ChatDev / chatdev /phase.py
ennet's picture
Duplicate from sp12138sp/ChatDev
594c559
raw
history blame
27.9 kB
import os
import re
from abc import ABC, abstractmethod
from camel.agents import RolePlaying
from camel.messages import ChatMessage
from camel.typing import TaskType, ModelType
from chatdev.chat_env import ChatEnv
from chatdev.statistics import get_info
from chatdev.utils import log_and_print_online, log_arguments
class Phase(ABC):
def __init__(self,
assistant_role_name,
user_role_name,
phase_prompt,
role_prompts,
phase_name,
model_type,
log_filepath):
"""
Args:
assistant_role_name: who receives chat in a phase
user_role_name: who starts the chat in a phase
phase_prompt: prompt of this phase
role_prompts: prompts of all roles
phase_name: name of this phase
"""
self.seminar_conclusion = None
self.assistant_role_name = assistant_role_name
self.user_role_name = user_role_name
self.phase_prompt = phase_prompt
self.phase_env = dict()
self.phase_name = phase_name
self.assistant_role_prompt = role_prompts[assistant_role_name]
self.user_role_prompt = role_prompts[user_role_name]
self.ceo_prompt = role_prompts["Chief Executive Officer"]
self.counselor_prompt = role_prompts["Counselor"]
self.timeout_seconds = 1.0
self.max_retries = 3
self.reflection_prompt = """Here is a conversation between two roles: {conversations} {question}"""
self.model_type = model_type
self.log_filepath = log_filepath
@log_arguments
def chatting(
self,
chat_env,
task_prompt: str,
assistant_role_name: str,
user_role_name: str,
phase_prompt: str,
phase_name: str,
assistant_role_prompt: str,
user_role_prompt: str,
task_type=TaskType.CHATDEV,
need_reflect=False,
with_task_specify=False,
model_type=ModelType.GPT_3_5_TURBO,
placeholders=None,
chat_turn_limit=10
) -> str:
"""
Args:
chat_env: global chatchain environment TODO: only for employee detection, can be deleted
task_prompt: user query prompt for building the software
assistant_role_name: who receives the chat
user_role_name: who starts the chat
phase_prompt: prompt of the phase
phase_name: name of the phase
assistant_role_prompt: prompt of assistant role
user_role_prompt: prompt of user role
task_type: task type
need_reflect: flag for checking reflection
with_task_specify: with task specify
model_type: model type
placeholders: placeholders for phase environment to generate phase prompt
chat_turn_limit: turn limits in each chat
Returns:
"""
if placeholders is None:
placeholders = {}
assert 1 <= chat_turn_limit <= 100
if not chat_env.exist_employee(assistant_role_name):
raise ValueError(f"{assistant_role_name} not recruited in ChatEnv.")
if not chat_env.exist_employee(user_role_name):
raise ValueError(f"{user_role_name} not recruited in ChatEnv.")
# init role play
role_play_session = RolePlaying(
assistant_role_name=assistant_role_name,
user_role_name=user_role_name,
assistant_role_prompt=assistant_role_prompt,
user_role_prompt=user_role_prompt,
task_prompt=task_prompt,
task_type=task_type,
with_task_specify=with_task_specify,
model_type=model_type,
)
# log_and_print_online("System", role_play_session.assistant_sys_msg)
# log_and_print_online("System", role_play_session.user_sys_msg)
# start the chat
_, input_user_msg = role_play_session.init_chat(None, placeholders, phase_prompt)
seminar_conclusion = None
# handle chats
# the purpose of the chatting in one phase is to get a seminar conclusion
# there are two types of conclusion
# 1. with "<INFO>" mark
# 1.1 get seminar conclusion flag (ChatAgent.info) from assistant or user role, which means there exist special "<INFO>" mark in the conversation
# 1.2 add "<INFO>" to the reflected content of the chat (which may be terminated chat without "<INFO>" mark)
# 2. without "<INFO>" mark, which means the chat is terminated or normally ended without generating a marked conclusion, and there is no need to reflect
for i in range(chat_turn_limit):
# start the chat, we represent the user and send msg to assistant
# 1. so the input_user_msg should be assistant_role_prompt + phase_prompt
# 2. then input_user_msg send to LLM and get assistant_response
# 3. now we represent the assistant and send msg to user, so the input_assistant_msg is user_role_prompt + assistant_response
# 4. then input_assistant_msg send to LLM and get user_response
# all above are done in role_play_session.step, which contains two interactions with LLM
# the first interaction is logged in role_play_session.init_chat
assistant_response, user_response = role_play_session.step(input_user_msg, chat_turn_limit == 1)
conversation_meta = "**" + assistant_role_name + "<->" + user_role_name + " on : " + str(
phase_name) + ", turn " + str(i) + "**\n\n"
# TODO: max_tokens_exceeded errors here
if isinstance(assistant_response.msg, ChatMessage):
# we log the second interaction here
log_and_print_online(role_play_session.assistant_agent.role_name,
conversation_meta + "[" + role_play_session.user_agent.system_message.content + "]\n\n" + assistant_response.msg.content)
if role_play_session.assistant_agent.info:
seminar_conclusion = assistant_response.msg.content
break
if assistant_response.terminated:
break
if isinstance(user_response.msg, ChatMessage):
# here is the result of the second interaction, which may be used to start the next chat turn
log_and_print_online(role_play_session.user_agent.role_name,
conversation_meta + "[" + role_play_session.assistant_agent.system_message.content + "]\n\n" + user_response.msg.content)
if role_play_session.user_agent.info:
seminar_conclusion = user_response.msg.content
break
if user_response.terminated:
break
# continue the chat
if chat_turn_limit > 1 and isinstance(user_response.msg, ChatMessage):
input_user_msg = user_response.msg
else:
break
# conduct self reflection
if need_reflect:
if seminar_conclusion in [None, ""]:
seminar_conclusion = "<INFO> " + self.self_reflection(task_prompt, role_play_session, phase_name,
chat_env)
if "recruiting" in phase_name:
if "Yes".lower() not in seminar_conclusion.lower() and "No".lower() not in seminar_conclusion.lower():
seminar_conclusion = "<INFO> " + self.self_reflection(task_prompt, role_play_session,
phase_name,
chat_env)
elif seminar_conclusion in [None, ""]:
seminar_conclusion = "<INFO> " + self.self_reflection(task_prompt, role_play_session, phase_name,
chat_env)
else:
seminar_conclusion = assistant_response.msg.content
log_and_print_online("**[Seminar Conclusion]**:\n\n {}".format(seminar_conclusion))
seminar_conclusion = seminar_conclusion.split("<INFO>")[-1]
return seminar_conclusion
def self_reflection(self,
task_prompt: str,
role_play_session: RolePlaying,
phase_name: str,
chat_env: ChatEnv) -> str:
"""
Args:
task_prompt: user query prompt for building the software
role_play_session: role play session from the chat phase which needs reflection
phase_name: name of the chat phase which needs reflection
chat_env: global chatchain environment
Returns:
reflected_content: str, reflected results
"""
messages = role_play_session.assistant_agent.stored_messages if len(
role_play_session.assistant_agent.stored_messages) >= len(
role_play_session.user_agent.stored_messages) else role_play_session.user_agent.stored_messages
messages = ["{}: {}".format(message.role_name, message.content.replace("\n\n", "\n")) for message in messages]
messages = "\n\n".join(messages)
if "recruiting" in phase_name:
question = """Answer their final discussed conclusion (Yes or No) in the discussion without any other words, e.g., "Yes" """
elif phase_name == "DemandAnalysis":
question = """Answer their final product modality in the discussion without any other words, e.g., "PowerPoint" """
# elif phase_name in [PhaseType.BRAINSTORMING]:
# question = """Conclude three most creative and imaginative brainstorm ideas from the whole discussion, in the format: "1) *; 2) *; 3) *; where '*' represents a suggestion." """
elif phase_name == "LanguageChoose":
question = """Conclude the programming language being discussed for software development, in the format: "*" where '*' represents a programming language." """
elif phase_name == "EnvironmentDoc":
question = """According to the codes and file format listed above, write a requirements.txt file to specify the dependencies or packages required for the project to run properly." """
else:
raise ValueError(f"Reflection of phase {phase_name}: Not Assigned.")
# Reflections actually is a special phase between CEO and counselor
# They read the whole chatting history of this phase and give refined conclusion of this phase
reflected_content = \
self.chatting(chat_env=chat_env,
task_prompt=task_prompt,
assistant_role_name="Chief Executive Officer",
user_role_name="Counselor",
phase_prompt=self.reflection_prompt,
phase_name="Reflection",
assistant_role_prompt=self.ceo_prompt,
user_role_prompt=self.counselor_prompt,
placeholders={"conversations": messages, "question": question},
need_reflect=False,
chat_turn_limit=1,
model_type=self.model_type)
if "recruiting" in phase_name:
if "Yes".lower() in reflected_content.lower():
return "Yes"
return "No"
else:
return reflected_content
@abstractmethod
def update_phase_env(self, chat_env):
"""
update self.phase_env (if needed) using chat_env, then the chatting will use self.phase_env to follow the context and fill placeholders in phase prompt
must be implemented in customized phase
the usual format is just like:
```
self.phase_env.update({key:chat_env[key]})
```
Args:
chat_env: global chat chain environment
Returns: None
"""
pass
@abstractmethod
def update_chat_env(self, chat_env) -> ChatEnv:
"""
update chan_env based on the results of self.execute, which is self.seminar_conclusion
must be implemented in customized phase
the usual format is just like:
```
chat_env.xxx = some_func_for_postprocess(self.seminar_conclusion)
```
Args:
chat_env:global chat chain environment
Returns:
chat_env: updated global chat chain environment
"""
pass
def execute(self, chat_env, chat_turn_limit, need_reflect) -> ChatEnv:
"""
execute the chatting in this phase
1. receive information from environment: update the phase environment from global environment
2. execute the chatting
3. change the environment: update the global environment using the conclusion
Args:
chat_env: global chat chain environment
chat_turn_limit: turn limit in each chat
need_reflect: flag for reflection
Returns:
chat_env: updated global chat chain environment using the conclusion from this phase execution
"""
self.update_phase_env(chat_env)
self.seminar_conclusion = \
self.chatting(chat_env=chat_env,
task_prompt=chat_env.env_dict['task_prompt'],
need_reflect=need_reflect,
assistant_role_name=self.assistant_role_name,
user_role_name=self.user_role_name,
phase_prompt=self.phase_prompt,
phase_name=self.phase_name,
assistant_role_prompt=self.assistant_role_prompt,
user_role_prompt=self.user_role_prompt,
chat_turn_limit=chat_turn_limit,
placeholders=self.phase_env,
model_type=self.model_type)
chat_env = self.update_chat_env(chat_env)
return chat_env
class DemandAnalysis(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
pass
def update_chat_env(self, chat_env) -> ChatEnv:
if len(self.seminar_conclusion) > 0:
chat_env.env_dict['modality'] = self.seminar_conclusion.split("<INFO>")[-1].lower().replace(".", "").strip()
return chat_env
class LanguageChoose(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas']})
def update_chat_env(self, chat_env) -> ChatEnv:
if len(self.seminar_conclusion) > 0 and "<INFO>" in self.seminar_conclusion:
chat_env.env_dict['language'] = self.seminar_conclusion.split("<INFO>")[-1].lower().replace(".", "").strip()
elif len(self.seminar_conclusion) > 0:
chat_env.env_dict['language'] = self.seminar_conclusion
else:
chat_env.env_dict['language'] = "Python"
return chat_env
class Coding(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
gui = "" if not chat_env.config.gui_design \
else "The software should be equipped with graphical user interface (GUI) so that user can visually and graphically use it; so you must choose a GUI framework (e.g., in Python, you can implement GUI via tkinter, Pygame, Flexx, PyGUI, etc,)."
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"gui": gui})
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env.update_codes(self.seminar_conclusion)
if len(chat_env.codes.codebooks.keys()) == 0:
raise ValueError("No Valid Codes.")
chat_env.rewrite_codes()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class ArtDesign(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env = {"task": chat_env.env_dict['task_prompt'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes()}
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env.proposed_images = chat_env.get_proposed_images_from_message(self.seminar_conclusion)
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class ArtIntegration(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env = {"task": chat_env.env_dict['task_prompt'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"images": "\n".join(
["{}: {}".format(filename, chat_env.proposed_images[filename]) for
filename in sorted(list(chat_env.proposed_images.keys()))])}
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env.update_codes(self.seminar_conclusion)
chat_env.rewrite_codes()
# chat_env.generate_images_from_codes()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class CodeComplete(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"unimplemented_file": ""})
unimplemented_file = ""
for filename in self.phase_env['pyfiles']:
code_content = open(os.path.join(chat_env.env_dict['directory'], filename)).read()
lines = [line.strip() for line in code_content.split("\n") if line.strip() == "pass"]
if len(lines) > 0 and self.phase_env['num_tried'][filename] < self.phase_env['max_num_implement']:
unimplemented_file = filename
break
self.phase_env['num_tried'][unimplemented_file] += 1
self.phase_env['unimplemented_file'] = unimplemented_file
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env.update_codes(self.seminar_conclusion)
if len(chat_env.codes.codebooks.keys()) == 0:
raise ValueError("No Valid Codes.")
chat_env.rewrite_codes()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class CodeReviewComment(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update(
{"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"images": ", ".join(chat_env.incorporated_images)})
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env.env_dict['review_comments'] = self.seminar_conclusion
return chat_env
class CodeReviewModification(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"comments": chat_env.env_dict['review_comments']})
def update_chat_env(self, chat_env) -> ChatEnv:
if "```".lower() in self.seminar_conclusion.lower():
chat_env.update_codes(self.seminar_conclusion)
chat_env.rewrite_codes()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
self.phase_env['modification_conclusion'] = self.seminar_conclusion
return chat_env
class CodeReviewHuman(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
print(
f"You can participate in the development of the software {chat_env.env_dict['task_prompt']}. Please input your feedback. (\"End\" to quit the involvement.)")
provided_comments = input()
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"comments": provided_comments})
def update_chat_env(self, chat_env) -> ChatEnv:
if "```".lower() in self.seminar_conclusion.lower():
chat_env.update_codes(self.seminar_conclusion)
chat_env.rewrite_codes()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class TestErrorSummary(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
chat_env.generate_images_from_codes()
(exist_bugs_flag, test_reports) = chat_env.exist_bugs()
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"test_reports": test_reports,
"exist_bugs_flag": exist_bugs_flag})
log_and_print_online("**[Test Reports]**:\n\n{}".format(test_reports))
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env.env_dict['error_summary'] = self.seminar_conclusion
chat_env.env_dict['test_reports'] = self.phase_env['test_reports']
return chat_env
def execute(self, chat_env, chat_turn_limit, need_reflect) -> ChatEnv:
self.update_phase_env(chat_env)
if "ModuleNotFoundError" in self.phase_env['test_reports']:
chat_env.fix_module_not_found_error(self.phase_env['test_reports'])
log_and_print_online(
f"Software Test Engineer found ModuleNotFoundError:\n{self.phase_env['test_reports']}\n")
pip_install_content = ""
for match in re.finditer(r"No module named '(\S+)'", self.phase_env['test_reports'], re.DOTALL):
module = match.group(1)
pip_install_content += "{}\n```{}\n{}\n```\n".format("cmd", "bash", f"pip install {module}")
log_and_print_online(f"Programmer resolve ModuleNotFoundError by:\n{pip_install_content}\n")
self.seminar_conclusion = "nothing need to do"
else:
self.seminar_conclusion = \
self.chatting(chat_env=chat_env,
task_prompt=chat_env.env_dict['task_prompt'],
need_reflect=need_reflect,
assistant_role_name=self.assistant_role_name,
user_role_name=self.user_role_name,
phase_prompt=self.phase_prompt,
phase_name=self.phase_name,
assistant_role_prompt=self.assistant_role_prompt,
user_role_prompt=self.user_role_prompt,
chat_turn_limit=chat_turn_limit,
placeholders=self.phase_env)
chat_env = self.update_chat_env(chat_env)
return chat_env
class TestModification(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"test_reports": chat_env.env_dict['test_reports'],
"error_summary": chat_env.env_dict['error_summary'],
"codes": chat_env.get_codes()
})
def update_chat_env(self, chat_env) -> ChatEnv:
if "```".lower() in self.seminar_conclusion.lower():
chat_env.update_codes(self.seminar_conclusion)
chat_env.rewrite_codes()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class EnvironmentDoc(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes()})
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env._update_requirements(self.seminar_conclusion)
chat_env.rewrite_requirements()
log_and_print_online("**[Software Info]**:\n\n {}".format(get_info(chat_env.env_dict['directory'],self.log_filepath)))
return chat_env
class Manual(Phase):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def update_phase_env(self, chat_env):
self.phase_env.update({"task": chat_env.env_dict['task_prompt'],
"modality": chat_env.env_dict['modality'],
"ideas": chat_env.env_dict['ideas'],
"language": chat_env.env_dict['language'],
"codes": chat_env.get_codes(),
"requirements": chat_env.get_requirements()})
def update_chat_env(self, chat_env) -> ChatEnv:
chat_env._update_manuals(self.seminar_conclusion)
chat_env.rewrite_manuals()
return chat_env