如何使用LangChain插件¶

为了便利大家结合文心大模型与LangChain开发应用，ERNIE Bot Agent扩展LangChain框架的功能，提供大语言模型（large language model）、聊天模型（chat model）、文本嵌入模型（text embedding model）等组件（这些组件的集合称为LangChain插件）。本文档将介绍ERNIE Bot Agent的LangChain插件的基础用法。

ERNIE Bot Agent的LangChain插件目前包含如下组件：

• ErnieBot：大语言模型，用于完成文本补全任务。
• ErnieBotChat：聊天模型，用于完成对话补全任务。
• ErnieEmbeddings：文本嵌入模型，用于生成文本的向量表示。

准备工作¶

安装erniebot-agent与langchain：

In [ ]:

!pip install erniebot-agent langchain

!pip install erniebot-agent langchain

根据ERNIE Bot 认证鉴权文档中的说明，获取AI Studio星河社区的access token。执行如下代码，填写access token并敲击回车键：

In [ ]:

import getpass

access_token = getpass.getpass(prompt="Access token: ")

import getpass access_token = getpass.getpass(prompt="Access token: ")

ErnieBot¶

ErnieBot是LangChain大语言模型组件，可用于完成文本补全任务。本文档仅介绍ErnieBot的用法，大家可以在LangChain官方文档了解关于大语言模型组件的更多信息。

创建一个ErnieBot对象：

In [ ]:

from erniebot_agent.extensions.langchain.llms import ErnieBot

from erniebot_agent.extensions.langchain.llms import ErnieBot

In [ ]:

llm = ErnieBot(aistudio_access_token=access_token)

llm = ErnieBot(aistudio_access_token=access_token)

基本使用¶

In [ ]:

question = "What does SFINAE mean in C++ template metaprogramming?"

print(llm(question))

question = "What does SFINAE mean in C++ template metaprogramming?" print(llm(question))

在chain中使用¶

In [ ]:

from langchain.prompts import PromptTemplate
from langchain.chains import LLMChain

from langchain.prompts import PromptTemplate from langchain.chains import LLMChain

In [ ]:

template = "Tell me a joke about {content}."

prompt = PromptTemplate(template=template, input_variables=["content"])

template = "Tell me a joke about {content}." prompt = PromptTemplate(template=template, input_variables=["content"])

In [ ]:

llm_chain = LLMChain(prompt=prompt, llm=llm)

llm_chain = LLMChain(prompt=prompt, llm=llm)

In [ ]:

content = "rabbits"

print(llm_chain.run(content=content))

content = "rabbits" print(llm_chain.run(content=content))

异步调用¶

In [ ]:

question = "Please write a Python program that checks if an integer is a prime number."

answer = await llm.agenerate([question])
print(answer)

question = "Please write a Python program that checks if an integer is a prime number." answer = await llm.agenerate([question]) print(answer)

流式回复¶

In [ ]:

question = "What is the difference between capybara and kiwi?"

for chunk in llm.stream(question):
    print(chunk, end="", flush=True)
print("")

question = "What is the difference between capybara and kiwi?" for chunk in llm.stream(question): print(chunk, end="", flush=True) print("")

ErnieBotChat¶

ErnieBotChat是LangChain聊天模型组件，可用于完成文本补全任务。本文档仅介绍ErnieBotChat的用法，大家可以在LangChain官方文档了解关于聊天模型模型组件的更多信息。

创建一个ErnieBotChat对象：

In [ ]:

from erniebot_agent.extensions.langchain.chat_models import ErnieBotChat

from erniebot_agent.extensions.langchain.chat_models import ErnieBotChat

In [ ]:

chat = ErnieBotChat(aistudio_access_token=access_token)

chat = ErnieBotChat(aistudio_access_token=access_token)

基本使用¶

In [ ]:

from langchain.chat_models.base import HumanMessage

from langchain.chat_models.base import HumanMessage

In [ ]:

message = HumanMessage(content="What does SFINAE mean in C++ template metaprogramming?")
print(chat([message]))

message = HumanMessage(content="What does SFINAE mean in C++ template metaprogramming?") print(chat([message]))

在chain中使用¶

In [ ]:

from langchain.prompts import ChatPromptTemplate

from langchain.prompts import ChatPromptTemplate

In [ ]:

message = "Tell me a joke about {content}."
prompt = ChatPromptTemplate.from_messages([("human", message)])

message = "Tell me a joke about {content}." prompt = ChatPromptTemplate.from_messages([("human", message)])

In [ ]:

chain = prompt | chat

chain = prompt | chat

In [ ]:

print(chain.invoke({"content": "rabbits"}))

print(chain.invoke({"content": "rabbits"}))

异步调用¶

In [ ]:

from langchain.chat_models.base import HumanMessage

from langchain.chat_models.base import HumanMessage

In [ ]:

message = HumanMessage(content="Please write a Python program that checks if an integer is a prime number.")

response = await chat.agenerate([[message]])
print(response)

message = HumanMessage(content="Please write a Python program that checks if an integer is a prime number.") response = await chat.agenerate([[message]]) print(response)

流式回复¶

In [ ]:

from langchain.chat_models.base import HumanMessage

from langchain.chat_models.base import HumanMessage

In [ ]:

message = HumanMessage(content="What is the difference between capybara and kiwi?")

for chunk in chat.stream([message]):
    print(chunk.content, end="", flush=True)
print("")

message = HumanMessage(content="What is the difference between capybara and kiwi?") for chunk in chat.stream([message]): print(chunk.content, end="", flush=True) print("")

ErnieEmbeddings¶

ErnieEmbeddings是LangChain文本嵌入模型组件，可用于生成文本的向量表示。本文档仅介绍ErnieEmbeddings的用法，大家可以在LangChain官方文档了解关于聊天模型模型组件的更多信息。

创建一个ErnieEmbeddings对象：

In [ ]:

from erniebot_agent.extensions.langchain.embeddings import ErnieEmbeddings

from erniebot_agent.extensions.langchain.embeddings import ErnieEmbeddings

In [ ]:

embeddings = ErnieEmbeddings(aistudio_access_token=access_token)

embeddings = ErnieEmbeddings(aistudio_access_token=access_token)

处理单段输入文本¶

In [ ]:

text = "This is a test document."

query_result = embeddings.embed_query(text)
print(len(query_result))

text = "This is a test document." query_result = embeddings.embed_query(text) print(len(query_result))

处理多段输入文本¶

In [ ]:

texts = ["doc1", "doc2"]

docs_result = embeddings.embed_documents(texts)
print(len(docs_result))
for res in docs_result:
    print(len(res))

texts = ["doc1", "doc2"] docs_result = embeddings.embed_documents(texts) print(len(docs_result)) for res in docs_result: print(len(res))