| 12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061 |
- # -*-coding:utf-8-*-
- import asyncio
- import os
- from langchain.chat_models import init_chat_model
- from langchain_core.messages import SystemMessage, HumanMessage
- llm = init_chat_model(
- model="deepseek-chat",
- model_provider="openai",
- api_key=os.environ["DEEPSEEK_API_KEY"],
- base_url="https://api.deepseek.com/v1",
- temperature=0,
- )
- # async def main():
- # messages = [
- # SystemMessage(content="你是一位专业的 Python 讲师，用一句话回答。"),
- # HumanMessage(content="什么是装饰器？"),
- # ]
- # response = await llm.ainvoke(messages) # 异步调用，返回一个 AIMessage
- # print(response.content) # 用 .text 取出文本（属性，不加括号！）
- # print("*" * 80)
- # print(response.text) # 用 .text 取出文本（属性，不加括号！）
- #
- # asyncio.run(main())
- from pydantic import BaseModel, Field
- # # ① 定义期望的输出结构（回顾 2.2：description 就是给大模型的填空指令）
- # class PersonInfo(BaseModel):
- # name: str = Field(description="姓名")
- # age: int = Field(description="年龄（整数）")
- # city: str = Field(description="所在城市")
- #
- # # ② 把模型绑定上去，得到一个「结构化输出版」的 llm
- # structured_llm = llm.with_structured_output(PersonInfo, method="function_calling")
- #
- # async def main():
- # messages = [
- # SystemMessage(content="你负责从文本中抽取人物信息。"),
- # HumanMessage(content="我叫小明，今年 25 岁，住在上海。"),
- # ]
- # # ③ 调用后直接返回 PersonInfo 对象，不是文本！
- # result: PersonInfo = await structured_llm.ainvoke(messages)
- # print(result.model_dump())
- # print(type(result)) # <class '__main__.PersonInfo'>
- # print("*"*80)
- # print(result.name) # 小明
- # print(result.age) # 25
- # print(result.city) # 上海
- #
- # asyncio.run(main())
- # steam输出
- async def main():
- messages = [HumanMessage(content="用三句话介绍一下 Python。")]
- async for chunk in llm.astream(messages): # 逐块接收
- print(chunk.text, flush=True) # 拼接打印，形成打字机效果
- asyncio.run(main())
|
