A large number of spaces and line breaks are returned

API Bugs
1

How to deal with this situation? A large number of spaces and line breaks are returned, which causes the length to exceed the limit.

2
  1. Use max_completion_tokens to limit your cost to the maximum ever expected from normal responses
  2. Use Chat Completion’s frequency_penalty with a small positive value, so the bad tokens become more discouraged the more of the same is written.
  3. Completely and exhaustively describe any JSON output format you ask for, especially when using a non-strict schema.
  4. Switch models to one where the symptom can’t easily be reproduced by others also (aka my detector says to get off gpt-4o-mini). :innocent:
3

We use o3-mini, max_completion_token is 50,000, because some questions and answers require long tokens. The frequency_penalty is currently 0. So I will set the frequency_penalty to 0.7

5

o3-mini not support this request param frequency_penalty ,then what should I do

6

Or are there any recommended prompt tips to reduce this problem?