I was building a hardware monitoring and metrics tool and realized that I have no way to test if I get the right information out from nvidia gpus via nvml metrics without using deployed gpus, so I made a mock gpu driver that allows you to present a gpu in a desired state as a real gpu to a linux system. It creates and overloads the libnvidia.so* files and simply construct a scenario from json that is hot-swappable when it is running.
It is a useful tool and easy to use if you're building monitoring or telemetry for nvidia gpus.