pdftoqa_generator.py

import json
import os
import re
import statistics

import gradio as gr
import pandas as pd
from langchain.document_loaders import OnlinePDFLoader
from langchain.text_splitter import (
    CharacterTextSplitter,
    RecursiveCharacterTextSplitter,
)
from tqdm import tqdm
from tempfile import NamedTemporaryFile
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

os.environ["OPENAI_API_KEY"] = "sk-"


def pdf_parser(uploaded_file):
    '''
    bytes_data = uploaded_file.read()
    with NamedTemporaryFile(delete=False) as tmp:  # open a named temporary file
        tmp.write(bytes_data)                      # Write data from the uploaded file into it
        pdf_loader = PyPDFLoader(tmp.name)        # <---- now it works!
    '''
    #pdf_loader = PyPDFLoader(file_path) only for file path offline
    pdf_loader=OnlinePDFLoader(uploaded_file.name) #https://huggingface.co/spaces/fffiloni/langchain-chat-with-pdf/blob/main/app.py
    
    documents = pdf_loader.load()
    documents_text = [d.page_content for d in documents]

    text_splitter = RecursiveCharacterTextSplitter(
        # Set a really small chunk size, just to show.
        chunk_size=600,
        chunk_overlap=200,
        length_function=len,
        is_separator_regex=False,
    )

    # Split the text into chunks
    texts = text_splitter.create_documents(documents_text)
    #os.remove(tmp.name)                            # remove temp file
    return texts


def qa_generator(texts):
    question_tokenizer = AutoTokenizer.from_pretrained(
        "potsawee/t5-large-generation-squad-QuestionAnswer"
    )
    question_model = AutoModelForSeq2SeqLM.from_pretrained(
        "potsawee/t5-large-generation-squad-QuestionAnswer"
    )

    question_answer_dic = {}
    for i in tqdm(texts):

        context = i.page_content
        try:
            inputs = question_tokenizer(context, return_tensors="pt")
            outputs = question_model.generate(**inputs, max_length=100)
            question_answer = question_tokenizer.decode(
                outputs[0], skip_special_tokens=False
            )
            question_answer = question_answer.replace(
                question_tokenizer.pad_token, ""
            ).replace(question_tokenizer.eos_token, "")
            question, answer = question_answer.split(question_tokenizer.sep_token)

            question_answer_dic[question] = answer
        except:
            print(i)

    qa_notes_df = pd.DataFrame(data=[], columns=["No", "Question", "Answer"])
    qa_notes_df["No"] = [i + 1 for i in range(0, len(question_answer_dic))]
    qa_notes_df["Question"] = [k for k in question_answer_dic.keys()]
    qa_notes_df["Answer"] = [a for a in question_answer_dic.values()]
    qa_notes_json = qa_notes_df.to_dict("records")

    return qa_notes_json