app.py

import os
import streamlit as st
from dotenv import load_dotenv
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.llms import llamacpp
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.callbacks import CallbackManager, StreamingStdOutCallbackHandler
from langchain.chains import create_history_aware_retriever, create_retrieval_chain, ConversationalRetrievalChain
from langchain.document_loaders import TextLoader
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_community.chat_message_histories.streamlit import StreamlitChatMessageHistory
from langchain.prompts import PromptTemplate
from langchain.vectorstores import Chroma
from utills import load_txt_documents, split_docs, load_uploaded_documents, retriever_from_chroma
from langchain.text_splitter import TokenTextSplitter, RecursiveCharacterTextSplitter
from langchain_community.document_loaders.directory import DirectoryLoader
from HTML_templates import css, bot_template, user_template
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain import hub
from langchain.retrievers import ContextualCompressionRetriever
from langchain.retrievers.document_compressors import LLMChainExtractor

lang_api_key = os.getenv("lang_api_key")

os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_ENDPOINT"] = "https://api.langchain.plus"
os.environ["LANGCHAIN_API_KEY"] = lang_api_key
os.environ["LANGCHAIN_PROJECT"] = "Chat with multiple PDFs"



def create_retriever_from_chroma(vectorstore_path="docs/chroma/", search_type='mmr', k=7, chunk_size=250, chunk_overlap=20):
    
    model_name = "Alibaba-NLP/gte-base-en-v1.5"
    model_kwargs = {'device': 'cpu',
                   "trust_remote_code" : 'True'}
    encode_kwargs = {'normalize_embeddings': True}
    embeddings = HuggingFaceEmbeddings(
        model_name=model_name,
        model_kwargs=model_kwargs,
        encode_kwargs=encode_kwargs
    )


    llm = llamacpp.LlamaCpp(
        model_path='qwen2-0_5b-instruct-q4_0.gguf',
        n_gpu_layers=0,
        temperature=0.1,
        top_p=0.9,
        n_ctx=22000,
        n_batch=2000,
        max_tokens=200,
        repeat_penalty=1.7,
        last_n_tokens_size = 1500,
        # callback_manager=callback_manager,
        verbose=False,
    )

    
    # Check if vectorstore exists
    if os.path.exists(vectorstore_path) and os.listdir(vectorstore_path):
        # Load the existing vectorstore
        vectorstore = Chroma(persist_directory=vectorstore_path,embedding_function=embeddings)
    else:
        # Load documents from the specified data path
        loader = DirectoryLoader('./data', glob="./*.txt", loader_cls=TextLoader)
        docs = loader.load()
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
        split_docs = text_splitter.split_documents(docs)

        

        # Create the vectorstore
        vectorstore = Chroma.from_documents(
            documents=split_docs, embedding=embeddings, persist_directory=vectorstore_path
        )
    
    
    retriever=vectorstore.as_retriever(search_type = search_type, search_kwargs={"k": k})

    compressor = LLMChainExtractor.from_llm(llm)

    compression_retriever = ContextualCompressionRetriever(
    base_compressor=compressor,
    base_retriever=retriever
)
    
    

    return compression_retriever






def main():


    if "messages" not in st.session_state:
        st.session_state["messages"] = [
        {"role": "assistant", "content": "Hi, I'm a chatbot who can search the web. How can I help you?"}
    ]


    st.set_page_config(page_title="Chat with multiple PDFs",
                       page_icon=":books:")
    st.write(css, unsafe_allow_html=True)

    
    st.header("Chat with multiple PDFs :books:")
    st.markdown("Hi, I am Qwen, chat mmodel, based on respublic of Lithuania law document. Write you question and press enter to start chat.")

    
    retriever = create_retriever_from_chroma(vectorstore_path="docs/chroma/", search_type='mmr', k=9, chunk_size=250, chunk_overlap=20)
    if user_question := st.text_input("Ask a question about your documents:"):
        handle_userinput(user_question,retriever)
        
    
    
    
    
 
    
    

    


def handle_userinput(user_question,retriever):
    st.session_state.messages.append({"role": "user", "content": user_question})
    st.chat_message("user").write(user_question)
    docs = retriever.invoke(user_question)

    with st.sidebar:
        st.subheader("Your documents")
        with st.spinner("Processing"):
            for doc in docs:
                st.write(f"Document: {doc}")
    
    doc_txt = [doc.page_content for doc in docs]
    
    rag_chain = create_conversational_rag_chain(retriever)
    response = rag_chain.invoke({"context": doc_txt, "question": user_question})
    st.session_state.messages.append({"role": "assistant", "content": response})
    st.chat_message("assistant").write(response)
    

                
            



def create_conversational_rag_chain(retriever):
    
    model_path = ('qwen2-0_5b-instruct-q4_0.gguf')

    callback_manager = CallbackManager([StreamingStdOutCallbackHandler()])

    llm = llamacpp.LlamaCpp(
        model_path = "qwen2-0_5b-instruct-q4_0.gguf",
        n_gpu_layers=0,
        temperature=0.4,
        top_p=0.9,
        n_ctx=22000,
        n_batch=2000,
        max_tokens=200,
        repeat_penalty=1.7,
        last_n_tokens_size = 200,
        # callback_manager=callback_manager,
        verbose=False,
    )

    prompt = hub.pull("rlm/rag-prompt")

    rag_chain = prompt | llm | StrOutputParser()


    return rag_chain


 

if __name__ == "__main__":
    main()