The htop for LLM inference see exactly where every GB of VRAM goes and get measured quantization savings.
By chatting or signing in you agree to the Terms and chat-message logging (revocable in History).