We need some support from yzma for this. Specifically `jinja` support. https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/main/chat_template.jinja https://github.com/ggml-org/llama.cpp/pull/17524