Amazing. I'll give it a shot on my 5090. I already tried using vLLM but it ran out of GPU memory. I guess it's likely Llama.cpp will work.