Update app.py

app.py

@@ -244,13 +244,12 @@ def extract_video_frames(video_path: str):
     log.info(f"clip_chunk: {clip_chunk.shape}, sync_chunk: {sync_chunk.shape}")
     return clip_chunk, sync_chunk, duration_sec
 
-
+@spaces.GPU
 def extract_features_cpu(clip_chunk, sync_chunk, caption):
     model = _MODELS["feature_extractor"]
     
     info = {}
     with torch.no_grad():
-        # videoprism 是 CPU
         clip_input = torch.from_numpy(clip_chunk).unsqueeze(0)
         video_feat, frame_embed, _, text_feat = \
             model.encode_video_and_text_with_videoprism(clip_input, [caption])