Update app.py

app.py

@@ -85,7 +85,7 @@ pipeline = DistilledPipeline(
     spatial_upsampler_path=SPATIAL_UPSAMPLER,
     gemma_root=GEMMA_ROOT,
     loras=[],
-#    quantization=QuantizationPolicy.fp8_cast(),
+    quantization=QuantizationPolicy.fp8_cast(),
 )
 
 # Preload all models so first request is fast.
@@ -112,7 +112,8 @@ ledger.spatial_upsampler = lambda: _spatial_upsampler
 print("All models preloaded!")
 
 
-@spaces.GPU(duration=120, size='xlarge')
+@spaces.GPU(duration=300)
+@torch.inference_mode()
 def generate_video(
     input_image,
     prompt: str,