what do you mean you don’t want an LLM built into the kernel that controls the scheduler?