format

handler.py

@@ -10,12 +10,13 @@ class EndpointHandler:
         Initialize the model
         """
         self.sign_ids = np.load(os.path.join(path, "sign_ids.npy"))
-        self.sign_embeddings = np.load(os.path.join(path, "vanilla_large-patch14_image_embeddings_normalized.npy"))
+        self.sign_embeddings = np.load(
+            os.path.join(path, "vanilla_large-patch14_image_embeddings_normalized.npy")
+        )
 
         hf_model_path = "openai/clip-vit-large-patch14"
         self.model = CLIPModel.from_pretrained(hf_model_path)
         self.tokenizer = CLIPTokenizer.from_pretrained(hf_model_path)
-    
 
     def __call__(self, data: Dict[str, Any]) -> List[float]:
         """
@@ -25,7 +26,9 @@ class EndpointHandler:
         Return:
               A :obj:`list` | `dict`: will be serialized and returned
         """
-        token_inputs = self.tokenizer([data["inputs"]],  padding=True, return_tensors="pt")
+        token_inputs = self.tokenizer(
+            [data["inputs"]], padding=True, return_tensors="pt"
+        )
         query_embed = self.model.get_text_features(**token_inputs)
         np_query_embed = query_embed.detach().cpu().numpy()[0]
         np_query_embed /= np.linalg.norm(np_query_embed)
@@ -37,7 +40,7 @@ class EndpointHandler:
         cos_similarites = w * (self.sign_embeddings @ np_query_embed)
         count_above_threshold = np.sum(cos_similarites > threshold)
         sign_id_arg_rankings = np.argsort(cos_similarites)[::-1]
-        
+
         threshold_id_arg_rankings = sign_id_arg_rankings[:count_above_threshold]
 
         result_sign_ids = self.sign_ids[threshold_id_arg_rankings]