Instruction: Discuss the concept of perplexity and its role in measuring the performance of LLMs.
Context: This question tests the candidate's understanding of how LLMs' performance is quantitatively assessed, focusing on a specific metric.
Official answer available
Preview the opening of the answer, then unlock the full walkthrough.
The way I'd explain it in an interview is this: Perplexity measures how well a language model predicts a sequence, so it is useful as a training-time or modeling-quality signal. Lower perplexity generally means the model is assigning higher probability to the observed...
easy
easy
medium
hard
hard