POSS: Position Specialist Generates Better Draft for Speculative Decoding

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Huang, Langlin, Huang, Chengsong, Leng, Jixuan, Huang, Di, Huang, Jiaxin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910986315759616
author Huang, Langlin
Huang, Chengsong
Leng, Jixuan
Huang, Di
Huang, Jiaxin
author_facet Huang, Langlin
Huang, Chengsong
Leng, Jixuan
Huang, Di
Huang, Jiaxin
contents Speculative decoding accelerates Large Language Model (LLM) inference by using a small draft model to predict multiple tokens, and a large target model to verify these tokens in parallel. Recent studies leverage the hidden state of the target model to enhance draft model prediction accuracy. However, existing methods suffer from the degrading quality of draft token predictions at later positions, due to error accumulation in draft model generated features. In this paper, we propose Position Specialists (PosS), which consist of multiple position-specialized draft layers to generate tokens at assigned position(s). Position specialists greatly improve token acceptance rate at later positions per drafting round, as each specialist only needs to focus on handling a certain level of draft model feature deviation. Experiment results on Llama-3-8B-Instruct and Llama-2-13B-chat across six datasets demonstrate that PosS effectively improves over baselines on average acceptance length and speed-up ratio. Our codebase is available at https://github.com/shrango/PosS.
format Preprint
id arxiv_https___arxiv_org_abs_2506_03566
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle POSS: Position Specialist Generates Better Draft for Speculative Decoding
Huang, Langlin
Huang, Chengsong
Leng, Jixuan
Huang, Di
Huang, Jiaxin
Computation and Language
Artificial Intelligence
Machine Learning
Speculative decoding accelerates Large Language Model (LLM) inference by using a small draft model to predict multiple tokens, and a large target model to verify these tokens in parallel. Recent studies leverage the hidden state of the target model to enhance draft model prediction accuracy. However, existing methods suffer from the degrading quality of draft token predictions at later positions, due to error accumulation in draft model generated features. In this paper, we propose Position Specialists (PosS), which consist of multiple position-specialized draft layers to generate tokens at assigned position(s). Position specialists greatly improve token acceptance rate at later positions per drafting round, as each specialist only needs to focus on handling a certain level of draft model feature deviation. Experiment results on Llama-3-8B-Instruct and Llama-2-13B-chat across six datasets demonstrate that PosS effectively improves over baselines on average acceptance length and speed-up ratio. Our codebase is available at https://github.com/shrango/PosS.
title POSS: Position Specialist Generates Better Draft for Speculative Decoding
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2506.03566