handler.py

from typing import Dict, List, Any
from transformers import CLIPTokenizer, CLIPModel
import numpy as np
import os
import torch


class EndpointHandler:
    def __init__(self, path="."):
        # load the model
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.model = CLIPModel.from_pretrained(path).to(self.device).eval()
        self.tokenizer = CLIPTokenizer.from_pretrained(path)

    def __call__(self, data: Dict[str, Any]) -> List[float]:
        """
         data args:
              inputs (:obj: `str` | `PIL.Image` | `np.array`)
              kwargs
        Return:
              A :obj:`list` | `dict`: will be serialized and returned
        """
        # compute the embedding of the input
        query = data["inputs"]
        inputs = self.tokenizer(query, padding=True, return_tensors="pt").to(
            self.device
        )
        with torch.no_grad():
            text_features = self.model.get_text_features(**inputs)

        text_features = text_features.cpu().detach().numpy()
        input_embedding = text_features[0]

        # normalize the embedding
        input_embedding /= np.linalg.norm(input_embedding)

        return input_embedding.tolist()