gpu full load

app.py

@@ -505,7 +505,8 @@ if __name__ == "__main__":
     from videox_fun.ui.controller import flow_scheduler_dict
     
     # Keep everything on GPU to avoid offload hooks that break ZeroGPU pickling
-    GPU_memory_mode = "model_full_load_and_qfloat8"
+    # and avoid qfloat8 wrapping (which introduces lambda hooks) to stay pickle-safe.
+    GPU_memory_mode = "model_full_load"
     compile_dit = False
     weight_dtype = torch.bfloat16
     server_name = "0.0.0.0"