47 lines
1.3 KiB
Python
47 lines
1.3 KiB
Python
|
import os
|
|||
|
|
from langchain_openai import ChatOpenAI
|
|||
|
|
from dotenv import load_dotenv
|
|||
|
|
|
|||
|
|
# 加载环境变量
|
|||
|
|
load_dotenv()
|
|||
|
|
|
|||
|
|
|
|||
|
|
def get_llm():
|
|||
|
|
"""
|
|||
|
|
初始化并返回 DeepSeek LLM 实例
|
|||
|
|
"""
|
|||
|
|
# 注意：这里的 ChatOpenAI 实际上是兼容 OpenAI 接口的类，用于连接 DeepSeek API
|
|||
|
|
llm = ChatOpenAI(
|
|||
|
|
base_url=os.getenv("base_url1"),
|
|||
|
|
api_key=os.getenv("DEEPSEEK_API_KEY1"),
|
|||
|
|
model="deepseek-chat",
|
|||
|
|
temperature=0, # 设置为0，确保结果稳定
|
|||
|
|
)
|
|||
|
|
return llm
|
|||
|
|
|
|||
|
|
|
|||
|
|
# 1. 初始化 LLM
|
|||
|
|
llm = get_llm()
|
|||
|
|
|
|||
|
|
# 2. 用户级 prompt
|
|||
|
|
user_prompt = "你是谁？请详细介绍一下你自己。"
|
|||
|
|
|
|||
|
|
print(f"--- 用户提问：{user_prompt} ---")
|
|||
|
|
print("--- LLM 流式回复开始 ---")
|
|||
|
|
|
|||
|
|
# 3. **核心修改：使用 .stream() 方法进行流式调用**
|
|||
|
|
# .stream() 返回一个迭代器，我们可以遍历它来获取分块的输出。
|
|||
|
|
response_stream = llm.stream(user_prompt)
|
|||
|
|
|
|||
|
|
# 4. 遍历并打印流式输出
|
|||
|
|
full_response = ""
|
|||
|
|
for chunk in response_stream:
|
|||
|
|
# chunk.content 包含当前流片段的内容
|
|||
|
|
# end="" 确保打印时不换行，模拟流式输出效果
|
|||
|
|
print(chunk.content, end="", flush=True)
|
|||
|
|
full_response += chunk.content
|
|||
|
|
|
|||
|
|
print("\n--- LLM 流式回复结束 ---")
|
|||
|
|
|
|||
|
|
# 5. 可选：打印完整的回复内容
|
|||
|
|
# print(f"\n完整的回复内容：{full_response}")
|