100% not an llm problem. Llama.cpp, which only recently started taking large amounts of ai code has had this problem for years.