Segmentation fault (core dumped)

[anvenle]

Thanks for your great work！When I execute train.py, I get an error Segmentation fault (core dumped).I tried to reinstall the virtual environment, but the same error message appeared：

my command python train.py -s /path/to/files, and error:

/home/lee/miniconda3/envs/ever2/lib/python3.10/site-packages/torch/utils/cpp_extension.py:1965: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation.
If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST'].
warnings.warn(
Segmentation fault (core dumped)

[N-Demir]

You've probably already figured this out but TORCH_CUDA_ARCH_LIST does need to be set to correspond with your system's GPU cuda architecture. For example for a T4 GPU the env var is export TORCH_CUDA_ARCH_LIST=7.5