Overcoming Joint Intractability with Lossless Hierarchical Speculative Decoding

ICLROral2026

Authors
Yuxuan Zhou, Fei Huang, Heng Li, Fengyi Wu, Tianyu Wang, jianwei zhang, Junyang Lin, Zhi-Qi Cheng
Affiliation
Baidu
Venue
ICLR 2026
Track
Oral

TL;DR

Hierarchical speculative decoding enables lossless long-prefix verification with substantially improved decoding efficiency.

Opening excerpt from the authors’ abstract. source

Read the paper

← All ICLR 2026 Oral papers · Browse the whole archive