Update speaker selector in GroupChat and update some notebooks (#688)

* Add speaker selection methods

* Update groupchat RAG

* Update seed to cache_seed

* Update RetrieveChat notebook

* Update parameter name

* Add test

* Add more tests

* Add mock to test

* Add mock to test

* Fix typo speaking

* Add gracefully exit manual input

* Update round_robin docstring

* Add method checking

* Remove participant roles

* Fix versions in notebooks

* Minimize installation overhead

* Fix missing lower()

* Add comments for try_count 3

* Update warning for n_agents < 3

* Update warning for n_agents < 3

* Add test_n_agents_less_than_3

* Add a function for manual select

* Update version in notebooks

* Fixed bugs that allow speakers to go twice in a row even when allow_repeat_speaker = False

---------

Co-authored-by: Adam Fourney <adamfo@microsoft.com>
This commit is contained in:
Li Jiang 2023-11-17 21:56:11 +08:00 committed by GitHub
parent 3ab8c97eb6
commit 370ebf5e00
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 2033 additions and 3721 deletions

View File

@ -40,7 +40,7 @@ jobs:
python -m pip install --upgrade pip wheel
pip install -e .
python -c "import autogen"
pip install -e. pytest
pip install -e. pytest mock
pip uninstall -y openai
- name: Install unstructured if not windows
if: matrix.os != 'windows-2019'

View File

@ -1,5 +1,6 @@
import logging
import sys
import random
from dataclasses import dataclass
from typing import Dict, List, Optional, Union
import re
@ -21,6 +22,13 @@ class GroupChat:
When set to True and when a message is a function call suggestion,
the next speaker will be chosen from an agent which contains the corresponding function name
in its `function_map`.
- speaker_selection_method: the method for selecting the next speaker. Default is "auto".
Could be any of the following (case insensitive), will raise ValueError if not recognized:
- "auto": the next speaker is selected automatically by LLM.
- "manual": the next speaker is selected manually by user input.
- "random": the next speaker is selected randomly.
- "round_robin": the next speaker is selected in a round robin fashion, i.e., iterating in the same order as provided in `agents`.
- allow_repeat_speaker: whether to allow the same speaker to speak consecutively. Default is True.
"""
agents: List[Agent]
@ -28,6 +36,10 @@ class GroupChat:
max_round: int = 10
admin_name: str = "Admin"
func_call_filter: bool = True
speaker_selection_method: str = "auto"
allow_repeat_speaker: bool = True
_VALID_SPEAKER_SELECTION_METHODS = ["auto", "manual", "random", "round_robin"]
@property
def agent_names(self) -> List[str]:
@ -55,13 +67,61 @@ class GroupChat:
def select_speaker_msg(self, agents: List[Agent]):
"""Return the message for selecting the next speaker."""
return f"""You are in a role play game. The following roles are available:
{self._participant_roles()}.
{self._participant_roles(agents)}.
Read the following conversation.
Then select the next role from {[agent.name for agent in agents]} to play. Only return the role."""
def manual_select_speaker(self, agents: List[Agent]) -> Agent:
"""Manually select the next speaker."""
print("Please select the next speaker from the following list:")
_n_agents = len(agents)
for i in range(_n_agents):
print(f"{i+1}: {agents[i].name}")
try_count = 0
# Assume the user will enter a valid number within 3 tries, otherwise use auto selection to avoid blocking.
while try_count <= 3:
try_count += 1
if try_count >= 3:
print(f"You have tried {try_count} times. The next speaker will be selected automatically.")
break
try:
i = input("Enter the number of the next speaker (enter nothing or `q` to use auto selection): ")
if i == "" or i == "q":
break
i = int(i)
if i > 0 and i <= _n_agents:
return agents[i - 1]
else:
raise ValueError
except ValueError:
print(f"Invalid input. Please enter a number between 1 and {_n_agents}.")
return None
def select_speaker(self, last_speaker: Agent, selector: ConversableAgent):
"""Select the next speaker."""
if self.speaker_selection_method.lower() not in self._VALID_SPEAKER_SELECTION_METHODS:
raise ValueError(
f"GroupChat speaker_selection_method is set to '{self.speaker_selection_method}'. "
f"It should be one of {self._VALID_SPEAKER_SELECTION_METHODS} (case insensitive). "
)
agents = self.agents
n_agents = len(agents)
# Warn if GroupChat is underpopulated
if n_agents < 2:
raise ValueError(
f"GroupChat is underpopulated with {n_agents} agents. "
"Please add more agents to the GroupChat or use direct communication instead."
)
elif n_agents == 2 and self.speaker_selection_method.lower() != "round_robin" and self.allow_repeat_speaker:
logger.warning(
f"GroupChat is underpopulated with {n_agents} agents. "
"It is recommended to set speaker_selection_method to 'round_robin' or allow_repeat_speaker to False."
"Or, use direct communication instead."
)
if self.func_call_filter and self.messages and "function_call" in self.messages[-1]:
# find agents with the right function_map which contains the function name
agents = [
@ -80,14 +140,20 @@ Then select the next role from {[agent.name for agent in agents]} to play. Only
f"No agent can execute the function {self.messages[-1]['name']}. "
"Please check the function_map of the agents."
)
else:
agents = self.agents
# Warn if GroupChat is underpopulated
n_agents = len(agents)
if n_agents < 3:
logger.warning(
f"GroupChat is underpopulated with {n_agents} agents. Direct communication would be more efficient."
)
# remove the last speaker from the list to avoid selecting the same speaker if allow_repeat_speaker is False
agents = agents if self.allow_repeat_speaker else [agent for agent in agents if agent != last_speaker]
if self.speaker_selection_method.lower() == "manual":
selected_agent = self.manual_select_speaker(agents)
if selected_agent:
return selected_agent
elif self.speaker_selection_method.lower() == "round_robin":
return self.next_agent(last_speaker, agents)
elif self.speaker_selection_method.lower() == "random":
return random.choice(agents)
# auto speaker selection
selector.update_system_message(self.select_speaker_msg(agents))
final, name = selector.generate_oai_reply(
self.messages
@ -99,26 +165,31 @@ Then select the next role from {[agent.name for agent in agents]} to play. Only
]
)
if not final:
# i = self._random.randint(0, len(self._agent_names) - 1) # randomly pick an id
# the LLM client is None, thus no reply is generated. Use round robin instead.
return self.next_agent(last_speaker, agents)
# If exactly one agent is mentioned, use it. Otherwise, leave the OAI response unmodified
mentions = self._mentioned_agents(name, agents)
if len(mentions) == 1:
name = next(iter(mentions))
else:
logger.warning(
f"GroupChat select_speaker failed to resolve the next speaker's name. This is because the speaker selection OAI call returned:\n{name}"
)
# Return the result
try:
return self.agent_by_name(name)
except ValueError:
logger.warning(
f"GroupChat select_speaker failed to resolve the next speaker's name. Speaker selection will default to the next speaker in the list. This is because the speaker selection OAI call returned:\n{name}"
)
return self.next_agent(last_speaker, agents)
def _participant_roles(self):
def _participant_roles(self, agents: List[Agent] = None) -> str:
# Default to all agents registered
if agents is None:
agents = self.agents
roles = []
for agent in self.agents:
for agent in agents:
if agent.system_message.strip() == "":
logger.warning(
f"The agent '{agent.name}' has an empty system_message, and may not work well with GroupChat."

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -29,10 +29,19 @@
"\n",
"AutoGen requires `Python>=3.8`. To run this notebook example, please install the [retrievechat] option.\n",
"```bash\n",
"pip install \"pyautogen[retrievechat] flaml[automl] qdrant_client[fastembed]\"\n",
"pip install \"pyautogen[retrievechat]~=0.2.0b5\" \"flaml[automl]\" \"qdrant_client[fastembed]\"\n",
"```"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"# %pip install \"pyautogen[retrievechat]~=0.2.0b5\" \"flaml[automl]\" \"qdrant_client[fastembed]\""
]
},
{
"attachments": {},
"cell_type": "markdown",
@ -165,7 +174,7 @@
" system_message=\"You are a helpful assistant.\",\n",
" llm_config={\n",
" \"timeout\": 600,\n",
" \"seed\": 42,\n",
" \"cache_seed\": 42,\n",
" \"config_list\": config_list,\n",
" },\n",
")\n",
@ -1224,7 +1233,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.11.6"
"version": "3.10.12"
}
},
"nbformat": 4,

View File

@ -24,7 +24,7 @@
"\n",
"AutoGen requires `Python>=3.8`. To run this notebook example, please install:\n",
"```bash\n",
"pip install pyautogen\n",
"pip install \"pyautogen~=0.2.0b5\"\n",
"```"
]
},
@ -34,7 +34,7 @@
"metadata": {},
"outputs": [],
"source": [
"# %pip install --quiet pyautogen~=0.1.0"
"# %pip install --quiet \"pyautogen~=0.2.0b5\""
]
},
{
@ -85,7 +85,7 @@
"\n",
"llm_config={\n",
" \"timeout\": 600,\n",
" \"seed\": 44, # change the seed for different trials\n",
" \"cache_seed\": 44, # change the seed for different trials\n",
" \"config_list\": autogen.config_list_from_json(\n",
" \"OAI_CONFIG_LIST\",\n",
" filter_dict={\"model\": [\"gpt-4-32k\"]},\n",

View File

@ -46,6 +46,7 @@ setuptools.setup(
"pre-commit",
"pytest-asyncio",
"pytest>=6.1.1",
"mock",
],
"blendsearch": ["flaml[blendsearch]"],
"mathchat": ["sympy", "pydantic==1.10.9", "wolframalpha"],

View File

@ -1,4 +1,6 @@
import pytest
import mock
import builtins
import autogen
import json
@ -8,7 +10,7 @@ def test_func_call_groupchat():
"alice",
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is alice sepaking.",
default_auto_reply="This is alice speaking.",
)
agent2 = autogen.ConversableAgent(
"bob",
@ -56,7 +58,7 @@ def test_chat_manager():
max_consecutive_auto_reply=2,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is alice sepaking.",
default_auto_reply="This is alice speaking.",
)
agent2 = autogen.ConversableAgent(
"bob",
@ -83,6 +85,150 @@ def test_chat_manager():
agent2.initiate_chat(group_chat_manager, message={"function_call": {"name": "func", "arguments": '{"x": 1}'}})
def _test_selection_method(method: str):
agent1 = autogen.ConversableAgent(
"alice",
max_consecutive_auto_reply=10,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is alice speaking.",
)
agent2 = autogen.ConversableAgent(
"bob",
max_consecutive_auto_reply=10,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is bob speaking.",
)
agent3 = autogen.ConversableAgent(
"charlie",
max_consecutive_auto_reply=10,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is charlie speaking.",
)
groupchat = autogen.GroupChat(
agents=[agent1, agent2, agent3],
messages=[],
max_round=6,
speaker_selection_method=method,
allow_repeat_speaker=False if method == "manual" else True,
)
group_chat_manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=False)
if method == "round_robin":
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
assert len(agent1.chat_messages[group_chat_manager]) == 6
assert len(groupchat.messages) == 6
assert [msg["content"] for msg in agent1.chat_messages[group_chat_manager]] == [
"This is alice speaking.",
"This is bob speaking.",
"This is charlie speaking.",
] * 2
elif method == "auto":
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
assert len(agent1.chat_messages[group_chat_manager]) == 6
assert len(groupchat.messages) == 6
elif method == "random":
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
assert len(agent1.chat_messages[group_chat_manager]) == 6
assert len(groupchat.messages) == 6
elif method == "manual":
for user_input in ["", "q", "x", "1", "10"]:
with mock.patch.object(builtins, "input", lambda _: user_input):
group_chat_manager.reset()
agent1.reset()
agent2.reset()
agent3.reset()
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
if user_input == "1":
assert len(agent1.chat_messages[group_chat_manager]) == 6
assert len(groupchat.messages) == 6
assert [msg["content"] for msg in agent1.chat_messages[group_chat_manager]] == [
"This is alice speaking.",
"This is bob speaking.",
"This is alice speaking.",
"This is bob speaking.",
"This is alice speaking.",
"This is bob speaking.",
]
else:
assert len(agent1.chat_messages[group_chat_manager]) == 6
assert len(groupchat.messages) == 6
elif method == "wrong":
with pytest.raises(ValueError):
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
def test_speaker_selection_method():
for method in ["auto", "round_robin", "random", "manual", "wrong", "RounD_roBin"]:
_test_selection_method(method)
def _test_n_agents_less_than_3(method):
agent1 = autogen.ConversableAgent(
"alice",
max_consecutive_auto_reply=10,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is alice speaking.",
)
agent2 = autogen.ConversableAgent(
"bob",
max_consecutive_auto_reply=10,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is bob speaking.",
)
# test two agents
groupchat = autogen.GroupChat(
agents=[agent1, agent2],
messages=[],
max_round=6,
speaker_selection_method=method,
allow_repeat_speaker=True if method == "random" else False,
)
group_chat_manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=False)
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
assert len(agent1.chat_messages[group_chat_manager]) == 6
assert len(groupchat.messages) == 6
if method != "random" or method.lower() == "round_robin":
assert [msg["content"] for msg in agent1.chat_messages[group_chat_manager]] == [
"This is alice speaking.",
"This is bob speaking.",
] * 3
# test one agent
groupchat = autogen.GroupChat(
agents=[agent1],
messages=[],
max_round=6,
speaker_selection_method="round_robin",
allow_repeat_speaker=False,
)
with pytest.raises(ValueError):
group_chat_manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=False)
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
# test zero agent
groupchat = autogen.GroupChat(
agents=[],
messages=[],
max_round=6,
speaker_selection_method="round_robin",
allow_repeat_speaker=False,
)
with pytest.raises(ValueError):
group_chat_manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=False)
agent1.initiate_chat(group_chat_manager, message="This is alice speaking.")
def test_n_agents_less_than_3():
for method in ["auto", "round_robin", "random", "RounD_roBin"]:
_test_n_agents_less_than_3(method)
def test_plugin():
# Give another Agent class ability to manage group chat
agent1 = autogen.ConversableAgent(
@ -90,7 +236,7 @@ def test_plugin():
max_consecutive_auto_reply=2,
human_input_mode="NEVER",
llm_config=False,
default_auto_reply="This is alice sepaking.",
default_auto_reply="This is alice speaking.",
)
agent2 = autogen.ConversableAgent(
"bob",
@ -185,8 +331,10 @@ def test_agent_mentions():
if __name__ == "__main__":
test_func_call_groupchat()
# test_func_call_groupchat()
# test_broadcast()
test_chat_manager()
# test_chat_manager()
# test_plugin()
test_speaker_selection_method()
test_n_agents_less_than_3()
# test_agent_mentions()