Update app.py

app.py

@@ -10,6 +10,7 @@ from ultralytics import YOLO
 from gtts import gTTS
 from PIL import Image
 from nltk.tokenize import sent_tokenize
+import spaces
 
 # Set device (use GPU if available, but don't initialize CUDA here)
 device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -25,6 +26,7 @@ client = OpenAI(api_key=api_key)
 # Download NLTK data
 nltk.download("punkt")
 
+@spaces.GPU
 # Lazy-load models to avoid initializing CUDA in the main process
 def load_yolo_model():
     return YOLO("yolov8s.pt")