Support Cuda from pip out of the box

facebookresearch / pytorch3d

PyTorch3D is FAIR's library of reusable components for deep learning with 3D data

https://pytorch3d.org/

Other

8.81k stars 1.32k forks source link

Support Cuda from pip out of the box #1640

Closed kodecreer closed 1 year ago

kodecreer commented 1 year ago

🚀 Feature

Pytorch normal is able to work perfectly fine with no issues with Cuda if the drivers are properly installed. However this is not the case with pytorch3d. It's very not flexible with the versioning of cuda versions.

Motivation

Pytorch is super easy to install. Pytorch shouldn't be so hard to install when

Pitch

Make it so I can just install it working perfectly fine without needing a FORCE_CUDA environment variable and not cause a bunch of issues when I pip to this git repo.

kodecreer commented 1 year ago

After some time figured out something consistent that should work for using pytorch3d with Anaconda. Download Cuda toolkit 11.8 for Anaconda(or any other version that is supported by mature pytorch) Download the GCC version compatible with the cuda version downloaded. Set CUDA_HOME to $CONDA_PREFIX Follow the rest of the instructions with building from source and it should build ok.