[LangChain] Building Custom Tool

Default

import os
os.environ['OPENAI_API_KEY'] = 'sk-...'

Building Custom Tools

1) Package Import

from langchain import OpenAI
from langchain.chat_models import ChatOpenAI
from langchain.chains.conversation.memory import ConversationBufferWindowMemory 

# callback cost 
from langchain.callbacks import get_openai_callback

 

2) llm model 설정

turbo_llm = ChatOpenAI(
    temperature = 0,
    model_name = 'gpt-3.5-turbo'
)

 

3) 검색 Langchain pakage load

from langchain.tools import DuckDuckGoSearchRun
from langchain.agents import Tool
from langchain.tools import BaseTool

 

4) Search (부가기능)

Define Custom Tool

search = DuckDuckGoSearchRun()

tools = [
    Tool(
        name = 'search',
        func = search.run,
        description="useful for when you need to answer questions about current events. You should ask targeted questions"
        # current event에 대해 도움이 되는 답을 해라. 목표로 하는 질문을 해야한다.        
    )
]

 

5) LLM 모델의 답변이 바뀌지 않도록 고정

어떤 답이 나와도 삶은 달걀로 답변 하는 함수 정의

def meaning_of_life(input=""):
    return '삶은 달걀'

 

life_tool = Tool(
    name='삶이란 무엇인가?',
    func= meaning_of_life,
    description="Useful for when you need to answer questions about the meaning of life. input should be MOL "
)

 

6) Agent 생성

from langchain.agents import initialize_agent

tools = [search, life_tool]

 

답변시 참조할 이전 memory 개수(k=3) 설정

memory = ConversationBufferWindowMemory(
    memory_key='chat_history',
    k=3,
    return_messages=True
)

conversational_agent = initialize_agent(
    agent = 'chat-conversational-react-description',
    tools = tools,
    llm = turbo_llm,
    verbose = True,
    max_iterations = 3,
    early_stopping_method = 'generate',
    memory = memory
)

 

7) Agent 실행

conversational_agent("삶이란 무엇인가?")

LLM이 항상 Agent Tool을 모니터링한 상태에서 Task를 정의한다.