app.py

import gradio as gr
from langchain_community.vectorstores import Chroma
from dotenv import load_dotenv
from langchain_huggingface import HuggingFaceEmbeddings, HuggingFaceEndpoint
from langchain_community.llms import HuggingFaceHub
from langchain.chains import ConversationalRetrievalChain
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.memory import ConversationBufferMemory
from langchain_community.document_loaders import PyPDFLoader
import os

load_dotenv()

# Get the Hugging Face API token from the .env file
hf_api_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")
if hf_api_token is None:
    raise ValueError("HUGGINGFACEHUB_API_TOKEN not found in .env file")

# Initialize the Hugging Face embedding model
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Initialize the LLaMA 2 model from Hugging Face Hub using the token from .env
llm = HuggingFaceEndpoint(
    repo_id="meta-llama/Llama-2-7b-hf",
    temperature=0.7,  # Specify temperature explicitly
    max_length=512,
    huggingfacehub_api_token=hf_api_token
)

# Initialize ChromaDB for storing and retrieving document embeddings
vectorstore = Chroma(embedding_function=embedding_model, persist_directory="chroma_db")

# Create a conversational chain with retrieval capabilities
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
qa_chain = ConversationalRetrievalChain.from_llm(llm, retriever=vectorstore.as_retriever(), memory=memory)

def upload_docs(docs):
    # Load and process the uploaded PDF documents
    loaded_docs = []
    for doc in docs:
        loader = PyPDFLoader(doc.name)
        loaded_docs.extend(loader.load())

    # Split documents into manageable chunks
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    texts = text_splitter.split_documents(loaded_docs)

    # Add documents to the vector store and persist them
    vectorstore.add_documents(texts)
    vectorstore.persist()
    return "PDF documents uploaded and processed successfully!"

def chat(query):
    # Process the query with the conversational chain and return the result
    response = qa_chain({"query": query})
    return response["result"]

# Gradio Interface
with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            doc_upload = gr.File(label="Upload your PDF documents", file_types=[".pdf"], multiple=True)
            upload_button = gr.Button("Upload")
            upload_button.click(upload_docs, inputs=doc_upload, outputs=gr.Textbox())
        with gr.Column():
            chat_input = gr.Textbox(label="Ask a question:")
            chat_output = gr.Textbox(label="Answer:")
            chat_button = gr.Button("Send")
            chat_button.click(chat, inputs=chat_input, outputs=chat_output)

demo.launch()