Identifying "Trigger Tokens" for Function Call

Has anyone uncovered the specific tokens OpenAI models use to recognize when to trigger API functions? I briefly saw them in a ChatGPT because of markdown error, but, sadly it does not reproduce. Knowing this sequence of tokens would be extremely helpful for fine-tuning, API interactions, and even direct ChatGPT prompts. in my case I am building a fine tune model and I need to hint the model when exactly it should call the API and when it should ignore it.

I appreciate any knowledge around this subject!