app.py

import os
import streamlit as st
from pathlib import Path
from io import StringIO

#for textfiles
from langchain.document_loaders import TextLoader
#text splitter
from langchain.text_splitter import CharacterTextSplitter
#for using HugginFace models & embeddings
from langchain.embeddings import HuggingFaceEmbeddings
from langchain import HuggingFaceHub
# Vectorstore: https://python.langchain.com/en/latest/modules/indexes/vectorstores.html
from langchain.vectorstores import FAISS
#facebook vectorization
from langchain.chains.question_answering import load_qa_chain
#load pdf
#vectorize db index with chromadb
from langchain.indexes import VectorstoreIndexCreator
from langchain.chains import RetrievalQA
from langchain.document_loaders import UnstructuredPDFLoader

os.environ["HUGGINGFACEHUB_API_TOKEN"] = st.secrets["hf_api_key"]


def init():
	global embeddings, llm, llm2, chain
	# Embeddings
	embeddings = HuggingFaceEmbeddings()
	llm=HuggingFaceHub(repo_id="declare-lab/flan-alpaca-large", model_kwargs={"temperature":0, "max_length":512})
	chain = load_qa_chain(llm, chain_type="stuff")

def pdf_file(txtFileObj):
	st.subheader('Uploaded PDF File:')
	st.write(txtFileObj.name)

	with open(txtFileObj.name, "wb") as f:
  		f.write(txtFileObj.getbuffer())

	loaders = [UnstructuredPDFLoader(txtFileObj.name)]
	index = VectorstoreIndexCreator(
    		embedding=embeddings,
    		text_splitter=CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)).from_loaders(loaders)
	
	chain = RetrievalQA.from_chain_type(llm=llm,
				     chain_type="stuff",
					 retriever=index.vectorstore.as_retriever(),
					 input_key="question")

	st.subheader('Enter query')
	query = st.text_input('Ask anything about the Document you uploaded')

	if (query):
		answer = chain.run(question=query)

		st.subheader('Answer')
		st.write(answer)

def text_file(txtFileObj):
	st.subheader('Uploaded Text File:')
	st.write(txtFileObj.name)

	#stringio = StringIO(txtFileObj.getvalue().decode("utf-8"))
	with open(txtFileObj.name, "wb") as f:
  		f.write(txtFileObj.getbuffer())	
	
	loader = TextLoader(txtFileObj.name)
	documents = loader.load()

	# Text Splitter
	text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=10)
	docs = text_splitter.split_documents(documents)

	db = FAISS.from_documents(docs, embeddings)

	st.subheader('Enter query')
	query = st.text_input('Ask anything about the Document you uploaded')

	if (query):
		docs = db.similarity_search(query)
		answer = chain.run(input_documents=docs, question=query)

		st.subheader('Answer')
		st.write(answer)

st.title('Document Q&A - Ask anything in your Document')
st.subheader('This application can be used to upload text(.txt) and PDF(.pdf) files and ask questions about their contents.')

init()

st.sidebar.subheader('Upload document')
uploaded_file = st.sidebar.file_uploader("Upload File",type=['txt','pdf'])

if uploaded_file and Path(uploaded_file.name).suffix == '.txt':
	st.sidebar.info(Path(uploaded_file.name))
	text_file(uploaded_file)

if uploaded_file and Path(uploaded_file.name).suffix == '.pdf':
	pdf_file(uploaded_file)

with st.sidebar.expander('File'):
    if (uploaded_file):
	    st.info(uploaded_file.name)
if os.path.exists('/content/'):
	st.info(os.listdir('/content/'))


# # PDFs
# # !wget https://github.com/fabiomatricardi/cdQnA/raw/main/PLC_mediumArticle.pdf
# # !wget https://github.com/fabiomatricardi/cdQnA/raw/main/BridgingTheGaap_fromMedium.pdf
# # !mkdir pdfs
# # !cp *pdf '/content/pdfs'

# # pdf_folder_path = '/content/pdfs'
# # os.listdir(pdf_folder_path)

# # loaders = [UnstructuredPDFLoader(os.path.join(pdf_folder_path, fn)) for fn in os.listdir(pdf_folder_path)]
# # loaders

# index = VectorstoreIndexCreator(
#     embedding=HuggingFaceEmbeddings(),
#     text_splitter=CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)).from_loaders(loaders)

# #Load llm with selected one
# llm2=HuggingFaceHub(repo_id="declare-lab/flan-alpaca-large", model_kwargs={"temperature":0, "max_length":512})
# #Prepare the pipeline
# from langchain.chains import RetrievalQA
# chain = RetrievalQA.from_chain_type(llm=llm2,
#                                     chain_type="stuff",
#                                     retriever=index.vectorstore.as_retriever(),
#                                     input_key="question")
# #get reply to our questions
# # chain.run('What is the difference between a PLC and a PC?')