app.py

import os
import streamlit as st
import pickle
import time
from langchain import OpenAI
from langchain.chains import RetrievalQAWithSourcesChain
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.document_loaders import UnstructuredURLLoader
# from langchain.embeddings import OpenAIEmbeddings
from langchain.embeddings import FakeEmbeddings
from langchain.llms import HuggingFaceHub
from langchain.chains import LLMChain
from langchain.vectorstores import FAISS

from dotenv import load_dotenv
# load_dotenv()  # take environment variables from .env (especially openai api key)
os.environ["HUGGINGFACEHUB_API_TOKEN"] = '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'

st.title("RockyBot: News Research Tool 📈")
st.sidebar.title("News Article URLs")

urls = []
for i in range(3):
    url = st.sidebar.text_input(f"URL {i+1}")
    urls.append(url)

process_url_clicked = st.sidebar.button("Process URLs")
file_path = "faiss_store_openai.pkl"

main_placeholder = st.empty()
llm = HuggingFaceHub( repo_id="google/flan-t5-xxl", model_kwargs={"temperature": 0.5, "max_length": 64} )

@st.cache
def process_urls(urls):
    """Processes the given URLs and saves the FAISS index to a pickle file."""

    # load data
    loader = UnstructuredURLLoader(urls=urls)

    # split data
    text_splitter = RecursiveCharacterTextSplitter(
        separators=['\n\n', '\n', '.', ','],
        chunk_size=1000
    )
    docs = text_splitter.split_documents(loader.load())

    # create embeddings and save it to FAISS index
    embeddings = FakeEmbeddings(size=1352)
    vectorstore_openai = FAISS.from_documents(docs, embeddings)

    # Save the FAISS index to a pickle file
    with open(file_path, "wb") as f:
        pickle.dump(vectorstore_openai, f)

if process_url_clicked:
    with st.progress(0.0):
        process_urls(urls)
        st.progress(100.0)

query = main_placeholder.text_input("Question: ")
if query:
    try:
        with open(file_path, "rb") as f:
            vectorstore = pickle.load(f)
            chain = RetrievalQAWithSourcesChain.from_llm(llm=llm, retriever=vector_store.as_retriever())
            result = chain({"question": query}, return_only_outputs=True)
            # result will be a dictionary of this format --> {"answer": "", "sources": [] }
            st.header("Answer")
            st.write(result["answer"])

            # Display sources, if available
            sources = result.get("sources", "")
            if sources:
                st.subheader("Sources:")
                sources_list = sources.split("\n")  # Split the sources by newline
                for source in sources_list:
                    st.write(source)
    except Exception as e:
        st.error(e)