Efficient Speculative Decoding for Llama at Scale: Challenges and Solutions
Bangsheng Tang, Carl Chengyan Fu, Fei Kou, Grigory Sizov, Haoci Zhang, Jason Park, Jiawen Liu, Jie You, Qirui Yang, Sachin Mehta, Shengyong Cai, Xiaodong Wang, Xingyu Liu, Yunlu Li, Yanjun Zhou, Wei Wei, Zhiwei Zhao, Zixi Qi, Adolfo Victoria, Aya Ibrahim, Bram Wasti, Changkyu Kim, Daniel Haziza, Fei Sun, Giancarlo Delfin, Emily Guo, Jialin Ouyang, Jaewon Lee, Jianyu Huang, Jeremy Reizenstein, Lu Fang, Quinn Zhu, Ria Verma, Vlad Mihailescu, Xingwen Guo, Yan Cui, Ye Hu, Yejin Lee
Abstract
Speculative decoding is a standard method for accelerating the inference speed of large language models. However, scaling it for production environments poses several engineering challenges, including efficiently implementing different operations (e.g., tree attention and multi-round speculative decoding) on GPU. In this paper, we detail the training and inference optimization techniques that we have implemented to enable EAGLE-based speculative decoding at a production scale for Llama models. With these changes, we achieve a new state-of-the-art inference latency for Llama models. For example, Llama4 Maverick decodes at a speed of about 4 ms per token (with a batch size of one) on 8 NVIDIA H100 GPUs, which is 10% faster than the previously best known method. Furthermore, for EAGLE-based speculative decoding, our optimizations enable us to achieve a speed-up for large batch sizes between 1.4x and 2.0x at production scale.