handler.py

from transformers import Blip2Processor, Blip2Model
from typing import Dict, List, Any
from PIL import Image
from transformers import pipeline
import requests
import torch

class EndpointHandler():
    def __init__(self, path=""):
        """
        path: 
        """
        # Preload all the elements you are going to need at inference.
        # pseudo:
        # self.model= load_model(path)
        #self.processor = Blip2Processor.from_pretrained(path)
        #self.pipeline = pipeline(model = path)
        self.path = path
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        #self.processor = Blip2Processor.from_pretrained(path)
        #self.model = Blip2Model.from_pretrained(path, torch_dtype=torch.float16)
        
    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
       data args:
            inputs (:obj: `str` | `PIL.Image` | `np.array`)
            kwargs
      Return:
            A :obj:`list` | `dict`: will be serialized and returned
        """
        inputs = data.pop("inputs", data)
        image_url = inputs['image_url']
        #image = Image.open(requests.get(image_url, stream=True).raw)
        #processed_image = self.processor(images=image, return_tensors="pt").to(self.device, torch.float16)

        #generated_ids = self.pipeline(**inputs)
        #generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()

        return image_url, self.path, self.device