Could you tell me how many tokens you used to represent the whole 3D scene before sending them to LLM? Thanks a lot.