Update handler.py

handler.py

@@ -62,13 +62,12 @@ class EndpointHandler():
             use_safetensors=True,
         ).to("cuda")
 
-        self.pidinet = PidiNetDetector.from_pretrained("lllyasviel/Annotators").to("cuda")
-
         self.pipeline.unet = torch.compile(self.pipeline.unet, mode="reduce-overhead", fullgraph=True)
         self.refiner.unet = torch.compile(self.refiner.unet, mode="reduce-overhead", fullgraph=True)
         self.pipeline.enable_model_cpu_offload()
         self.refiner.enable_model_cpu_offload()
-        self.pidinet.enable_model_cpu_offload()
+
+        self.pidinet = PidiNetDetector.from_pretrained("lllyasviel/Annotators").to("cuda")
 
     def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
         """