summarize.py

import streamlit as st
import pdfplumber
from PIL import Image
import pytesseract
#from transformers import pipeline
import io

import os
from dotenv import load_dotenv

from groqSummarizer import GroqSummarizer

# SwedishBeagle-dare
from transformers import AutoTokenizer
import transformers
import torch

from streamlit import session_state as ss
from streamlit_pdf_viewer import pdf_viewer


class Summarizer:

    def __init__(self, model = "groq"):
        self.model = model

    def run_app(self):
        # For displaying pdf
        if 'pdf_ref' not in ss:
            ss.pdf_ref = None
        
        uploaded_file = st.file_uploader("Upload an Image or PDF", type=["jpg", "jpeg", "png", "pdf"], key="file")
    
        if uploaded_file is not None:
            
            if uploaded_file.type == "application/pdf":
                with st.spinner("Extracting text from PDF..."):
                    text = self.extract_text_from_pdf(uploaded_file)


                if ss.file:
                    ss.pdf_ref = ss.file

            else:
                image = Image.open(uploaded_file)
                with st.spinner("Extracting text from image..."):
                    text = self.extract_text_from_image(image)
    
            if text:
                with st.spinner("Summarizing text..."):
                    summary = self.summarize_using_groq(text)
                st.subheader("Summary")
                st.write(summary)
        
            with st.expander("Extracted Text", expanded = False):
                st.write(text)

            if ss.pdf_ref:
                    st.subheader("Original pdf")
                    binary_data = ss.pdf_ref.getvalue()
                    pdf_viewer(input=binary_data, width=700)


    # Function to extract text from an image
    def extract_text_from_image(self, image):
        text = pytesseract.image_to_string(image)
        return text
    
    # Function to extract text from a PDF
    def extract_text_from_pdf(self, pdf):
        text = ""
        with pdfplumber.open(pdf) as pdf_file:
            for page in pdf_file.pages:
                text += page.extract_text()
        return text

    def shorten_text(self, text, max_tokens):
        tokens = text.split(" ")
        if len(tokens) > max_tokens:
            tokens = tokens[:max_tokens]
            text = " ".join(tokens)
        print("Shortened text to " + str(max_tokens) + " tokens")
        return text

    def summarize_using_groq(self, text):
        # Decrease the number of tokens if the response is 429, i.e. too many tokens in the request
        # 
        # https://context.ai/compare/llama3-70b-instruct-v1/gpt-4
        # ^^ Säger att max tokens är 8000, men efter tester så verkar det vara 
        # närmare 2000 om man räknar att tokens är separerade med blanksteg. 
        # (Detta är inte ett helt korrekt sätt att räkna det)
        
        # max_tokens = 8000
        max_tokens = 2000

        while True:
            try:
                gs = GroqSummarizer()
                return gs.summarize(text)
            except Exception as e:
                if e.response.status_code == 429:
                    text = self.shorten_text(text, max_tokens)
                    max_tokens = int(max_tokens * 0.9)
                else:
                    return "Error: " + str(e)

    def summarize_using_swedishbeagle(self, text):
        # https://huggingface.co/FredrikBL/SwedishBeagle-dare

        model = "FredrikBL/SwedishBeagle-dare"
        messages = [
            {
                "role": "system", 
                "content": "You summarize texts that the users sends"
            }, 
            {
                "role": "user", 
                "content": text
            }
        ]

        tokenizer = AutoTokenizer.from_pretrained(model)
        prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
        pipeline = transformers.pipeline(
            "text-generation",
            model=model,
            torch_dtype=torch.float16,
            device_map="auto",
        )

        outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
        return outputs[0]["generated_text"]

    def summarize(self, text):
        if(self.model == "groq"):
            return self.summarize_using_groq(text)
        elif(self.model == "SwedishBeagle-dare"):
            return self.summarize_using_swedishbeagle(text)
        

# Streamlit app
def main():
    # Models:
    # - groq
    # - SwedishBeagle-dare
    summarizer = Summarizer(model = "groq")
    summarizer.run_app()

if __name__ == "__main__":
    main()