环境准备
- 依赖(Python >= 3.11):
langchain、langchain-openai、dotenv .env 环境变量:OPENAI_API_KEY:模型供应商密钥OPENAI_BASE_URL:模型供应商请求地址OPENAI_MODEL:模型名称
检查环境变量
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
| import os
import langchain from dotenv import load_dotenv
load_dotenv()
api_key = { "OPENAI_API_KEY": os.getenv("OPENAI_API_KEY"), "OPENAI_BASE_URL": os.getenv("OPENAI_BASE_URL") }
print(f"LangChain 版本:{langchain.__version__}") print("OPENAI_API_KEY:", "已配置" if api_key["OPENAI_API_KEY"] else "未配置") print("OPENAI_BASE_URL:", "已配置" if api_key["OPENAI_BASE_URL"] else "未配置")
|
模型调用
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
| import os from dotenv import load_dotenv from langchain_openai import ChatOpenAI
load_dotenv()
llm = ChatOpenAI( base_url=os.getenv("OPENAI_BASE_URL"), api_key=os.getenv("OPENAI_API_KEY"), model=os.getenv("OPENAI_MODEL") )
response = llm.invoke("做个自我介绍让我认识你")
print(response)
|
后续示例均使用以下模型初始化:
1 2 3 4 5 6
| model = ChatOpenAI( base_url=os.getenv("OPENAI_BASE_URL"), api_key=os.getenv("OPENAI_API_KEY"), model=os.getenv("OPENAI_MODEL"), temperature=0.7, )
|
提示词模板
PromptTemplate 提示词模板
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
| from langchain_core.prompts import PromptTemplate
template = PromptTemplate.from_template("你是{domain}专家,你擅长{skill},问题:{query}")
chain = template | model
response = chain.invoke({ "domain": "金融", "skill": "金融知识", "query": "你知道哪些金融机构?" })
print(response)
|
1 2 3 4 5 6 7 8 9 10 11
| formatted_template = template.format_prompt( domain="数学", skill="计算", query="100+125=?" )
response = model.invoke(formatted_template)
print(response.content)
|
ChatPromptTemplate 提示词对话模板
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
| from langchain_core.prompts import ChatPromptTemplate
chat_template = ChatPromptTemplate.from_messages( [ ("system", "你是{domain}专家,你擅长{skill}"), ("human", "{input}") ] )
chain = chat_template | model
response = chain.invoke({ "domain": "金融", "skill": "金融知识", "input": "你知道哪些金融机构?" })
print(response.content)
|
1 2 3 4 5 6 7 8 9 10 11
| format = chat_template.format_messages( domain="金融", skill="金融知识", input="你知道哪些金融机构?" )
response = model.invoke(format)
print(response.content)
|
输出解析器
不使用输出解析器
链的输出为 Message 对象,需要 .content 取内容:
1 2 3 4 5 6 7 8 9 10 11 12 13
| chain = chat_template | model
reviews = [ {"domain": "数学", "skill": "计算", "input": "110 * 3 - 100"}, {"domain": "数学", "skill": "计算", "input": "100 * 2 - 100"}, {"domain": "数学", "skill": "计算", "input": "30 * 3 - 100"} ]
for review in reviews: response = chain.invoke(review) print(f"问题:{review}") print(f"内容:{response.content}")
|
StrOutputParser
StrOutputParser 将模型的输出转换为字符串:
1 2 3 4 5 6 7 8 9
| from langchain_core.output_parsers import StrOutputParser
chain = chat_template | model | StrOutputParser()
for review in reviews: response = chain.invoke(review) print(f"问题:{review}") print(f"内容:{response}")
|
CommaSeparatedListOutputParser
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
| from langchain_core.output_parsers import CommaSeparatedListOutputParser
list_parser = CommaSeparatedListOutputParser()
format_instrurtions = list_parser.get_format_instructions() print(format_instrurtions)
list_prompt = ChatPromptTemplate.from_template( "你是{domain}专家,你擅长{skill},输出:\n{format_instrurtions}" )
list_chain = list_prompt | model | list_parser
result = list_chain.invoke( { "domain": "信息收集", "skill": "在互联网查询信息,找出五门热门编程语言", "format_instrurtions": format_instrurtions } ) print(result)
|
并行链 RunnableParallel
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
| from langchain_core.runnables import RunnableParallel
translate_prompt = ChatPromptTemplate.from_template("将下面的中文翻译成英文:{text}") summarize_prompt = ChatPromptTemplate.from_template("将下面的内容总结为一句话,不超过50字:{text}")
step1 = translate_prompt | model | StrOutputParser() step2 = summarize_prompt | model | StrOutputParser()
parallel_chain = RunnableParallel[dict]( { "translate": step1, "summary": step2 } )
results = parallel_chain.invoke( { "text": "黄金今年大起大落" } ) print(results)
|
记忆
InMemoryChatMessageHistory
1 2 3 4 5 6 7 8
| from langchain_core.chat_history import InMemoryChatMessageHistory from langchain_core.messages import HumanMessage
chat_history = InMemoryChatMessageHistory() chat_history.add_message(HumanMessage(content="你好")) chat_history.add_message(HumanMessage(content="做个自我介绍")) chat_history.add_message(HumanMessage(content="具体说说")) print(chat_history.messages)
|