Is there a table which shows the supported cuda version for every pytorch version

Thank you for your reply.

If I have to build some additional libraries, such as detectron2, do I need a local CUDA toolkit?