assign8 / src /streamlit_app.py
Separate2's picture
Update src/streamlit_app.py
a587b94 verified
Raw
History Blame Contribute Delete
10.8 kB
import streamlit as st
from dotenv import load_dotenv
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
from htmlTemplates import css, bot_template, user_template
# ν…μŠ€νŠΈ μŠ€ν”Œλ¦¬ν„°
from langchain_text_splitters import CharacterTextSplitter, RecursiveCharacterTextSplitter
# λ²‘ν„°μŠ€ν† μ–΄/μž„λ² λ”©/LLM
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
# λ‘œλ”λ“€ (pebblo/pwd λŒλ €μ˜€μ§€ μ•Šκ²Œ μ„œλΈŒλͺ¨λ“ˆλ‘œ)
from langchain_community.document_loaders.pdf import PyPDFLoader
from langchain_community.document_loaders.text import TextLoader
from langchain_community.document_loaders.csv_loader import CSVLoader
from langchain_community.document_loaders.json_loader import JSONLoader
import tempfile # μž„μ‹œ νŒŒμΌμ„ μƒμ„±ν•˜κΈ° μœ„ν•œ λΌμ΄λΈŒλŸ¬λ¦¬μž…λ‹ˆλ‹€.
import os
import json
from langchain.docstore.document import Document
from langchain_groq import ChatGroq
# PDF λ¬Έμ„œλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_pdf_text(pdf_docs):
temp_dir = tempfile.TemporaryDirectory() # μž„μ‹œ 디렉토리λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
temp_filepath = os.path.join(temp_dir.name, pdf_docs.name) # μž„μ‹œ 파일 경둜λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
with open(temp_filepath, "wb") as f: # μž„μ‹œ νŒŒμΌμ„ λ°”μ΄λ„ˆλ¦¬ μ“°κΈ° λͺ¨λ“œλ‘œ μ—½λ‹ˆλ‹€.
f.write(pdf_docs.getvalue()) # PDF λ¬Έμ„œμ˜ λ‚΄μš©μ„ μž„μ‹œ νŒŒμΌμ— μ”λ‹ˆλ‹€.
pdf_loader = PyPDFLoader(temp_filepath) # PyPDFLoaderλ₯Ό μ‚¬μš©ν•΄ PDFλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
pdf_doc = pdf_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
return pdf_doc # μΆ”μΆœν•œ ν…μŠ€νŠΈλ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
# TXT λ¬Έμ„œλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_text_file(docs):
# Streamlit UploadedFile -> str
# TextLoaderλŠ” 파일 경둜λ₯Ό 인자둜 λ°›μœΌλ―€λ‘œ, PDF λ‘œλ”μ™€ μœ μ‚¬ν•˜κ²Œ μž„μ‹œ νŒŒμΌμ„ 생성해야 ν•©λ‹ˆλ‹€.
temp_dir = tempfile.TemporaryDirectory()
temp_filepath = os.path.join(temp_dir.name, docs.name)
with open(temp_filepath, "wb") as f:
f.write(docs.getvalue())
text_loader = TextLoader(temp_filepath) # TextLoaderλ₯Ό μ‚¬μš©ν•΄ TXTλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
text_doc = text_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
return text_doc
# CSV λ¬Έμ„œλ‘œλΆ€ν„° ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_csv_file(docs):
# CSVLoaderλŠ” 파일 경둜λ₯Ό 인자둜 λ°›μœΌλ―€λ‘œ, μž„μ‹œ νŒŒμΌμ„ 생성해야 ν•©λ‹ˆλ‹€.
temp_dir = tempfile.TemporaryDirectory()
temp_filepath = os.path.join(temp_dir.name, docs.name)
with open(temp_filepath, "wb") as f:
f.write(docs.getvalue())
# CSVLoaderλ₯Ό μ‚¬μš©ν•΄ CSVλ₯Ό λ‘œλ“œν•©λ‹ˆλ‹€.
# headerκ°€ 있고, ν•˜λ‚˜μ˜ Document둜 ν†΅ν•©ν•˜λŠ” 것이 μ•„λ‹ˆλΌ 각 행을 Document둜 λ‘œλ“œν•©λ‹ˆλ‹€.
csv_loader = CSVLoader(temp_filepath, encoding="utf8")
csv_doc = csv_loader.load() # ν…μŠ€νŠΈλ₯Ό μΆ”μΆœν•©λ‹ˆλ‹€.
return csv_doc
def get_json_file(file) -> list[Document]:
# Streamlit UploadedFile -> str
raw = file.getvalue().decode("utf-8", errors="ignore")
data = json.loads(raw)
docs = []
# μ˜ˆμ „ jq κ²½λ‘œκ°€ '.scans[].relationships'μ˜€λ‹€λ©΄, λ™μΌν•œ 의미둜 νŒŒμ‹±:
# μ‘΄μž¬ν•˜λ©΄ κ·Έκ²ƒλ§Œ 뽑고, μ—†μœΌλ©΄ ν†΅μœΌλ‘œ λ¬Έμ„œν™”
def add_doc(x):
docs.append(Document(page_content=json.dumps(x, ensure_ascii=False)))
if isinstance(data, dict) and "scans" in data and isinstance(data["scans"], list):
for s in data["scans"]:
rels = s.get("relationships", [])
if isinstance(rels, list) and rels:
for r in rels:
add_doc(r)
if not docs: # κ·Έλž˜λ„ λͺ» λ½‘μ•˜μœΌλ©΄ 전체λ₯Ό ν•˜λ‚˜λ‘œ
add_doc(data)
elif isinstance(data, list):
for item in data:
add_doc(item)
else:
add_doc(data)
return docs
# λ¬Έμ„œλ“€μ„ μ²˜λ¦¬ν•˜μ—¬ ν…μŠ€νŠΈ 청크둜 λ‚˜λˆ„λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_text_chunks(documents):
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000, # 청크의 크기λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
chunk_overlap=200, # 청크 μ‚¬μ΄μ˜ 쀑볡을 μ§€μ •ν•©λ‹ˆλ‹€.
length_function=len # ν…μŠ€νŠΈμ˜ 길이λ₯Ό μΈ‘μ •ν•˜λŠ” ν•¨μˆ˜λ₯Ό μ§€μ •ν•©λ‹ˆλ‹€.
)
documents = text_splitter.split_documents(documents) # λ¬Έμ„œλ“€μ„ 청크둜 λ‚˜λˆ•λ‹ˆλ‹€.
return documents # λ‚˜λˆˆ 청크λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
# ν…μŠ€νŠΈ μ²­ν¬λ“€λ‘œλΆ€ν„° 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def get_vectorstore(text_chunks):
# μ›ν•˜λŠ” μž„λ² λ”© λͺ¨λΈμ„ λ‘œλ“œν•©λ‹ˆλ‹€.
embeddings = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L12-v2',
model_kwargs={'device': 'cpu'}) # μž„λ² λ”© λͺ¨λΈμ„ μ„€μ •ν•©λ‹ˆλ‹€.
vectorstore = FAISS.from_documents(text_chunks, embeddings) # FAISS 벑터 μŠ€ν† μ–΄λ₯Ό μƒμ„±ν•©λ‹ˆλ‹€.
return vectorstore # μƒμ„±λœ 벑터 μŠ€ν† μ–΄λ₯Ό λ°˜ν™˜ν•©λ‹ˆλ‹€.
def get_conversation_chain(vectorstore):
# Groq LLM
llm = ChatGroq(
groq_api_key=os.environ.get("GROQ_API_KEY"),
model_name="llama-3.1-8b-instant",
temperature=0.75, # ν•„μš”μ— 맞게 νŠœλ‹
max_tokens=512 # μ»¨ν…μŠ€νŠΈ 초과 λ°©μ§€μš© (ν•„μš”μ‹œ μ‘°μ •)
)
memory = ConversationBufferMemory(
memory_key="chat_history",
return_messages=True
)
retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
conversation_chain = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=retriever,
memory=memory,
)
return conversation_chain
# μ‚¬μš©μž μž…λ ₯을 μ²˜λ¦¬ν•˜λŠ” ν•¨μˆ˜μž…λ‹ˆλ‹€.
def handle_userinput(user_question):
print('user_question => ', user_question)
# λŒ€ν™” 체인을 μ‚¬μš©ν•˜μ—¬ μ‚¬μš©μž μ§ˆλ¬Έμ— λŒ€ν•œ 응닡을 μƒμ„±ν•©λ‹ˆλ‹€.
response = st.session_state.conversation({'question': user_question})
# λŒ€ν™” 기둝을 μ €μž₯ν•©λ‹ˆλ‹€.
st.session_state.chat_history = response['chat_history']
for i, message in enumerate(st.session_state.chat_history):
if i % 2 == 0:
st.write(user_template.replace(
"{{MSG}}", message.content), unsafe_allow_html=True)
else:
st.write(bot_template.replace(
"{{MSG}}", message.content), unsafe_allow_html=True)
def main():
load_dotenv()
st.set_page_config(page_title="Basic_RAG_AI_Chatbot_with_Llama",
page_icon=":books:")
st.write(css, unsafe_allow_html=True)
if "conversation" not in st.session_state:
st.session_state.conversation = None
if "chat_history" not in st.session_state:
st.session_state.chat_history = None
st.header("Basic_RAG_AI_Chatbot_with_Llama3 :books:")
user_question = st.text_input("Ask a question about your documents:")
if user_question:
handle_userinput(user_question)
with st.sidebar:
st.subheader("Your documents")
docs = st.file_uploader(
"Upload your Files here and click on 'Process'", accept_multiple_files=True)
# --- PDF 처리 λ²„νŠΌ ---
if st.button("Process[PDF]"):
with st.spinner("Processing"):
# get pdf text
doc_list = []
for file in docs:
print('file - type : ', file.type)
if file.type in ['application/octet-stream', 'application/pdf']:
# file is .pdf
doc_list.extend(get_pdf_text(file))
else:
st.error("PDF 파일이 μ•„λ‹™λ‹ˆλ‹€.")
if not doc_list:
st.error("처리 κ°€λŠ₯ν•œ λ¬Έμ„œλ₯Ό μ°Ύμ§€ λͺ»ν–ˆμŠ΅λ‹ˆλ‹€.")
st.stop()
text_chunks = get_text_chunks(doc_list)
vectorstore = get_vectorstore(text_chunks)
st.session_state.conversation = get_conversation_chain(vectorstore)
# --- TXT 처리 λ²„νŠΌ (μž‘μ—… 3 μΆ”κ°€) ---
if st.button("Process[TXT]"):
with st.spinner("Processing"):
# get txt text
doc_list = []
for file in docs:
print('file - type : ', file.type)
if file.type == 'text/plain':
# file is .txt
doc_list.extend(get_text_file(file))
else:
st.error("TXT 파일이 μ•„λ‹™λ‹ˆλ‹€.")
if not doc_list:
st.error("처리 κ°€λŠ₯ν•œ λ¬Έμ„œλ₯Ό μ°Ύμ§€ λͺ»ν–ˆμŠ΅λ‹ˆλ‹€.")
st.stop()
text_chunks = get_text_chunks(doc_list)
vectorstore = get_vectorstore(text_chunks)
st.session_state.conversation = get_conversation_chain(vectorstore)
# --- CSV 처리 λ²„νŠΌ (μž‘μ—… 3 μΆ”κ°€) ---
if st.button("Process[CSV]"):
with st.spinner("Processing"):
# get csv text
doc_list = []
for file in docs:
print('file - type : ', file.type)
if file.type == 'text/csv':
# file is .csv
doc_list.extend(get_csv_file(file))
else:
st.error("CSV 파일이 μ•„λ‹™λ‹ˆλ‹€.")
if not doc_list:
st.error("처리 κ°€λŠ₯ν•œ λ¬Έμ„œλ₯Ό μ°Ύμ§€ λͺ»ν–ˆμŠ΅λ‹ˆλ‹€.")
st.stop()
text_chunks = get_text_chunks(doc_list)
vectorstore = get_vectorstore(text_chunks)
st.session_state.conversation = get_conversation_chain(vectorstore)
# --- JSON 처리 λ²„νŠΌ ---
if st.button("Process[JSON]"):
with st.spinner("Processing"):
# get json text
doc_list = []
for file in docs:
print('file - type : ', file.type)
if file.type == 'application/json':
# file is .json
doc_list.extend(get_json_file(file))
else:
st.error("JSON 파일이 μ•„λ‹™λ‹ˆλ‹€.")
if not doc_list:
st.error("처리 κ°€λŠ₯ν•œ λ¬Έμ„œλ₯Ό μ°Ύμ§€ λͺ»ν–ˆμŠ΅λ‹ˆλ‹€.")
st.stop()
text_chunks = get_text_chunks(doc_list)
vectorstore = get_vectorstore(text_chunks)
st.session_state.conversation = get_conversation_chain(vectorstore)
if __name__ == '__main__':
main()