EXPERIMENTAL, I'm just a hobbyist, this might be crap. Works with a fairly recent version of ik_llama.cpp only.
GGUF from ubergarm/Qwen3.6-27B-GGUF
MTP layer from Radamanthys11/Qwen3.6-27B-MTP-Q8_0-GGUF
combined with a script by tschunschi: https://huggingface.co/ubergarm/Qwen3.6-27B-GGUF/discussions/2#69f537e8dc0f21b75e58123d
Thanks to all of the above!
Settings to enable:
-mtp --draft-max 2 --draft-p-min 0.0
--draft-max 3 might be good also