How do you write your chat-completion queries?

[dpkirchner]

I'm tinkering with the chat completion route, trying to figure out the right way to specify a system prompt that will apply to the following user prompt. Nitro seems to use chatml formatting, e.g. [{"role": "system", "content": "Some system prompt"}, {"role": "user", "content": "What the user sent"}]. I'm not 100% sure this is working for me but I need to keep testing.

While investigating this I found the pre_prompt, system_prompt, user_prompt, and ai_prompt parameters that apply to the loadmodel route: https://nitro.jan.ai/api-reference#tag/Load-Model/operation/loadmodel . How are these used? Should my requests to the chat completion route look like what I pasted above, with a separate object per role and message, as found on https://nitro.jan.ai --or-- something more like:

[
  {
    "role": "user",
    "content": "ASSISTANT's RULE: Some system prompt\n\nUSER: What the user sent"
  }
]

as the API documentation seems to imply? If this latter option is incorrect, could you explain what those *_prompt parameters actually do?