SuperExpert / agents /meta_agent.py
JarvisChan630's picture
fix bugs
fc95199
raw
history blame
16.3 kB
import json
import logging
from termcolor import colored
from datetime import datetime
from typing import Any, Dict, Union, List
from typing import TypedDict, Annotated
from langgraph.graph.message import add_messages
from agents.base_agent import BaseAgent
from utils.read_markdown import read_markdown_file
from tools.advanced_scraper import scraper
from tools.google_serper import serper_search
from utils.logging import log_function, setup_logging
from utils.message_handling import get_ai_message_contents
from prompt_engineering.guided_json_lib import guided_json_search_query, guided_json_best_url, guided_json_router_decision
setup_logging(level=logging.DEBUG)
logger = logging.getLogger(__name__)
class MessageDict(TypedDict):
role: str
content: str
class State(TypedDict):
meta_prompt: Annotated[List[MessageDict], add_messages]
conversation_history: Annotated[List[dict], add_messages]
user_input: Annotated[List[str], add_messages]
router_decision: bool
chat_limit: int
chat_finished: bool
recursion_limit: int
state: State = {
"meta_prompt": [],
"conversation_history": [],
"user_input": [],
"router_decision": None,
"chat_limit": None,
"chat_finished": False,
"recursion_limit": None
}
# class State(TypedDict):
# meta_prompt: Annotated[List[MessageDict], add_messages]
# conversation_history: Annotated[List[dict], add_messages]
# user_input: Annotated[List[str], add_messages]
# router_decision: bool
# chat_limit: int
# chat_finished: bool
# state: State = {
# "meta_prompt": [],
# "conversation_history": [],
# "user_input": [],
# "router_decision": None,
# "chat_limit": None,
# "chat_finished": False
# }
# def chat_counter(state: State) -> State:
# chat_limit = state.get("chat_limit")
# if chat_limit is None:
# chat_limit = 0
# chat_limit += 1
# state["chat_limit"] = chat_limit
# return state
# def chat_counter(state: State) -> State:
# chat_limit = state.get("chat_limit")
# if chat_limit is None:
# chat_limit = 0
# chat_limit += 1
# state["chat_limit"] = chat_limit
# return chat_limit
def routing_function(state: State) -> str:
if state["router_decision"]:
return "no_tool_expert"
else:
return "tool_expert"
def set_chat_finished(state: State) -> bool:
state["chat_finished"] = True
final_response = state["meta_prompt"][-1].content
print(colored(f"\n\n Meta Agent 🧙‍♂️: {final_response}", 'cyan'))
return state
class MetaExpert(BaseAgent[State]):
def __init__(self, model: str = None, server: str = None, temperature: float = 0,
model_endpoint: str = None, stop: str = None):
super().__init__(model, server, temperature, model_endpoint, stop)
self.llm = self.get_llm(json_model=False)
def get_prompt(self, state:None) -> str:
system_prompt = read_markdown_file('prompt_engineering/meta_prompt.md')
return system_prompt
def process_response(self, response: Any, user_input: str = None, state: State = None) -> Dict[str, List[MessageDict]]:
user_input = None
updates_conversation_history = {
"meta_prompt": [
{"role": "user", "content": f"{user_input}"},
{"role": "assistant", "content": str(response)}
]
}
return updates_conversation_history
@log_function(logger)
def get_conv_history(self, state: State) -> str:
conversation_history = state.get("conversation_history", [])
expert_message_history = get_ai_message_contents(conversation_history)
print(f"Expert Data Collected: {expert_message_history}")
expert_message_history = f"Expert Data Collected: <Ex>{expert_message_history}</Ex>"
return expert_message_history
def get_user_input(self) -> str:
user_input = input("Enter your query: ")
return user_input
def get_guided_json(self, state: State) -> Dict[str, Any]:
pass
def use_tool(self) -> Any:
pass
@log_function(logger)
def run(self, state: State) -> State:
# counter = chat_counter(state)
user_input = state.get("user_input")
state = self.invoke(state=state, user_input=user_input)
return state
class NoToolExpert(BaseAgent[State]):
def __init__(self, model: str = None, server: str = None, temperature: float = 0,
model_endpoint: str = None, stop: str = None):
super().__init__(model, server, temperature, model_endpoint, stop)
self.llm = self.get_llm(json_model=False)
def get_prompt(self, state) -> str:
# print(f"\nn{state}\n")
system_prompt = state["meta_prompt"][-1].content
return system_prompt
def process_response(self, response: Any, user_input: str = None, state: State = None) -> Dict[str, Union[str, dict]]:
updates_conversation_history = {
"conversation_history": [
{"role": "user", "content": user_input},
{"role": "assistant", "content": f"{str(response)}"}
]
}
return updates_conversation_history
def get_conv_history(self, state: State) -> str:
pass
def get_user_input(self) -> str:
pass
def get_guided_json(self, state: State) -> Dict[str, Any]:
pass
def use_tool(self) -> Any:
pass
# @log_function(logger)
def run(self, state: State) -> State:
# chat_counter(state)
user_input = state["meta_prompt"][1].content
state = self.invoke(state=state, user_input=user_input)
return state
class ToolExpert(BaseAgent[State]):
def __init__(self, model: str = None, server: str = None, temperature: float = 0,
model_endpoint: str = None, stop: str = None):
super().__init__(model, server, temperature, model_endpoint, stop)
self.llm = self.get_llm(json_model=False)
def get_prompt(self, state) -> str:
system_prompt = state["meta_prompt"][-1].content
return system_prompt
def process_response(self, response: Any, user_input: str = None, state: State = None) -> Dict[str, Union[str, dict]]:
updates_conversation_history = {
"conversation_history": [
{"role": "user", "content": user_input},
{"role": "assistant", "content": f"{str(response)}"}
]
}
return updates_conversation_history
def get_conv_history(self, state: State) -> str:
pass
def get_user_input(self) -> str:
pass
def get_guided_json(self, state: State) -> Dict[str, Any]:
pass
# Use Serper to search
def use_tool(self, mode: str, tool_input: str, doc_type: str = None) -> Any:
if mode == "serper":
results = serper_search(tool_input, self.location)
return results
elif mode == "scraper":
results = scraper(tool_input, doc_type)
return results
# @log_function(logger)
def run(self, state: State) -> State:
# counter = chat_counter(state)
refine_query_template = """
Given the response from your manager.
# Response from Manager
{manager_response}
**Return the following JSON:**
{{"search_query": The refined google search engine query that aligns with the response from your managers.}}
"""
best_url_template = """
Given the serper results, and the instructions from your manager. Select the best URL
# Manger Instructions
{manager_response}
# Serper Results
{serper_results}
**Return the following JSON:**
{{"best_url": The URL of the serper results that aligns most with the instructions from your manager.,
"pdf": A boolean value indicating whether the URL is a PDF or not. This should be True if the URL is a PDF, and False otherwise.}}
"""
user_input = state["meta_prompt"][-1].content
state = self.invoke(state=state, user_input=user_input)
full_query = state["conversation_history"][-1].get("content")
refine_query = self.get_llm(json_model=True)
refine_prompt = refine_query_template.format(manager_response=full_query)
input = [
{"role": "user", "content": full_query},
{"role": "assistant", "content": f"system_prompt:{refine_prompt}"}
]
if self.server == 'vllm':
guided_json = guided_json_search_query
refined_query = refine_query.invoke(input, guided_json)
else:
refined_query = refine_query.invoke(input)
refined_query_json = json.loads(refined_query)
refined_query = refined_query_json.get("search_query")
serper_response = self.use_tool("serper", refined_query)
best_url = self.get_llm(json_model=True)
best_url_prompt = best_url_template.format(manager_response=full_query, serper_results=serper_response)
input = [
{"role": "user", "content": serper_response},
{"role": "assistant", "content": f"system_prompt:{best_url_prompt}"}
]
if self.server == 'vllm':
guided_json = guided_json_best_url
best_url = best_url.invoke(input, guided_json)
else:
best_url = best_url.invoke(input)
best_url_json = json.loads(best_url)
best_url = best_url_json.get("best_url")
doc_type = best_url_json.get("pdf")
if doc_type == "True" or doc_type == True:
doc_type = "pdf"
else:
doc_type = "html"
scraper_response = self.use_tool("scraper", best_url, doc_type)
updates = self.process_response(scraper_response, user_input)
for key, value in updates.items():
state = self.update_state(key, value, state)
return state
class Router(BaseAgent[State]):
def __init__(self, model: str = None, server: str = None, temperature: float = 0,
model_endpoint: str = None, stop: str = None):
super().__init__(model, server, temperature, model_endpoint, stop)
self.llm = self.get_llm(json_model=True)
def get_prompt(self, state) -> str:
system_prompt = state["meta_prompt"][-1].content
return system_prompt
def process_response(self, response: Any, user_input: str = None, state: State = None) -> Dict[str, Union[str, dict]]:
updates_conversation_history = {
"router_decision": [
{"role": "user", "content": user_input},
{"role": "assistant", "content": f"<Ex>{str(response)}</Ex> Todays date is {datetime.now()}"}
]
}
return updates_conversation_history
def get_conv_history(self, state: State) -> str:
pass
def get_user_input(self) -> str:
pass
def get_guided_json(self, state: State) -> Dict[str, Any]:
pass
def use_tool(self, tool_input: str, mode: str) -> Any:
pass
# @log_function(logger)
def run(self, state: State) -> State:
# router_template = """
# Given these instructions from your manager.
# # Response from Manager
# {manager_response}
# **Return the following JSON:**
# {{""router_decision: Return the next agent to pass control to.}}
# **strictly** adhere to these **guidelines** for routing.
# If your manager's response suggests a tool might be required to answer the query, return "tool_expert".
# If your manager's response suggests no tool is required to answer the query, return "no_tool_expert".
# If your manager's response suggest they have provided a final answer, return "end_chat".
# """
# chat_counter(state)
router_template = """
Given these instructions from your manager.
# Response from Manager
{manager_response}
**Return the following JSON:**
{{""router_decision: Return the next agent to pass control to.}}
**strictly** adhere to these **guidelines** for routing.
If your manager's response suggests the Expert Internet Researcher or the suggests the internet might be required, return "tool_expert".
If your manager's response suggests that the internet is not required, return "no_tool_expert".
If your manager's response suggest they have provided a final answer, return "end_chat".
"""
system_prompt = router_template.format(manager_response=state["meta_prompt"][-1].content)
input = [
{"role": "user", "content": ""},
{"role": "assistant", "content": f"system_prompt:{system_prompt}"}
]
router = self.get_llm(json_model=True)
if self.server == 'vllm':
guided_json = guided_json_router_decision
router_response = router.invoke(input, guided_json)
else:
router_response = router.invoke(input)
router_response = json.loads(router_response)
router_response = router_response.get("router_decision")
state = self.update_state("router_decision", router_response, state)
return state
# Example usage
if __name__ == "__main__":
from langgraph.graph import StateGraph
# For Claude
agent_kwargs = {
"model": "claude-3-haiku-20240307",
"server": "claude",
"temperature": 0.5
}
agent_kwargs = {
"model": "gpt-4o-mini",
"server": "openai",
"temperature": 0.1
}
# Ollama
# agent_kwargs = {
# "model": "phi3:instruct",
# "server": "ollama",
# "temperature": 0.5
# }
# Groq
# agent_kwargs = {
# "model": "mixtral-8x7b-32768",
# "server": "groq",
# "temperature": 0.5
# }
# # Gemnin - Not currently working, I will be debugging this soon.
# agent_kwargs = {
# "model": "gemini-1.5-pro",
# "server": "gemini",
# "temperature": 0.5
# }
# # Vllm
# agent_kwargs = {
# "model": "meta-llama/Meta-Llama-3-70B-Instruct",
# "server": "vllm",
# "temperature": 0.5,
# "model_endpoint": "https://vpzatdgopr2pmx-8000.proxy.runpod.net/",
# }
tools_router_agent_kwargs = agent_kwargs.copy()
tools_router_agent_kwargs["temperature"] = 0
def routing_function(state: State) -> str:
decision = state["router_decision"]
print(colored(f"\n\n Routing function called. Decision: {decision}", 'red'))
return decision
graph = StateGraph(State)
graph.add_node("meta_expert", lambda state: MetaExpert(**agent_kwargs).run(state=state))
graph.add_node("router", lambda state: Router(**tools_router_agent_kwargs).run(state=state))
graph.add_node("no_tool_expert", lambda state: NoToolExpert(**agent_kwargs).run(state=state))
graph.add_node("tool_expert", lambda state: ToolExpert(**tools_router_agent_kwargs).run(state=state))
graph.add_node("end_chat", lambda state: set_chat_finished(state))
graph.set_entry_point("meta_expert")
graph.set_finish_point("end_chat")
graph.add_edge("meta_expert", "router")
graph.add_edge("tool_expert", "meta_expert")
graph.add_edge("no_tool_expert", "meta_expert")
graph.add_conditional_edges(
"router",
lambda state: routing_function(state),
)
workflow = graph.compile()
while True:
query = input("Ask me anything: ")
if query.lower() == "exit":
break
# current_time = datetime.now()
recursion_limit = 30
state["recursion_limit"] = recursion_limit
state["user_input"] = query
limit = {"recursion_limit": recursion_limit}
for event in workflow.stream(state, limit):
pass