Recurrent Drafter for Efficient Speculative Decoding in Large Language Models

Enhancing large language model efficiency through a recurrent drafter approach.