Fetching the paper…

Dynamic Speculation Lookahead Accelerates Speculative Decoding of Large Language Models · Around