Count the tokens in your prompt and expected completion and see how much of the model context window they consume.
超出上限需截断或摘要历史。