This story has moved to /story/efficient-rwkv-based-representation-learning-for-3d-point-clouds/.