app.py

import streamlit as st
from langchain_community.document_loaders.pdf import PyPDFDirectoryLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain_community.embeddings import HuggingFaceInstructEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.chains import ConversationalRetrievalChain
from langchain_community.llms import HuggingFaceHub
from langchain.memory import ConversationBufferMemory

def make_vectorstore(embeddings):
    
    loader = PyPDFDirectoryLoader("data")   
    documents = loader.load()
    text_splitter = CharacterTextSplitter(chunk_size=1400, chunk_overlap=0)
    texts = text_splitter.split_documents(documents)
    docsearch = FAISS.from_documents(texts, embeddings)
    
    return docsearch

def get_qa(vectorstore, llmb):

    qa = RetrievalQA.from_chain_type(
        llm=llmb,
        chain_type="stuff", 
        retriever=vectorstore.as_retriever())
    
    return qa

def get_response(qa, query):
    
    response = qa.run(query)
    
    return response

def main():
    
    st.title("BetterZila RAG Enabled LLM")
    llm = HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":0.5, "max_length":512}, huggingfacehub_api_token = st.secrets["hf_token"])
    embeddings = HuggingFaceInstructEmbeddings(model_name="google/t5-v1_1-xl", model_kwargs = {'device': 'cpu'})
    vectorstore = make_vectorstore(embeddings)
    qa = get_qa(vectorstore, llm)
    queries = ["Can you give me an example from history where the enemy was crushed totally from the book?", "What's the point of making myself less accessible?", "Can you tell me the story of Queen Elizabeth I from this 48 laws of power book?"]
    for query in queries:
        st.subheader(f"Query: {query}")
        response = get_response(qa, query)
        st.write(query)
        st.write(response)
    st.success("Responses generated!")
    
if __name__ == "__main__":
    main()