知识卡片
重复试验要记录模型和请求条件
多次请求可观察回答变异,需保留模型版本、参数、提示和结果,并定义成功规则。隔周变好可能来自服务更新或随机性,不能直接归因为这位用户的反馈已训练模型;只挑最好的一次会高估可靠性。
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
Fortunately, if you set their temperature to a nonzero value, LLMs will respond differently each time they receive the same or similar prompts. One approach to creating LLM responses that meet your requirements is the trial-and-error approach: prompt the LLM several times and rank or score the generated responses based on the goals of your project or conversation manager. See figure 1.7 for a bit
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:摘录中关于重复试验要记录模型和请求条件的定义、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:Fortunately, if you set their temperature to a nonzero value, LLMs will respond differently each time they receive the same or similar prompts. One approach to creating LLM responses that meet your requirements is the trial-and-error approach: prompt the LLM several times and rank or score the generated responses based on the goals of your project or conversation manager. See figure 1.7 for a bit