Is the model self hosted? Or are you using a custom pre trained LLM like Llama or OpenAI?