auditqa/doc_process.py

import glob
import os
from langchain.text_splitter import RecursiveCharacterTextSplitter, SentenceTransformersTokenTextSplitter
from transformers import AutoTokenizer
from langchain_community.document_loaders import PyMuPDFLoader
path_to_data = "./data/"


def process_pdf():
    files = {'ABC':'./data/MWTS2021.pdf',
            'XYZ':'./data/Consolidated2021.pdf'}
    docs = {}
    for file,value in files.items():
        try:
            docs[file] = PyMuPDFLoader(value).load()
        except Exception as e:
            print("Exception: ", e)

    
    # text splitter based on the tokenizer of a model of your choosing
    # to make texts fit exactly a transformer's context window size
    # langchain text splitters: https://python.langchain.com/docs/modules/data_connection/document_transformers/
    chunk_size = 256
    text_splitter = RecursiveCharacterTextSplitter.from_huggingface_tokenizer(
            AutoTokenizer.from_pretrained("BAAI/bge-small-en-v1.5"),
            chunk_size=chunk_size,
            chunk_overlap=int(chunk_size / 10),
            add_start_index=True,
            strip_whitespace=True,
            separators=["\n\n", "\n", ".", " ", ""],
    )