handler.py

import io
from typing import Dict, List, Any
from transformers import LayoutLMv3ForSequenceClassification, LayoutLMv3FeatureExtractor, LayoutLMv3Tokenizer, LayoutLMv3Processor
import torch
from subprocess import run
from PIL import Image

# install tesseract-ocr and pytesseract
run("apt install -y tesseract-ocr", shell=True, check=True)
run("python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cpu/torch1.10/index.html", shell=True, check=True)

# set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


class EndpointHandler:
    def __init__(self, path=""):
        # load model and processor from path
        self.FEATURE_EXTRACTOR = LayoutLMv3FeatureExtractor()
        self.TOKENIZER = LayoutLMv3Tokenizer.from_pretrained("microsoft/layoutlmv3-base")
        self.PROCESSOR = LayoutLMv3Processor(self.FEATURE_EXTRACTOR, self.TOKENIZER)
        self.MODEL = LayoutLMv3ForSequenceClassification.from_pretrained("OtraBoi/document_classifier_testing").to(device)

    def __call__(self, data: Dict[str, bytes]):
        #image = Image.open(io.BytesIO(data["inputs"])).convert("RGB")
        image = data.pop("inputs", data)
        encoding = self.PROCESSOR(image, return_tensors="pt", padding="max_length", truncation=True)
        
        for k,v in encoding.items():
            encoding[k] = v.to(self.MODEL.device)

        # run prediction
        with torch.inference_mode():
            outputs = self.MODEL(**encoding)
            logits = outputs.logits
            predicted_class_idx = logits.argmax(-1).item()
        
        return self.MODEL.config.id2label[predicted_class_idx]