Your question is Distilling Models for Real-Time Inference. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
What are the challenges in distilling large models for lower-latency, real-time inference?