Shijing Hu — accepted AI-conference papers

2026

Bridging Draft Policy Misalignment: Group Tree Optimization for Speculative Decoding

Shijing Hu, Jingyang Li, Zhihui Lu, Pan Zhou

ICLR 2026poster

Speculative decoding accelerates large language model (LLM) inference by letting a lightweight draft model propose multiple tokens that the target model verifies in parallel. Yet existing training objectives optimize only a single greedy draft path, while decoding follows a tree policy that re-ranks…

Cited by 0 Source ↗Scholar ↗

2025

GRIFFIN: Effective Token Alignment for Faster Speculative Decoding

Shijing Hu, Jingyang Li, Xingyu Xie, Zhihui Lu, Kim-chuan Toh, Pan Zhou

NeurIPS 2025poster

Speculative decoding accelerates inference in large language models (LLMs) by generating multiple draft tokens simultaneously. However, existing methods often struggle with token misalignment between the training and decoding phases, limiting their performance. To address this, we propose GRIFFIN, a…

Cited by 0 Source ↗code ↗Scholar ↗