app.py

import streamlit as st
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.document_loaders import PyPDFLoader
from transformers import T5Tokenizer, T5ForConditionalGeneration
from transformers import pipeline
import torch
import base64
import time
from PIL import Image

# Model and tokenizer
model_checkpoint = "MBZUAI/LaMini-Flan-T5-783M"
model_tokenizer = T5Tokenizer.from_pretrained(model_checkpoint)
model = T5ForConditionalGeneration.from_pretrained(model_checkpoint)

# File loader and preprocessing
def preprocess_pdf(file):
    loader = PyPDFLoader(file)
    pages = loader.load_and_split()
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=170, chunk_overlap=70)
    texts = text_splitter.split_documents(pages)
    final_text = ""
    for text in texts:
        final_text = final_text + text.page_content
    return final_text

@st.cache_data
def language_model_pipeline(filepath):
    summarization_pipeline = pipeline(
        'summarization',
        model=model,
        tokenizer=model_tokenizer,
        max_length=500,
        min_length=32
    )
    input_text = preprocess_pdf(filepath)
    summary_result = summarization_pipeline(input_text)
    summarized_text = summary_result[0]['summary_text']
    return summarized_text

# User interface
title = st.title("PDF Summarization")
uploaded_file = st.file_uploader('Upload your PDF file', type=['pdf'])

if uploaded_file is not None:
    st.success("File uploaded")

    if st.button("Summarize"):
        with st.spinner("Summarizing..."):
            time.sleep(10)

            filepath = uploaded_file.name
            with open(filepath, "wb") as temp_file:
                temp_file.write(uploaded_file.read())

            summarized_result = language_model_pipeline(filepath)
            st.success("Summary:")
            st.write(summarized_result)