handler.py

from typing import Dict, List, Any
from transformers import AutoProcessor, AutoTokenizer, CLIPModel
from PIL import Image
import requests

class EndpointHandler():

    def __init__(self, path=""):
        self.model = CLIPModel.from_pretrained(path)
        self.processor = AutoProcessor.from_pretrained(path)
        self.tokenizer = AutoTokenizer.from_pretrained(path)

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
       data args:
            inputs (:obj: `str`)
            date (:obj: `str`)
      Return:
            A :obj:`list` | `dict`: will be serialized and returned
        """
        # get inputs
        inputs = data.pop("inputs", data) # if type is image, data is url
        params = data.pop("parameters", data)

        if params['datatype'] == 'image':
            image = Image.open(requests.get(inputs, stream=True).raw)
            # embed image
            features = self.embed_image(image)
        elif params['datatype'] == 'text':
            # embed text
            features = self.embed_text(inputs)

        # return features
        return {"features": features[0]}
    
    def embed_text(self, text):
        inputs = self.tokenizer(text, padding=True, return_tensors="pt")
        text_features = self.model.get_text_features(**inputs)
        # normalize
        text_features = text_features / text_features.norm(dim=-1, keepdim=True)
        # make a list of text features
        text_features = text_features.tolist()
        return text_features
    
    def embed_image(self, image):
        inputs = self.processor(images=image, return_tensors="pt")
        image_features = self.model.get_image_features(**inputs)
        # normalize
        image_features = image_features / image_features.norm(dim=-1, keepdim=True)
        # make a list of image features
        image_features = image_features.tolist()
        return image_features