A fork of
https://huggingface.co/dnhkng/RYS-Qwen3.5-27B-FP8-M .
A single layer being added didn't work with llama.cpp, as Qwen3.5 uses full_attention_interval = 4, expecting every 4th layer to be a full attention layer.
So I added 4 duplicate layers, 31-35, and made it a gguf.
