知识卡片
相似度度量应与训练一致
点积受方向和模长影响,余弦关注归一方向,欧氏距离比较空间距离。应采用训练或模型说明所对应的度量;任意替换可能改变排序,尤其原模型通过模长表达信息时。
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
- · 表示学习与嵌入
原书出处 · 1 处
- · 查看原书依据
Similarity Measures Commonly used similarity measures include dot product, cosine similarity, and Euclidean distance. Refer to the Pinecone tutorial on similarity measures if you need a backgrounder. While using embedding models, use the similarity measure that was used to train the model. You will find this information in the model card or Hugging Face model hub page. Tip If you set normalize_embeddings to True as
参考来源
- 位置:《Designing Large Language Model Applications》第11章《表示学习与嵌入》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch11.html)
- 结论依据:摘录中关于相似度度量应与训练一致的定义、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:Similarity Measures Commonly used similarity measures include dot product, cosine similarity, and Euclidean distance. Refer to the Pinecone tutorial on similarity measures if you need a backgrounder. While using embedding models, use the similarity measure that was used to train the model. You will find this information in the model card or Hugging Face model hub page. Tip If you set normalize_embeddings to True as