LiLiCorr: Lightweight Likelihood Correlation of Parallel Drafts for Speculative Decoding

Method and contribution

Scores adjacent compatibility among top-K parallel draft candidates and selects a draft through a cheap greedy walk.

Uses the candidate–edge–path pattern with local training and inference tailored to speculative prefix acceptance.

Read in context

Structured Decoding: Higher-Order CRFs, Diffusion Language Models, and Candidate Lattices →

Citation

Matan Rusanovsky, Yoav Miron, Roy Uziel, Omer Belhasin, Hao Guo, Ran Zilberstein, Maor Ashkenazi, Michael Elad. LiLiCorr: Lightweight Likelihood Correlation of Parallel Drafts for Speculative Decoding. arXiv 2026.

@misc{rusanovsky2026lilicorr,
  title = {{LiLiCorr: Lightweight Likelihood Correlation of Parallel Drafts for Speculative Decoding}},
  author = {Rusanovsky, Matan and Miron, Yoav and Uziel, Roy and Belhasin, Omer and Guo, Hao and Zilberstein, Ran and Ashkenazi, Maor and Elad, Michael},
  year = {2026},
  url = {https://arxiv.org/abs/2608.20530},
  eprint = {2608.20530},
  archivePrefix = {arXiv}
}