Upload to debug

app.py

@@ -75,6 +75,13 @@ import spaces
 # Load model once globally
 #ckpt_path = "ckpts/ep215_0906.oat.ckpt"
 #model = AVSEModule.load_from_checkpoint(ckpt_path)
+avse_model = AVSEModule()
+#avse_state_dict = torch.load("ckpts/ep215_0906.oat.ckpt")
+avse_state_dict = torch.load("ckpts/ep220_0908.oat.ckpt")
+avse_model.load_state_dict(avse_state_dict, strict=True)
+avse_model.to("cuda")
+avse_model.eval()
+
 CHUNK_SIZE_AUDIO = 2 * 48000  # 3 sec at 16kHz
 CHUNK_SIZE_VIDEO = 2 * 75     # 25fps × 3 sec
 
@@ -159,15 +166,7 @@ def extract_resampled_audio(video_path, target_sr=16000):
 def yolo_detection(frame, verbose=False):
     return model(frame, verbose=verbose)[0]
 
-@spaces.GPU
 def extract_faces(video_file):
-    avse_model = AVSEModule()
-    #avse_state_dict = torch.load("ckpts/ep215_0906.oat.ckpt")
-    avse_state_dict = torch.load("ckpts/ep220_0908.oat.ckpt")
-    avse_model.load_state_dict(avse_state_dict, strict=True)
-    avse_model.to("cuda")
-    avse_model.eval()
-
     cap = cv2.VideoCapture(video_file)
     fps = cap.get(cv2.CAP_PROP_FPS)
     frames = []