Would be great to have pre-built binaries for each release (Windows/Linux/macOS) with Vulkan, CUDA and Metal support, like llama.cpp does. Saves a lot of time not having to compile from source every update.