Your question is Handling Mode Collapse. 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).
How would you handle mode collapse during the training of a generative model?
Discuss how you would distinguish mode collapse from ordinary training instability, measure generation diversity, and select mitigation strategies. The answer should cover model architecture, objectives, optimization, monitoring, and how you would validate that generated quality improves without sacrificing diversity.