Your question is Handle OOV Words in Generation. 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).
You're working with a generative language model that needs to produce fluent text even when it encounters rare, misspelled, newly coined, or domain specific terms that were not common in training. You need an approach that keeps generation stable when vocabulary coverage is imperfect.
How would you handle out-of-vocabulary (OOV) words in a generative model?