I haven't tested anything yet.
This project uses parts of llama.cpp combined with a Vulkan backend in order to provide acceleration on any modern AMD GPU.
It's the easiest way to use text generation locally.
llama.cpp is an amazing project with different GPU backends to choose from.
The OpenCL backend just works on most GPUs.
A Vulkan backend is work in progress and will eventually replace the OpenCL one. Just like the previous backend it will just work on most hardware.
It should work on Windows using a GPU supported by ROCm.