pytorch.org/whl/torch/ for the torch that is compatible with your CUDA version. For example torch-1.12.0+cu116-cp310-cp310-linux_x86_64.whl is compatible with CUDA 11.6 and python 3.10.

These steps helped me solve the bias_act_plugin and upfirdn2d_plugin problems.