Files
sci-gui-agent-benchmark/mm_agents/coact/autogen/agentchat/group/patterns/round_robin.py
2025-07-31 10:35:20 +08:00

118 lines
3.8 KiB
Python

# Copyright (c) 2023 - 2025, AG2ai, Inc., AG2ai open-source projects maintainers and core contributors
#
# SPDX-License-Identifier: Apache-2.0
from typing import TYPE_CHECKING, Any, Optional, Tuple, Union
from ..context_variables import ContextVariables
from ..targets.transition_target import AgentTarget, TransitionTarget
from .pattern import Pattern
if TYPE_CHECKING:
from ...conversable_agent import ConversableAgent
from ...groupchat import GroupChat, GroupChatManager
from ..group_tool_executor import GroupToolExecutor
class RoundRobinPattern(Pattern):
"""RoundRobinPattern implements a round robin with handoffs between agents."""
def _generate_handoffs(
self,
initial_agent: "ConversableAgent",
agents: list["ConversableAgent"],
user_agent: Optional["ConversableAgent"],
) -> None:
"""Generate handoffs between agents in a round-robin fashion."""
# Create a list of the agents and the user_agent but put the initial_agent first
agent_list = [initial_agent]
# Add the rest of the agents, excluding the initial_agent and user_agent
for agent in agents:
if agent != initial_agent and (user_agent is None or agent != user_agent):
agent_list.append(agent)
# Add the user_agent last if it exists
if user_agent is not None:
agent_list.append(user_agent)
# Create handoffs in a round-robin fashion
for i, agent in enumerate(agent_list):
# Last agent hands off to the first agent
# Otherwise agent hands off to the next one
handoff_target = agent_list[0] if i == len(agent_list) - 1 else agent_list[i + 1]
agent.handoffs.set_after_work(target=AgentTarget(agent=handoff_target))
def prepare_group_chat(
self,
max_rounds: int,
messages: Union[list[dict[str, Any]], str],
) -> Tuple[
list["ConversableAgent"],
list["ConversableAgent"],
Optional["ConversableAgent"],
ContextVariables,
"ConversableAgent",
TransitionTarget,
"GroupToolExecutor",
"GroupChat",
"GroupChatManager",
list[dict[str, Any]],
Any,
list[str],
list[Any],
]:
"""Prepare the group chat for organic agent selection.
Ensures that:
1. The group manager has a valid LLM config
2. All agents have appropriate descriptions for the group manager to use
Args:
max_rounds: Maximum number of conversation rounds.
messages: Initial message(s) to start the conversation.
Returns:
Tuple containing all necessary components for the group chat.
"""
# Use the parent class's implementation to prepare the agents and group chat
(
agents,
wrapped_agents,
user_agent,
context_variables,
initial_agent,
group_after_work,
tool_executor,
groupchat,
manager,
processed_messages,
last_agent,
group_agent_names,
temp_user_list,
) = super().prepare_group_chat(
max_rounds=max_rounds,
messages=messages,
)
# Create the handoffs between agents
self._generate_handoffs(initial_agent=initial_agent, agents=agents, user_agent=user_agent)
# Return all components with our group_after_work
return (
agents,
wrapped_agents,
user_agent,
context_variables,
initial_agent,
group_after_work,
tool_executor,
groupchat,
manager,
processed_messages,
last_agent,
group_agent_names,
temp_user_list,
)