is there a difference between pytorch version 1.13.1 and 1.13.1+cu117 (assuming my cuda version is 11.7)?
Your locally installed CUDA version won’t be used since the PyTorch binaries ship with their own CUDA dependencies. 1.13.1+cu117 indicates a wheel with CUDA 11.7 dependencies while 1.13.1 indicates either a CPU-only build (current versions should show +cpu but unsure if that was always the case) or a conda binary.
Could you show me how to install PyTorch with a locally installed CUDA version? I tried it and failed because torch.cuda.is_available() returned False . Instead, installing it with pip install torch==1.13.1+cu117 --extra-index-url https://download.pytorch.org/whl/cu117 is working (torch.cuda.is_available() returned ‘True’).
You could need to build from source as described here.