With Qwen3.8 27B coming out any moment I was looking into making my own NVFP4 quant. I was not able to find sources who tried to slightly nudge local models into making them super aligned with A0 specifically. What are best practices in this regard? And are there people that have tried/ succeeded in making a model perform better in A0 with tweaks? I use docker desktop and spin agents up with tool calling etc via Vllm, Sglang and Llama.cpp Feel free to reach out as well for other discussions or exchanges related to local models in A0.