当前位置：首页 > news >正文

AI Agent设计模式二：Parallelization

news 2025/7/18 12:33:37

概念：并行任务执行引擎

✅ 优点：提升吞吐量，充分利用多核资源
❌ 缺点：复杂度高，存在竞态条件风险

在这里插入图片描述

from langchain_openai import ChatOpenAI
from langgraph.graph import StateGraph, START, END
from typing import TypedDict
import os

# 初始化模型
client = ChatOpenAI(
    model="deepseek-r1",
    openai_api_key=os.environ["BAILIAN_API_KEY"],
    openai_api_base="https://dashscope.aliyuncs.com/compatible-mode/v1",
    streaming=False  # 禁用流式传输
)

# 定义实体类
class PhilosophyState(TypedDict):
    topic: str
    part0: str
    part1: str
    part2: str
    combined_output: str

meta_prompt = """
    针对如下问题进行思考，并得出结论。
    问题如下：{topic}
    你分析的角度如下：{aspect}
"""

# 道家观点解释
def part0_interpreter(state: PhilosophyState):
    print(f"道家视点解释开始 ：{state['topic']}")

    prompt = meta_prompt.format(topic = state['topic'], aspect = "道家")
    response = client.invoke(prompt)

    print(f"道家观点：{response}")
    return {'part0': response}

# 儒学观点解释
def part1_interpreter(state: PhilosophyState):
    print(f"儒学观点解释开始 ：{state['topic']}")
    prompt = meta_prompt.format(topic=state['topic'], aspect="儒学")
    response = client.invoke(prompt)
    print(f"儒学观点：{response}")
    return {'part1': response}

# 法学观点解释
def part2_interpreter(state: PhilosophyState):
    print(f"法学观点解释开始 ：{state['topic']}")
    prompt = meta_prompt.format(topic=state['topic'], aspect="法学")
    response = client.invoke(prompt)
    print(f"法学观点：{response}")
    return {'part2': response}

def aggregate_results(state: PhilosophyState):
    combined = f"{state['part0']}\n{state['part1']}\n{state['part2']}"
    return {'combined_output': combined }

# 创建工作流
workflow = StateGraph(PhilosophyState)

# 添加节点
workflow.add_node("part0_interpreter", part0_interpreter)
workflow.add_node("part1_interpreter", part1_interpreter)
workflow.add_node("part2_interpreter", part2_interpreter)
workflow.add_node("aggregate_results", aggregate_results)

# 添加节点边
workflow.add_edge(START, "part0_interpreter")
workflow.add_edge(START, "part1_interpreter")
workflow.add_edge(START, "part2_interpreter")
workflow.add_edge("part0_interpreter", "aggregate_results")
workflow.add_edge("part1_interpreter", "aggregate_results")
workflow.add_edge("part2_interpreter", "aggregate_results")
workflow.add_edge("aggregate_results", END)

# 编译工作流
app = workflow.compile()

result = app.invoke({"topic": "治国之道在于平衡各方利益"})
print(result)