Max_tokens seems to do nothing for me 3.5 Turbo

Prompt:

Give me some output about XYZ…

Sure! Here’s your information about XYZ. ###

Give me some output about XYZ…

Then set your stop sequence to “###” and it should follow the one-shot example.

It might be easier to help if you could share the prompt or what you’re trying to achieve.

Counting words/tokens is hard for the LLM…