ParentFull threadandhuman·Isn’t llama.cpp’s implementation of Qwen 3.5 better, or am I misinformed?View on HN