Currently the original TabPFN installation assumes that the user may have a GPU to spare and installs a bunch of cuda libraries. This makes the overall installation very bulky, and pointless if the user does not have a GPU. A solution is to create a tabpfn fork and install this dependency directly from source, supporting both options.
Currently the original TabPFN installation assumes that the user may have a GPU to spare and installs a bunch of cuda libraries. This makes the overall installation very bulky, and pointless if the user does not have a GPU. A solution is to create a tabpfn fork and install this dependency directly from source, supporting both options.