handler.py

from typing import Dict, List, Any
import numpy as np
from transformers import CLIPProcessor, CLIPModel
from PIL import Image
from io import BytesIO
import base64

class EndpointHandler():
    def __init__(self, path=""):
        # Preload all the elements you we need at inference.
        self.model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
        self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")


    def __call__(self, data):
        images = list(map(file_to_image, [data]))
        inputs = self.processor(images=images, return_tensors="jax", padding=True) # converts the images into model-acceptable inputs and applies padding
        emb = self.model.get_image_features(**inputs)
        return { "embeddings": emb }

def file_to_image(file):
  image = Image.open(BytesIO(file)).convert("RGB")
  return image