File size: 11,337 Bytes
4ad488c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 |
"""
Filename: MetaGPT/examples/debate.py
Created Date: Tuesday, September 19th 2023, 6:52:25 pm
Author: garylin2099
@Modified By: mashenquan, 2023-11-1. In accordance with Chapter 2.1.3 of RFC 116, modify the data type of the `send_to`
value of the `Message` object; modify the argument type of `get_by_actions`.
"""
import asyncio
import platform
from typing import Any
import fire
from metagpt.actions import Action, UserRequirement
from metagpt.logs import logger
from metagpt.roles import Role
from metagpt.schema import Message
from research_actions import CollectLinks, WebBrowseAndSummarize, ConductResearch
from metagpt.roles import Role
from metagpt.roles.role import RoleReactMode
import asyncio
import re
class RequestResearch(Action):
"""Action: Request research from researcher"""
PROMPT_TEMPLATE: str = """
## BACKGROUND
You are {name}, a {profile} preparing for a debate on: {topic}
## TASK
What specific aspect of this topic would you like to research to strengthen your position?
Provide ONE specific research query (1-2 sentences) that would help you in this debate.
Focus on facts, statistics, examples, or evidence that would support your {profile} perspective.
"""
name: str = "RequestResearch"
async def run(self, name: str, profile: str, topic: str) -> str:
prompt = self.PROMPT_TEMPLATE.format(name=name, profile=profile, topic=topic)
rsp = await self._aask(prompt)
return rsp
class SpeakAloud(Action):
"""Action: Speak out aloud in a debate (quarrel)"""
PROMPT_TEMPLATE: str = """
## BACKGROUND
Suppose you are {name}, you are in a debate with {opponent_name1} and {opponent_name2}. You are debating the topic:
{idea}
## DEBATE HISTORY
Previous rounds:
{context}
## RESEARCH INFORMATION
{research_info}
## YOUR TURN
{instruction}
"""
name: str = "SpeakAloud"
async def run(self, context: str, name: str, opponent_name1: str, opponent_name2: str, idea: str = "", profile: str = "", round_num: int = 1, research_info: str = "") -> str:
if round_num <= 3:
instruction = f"This is round {round_num} of 3 opening rounds. You should ONLY state your view on the topic, give your arguments and how you logically and rigorously arrived at your views. Do NOT rebut or respond to any of your opponents' arguments yet. Your viewpoint should be clear, concise, and extremely stereotypical of a {profile}. MANDATORY: Use specific facts, statistics, and evidence from your research information to support your arguments. Include proper citations in your response using [Source: URL or description] format."
else:
instruction = f"This is round {round_num}. You should defend your arguments, and attack and directly rebut your opponents' arguments if they differ from yours. Craft a strong, logically rigorous response in {name}'s rhetoric and viewpoints. MANDATORY: Support your arguments with specific evidence from your research and include citations using [Source: URL or description] format."
prompt = self.PROMPT_TEMPLATE.format(context=context, name=name, opponent_name1=opponent_name1, opponent_name2=opponent_name2, idea=idea, profile=profile, instruction=instruction, research_info=research_info)
rsp = await self._aask(prompt)
return rsp
class EvaluateDebate(Action):
"""Action: Evaluate and summarize a debate into concise recommendations"""
PROMPT_TEMPLATE: str = """
## ROLE
You are a neutral evaluator analyzing a debate to provide clear, actionable recommendations.
## DEBATE TOPIC
{topic}
## DEBATE CONTENT
{debate_content}
## TASK
Write a concise evaluation (200-300 words) that:
1. Summarizes key arguments from all participants
2. Identifies core trade-offs and considerations
3. Provides balanced, practical recommendations for decision-makers
4. Focuses on actionable solutions
Your response should be structured and help stakeholders make informed decisions.
"""
name: str = "EvaluateDebate"
async def run(self, topic: str, debate_content: str) -> str:
prompt = self.PROMPT_TEMPLATE.format(topic=topic, debate_content=debate_content)
rsp = await self._aask(prompt)
return rsp
class DebateEvaluator(Role):
name: str = "Evaluator"
profile: str = "Neutral Analyst"
def __init__(self, **data: Any):
super().__init__(**data)
self.set_actions([EvaluateDebate])
async def evaluate(self, topic: str, debate_messages: list) -> str:
todo = EvaluateDebate()
debate_content = "\n\n".join(f"{msg.sent_from}: {msg.content}" for msg in debate_messages)
evaluation = await todo.run(topic=topic, debate_content=debate_content)
return evaluation
class Researcher(Role):
"""Researcher role with three actions"""
name: str = "Researcher"
profile: str = "Research Assistant"
def __init__(self, **data):
super().__init__(**data)
self.set_actions([CollectLinks, WebBrowseAndSummarize, ConductResearch])
self._set_react_mode(RoleReactMode.BY_ORDER.value, len(self.actions))
async def research_topic(self, topic: str) -> str:
"""Conduct quick research on a topic"""
# Collect links (limited to 2 queries, 2 URLs each)
collect_action = CollectLinks()
links = await collect_action.run(topic, decomposition_nums=2, url_per_query=2)
# Browse and summarize (max 4 URLs total)
browse_action = WebBrowseAndSummarize()
summaries = []
url_count = 0
for query, urls in links.items():
if urls and url_count < 4:
remaining_urls = 4 - url_count
limited_urls = urls[:remaining_urls]
result = await browse_action.run(*limited_urls, query=query)
summaries.extend(result.values())
url_count += len(limited_urls)
# Conduct research
research_action = ConductResearch()
content = "\n---\n".join(summaries)
report = await research_action.run(topic, content)
return report
class Debator(Role):
name: str = ""
profile: str = ""
opponent_name1: str = ""
opponent_name2: str = ""
research_info: str = ""
research_count: int = 0
max_research: int = 1
def __init__(self, **data: Any):
super().__init__(**data)
self.set_actions([SpeakAloud])
self._watch([UserRequirement, SpeakAloud])
async def request_research(self, topic: str, researcher: Researcher) -> str:
"""Request research from the researcher"""
if self.research_count >= self.max_research:
return "Research limit reached"
request_action = RequestResearch()
query = await request_action.run(name=self.name, profile=self.profile, topic=topic)
# Get research report
research_result = await researcher.research_topic(query)
self.research_info += f"\n\nResearch Query: {query}\nResearch Result: {research_result}"
self.research_count += 1
return research_result
async def _observe(self) -> int:
await super()._observe()
# accept messages sent (from opponent) to self, disregard own messages from the last round
self.rc.news = [msg for msg in self.rc.news if self.name in msg.send_to or msg.send_to == {self.name}]
return len(self.rc.news)
async def _act(self) -> Message:
logger.info(f"{self._setting}: to do {self.rc.todo}({self.rc.todo.name})")
todo = self.rc.todo
memories = self.get_memories()
context = "\n".join(f"{msg.sent_from}: {msg.content}" for msg in memories)
# Get the debate topic from the first message
topic = self.get_memories()[0].content if self.get_memories() else "debate topic"
# Count how many times this speaker has spoken
speaker_turns = len([m for m in memories if m.sent_from == self.name])
rsp = await todo.run(context=context, name=self.name, opponent_name1=self.opponent_name1, opponent_name2=self.opponent_name2, idea=topic, profile=self.profile, round_num=speaker_turns + 1, research_info=self.research_info)
msg = Message(
content=rsp,
role=self.profile,
cause_by=type(todo),
sent_from=self.name,
send_to={self.opponent_name1, self.opponent_name2}
)
self.rc.memory.add(msg)
return msg
async def debate(idea: str, investment: float = 3.0, n_round: int = 5):
"""Run a debate and evaluate the results"""
School = Debator(name="Principal", profile="School", opponent_name1="John", opponent_name2 = "Mom")
Student = Debator(name="John", profile="Student", opponent_name1="Mom", opponent_name2 = "Principal")
Parent = Debator(name="Mom", profile="Parent", opponent_name1="Principal", opponent_name2 = "John")
# Create researcher
researcher = Researcher()
logger.info(f"Starting debate on: {idea}")
# Research phase - each debater gets 1 research only
debaters = [School, Student, Parent]
logger.info(f"\n=== Research Phase ===")
for debater in debaters:
logger.info(f"{debater.name} requesting research...")
await debater.request_research(idea, researcher)
# Start with Principal responding to the topic
current_speaker = School
second_speaker = Student
third_speaker = Parent
# Store all debate messages for evaluation
all_messages = []
# Initial message to start the debate
msg = Message(content=idea, role="user", send_to={"Principal"}, sent_from="User")
for round_num in range(n_round):
logger.info(f"\n=== Round {round_num + 1} ===\n{current_speaker.name}'s turn:")
# Current speaker responds
response = await current_speaker.run(msg)
logger.info(f"{current_speaker.name}: {response.content}")
# Store the response for evaluation
all_messages.append(response)
# Switch speakers for next round
current_speaker, second_speaker, third_speaker = third_speaker, current_speaker, second_speaker
msg = response
# Evaluate the debate
logger.info("\n=== EVALUATING DEBATE ===")
evaluator = DebateEvaluator()
evaluation = await evaluator.evaluate(idea, all_messages)
logger.info(f"\n=== EVALUATION RESULTS ===\n{evaluation}")
return evaluation
def main(idea: str, investment: float = 3.0, n_round: int = 6):
"""
:param idea: Debate topic, such as "Topic: The U.S. should commit more in climate change fighting"
or "Trump: Climate change is a hoax"
:param investment: contribute a certain dollar amount to watch the debate
:param n_round: maximum rounds of the debate
:return:
"""
if platform.system() == "Windows":
asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())
asyncio.run(debate(idea, investment, n_round))
if __name__ == "__main__":
fire.Fire(main) # run as python debate.py --idea="TOPIC" --investment=3.0 --n_round=5 |