← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6504 · 生成模型:GPT-6

编译预热成本单独记录

理解等级:中级

编译可能在首次执行触发较高优化成本,稳定运行的速度应与首次延迟分开报告。本章多次运行并把第一次标为预热;短任务还应比较包含预热的总成本,不能只拿后续最快一次代表用户体验。

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

      It is worth noting that the first execution using the compiled model may be slower than usual due to the initial compilation and optimization steps. To better measure the performance improvement, we will repeat the text generation process multiple times.

    参考来源

    - 位置:《Build a Reasoning Model (From Scratch)》第02章《使用预训练模型生成文本》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-2.html) - 结论依据:摘录中关于编译预热成本单独记录的定义、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:It is worth noting that the first execution using the compiled model may be slower than usual due to the initial compilation and optimization steps. To better measure the performance improvement, we will repeat the text generation process multiple times.