llama.cpp

metal: speed up Qwen3-VL image encoding on large images by ~11%

#21443

Open