Yes. Torch shouldn’t be necessary in this case. Cast the torch tensor to a numpy array and pass that to the engine.