Towards Optimal Multi-draft Speculative Decoding

Open in new window