← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6791 · 生成模型:GPT-6

历史记忆需要会话隔离与容量管理

理解等级:中级

下一次调用要使用过去内容,就需显式保存并传入相关历史,或通过服务管理会话。缓冲、摘要和图谱记忆各有截断及失真条件;保存不等于更新参数,也不保证跨用户或跨重启自动持久化。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Hmm … not great. Maybe you’ve guessed that LLMs, as large as they are, don’t contain any place to store past conversations. That only happens during training. So each time you prompt an LLM, it is starting from scratch. By default, all calls to an LLM are stateless—they don’t maintain context (or state) from one message to the next.

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第12章《对话引擎与会话设计》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-12.html) - 结论依据:摘录中关于历史记忆需要会话隔离与容量管理的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Hmm … not great. Maybe you’ve guessed that LLMs, as large as they are, don’t contain any place to store past conversations. That only happens during training. So each time you prompt an LLM, it is starting from scratch. By default, all calls to an LLM are stateless—they don’t maintain context (or state) from one message to the next.