I've been tinkering around with LLM for a while now, started with LM Studio like probably all of us and wanted to go into headless selhosted model so that I can use my macbook and still use my AI models. I've been using Qwen 3.6 (and 3.5) 27B on my main computer which has a Ryzen 7 3800X, a 7900XT, 32Gb of RAM and that thing was pretty sloooooow even with MTP enabled. You can probably call this a skill issue as I'm not familiar with llama.cpp forest of arguments yet despite reading the docum...