It can be useful to (ab)use python's dynamic object model to add functions to existing objects in order to better interrogate and debug a pre-trained pytorch model.

Integers overflow silently in a lot of popular python libraries (including numpy, scipy, pandas, pytorch, tensorflow).