Fast-DLLM: Training-Free Acceleration of Diffusion LLM