179 questions
No questions match those filters.
When would you distill rather than build a funnel?
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThey solve different shapes of problem. Distillation makes one expensive model cheaper everywhere it runs — the right choice when almost every request genuinely needs the full capability, so there’s no easy volume to divert away from the expensive model in the first place.
A funnel exploits the fact that most requests don’t need the expensive model at all — right when task difficulty is unevenly distributed across traffic, which in enterprise workloads it usually is: most cases are routine, a small fraction are genuinely hard.
They compose rather than compete — you can distill a model that sits inside one layer of a funnel. And in regulated settings the funnel has a second advantage beyond cost: the cheap, deterministic early layers are far more explainable to an auditor than “a smaller version of the same neural network.”