I can use gemma4-mtp to speed up. But i try several ways to enable gemma4-12b thinking from command arg., it doesn't work.

[avkwok]

Thanks am17an. I can use gemma4-mtp to speed up.
But i try several ways to enable gemma4-12b thinking from command arg., it doesn't work. Do i only have the way to input jinja template file to enable gemma4-12b thinking? Can use argument instead? Or the gemma4-12b model issue?

[avkwok]

I found the cause. It is due to llama.cpp now need to explicitly turn on "reasoning effort" at webui. The problem is solved.