Update handler.py

handler.py

@@ -154,6 +154,13 @@ class EndpointHandler:
 
             #apply_teacache(self.image_to_video)
 
+            # Compilation requires some time to complete, so it is best suited for
+            # situations where you prepare your pipeline once and then perform the 
+            # same type of inference operations multiple times.
+            # For example, calling the compiled pipeline on a different image size
+            # triggers compilation again which can be expensive.
+            self.image_to_video.unet = torch.compile(self.image_to_video.unet, mode="reduce-overhead", fullgraph=True)
+
         else:
             # Initialize models with bfloat16 precision
             self.text_to_video = LTXPipeline.from_pretrained(
@@ -163,6 +170,14 @@ class EndpointHandler:
             
             #apply_teacache(self.text_to_video)
 
+            # Compilation requires some time to complete, so it is best suited for
+            # situations where you prepare your pipeline once and then perform the 
+            # same type of inference operations multiple times.
+            # For example, calling the compiled pipeline on a different image size
+            # triggers compilation again which can be expensive.
+            self.text_to_video.unet = torch.compile(self.text_to_video.unet, mode="reduce-overhead", fullgraph=True)
+
+
         # Initialize LoRA tracking
         self._current_lora_model = None