LayerSkip enables early exit inference and self-speculative decoding for large language models to...