Will a model trained using the CUDA 10.1 version of pytorch work properly on a pytorch version using CUDA 11?

Basically what the title asks. Is this guaranteed to work and is this documented somewhere? Also, will the pytorch cuda 10.1 version work on a system with cuda 11 installed?

PyTorch conda binaries and pip wheels ship with their own CUDA (, cudnn, NCCL, etc.) runtime and your local CUDA toolkit will not be used unless you build PyTorch from source or any custom CUDA extension.

Yes, CUDA11 should work, if it supports your GPU architecture. If not, please create a topic here or an issue on GitHub and we’ll take a look.