app.py

import gradio as gr
import PyPDF2
import io
import requests
import torch
from transformers import AutoTokenizer, AutoModelForQuestionAnswering

# Download and load pre-trained model and tokenizer
model_name = "distilbert-base-cased-distilled-squad"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForQuestionAnswering.from_pretrained(model_name)

def answer_questions(pdf_file, questions):
    # Load PDF file and extract text
    pdf_reader = PyPDF2.PdfFileReader(io.BytesIO(pdf_file.read()))
    text = ""
    for i in range(pdf_reader.getNumPages()):
        page = pdf_reader.getPage(i)
        text += page.extractText()
    text = text.strip()

    answers = []
    for question in questions:
        # Tokenize question and text
        input_ids = tokenizer.encode(question, text)

        # Perform question answering
        outputs = model(torch.tensor([input_ids]), return_dict=True)
        answer_start = outputs.start_logits.argmax().item()
        answer_end = outputs.end_logits.argmax().item()
        answer = tokenizer.convert_tokens_to_string(tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end+1]))
        
        answers.append(answer)
    
    return answers

inputs = [
    gr.inputs.File(label="PDF document"),
    gr.inputs.Textbox(label="Questions (one per line)", type="textarea")
]

outputs = gr.outputs.Textarea(label="Answers")

gr.Interface(fn=answer_questions, inputs=inputs, outputs=outputs, title="PDF Question Answering Tool", 
             description="Upload a PDF document and ask multiple questions. The app will use a pre-trained model to find the answers.").launch()