Update handler.py

handler.py

@@ -50,14 +50,14 @@ class EndpointHandler():
             image = None
             mask_image = None 
 
-        #self.pipe.enable_xformers_memory_efficient_attention()
+        self.pipe.enable_xformers_memory_efficient_attention()
         
         # run inference pipeline
         out = self.pipe(prompt=prompt, image=image, mask_image=mask_image)
 
         image = out.images[0].resize((1024, 1024))
 
-        #self.pipe2.enable_xformers_memory_efficient_attention()
+        self.pipe2.enable_xformers_memory_efficient_attention()
 
         image = self.pipe2(
             prompt=prompt,
@@ -69,7 +69,7 @@ class EndpointHandler():
             output_type="latent",  # let's keep in latent to save some VRAM
         ).images[0]
         
-        #self.pipe3.enable_xformers_memory_efficient_attention()
+        self.pipe3.enable_xformers_memory_efficient_attention()
         
         image = self.pipe3(
             prompt=prompt,