Update app.py

app.py

@@ -34,6 +34,7 @@ xflux_pipeline = XFluxPipeline(Config.name, Config.device, Config.offload)
 xflux_pipeline.to(device='cuda' if torch.cuda.is_available() else 'cpu')
 
 @spaces.GPU
+@torch.inference_mode()
 def generate(**kwargs): 
     torch.multiprocessing.set_start_method('spawn', force=True)
     return xflux_pipeline.gradio_generate(**kwargs)