终止条件#
在前一节中，我们探讨了如何定义代理，并将它们组织成可以通过通信（对话）解决任务的团队。然而，对话可能永远持续下去，在许多情况下，我们需要知道_何时_停止它们。这就是终止条件的作用。
AgentChat 通过提供基础 TerminationCondition 类和几个继承自它的实现来支持多种终止条件。
终止条件是一个可调用对象，它接收自上次调用该条件以来的 ChatMessage 对象序列，如果对话应该终止则返回 StopMessage，否则返回 None。一旦达到终止条件，必须在再次使用之前重置它。
关于终止条件的一些重要注意事项：
它们是有状态的，必须在再次使用之前重置。
它们可以使用 AND 和 OR 运算符组合。
它们由团队实现/执行，而不是由代理执行。代理可以通过发送 StopMessage 来发出或请求终止，但团队负责执行它。
首先，让我们定义一个只有一个代理的简单团队，然后探索如何应用多个终止条件来指导结果行为。
import logging
from autogen_agentchat import EVENT_LOGGER_NAME
from autogen_agentchat.agents import CodingAssistantAgent
from autogen_agentchat.logging import ConsoleLogHandler
from autogen_agentchat.task import MaxMessageTermination, TextMentionTermination
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_core.components.models import OpenAIChatCompletionClient
logger = logging.getLogger(EVENT_LOGGER_NAME)
logger.addHandler(ConsoleLogHandler())
logger.setLevel(logging.INFO)
model_client = OpenAIChatCompletionClient(
model="gpt-4o",
temperature=1,
api_key="sk-", # Optional if you have an OPENAI_API_KEY env variable set.
)
writing_assistant_agent = CodingAssistantAgent(
name="writing_assistant_agent",
system_message="You are a helpful assistant that solve tasks by generating text responses and code.",
model_client=model_client,
)
/tmp/ipykernel_326523/3735405718.py:21: DeprecationWarning: CodingAssistantAgent is deprecated. Use AssistantAgent instead.
writing_assistant_agent = CodingAssistantAgent(
MaxMessageTermination#
最简单的终止条件是 MaxMessageTermination 条件，它在固定数量的消息后终止对话。
max_msg_termination = MaxMessageTermination(max_messages=3)
round_robin_team = RoundRobinGroupChat([writing_assistant_agent], termination_condition=max_msg_termination)
round_robin_team_result = await round_robin_team.run(task="Write a unique, Haiku about the weather in Paris")
{"timestamp": "2024-12-06T17:57:10.104944", "message": "{\n \"source\": \"user\",\n \"models_usage\": null,\n \"content\": \"Write a unique, Haiku about the weather in Paris\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:10.104944", "message": "{\n \"source\": \"user\",\n \"models_usage\": null,\n \"content\": \"Write a unique, Haiku about the weather in Paris\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:10.104944", "message": "{\n \"source\": \"user\",\n \"models_usage\": null,\n \"content\": \"Write a unique, Haiku about the weather in Paris\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.015792", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 38,\n \"completion_tokens\": 19\n },\n \"content\": \"Rain whispers softly, \\nEiffel towers through the mist\u2014 \\nCity wrapped in gray. \"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.015792", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 38,\n \"completion_tokens\": 19\n },\n \"content\": \"Rain whispers softly, \\nEiffel towers through the mist\u2014 \\nCity wrapped in gray. \"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.015792", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 38,\n \"completion_tokens\": 19\n },\n \"content\": \"Rain whispers softly, \\nEiffel towers through the mist\u2014 \\nCity wrapped in gray. \"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.737698", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 66,\n \"completion_tokens\": 19\n },\n \"content\": \"Rain whispers softly, \\nEiffel towers through the mist\u2014 \\nCity wrapped in gray. \"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.737698", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 66,\n \"completion_tokens\": 19\n },\n \"content\": \"Rain whispers softly, \\nEiffel towers through the mist\u2014 \\nCity wrapped in gray. \"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.737698", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 66,\n \"completion_tokens\": 19\n },\n \"content\": \"Rain whispers softly, \\nEiffel towers through the mist\u2014 \\nCity wrapped in gray. \"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:57:11.738580", "message": "{\n \"source\": \"MaxMessageTermination\",\n \"models_usage\": null,\n \"content\": \"Maximum number of messages 3 reached, current message count: 3\"\n}", "type": "StopMessage"}
{"timestamp": "2024-12-06T17:57:11.738580", "message": "{\n \"source\": \"MaxMessageTermination\",\n \"models_usage\": null,\n \"content\": \"Maximum number of messages 3 reached, current message count: 3\"\n}", "type": "StopMessage"}
{"timestamp": "2024-12-06T17:57:11.738580", "message": "{\n \"source\": \"MaxMessageTermination\",\n \"models_usage\": null,\n \"content\": \"Maximum number of messages 3 reached, current message count: 3\"\n}", "type": "StopMessage"}
我们看到对话在代理发送指定数量的消息后终止。
StopMessageTermination#
在这种情况下，如果任何代理发送 StopMessage，团队就会终止对话。那么，代理什么时候会发送 StopMessage 呢？通常，这是在代理的 on_message 方法中实现的，代理可以检查传入的消息，并根据某些条件决定发送 StopMessage。
这里一个常见的模式是提示代理（或参与对话的某个代理）在其响应中发出特定的文本字符串，这可以用来触发终止条件。
事实上，如果你查看 AgentChat 提供的默认 CodingAssistantAgent 类的代码实现，你会发现两件事：
默认的
system_message指示代理在认为任务完成时以"terminate"一词结束其响应在
on_message方法中，代理检查传入的消息是否包含文本"terminate"，如果包含则返回StopMessage。
writing_assistant_agent = CodingAssistantAgent(
name="writing_assistant_agent",
system_message="You are a helpful assistant that solve tasks by generating text responses and code. Respond with TERMINATE when the task is done.",
model_client=model_client,
)
text_termination = TextMentionTermination("TERMINATE")
round_robin_team = RoundRobinGroupChat([writing_assistant_agent], termination_condition=text_termination)
round_robin_team_result = await round_robin_team.run(task="Write a unique, Haiku about the weather in Paris")
/tmp/ipykernel_326523/3843252945.py:1: DeprecationWarning: CodingAssistantAgent is deprecated. Use AssistantAgent instead.
writing_assistant_agent = CodingAssistantAgent(
{"timestamp": "2024-12-06T17:58:44.631351", "message": "{\n \"source\": \"user\",\n \"models_usage\": null,\n \"content\": \"Write a unique, Haiku about the weather in Paris\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:58:44.631351", "message": "{\n \"source\": \"user\",\n \"models_usage\": null,\n \"content\": \"Write a unique, Haiku about the weather in Paris\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:58:44.631351", "message": "{\n \"source\": \"user\",\n \"models_usage\": null,\n \"content\": \"Write a unique, Haiku about the weather in Paris\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:58:45.489890", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 48,\n \"completion_tokens\": 25\n },\n \"content\": \"Cobblestones glisten, \\nParis whispers in the rain\u2014 \\nClouds dance with the Seine. \\n\\nTERMINATE\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:58:45.489890", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 48,\n \"completion_tokens\": 25\n },\n \"content\": \"Cobblestones glisten, \\nParis whispers in the rain\u2014 \\nClouds dance with the Seine. \\n\\nTERMINATE\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:58:45.489890", "message": "{\n \"source\": \"writing_assistant_agent\",\n \"models_usage\": {\n \"prompt_tokens\": 48,\n \"completion_tokens\": 25\n },\n \"content\": \"Cobblestones glisten, \\nParis whispers in the rain\u2014 \\nClouds dance with the Seine. \\n\\nTERMINATE\"\n}", "type": "TextMessage"}
{"timestamp": "2024-12-06T17:58:45.490716", "message": "{\n \"source\": \"TextMentionTermination\",\n \"models_usage\": null,\n \"content\": \"Text 'TERMINATE' mentioned\"\n}", "type": "StopMessage"}
{"timestamp": "2024-12-06T17:58:45.490716", "message": "{\n \"source\": \"TextMentionTermination\",\n \"models_usage\": null,\n \"content\": \"Text 'TERMINATE' mentioned\"\n}", "type": "StopMessage"}
{"timestamp": "2024-12-06T17:58:45.490716", "message": "{\n \"source\": \"TextMentionTermination\",\n \"models_usage\": null,\n \"content\": \"Text 'TERMINATE' mentioned\"\n}", "type": "StopMessage"}