handler.py

from typing import Dict, Any, List
import os
import base64
current_dir = os.getcwd()
os.environ['HF_HOME'] = os.path.join(current_dir)
os.environ['PAGINATE_OUTPUT']='True'
from marker.convert import convert_single_pdf
from marker.logger import configure_logging
from marker.models import load_all_models
from marker.output import save_markdown
from io import BytesIO
class EndpointHandler:
    def __init__(self, path=""):
        # Initialize the OCR model 
        self.models = load_all_models()
        self.file_location = "input/temp.pdf"
        os.makedirs("input", exist_ok=True)

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
        data args:
            inputs (:obj: dict): A dictionary containing the inputs.
                max_pages (:obj: int): The maximum number of pages to process.
                file (:obj: str): The base64-encoded PDF file content.
        Return:
            A list of dictionaries containing the extracted text.
        """
        inputs = data.get("inputs", {})
        file_content = inputs.get("file")
        max_pages = inputs.get("max_pages", None)

        # Decode the base64-encoded file content
        file_bytes = base64.b64decode(file_content)
        self.upload_file(BytesIO(file_bytes))

        pdf_path = self.file_location

        # Perform OCR on the input PDF
        extracted_text, _, _ = convert_single_pdf(pdf_path, self.models, max_pages=max_pages, langs=["vi"])
        # Return the extracted text
        return [{"extracted_text": extracted_text}]

    def upload_file(self, file: BytesIO, max_pages: int = None):
        with open(self.file_location, "wb") as f:
            f.write(file.read())
        return True