That reason is that they replaced Lua with Python as the high-level language that wrapped Torch's core, and needed to differentiate that from the original Torch. But it seems as though you believe the "py" prefix means the correct design decision for the project is to replace ever more parts of torch with Python. Perhaps you could elaborate more on your thinking there?

I am concerned that moving more stuff up into Python is a mistake. It limits support for other languages, like Rust, which speak to the C++ core. Also, executing Python is slower, so limits what can be done by the framework before being considered “too slow.”

Moving a bit to a high level language seems like a win, but when that inspires moving large parts of a big project to high-level languages, I’ve seen unfortunate results. It seems each piece in a high level language often imposes non-obvious costs on all the pieces.

This is nothing new. Way back in the day, Netscape gave up on Javagator, and Microsoft “reset” Windows longhorn to rip out all the c#. Years of work by large teams thrown away.