llama.cpp

92bb84f7 - mtmd: allow QwenVL to process larger image by default (#17020)