Update handler.py

handler.py

@@ -42,13 +42,13 @@ class EndpointHandler():
         #self.pipe = self.pipe.to(device)
 
         self.pipe2 = AutoPipelineForInpainting.from_pretrained("stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
-        self.pipe2.enable_model_cpu_offload()
+        #self.pipe2.enable_model_cpu_offload()
         self.pipe2.enable_xformers_memory_efficient_attention()
         
-        #self.pipe2.to("cuda")
+        self.pipe2.to("cuda")
 
         self.pipe3 = AutoPipelineForImage2Image.from_pipe(self.pipe2)
-        self.pipe3.enable_model_cpu_offload()
+        #self.pipe3.enable_model_cpu_offload()
         self.pipe3.enable_xformers_memory_efficient_attention()