Get all requested max tokens with gpt-3.5-turbo-instruct