It's not just about "raw-flop performance" though; it affects even basic things like creating data-loaders that run in the background while your main thread is

It's not just about "raw-flop performance" though; it affects even basic things like creating data-loaders that run in the background while your main thread is doing some hard ML crunching. Every DL library comes with its own C++ backend that does this for now, but it's annoyingly inflexible. And dealing with GIL is a nightmare if you're dealing with mixed Python code.