test moving feature extractor output to torch device

app.py

@@ -335,7 +335,7 @@ def load_img_from_URL(URL):
 
 @spaces.GPU
 def embed_img(input_image):
-    tokens = pipe.feature_extractor(input_image)
+    tokens = pipe.feature_extractor(input_image).to(torch_device)
     img_model = pipe.image_encoder.to(torch_device)
     with torch.no_grad():
         embeds = img_model(torch.tensor(tokens.pixel_values[0]).unsqueeze(0))