| from typing import Dict, List, Any | |
| from transformers import CLIPTokenizer, CLIPModel | |
| import numpy as np | |
| import os | |
| class EndpointHandler: | |
| def __init__(self, path=""): | |
| self.model = CLIPModel.from_pretrained("openai/clip-vit-large-patch14") | |
| self.tokenizer = CLIPTokenizer.from_pretrained("openai/clip-vit-large-patch14") | |
| self.artwork_urls = np.load(os.path.join(path, "artwork_urls.npy"), allow_pickle=True) | |
| self.embeddings = np.load(os.path.join(path, "embeddings.npy"), allow_pickle=True) | |
| 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 | |
| """ | |
| inputs = self.tokenizer(data["inputs"], padding=True, return_tensors="pt") | |
| text_features = self.model.get_text_features(**inputs) | |
| text_features = text_features.detach().numpy() | |
| input_embedding = text_features[0] | |
| input_embedding = input_embedding / np.linalg.norm(input_embedding) | |
| cos_score = self.embeddings @ input_embedding | |
| top_10 = cos_score.argsort()[-100:][::-1] | |
| return self.artwork_urls[top_10].tolist() | |