When we call the open ai models, we do a system role, user role message. the roles like “System” and “user” does those count in the input token. Also the and tokens, are they counted too as part of output tokens.
Yes. When you send a message to an OpenAI model, every part of your message, including the “system role” and “user role”, is tokenized and counts towards the input token limit.
Similarly, the model’s response is also tokenized. Every part of the response, including these role labels and any other text or characters, counts towards the output token limit.