英文:
Manage search results in Langchain
问题
我是Langchain的新手。我想创建一个应用程序,使用Langchain的搜索代理,如serpapi,在互联网上找到一些有用的信息。但大多数时候,当我需要参考信息时,它返回无关的来源。我想要管理它的搜索内容。我该如何在Python中实现这个?我已经搜索过解决方案,但找到的自定义工具都没有帮助。
英文:
I am kind of new to Langchain. I wanted to create an app that uses Langchain's search agents like serpapi and find some useful information on the internet. But most of the times, when I want references from it, it returns irrelevant sources. I want to manage what it searches. How can I do that in python?
I have searched for the solutions and found custom tools but they were not helpful.
答案1
得分: 1
搜索结果依赖于您的LLM用于搜索的关键词。LLM根据您提供的信息(提示)选择要搜索的关键词。所以,您在这里有三个选项:
- 优化您的提示,以便您的LLM可以正确跟随。也许可以给出一些示例(以防需要少量示例来描述反应)。
- 在收到搜索结果后,尝试要求您的LLM提取相关信息并优化结果。您可以查看这个解释。
- 如果您正在使用本地LLM,可以尝试更大的LLM。据我们所知,更大的模型具有更好的推理能力。
英文:
The search results depend on the keyword your LLM used to search. The LLM chooses keywords to search based on your provided information (prompt). So, you have three options here:
- Optimize your prompt so that your LLM can follow correctly. Maybe
give it some examples (in case of a few-shot react description). - After receiving a search result, try to ask your LLM to extract
relevant information and refine the result. You can check this
explanation. - In case you're using local LLM, try a bigger LLM. As far as we know,
a bigger model has a better reasoning ability.
答案2
得分: 0
唯一操纵搜索结果的方式是使用提示。明确指定目标,添加示例,不要添加矛盾的观点。
英文:
the only way to manipulate the search result is using the prompt. Specify the goal clearly, add examples, Do not add contradictory ideas.
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库,让每个人都能够通过互相帮助和分享经验来进步。
评论