랭체인 PDF에 기반한 챗봇 만들기

[주인장]

import streamlit as st

from PyPDF2 import PdfReader

from langchain_community.embeddings import HuggingFaceEmbeddings

from langchain_community.chat_models import ChatOllama

from langchain.chains import ConversationalRetrieval‎Chain

from langchain.memory import ConversationBufferWindowMemory

from langchain.vectorstores import FAISS

from langchain.text_splitter import RecursiveCharacterTextSplitter

# PDF 문서에서 텍스트 추출

def get_pdf_text(pdf_docs):

    text = ""

    for pdf in pdf_docs:

        pdf_reader = PdfReader(pdf)

        for page in pdf_reader.pages:

            text += page.extract_text()

    return text

# 텍스트를 청크로 나누기

def get_text_chunks(text):

    text_splitter = RecursiveCharacterTextSplitter(

        separators=["\n"],

        chunk_size=1000,

        chunk_overlap=200,

        length_function=len

    )

    chunks = text_splitter.split_text(text)

    return chunks

# 임베딩 처리 및 벡터스토어 생성

def get_vectorstore(text_chunks):

    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")

    vectorstore = FAISS.from_texts(texts=text_chunks, embedding=embeddings)

    return vectorstore

# 챗봇 대화 체인 생성

def get_conversation_chain(vectorstore):

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

    conversation_chain = ConversationalRetrieval‎Chain.from_llm(

        llm=ChatOllama(model="llama3-ko", temperature=0),

        retriever=vectorstore.as_retriever(),

        memory=memory,

        get_chat_history=lambda h: h,

    )

    return conversation_chain

# Streamlit 인터페이스

st.title("📄 PDF 기반 챗봇 (llama3-ko)")

st.divider()

user_uploads = st.file_uploader("파일을 업로드해주세요~", accept_multiple_files=True)

if user_uploads is not None:

    if st.button("Upload"):

        with st.spinner("처리중.."):

            raw_text = get_pdf_text(user_uploads)

            text_chunks = get_text_chunks(raw_text)

            vectorstore = get_vectorstore(text_chunks)

            st.session_state.conversation = get_conversation_chain(vectorstore)

if user_query := st.chat_input("질문을 입력해주세요~"):

    if 'conversation' in st.session_state:

        result = st.session_state.conversation({

            "question": user_query,

            "chat_history": st.session_state.get('chat_history', [])

        })

        response = result["answer"]

    else:

        response = "먼저 문서를 업로드해주세요~."

    with st.chat_message("assistant"):

        st.write(response)