what I have in mind:
- 23178 is the cumulative sum of tokens sent to the LLM, for instance if during the conversation I have 24 calls to the llm with a window of 1k tokens
- 327 is the number of tokens answered by the llm during the full conversation