Co-Designing AI Models Using Speculative Decoding for Faster LLM Inference (developer.nvidia.com)