I have a question on the ideal range for completion lengths regarding my prompt completion data set for fine-tuning gpt3. Basically my goal is to be able to have my model respond to the user with responses that are around 400 to 600 characters on average. The average prompts that I plan on using when interacting with it will be in the form of questions and will be around 40 to 90 characters.
What should I aim for regarding my ideal length of completions? Should I do shorter completions like around the 200 to 400 range and then use a script to elicit longer responses or should I have a majority of the completions be in the 400 to 600 range?
(For reference the bot is emulating the personality of a podcast host and is trained off of segments of transcribed podcast episodes)