app.py

import os 
import gradio as gr
import re
from langchain.embeddings.base import Embeddings
from typing import List
from sentence_transformers import SentenceTransformer
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.prompts import PromptTemplate
from langchain_community.llms.huggingface_hub import HuggingFaceHub
from read_photodocument import convert_PDF_to_Text,convert_image_to_pdf
from doctr.io import DocumentFile
from doctr.models import ocr_predictor
import contextlib
from langchain.schema import Document
from langchain.text_splitter import CharacterTextSplitter
from langchain.chains.summarize import load_summarize_chain
import logging

logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s %(levelname)s %(message)s",
    datefmt="%m/%d/%Y %I:%M:%S",
)

DEVICE = 'cpu'
FILE_EXT = ['pdf','jpg','jpeg']
DEFAULT_SYSTEM_PROMPT = "As an intelligent AI your task is to extract text from the pdf containing image and create a summary and higlight vital point within it ."

MAX_NEW_TOKENS = 2048
DEFAULT_TEMPERATURE = 0.1
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = 2048

embedding_modelPath = 'multi-qa-mpnet-base-dot-v1'# "sentence-transformers/all-MiniLM-l6-v2"
local_embeddings = HuggingFaceEmbeddings(model_name=embedding_modelPath,model_kwargs = {'device':'cpu'},encode_kwargs = {'normalize_embeddings': False})


with contextlib.redirect_stdout(None):
        ocr_model = ocr_predictor(
        "db_resnet50",
        "crnn_mobilenet_v3_large",
        pretrained=True,
        assume_straight_pages=True,
    )

def loading_file():
    return "Loading..."


def summarize_data(docs,llm_model,chain_type='refine'):
    prompt_template = """
    Write a concise summary of the following text pointwise without repeating sentences:
    {text}
    CONCISE SUMMARY:
    """
    refine_template = (
        "Your job is to produce a final summary in points.\n"
        "Existing summary up to a certain point: {existing_answer}\n"
        "write the details of summary pointwise and avoid repetion."
    )
    
    prompt = PromptTemplate.from_template(prompt_template)
    refine_prompt = PromptTemplate.from_template(refine_template)
    

    chain = load_summarize_chain(llm=llm_model,
                            chain_type=chain_type,
                            question_prompt=prompt,
                            # refine_prompt=,
                            return_intermediate_steps=False,
                            input_key="input_documents",
                            output_key="output_text",
                            )
    summary = chain({"input_documents": docs}, return_only_outputs=True)
    output_text  = summary["output_text"].replace('\n',' ')
    
    consice_sumary = re.search("CONCISE SUMMARY:.*\.*$", output_text).group(0)
    
    return consice_sumary.replace('  ','\n')
    
    # matches = re.finditer(regex, output_text, re.DOTALL)
    # for matchNum, match in enumerate(matches, start=1):
    #     for groupNum in range(0, len(match.groups())):
    #         groupNum = groupNum + 1
    #         lines = match.group(groupNum).strip().split("\n")
    


def process_documents(texts,data_chunk=1000,chunk_overlap=10):
    text_splitter = CharacterTextSplitter(
        separator="\n",
        chunk_size=data_chunk,
        chunk_overlap=chunk_overlap,
        length_function=len
    )

    texts = text_splitter.split_text(texts)
    docs = [Document(page_content=txt) for txt in texts]
    return docs

def get_hugging_face_model(model_id='tiiuae/falcon-7b-instruct',temperature=0.01,max_tokens=4096,API_key=None):
    llm = HuggingFaceHub(
        huggingfacehub_api_token =API_key ,
        repo_id=model_id, 
        model_kwargs={"temperature":temperature, "max_new_tokens":max_tokens}
        )
    return llm


def document_loader(temperature,max_tokens,api_key,model_name,file_path):
    model = get_hugging_face_model(model_id=model_name,API_key=api_key,temperature=temperature,max_tokens=max_tokens)
    converted_txt = None
    if file_path.endswith('.pdf'):
        conversion_stats = convert_PDF_to_Text(PDF_file=file_path,ocr_model=ocr_model)
        converted_txt = conversion_stats["converted_text"]
        num_pages = conversion_stats["num_pages"]
        was_truncated = conversion_stats["truncated"]
        print("Converted text {}\nNum Pages;{}".format(converted_txt,num_pages))
    
    elif file_path.endswith('.jpg') or file_path.endswith('.jpeg'):
        conversion_stats = convert_image_to_pdf(file_path,ocr_model)
        converted_txt = conversion_stats["converted_text"]
        num_pages = conversion_stats["num_pages"]
        was_truncated = conversion_stats["truncated"]
        print("Converted text {}\nNum Pages;{}".format(converted_txt,num_pages))

    else:
        return ("Invalid Format ....")
    if converted_txt:
        print("Document Processed ..")
        texts = process_documents(texts=converted_txt)
        lines = summarize_data(docs=texts,llm_model=model)

        return lines 
    else:
        return "Error in Processsing document "



iface = gr.Interface(
    fn = document_loader,
    inputs = [gr.Slider(0.01, 0.1, value=0.01, step=0.01 , label="temperature", info="Choose between 0.01 to 0.1"),
        gr.Slider(512,MAX_INPUT_TOKEN_LENGTH,value=1024,step=512,label="max new tokens",info='Max new tokens'),
        gr.Textbox(label="Add API key", type="password"),
        gr.Dropdown(['tiiuae/falcon-7b-instruct','mistralai/Mistral-7B-v0.1'],label='Large Language Model',info='LLM Service'),
        "file"
    ],
    outputs="text",
    description ="Summarize your PDF Document having Image  • HuggingFace",
)

iface.launch()