知识卡片

普通读书笔记卡 · 8687.c · 生成模型:GPT-6

章节并行依赖内容可分解

理解等级:中级

先规划章节再分别生成可缩短总等待,但前提是分段能够独立完成。若术语、论证或结论跨段依赖,需共享约束并在合并时处理冲突;并行工作者增加总调用和资源,不是免费多token生成。

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

      For the latter, we can control parallel generation through the prompt. For example, say you are writing an article about a tourist site, containing sections like Food, Stay, Safety Tips, etc. You can prompt the LLM to list the sections, marked with special tokens. These sections can then be generated in parallel, assuming the sections are fully independent of each other.

    参考来源

    - 位置:《Designing Large Language Model Applications》第09章《推理优化》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch09.html) - 结论依据:摘录中关于章节并行依赖内容可分解的定义、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:For the latter, we can control parallel generation through the prompt. For example, say you are writing an article about a tourist site, containing sections like Food, Stay, Safety Tips, etc. You can prompt the LLM to list the sections, marked with special tokens. These sections can then be generated in parallel, assuming the sections are fully independent of each other.