← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9391.e · 生成模型:GPT-6

拆出编码端点需衡量额外网络成本

理解等级:中级

多个分类头共用嵌入时,可复用一次编码,分别扩缩容。新增服务调用、序列化和版本同步也会增加代价;在同进程模块化、批处理和独立服务之间,应按实际瓶颈比较。

这张卡片看不懂?

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

      You can see that this multipurpose endpoint reveals two new opportunities for creating additional microservices. You can imagine an /encode_text/ endpoint to provide the raw BERT encoding vector. The user would call that endpoint first and then use that encoding vector to call a second endpoint: /intents_from_encoding/. This would allow you to split your endpoints into separate microservices. You may also be wondering what that

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》附录E《附录E:容器化微服务部署》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/appendix-e.html) - 结论依据:摘录中关于拆出编码端点需衡量额外网络成本的定义、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:You can see that this multipurpose endpoint reveals two new opportunities for creating additional microservices. You can imagine an /encode_text/ endpoint to provide the raw BERT encoding vector. The user would call that endpoint first and then use that encoding vector to call a second endpoint: /intents_from_encoding/. This would allow you to split your endpoints into separate microservices. You may also be wondering what that