知识卡片
领域选择可用表示相似度辅助
可用代表性域内文本的嵌入或域分类器筛选候选数据,帮助限定覆盖范围。表示聚类只是分布代理,可能偏向文风或常见主题;语义相近不能自动证明符合业务领域定义。
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
- · 进阶微调技术
原书出处 · 1 处
- · 查看原书依据
text that is most representative of the domain. In the same paper, Aharoni et al. introduced domain-data selection techniques based on sentence representations generated through the LLMs. One way to select data representative of the domain is to use embedding similarity with gold-truth in-domain data. Another way is to fine-tune a domain classifier that is trained on gold-truth in-domain data and randomly sampled negative examples.
参考来源
- 位置:《Designing Large Language Model Applications》第07章《进阶微调技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch07.html)
- 结论依据:摘录中关于领域选择可用表示相似度辅助的定义、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:text that is most representative of the domain. In the same paper, Aharoni et al. introduced domain-data selection techniques based on sentence representations generated through the LLMs. One way to select data representative of the domain is to use embedding similarity with gold-truth in-domain data. Another way is to fine-tune a domain classifier that is trained on gold-truth in-domain data and randomly sampled negative examples.