Comment by ranger_danger

Update: here is a patched llama.cpp and quantized model for desktop use: https://github.com/solwyc/talkie-1930-13b-it-q5