the config.py currently has to be edited before the package can be used. Working with pip install, and virtual environments I tend to install the same package multiple times, and don't want to always edit this file by hand (makes automation difficult).
this MR gives the users a mechanism to specify the configuration outside of the source, in the user home directory.
to run the code on platforms without cuda (e.g. Apple M2), it would be nice if the pytorch code would not hard-code the device it runs on.
in the core package I modified the code so the demo load_torch.py can run without cuda.
This MR addresses two things
the config.py currently has to be edited before the package can be used. Working with pip install, and virtual environments I tend to install the same package multiple times, and don't want to always edit this file by hand (makes automation difficult).
to run the code on platforms without cuda (e.g. Apple M2), it would be nice if the pytorch code would not hard-code the device it runs on.
load_torch.py
can run without cuda.