知识卡片
紧耦合摘要用答案收益筛选
可用带金标训练样本,比较候选摘要加入后正确答案的条件似然,挑选有利摘要训练压缩器。该目标优化下游回答而非人类可读性,可能依赖特定生成器;换模型后需重新评估。
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
- · 检索增强生成
原书出处 · 1 处
- · 查看原书依据
While you can leverage the LLM’s zero-shot capabilities for both extractive and abstractive summarization, it is more effective (albeit expensive) to fine-tune them so that the summaries generated are specifically optimized to enable the LLM to generate the correct answer. We will call these tightly-coupled summarizers.
参考来源
- 位置:《Designing Large Language Model Applications》第12章《检索增强生成》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch12.html)
- 结论依据:摘录中关于紧耦合摘要用答案收益筛选的定义、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:While you can leverage the LLM’s zero-shot capabilities for both extractive and abstractive summarization, it is more effective (albeit expensive) to fine-tune them so that the summaries generated are specifically optimized to enable the LLM to generate the correct answer. We will call these tightly-coupled summarizers.