Basically, I have a prompt that prevents certain punctuation marks from appearing in the LLM's output (no exclamation marks). When I use that prompt in OpenAI playground, the LLM doesn't produce the exclamation marks at all. When I use the same prompt with the same model and the same parameters through VAPI's web UI, the LLM DOES produce exclamation marks, always. This can be seen via token-by-token output in the logs.