知识卡片
专用模型接口需保留任务约束
将复杂请求分成子任务后派给专门模型,可利用领域适配降低成本。但子任务应包含必要上下文、输出约定和前置依赖,最终整合还需检验原问题,不是各模型分别完成就自动整体成功。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
Task-Specialized LLMs Yet another way of organizing multi-LLM architectures is to deploy a variety of task-specific LLMs, each of them specialized in solving a particular type of task or subtask. Given a complex user query, a relatively powerful LLM can be used to decompose the query into its constituent subtasks. A router can then assign each of these subtasks to the specialized model most equipped
参考来源
- 位置:《Designing Large Language Model Applications》第13章《设计模式与系统架构》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch13.html)
- 结论依据:摘录中关于专用模型接口需保留任务约束的定义、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:Task-Specialized LLMs Yet another way of organizing multi-LLM architectures is to deploy a variety of task-specific LLMs, each of them specialized in solving a particular type of task or subtask. Given a complex user query, a relatively powerful LLM can be used to decompose the query into its constituent subtasks. A router can then assign each of these subtasks to the specialized model most equipped