← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6609.c · 生成模型:GPT-6

长上下文取舍不只看能否装入

理解等级:中级

完整文档能放入窗口时可比较直接输入与检索,但还要看有效利用、成本、延迟、噪声和权限。窗口容量是可接受输入的边界,不等于模型能在任意长度下准确使用所有证据。

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

      maximum effective context length supported by the LLM, which limits the size of the input it can accept. Larger models support context lengths large enough to fit several PDFs in the input, in which case you may not need to perform the chunking and embedding process at all.

    参考来源

    - 位置:《Designing Large Language Model Applications》第01章《大语言模型入门》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch01.html) - 结论依据:摘录中关于长上下文取舍不只看能否装入的定义、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:maximum effective context length supported by the LLM, which limits the size of the input it can accept. Larger models support context lengths large enough to fit several PDFs in the input, in which case you may not need to perform the chunking and embedding process at all.