Metaphor（EXA） 基于大语言模型的搜索引擎

---

---

关于 Metaphor

Metaphor是基于大语言模型的搜索引擎，允许用户使用完整的句子和自然语言搜索，还可以模拟人们在互联网上分享和谈论链接的方式进行查询内容。
Metaphor同时还能与LLMs结合使用，允许LLMs连接互联网，查询自己知识库外部的内容。

• 官网：https://exa.ai
• github : https://github.com/NorthBit/Metaphor

---

相关教程

• Metaphor：改变搜索方式，搜索引擎革命，基于大语言模型的AI搜索引擎(1min)
  https://www.bilibili.com/video/BV1om4y1M7XM/
  https://www.1aibox.com/sites/2222.html
• LLM之RAG实战（二）：使用LlamaIndex + Metaphor实现知识工作自动化
  https://mp.weixin.qq.com/s/RsaqhgB9VmMtXDc5UW-1vQ

---

使用示例

安装环境

!pip install exa_py
!pip install langchain 
!pip install llama_hub  
!pip install llama_index   

---

• api_key 为从官网 https://exa.ai 注册后，个人的 key
• 过去使用 MetaphorToolSpec，后面改为使用 ExaToolSpec

from llama_hub.tools.exa.base import ExaToolSpec 
metaphor_tool = ExaToolSpec( api_key="xxx" )

metaphor_tool_list = metaphor_tool.to_tool_list() 

for tool in metaphor_tool_list:
    print(tool.metadata.name)


'''
search
retrieve_documents
search_and_retrieve_documents
search_and_retrieve_highlights
find_similar
current_date
'''

metaphor_tool.search('machine learning transformers', num_results=3)
'''
[Exa Tool] Autoprompt: Here's an insightful article on machine learning transformers:
[{'title': 'On the potential of Transformers in Reinforcement Learning',
  'url': 'https://lorenzopieri.com/rl_transformers/',
  'id': 'snjxQqPfYVWKEmVXunkqhg'},
 {'title': 'How Transformers work in deep learning and NLP: an intuitive introduction | AI Summer',
  'url': 'https://theaisummer.com/transformer/',
  'id': 'kzGcf36vWzWeyL28Az1fNQ'},
 {'title': 'Transformers in Computer Vision: Farewell Convolutions!',
  'url': 'https://towardsdatascience.com/transformers-in-computer-vision-farewell-convolutions-f083da6ef8ab?gi=a1d0a9a2896c',
  'id': 'xF0V2oTHAfVmfS5UvJIFDA'}]
'''

---

# 使用 Mataphor 设置 OpenAI Agent

from llama_index.agent import OpenAIAgent
agent = OpenAIAgent.from_tools(
    metaphor_tool_list,
    verbose=True,
)

# 直接查询
agent.chat('what are the best restaurants in torando?')

---

from llama_index.tools.tool_spec.load_and_search.base import LoadAndSearchToolSpec

# The search_and_retrieve_documents tool is the third in the tool list, as seen above
wrapped_retrieve = LoadAndSearchToolSpec.from_defaults(
  metaphor_tool_list[2],
)

# Just pass the wrapped tools and the get_date utility
agent = OpenAIAgent.from_tools(
  [*wrapped_retrieve.to_tool_list(), metaphor_tool_list[4]],
  verbose=True,
)
agent.chat('Can you summarize everything published in the last month regarding news on superconductors')

---

伊织 2024-01-27（六）