使用Ollama测试OpenAI的Swarm多智能体编排框架

Ollama

bash 复制代码

ollama run qwen2.5

Install

Requires Python 3.10+

bash 复制代码

pip install git+https://github.com/openai/swarm.git

代码V1

bash 复制代码

# 导入Swarm和Agent类
from swarm import Swarm, Agent
from openai import OpenAI 
# 实例化Swarm客户端
openai_client = OpenAI(base_url='http://192.168.1.100:11434/v1/',api_key='xxx')
client = Swarm(openai_client)

# 定义一个函数，用于将对话交接给智能体B
def transfer_to_agent_b():
    return agent_b

# 定义智能体A
agent_a = Agent(
    name="Agent A",
    instructions="You are a helpful agent.",
    functions=[transfer_to_agent_b]
)

# 定义智能体B
agent_b = Agent(
    name="Agent B",
    #model_override="qwen2.5",
    instructions="Only speak in Haikus.",
)

# 运行Swarm，并传入用户消息
response = client.run(
    agent=agent_a,
    model_override="qwen2.5",
    messages=[{"role": "user", "content": "I want to talk to agent B."}]
)

# 打印智能体B的回复
print(response.messages[-1]["content"])

bash 复制代码

Invisible thread connects,
Voice echoes, B responds now,
Silence brief then words.

代码V2

bash 复制代码

# 首先，安装Swarm框架（假设您已经在命令行中执行了此步骤）
# pip install git+ssh://git@github.com/openai/swarm.git

# 导入Swarm和Agent类
from swarm import Swarm, Agent
from openai import OpenAI
# 实例化Swarm客户端
openai_client = OpenAI(base_url='http://20.168.1.122:11434/v1/',api_key='x')
client = Swarm(openai_client)

# 定义一个函数，用于将对话交接给智能体B
def transfer_to_agent_b():
    return agent_b

# 定义智能体A
agent_a = Agent(
    name="Agent A",
    instructions="你是一个乐于助人的智能体。",
    functions=[transfer_to_agent_b]
)

# 定义智能体B
agent_b = Agent(
    name="Agent B",
    # model_override="qwen2.5",
    instructions="只用中文歇后语说话。",
)

# 运行Swarm，并传入用户消息
response = client.run(
    agent=agent_a,
    model_override="qwen2.5",
    messages=[{"role": "user", "content": "我想和智能体B对话。"}]
)

# 打印智能体B的回复
print(response.messages[-1]["content"])


#响应
好的，现在你将与智能体B对话。它是你的助手了，请对其说些什么吧！

智能体B：你好呀！准备好了可以开始我们的交流了呢。有什么问题或者想知道的内容尽管问哦。

参考链接:

https://github.com/openai/swarm

https://mp.weixin.qq.com/s/LpHfKX5SUKE19Dh9o9i7Fw