Update handler.py

handler.py

@@ -21,10 +21,10 @@ class EndpointHandler:
         # quantize_(self.pipe.text_encoder, float8_weight_only())
         # quantize_(self.pipe.transformer, float8_dynamic_activation_float8_weight())
         self.pipe.transformer = torch.compile(
-            self.pipe.transformer, mode="reduce-overhead",
+            self.pipe.transformer, mode="max-autotune-no-cudagraphs",
         )
         self.pipe.vae = torch.compile(
-            self.pipe.vae, mode="reduce-overhead",
+            self.pipe.vae, mode="max-autotune-no-cudagraphs",
         )
 
     def __call__(self, data: Dict[str, Any]) -> Image.Image: