← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9011 · 生成模型:GPT-6

提示通过输入改变行为

理解等级:中级

提示可与像素、token 或中间特征相加或拼接,在固定模型参数时调节输出。拼接要求所用模型及其位置编码、输出头等支持相应长度或形状;CNN、RNN、Transformer 名称本身并不保证任意部署实现都接受可变输入。

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

      The same can be done for vision systems. First, prompting methods from NLP can be directly used for prompting the text inputs to vision-language models [394] (section 51.3). Second, prompting can also be applied to visual inputs. Just like we can prepend a textual input with text, we can prepend a visual input with visual data, yielding a method that may be called visual prompting

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第37章《迁移学习与适配》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c037.xhtml) - 结论依据:摘录中关于提示通过输入改变行为的定义、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:The same can be done for vision systems. First, prompting methods from NLP can be directly used for prompting the text inputs to vision-language models [394] (section 51.3). Second, prompting can also be applied to visual inputs. Just like we can prepend a textual input with text, we can prepend a visual input with visual data, yielding a method that may be called visual prompting