app.py

import streamlit as st
from PyPDF2 import PdfReader
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Cache the model and tokenizer loading
@st.cache_resource
def load_model():
    tokenizer = AutoTokenizer.from_pretrained("himmeow/vi-gemma-2b-RAG")
    model = AutoModelForCausalLM.from_pretrained(
        "himmeow/vi-gemma-2b-RAG",
        device_map="auto",
        torch_dtype=torch.bfloat16
    )
    if torch.cuda.is_available():
        model.to("cuda")
    return tokenizer, model

# Cache the text extraction from PDFs
@st.cache_data
def extract_text_from_pdfs(files):
    pdf_text = ""
    for file in files:
        reader = PdfReader(file)
        for page_num in range(len(reader.pages)):
            page = reader.pages[page_num]
            pdf_text += page.extract_text() + "\n"
    return pdf_text

# Load the model and tokenizer
tokenizer, model = load_model()

# Sidebar for PDF file upload
st.sidebar.title("📂 Upload PDFs")
uploaded_files = st.sidebar.file_uploader("Upload PDF files", type="pdf", accept_multiple_files=True)

# Initialize session state
if "history" not in st.session_state:
    st.session_state.history = []

# Extract text from PDFs and maintain session state
if uploaded_files:
    if "pdf_text" not in st.session_state:
        st.session_state.pdf_text = extract_text_from_pdfs(uploaded_files)

# Main interface
st.title("💬 RAG PDF Chatbot")
st.markdown("Ask questions based on the uploaded PDF documents.")

# Input for user query
query = st.text_input("Enter your question:")

# Process and respond to user query
if st.button("Submit"):
    if uploaded_files and query:
        with st.spinner("Generating response..."):
            # Prepare the input data
            prompt = """
            ### Instruction and Input:
            Based on the following context/document:
            {}
            Please answer the question: {}

            ### Response:
            """.format(st.session_state.pdf_text, query)

            # Encode the input text
            input_ids = tokenizer(prompt, return_tensors="pt")

            # Use GPU for input ids if available
            if torch.cuda.is_available():
                input_ids = input_ids.to("cuda")

            # Generate text using the model
            outputs = model.generate(
                **input_ids,
                max_new_tokens=500,  # Limit tokens to speed up generation
                no_repeat_ngram_size=3,  # Avoid repetition
                do_sample=True,  # Sampling for variability
                temperature=0.7  # Control randomness
            )

            # Decode and display the results
            response = tokenizer.decode(outputs[0], skip_special_tokens=True)
            st.session_state.history.append({"question": query, "answer": response})

# Display chat history
if st.session_state.history:
    for i, qa in enumerate(reversed(st.session_state.history), 1):
        st.markdown(f"**Q{i}:** {qa['question']}")
        st.markdown(f"**A{i}:** {qa['answer']}")

# Footer with author information
st.sidebar.markdown("### Created by: [Engr. Hamesh Raj](https://www.linkedin.com/in/datascientisthameshraj/)")
st.sidebar.markdown("## 🗂️ RAG PDF Chatbot")