H hao-ai-lab.github.io Consistency LLM: converting LLMs to parallel decoders accelerates inference 3.5x 461 points zhisbug 2 years ago 98 comments
