GPT_1(노마드코더)

주피터 노트북 사용법

 

파일확장자 (.ipynb)

 

커널선택 -> 생성했던 가상환경 선택

 

프린트를 계속해서 찍어보지 않아도 됨

 

========================================================================================

 

1. LangChain 사용

 

LLM 모델 호출 (text-davinci)

chat 모델 호출 (gpt-3.5-turbo) <- text-davinci의 1/10 가격, chat에 최적화된 모델, 더 최신 모델임

from langchain.llms.openai import OpenAI
from langchain.chat_models import ChatOpenAI, ChatAnthropic

llm = OpenAI()
chat = ChatOpenAI()

a = llm.predict("How many planets are there?")
b = chat.predict("How many planets are there?")

a, b

 

결과 이미지

 

OpenAI와 ChatOpenAI는 기본적으로 환경변수를 OPENAI_API_KEY를 사용함 

 

다른 변수를 사용할 경우 아래 오류가 나올 수 있음

ValidationError: 1 validation error for OpenAI __root__ Did not find openai_api_key, please add an environment variable `OPENAI_API_KEY` which contains it, or pass `openai_api_key` as a named parameter. (type=value_error)

 

========================================================================================

 

2. gpt-3.5-turbo 모델을 사용할 예정 주피터 노트북에서 아래와 같이 코드 실행

from langchain.chat_models import ChatOpenAI


chat = ChatOpenAI(
    temperature=0.1
)

 

SystemMessage는 역할 부여

AIMessage는 gpt

HumanMessage는 질문

아래 코드를 실행하면 대화처럼 다음 답변이 나옴

from langchain.schema import HumanMessage, AIMessage, SystemMessage

messages = [
    SystemMessage(content="You are a geography expert. And you only reply in Italian."),
    AIMessage(content="Ciao, mi chiamo Paolo!"),
    HumanMessage(content="What is the distance between Mexico and Thailand. Also, what is your name?"),
]

chat.predict_messages(messages)

gpt 답변

========================================================================================

 

3. prompt template

 

template를 만들고 format 한 후 predict 메소드 호출

from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate, ChatPromptTemplate

chat = ChatOpenAI(
    temperature=0.1
)

# 템플릿 작성
template = ChatPromptTemplate.from_messages([
    ("system", "You are a geography expert. And you only reply in {language}."),
    ("ai", "Ciao, mi chiamo {name}!"),
    ("human", "What is the distance between {country_a} and {country_b}. Also, what is your name?"),
])
# format
prompt = template.format_messages(
    language = "Greek",
    name = "Socrates",
    country_a = "Mexico",
    country_b = "Thailand",
)
# 출력
chat.predict_messages(prompt)

 

결과 화면

========================================================================================

4. LangChain expression language(표현언어)

 

chain에 template, chat, 출력형식을 정해주고 invoke 함수를 사용해 코드를 좀 더 단순화할 수 있음

from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate, ChatPromptTemplate
from langchain.schema import BaseOutputParser

chat = ChatOpenAI(
    temperature=0.1
)

class CommaOutputParser(BaseOutputParser):

    def parse(self, text):
        items =  text.strip().split(",")
        return list(map(str.strip, items))
        
        
template = ChatPromptTemplate.from_messages([
    ("system", "You are a list generating machine. Everything you are asked will be answered with a comma separated list of max {max_items} in lowercase. Do Not reply with anything else."),
    ("human", "{question}"),
])

chain = template | chat | CommaOutputParser()


chain.invoke({
    "max_items" : 5,
    "question" : "what are the poketmons?"
})

결과화면

========================================================================================

5. LangChain expression language(표현언어) 심화

 

 

chain = template | chat | CommaOutputParser()

 

template형식 -> chat의 값 -> CommaOutputParse 출력형식

 

 

from langchain.chat_models import ChatOpenAI
from langchain.prompts import ChatPromptTemplate
from langchain.callbacks import StreamingStdOutCallbackHandler

chat = ChatOpenAI(
    temperature=0.1,
    # 실시간으로 gpt의 답변이 생성되는 것을 확인 가능
    streaming=True,
    callbacks=[StreamingStdOutCallbackHandler()]
)


chef_prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a world-class international chef. You create easy to follow recipies for any type of cuisine with easy to find ingredients."),
    ("human", "I want to cook {cuisine} food."),
])

chef_chain = chef_prompt | chat 

veg_chef_prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a vegetarian chef specialized on making traditional recipes vegetarian. You find alternative ingredients and explain their preparation. You don't radically modify the recipe. If there is no alternative for a food just say you don't know how to recipe it."),
    ("human", "{recipe}"),
])

veg_chain = veg_chef_prompt | chat

final_chain ={"recipe":chef_chain} | veg_chain

final_chain.invoke({
    "cuisine" : "indian"
})

 

final_chain ={"recipe":chef_chain} | veg_chain

final_chain.invoke({
    "cuisine" : "indian"
})

 

해당 코드를 통해 chef_prompt에서 받은 chef_chain의 레시피를 veg_chain의 recipe 변수 값에 담아

 

두 템플릿 형식을 하나의 답변으로 받을 수 있다.

 

결과 화면