How often do users stop a generation halfway?

What’s the average number of turns per conversation?

What’s the average number of tokens per input?

Are users
using your application for more complex tasks, or are they
learning to be more concise with their prompts?
What’s the average number of tokens per output?

Are some
models more verbose than others? Are certain types of
queries more likely to result in lengthy answers?

What’s the model’s output token distribution?

How has it
changed over time? Is the model getting more or less diverse?