Update handler.py

handler.py

@@ -143,19 +143,23 @@ class EndpointHandler:
         Args:
             model_path: Path to LTX model weights
         """
+        print("EndpointHandler.__init__(): initializing..")
         # Enable TF32 for potential speedup on Ampere GPUs
         #torch.backends.cuda.matmul.allow_tf32 = True
 
         # use distilled weights
-        model_path = "/repository/ltxv-2b-0.9.6-distilled-04-25.safetensors"
+        model_path = Path("/repository/ltxv-2b-0.9.6-distilled-04-25.safetensors")
 
+        print("EndpointHandler.__init__(): initializing LTXVideoTransformer3DModel..")
         transformer = LTXVideoTransformer3DModel.from_single_file(
             model_path, torch_dtype=torch.bfloat16
         )
 
+        print("EndpointHandler.__init__(): initializing AutoencoderKLLTXVideo..")
         vae = AutoencoderKLLTXVideo.from_single_file(model_path, torch_dtype=torch.bfloat16)
 
         if support_image_prompt:
+            print("EndpointHandler.__init__(): initializing LTXImageToVideoPipeline..")
             self.image_to_video = LTXImageToVideoPipeline.from_pretrained(
                 "/repository",
                 transformer=transformer,
@@ -173,6 +177,7 @@ class EndpointHandler:
             #self.image_to_video.unet = torch.compile(self.image_to_video.unet, mode="reduce-overhead", fullgraph=True)
 
         else:
+            print("EndpointHandler.__init__(): initializing LTXPipeline..")
             # Initialize models with bfloat16 precision
             self.text_to_video = LTXPipeline.from_pretrained(
                 "/repository",
@@ -227,7 +232,7 @@ class EndpointHandler:
             #     magic_number = pickle_module.load(f, **pickle_load_args)
             #                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
             # _pickle.UnpicklingError: invalid load key, '<'.
-            enable_mmaudio=True,
+            enable_mmaudio=False,
         )
 
         # Determine if TeaCache is already installed or not
@@ -319,7 +324,10 @@ class EndpointHandler:
                 - content-type: MIME type
                 - metadata: Generation metadata
         """
+        print("__call__(): inputs = data.get('inputs', dict())")
         inputs = data.get("inputs", dict())
+        print("inputs = ")
+        print(inputs)
         
         input_prompt = inputs.get("prompt", "")
         input_image = inputs.get("image")
@@ -360,14 +368,14 @@ class EndpointHandler:
             quality=params.get("quality", GenerationConfig.quality),
             
             # TeaCache settings
-            enable_teacache=params.get("enable_teacache", True),
+            enable_teacache=params.get("enable_teacache", False),
 
             # values: 0 (original), 0.03 (1.6x speedup), 0.05 (2.1x speedup).
             teacache_threshold=params.get("teacache_threshold", 0.05),
 
             
             # Add enhance-a-video settings
-            enable_enhance_a_video=params.get("enable_enhance_a_video", True),
+            enable_enhance_a_video=params.get("enable_enhance_a_video", False),
             enhance_a_video_weight=params.get("enhance_a_video_weight", 5.0),
 
             # LoRA settings