handler.py

import base64
from typing import Any, Dict
from transformers import BlipProcessor, BlipForConditionalGeneration
from PIL import Image
from io import BytesIO
import torch
import logging


logging.basicConfig(level=logging.DEBUG)
logger = logging.getLogger(__name__)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')


class EndpointHandler():
    def __init__(self, path=""):
        logger.debug("Initializing model and processor.")
        self.model = BlipForConditionalGeneration.from_pretrained(
            "quadranttechnologies/qhub-blip-image-captioning-finetuned").to(device)
        self.processor = BlipProcessor.from_pretrained("quadranttechnologies/qhub-blip-image-captioning-finetuned")
        self.model.eval()
        self.model = self.model.to(device).to(device)

    def __call__(self, data: Any) -> Dict[str, Any]:
        """
        Args:
            data (:obj:):
                includes the input data and the parameters for the inference.
        Return:
            A :obj:`dict`:. The object returned should be a dict of one list like {"descriptions": ["Description of the image"]} containing :
                - "description": A string corresponding to the generated description.
        """
        logger.debug(f"Received data keys: {data.keys()}")

        image_base64 = data["inputs"].get("image")
        image_data = base64.b64decode(image_base64)

        # Convert image data to PIL Image
        images = Image.open(BytesIO(image_data))

        # Optional text input
        text = data["inputs"].get("text", "")
        parameters = data.pop("parameters", {})

        processed_image = self.processor(images=images, text=text, return_tensors="pt")
        processed_image["pixel_values"] = processed_image["pixel_values"].to(device)
        processed_image = {**processed_image, **parameters}

        with torch.no_grad():
            out = self.model.generate(
                **processed_image
            )
        description = self.processor.batch_decode(out, skip_special_tokens=True)

        return {"description": description}