-
-
Notifications
You must be signed in to change notification settings - Fork 23
Description
Checklist
- I updated to the latest version of the theme (
Settings→Appearance→Check for updates). - I made sure that the issue is related to this theme by checking that the problem does not occur with the default theme.
- I have searched the existing issues for this theme.
- In case I am missing a UI element, I have checked that it cannot be re-enabled with the Style Settings plugin.
- The bug is about Obsidian desktop, Obsidian mobile is not supported by this theme.
- My Obsidian installer version is 1.1.9 or later. If not, please update first. You can find the version in the
Abouttab of the Obsidian settings.
Bug description
Hello author, when I insert an image and adjust its width value, it does not change; only when I click on it and hold does the image display at its normal size.
I really love this theme and it makes me happy when I'm studying. When I was troubleshooting the issue, I found that the image could be resized normally when the image size was set to a value below 300-400, but it could not be enlarged above this value. I also searched for related elements in the developer tools of Obsidian, but I couldn't find any related issues. When I tried changing the theme, the problem disappeared, so I think there might be a problem somewhere. I really like this theme, and I don't want to switch to another one. I hope I can continue to use it. Thank you to the author.
Relevant screenshot
2025-08-07.15.28.50.mov
To reproduce
No response
Plugins involved
No response
Sample text
提示词的基本概念
通过提供简单的 prompts 给大模型,使其输出大量的结果反馈给你,但是输出的内容质量和数量与提示词有关,那么不同的提示词也会产生不同的效果。所以提示词需要包含一些特定的要素,从而在大模型的回答中获得想要的且准确的结果。
提示词的四个要素
提示词可以包含以下任意要素
指令:想要模型执行的特定任务或指令
上下文:包含外部信息和额外的上下文信息,引导语言模型更好地响应提问
输入数据: 用户输入的内容或问题
输出提示:指定输出的类型或格式
提示词示例:文本分类任务
请将文本分为中性、否定或肯定
文本:我觉得食物还可以。
情绪:
其中第一行是指令;第二行是输入数据;第三行是输出提示
提示词所需的格式取决于想要大模型完成的任务类型,并不是所有要素都必须包含
提示词设计
指令
指令最好放在一段提示词的最开头,用来指示模型进行各种简单的任务,例如“写入”、“分类”、“总结”、“翻译”、“排序”等,给各种简单任务设计有效的提示。
当然还需要通过大量实践来找出最适合当前应用需求的提示词。以不同的关键词,上下文和数据来试验不同的指令,看看什么样的是最适合当前特定用例和任务的
具体性
要非常具体地说明希望执行的指令和任务,提示越详细,结果越准确。但同时提示词不能太长。
提示技术
零样本提示
就是不给上下文的提示词,不向模型提供任何示例
比如说
将文本分类为中性、负面或正面。
文本:我认为这次假期还可以。
情感:
输出:
中性
少样本提示
在一些较高要求的需求和场景中,零样本提示可能不足以满足模型的准确生成,在复杂的任务中的表现一般。而少样本提示则是提供一段上下文的内容,并且让模型学习这段上下文。
例子:在句子中正确使用一个新词。
“whatpu”是坦桑尼亚的一种小型毛茸茸的动物。一个使用whatpu这个词的句子的例子是:
我们在非洲旅行时看到了这些非常可爱的whatpus。
“farduddle”是指快速跳上跳下。举一个使用farduddle这个词的句子的例子:
输出:
当我们赢得比赛时,我们都开始farduddle。
当我们给模型提供一个样本(上下文)时,模型已经学会了如何执行任务并正确输出,那么当面临更加复杂和困难的任务时,我们可以增加样本数量,使模型学习地更加准确
但是当遇到处理复杂的推理类任务时,即使增加样本和上下文数量,输出可能也不准确。
比如说这个示例
提示:
这组数字中的奇数加起来是一个偶数:4、8、9、15、12、2、1。
A:答案是False。
这组数字中的奇数加起来是一个偶数:17、10、19、4、8、12、24。
A:答案是True。
这组数字中的奇数加起来是一个偶数:16、11、14、4、8、13、24。
A:答案是True。
这组数字中的奇数加起来是一个偶数:17、9、10、12、13、4、2。
A:答案是False。
这组数字中的奇数加起来是一个偶数:15、32、5、13、82、7、1。
A:
输出:
答案是True。
模型给的是错误答案。所以在这种复杂的推理类任务中,给出多个样本提示依然不能提高模型的正确率。所以我们将引入链式思考提示(CoT Chain-of-Thought Prompting)
CoT 链式思考提示
链式思考指的是
![[Pasted image 20250807151545.png|1350]]
Obsidian version
1.8.10
Installer version
1.8.9
In which mode(s) does the problem occur? (if applicable)
- Source Mode
- Reading Mode
- Live Preview
Which operating system?
- macOS
- Linux
- Windows
