Update app.py

app.py

@@ -7,6 +7,7 @@ from scipy.spatial import cKDTree
 import numpy as np
 import math
 import time
+import spaces
 
 # Model initialization
 model = None
@@ -26,7 +27,7 @@ def load_model():
             attn_implementation='sdpa',
             torch_dtype=torch.bfloat16
         )
-        model = model.eval().cuda()
+        model = model.eval()
         tokenizer = AutoTokenizer.from_pretrained(
             'openbmb/MiniCPM-V-4_5',
             trust_remote_code=True
@@ -84,6 +85,7 @@ def encode_video(video_path, choose_fps=3, force_packing=None):
     
     return frames, frame_ts_id_group, video_duration, len(frame_idx), packing_nums
 
+@spaces.GPU(duration=60)
 def process_video_and_question(video, question, fps, force_packing, history):
     if video is None:
         gr.Warning("Please upload a video first.")
@@ -96,6 +98,7 @@ def process_video_and_question(video, question, fps, force_packing, history):
     try:
         # Load model if not already loaded
         model, tokenizer = load_model()
+        model = model.cuda()
         
         # Encode video
         gr.Info(f"Processing video with {fps} FPS...")