Transformer-based models have rapidly spread from text to speech, vision, and other modalities. This has created challenges for ...