Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Jikai, Tian, Zhenxu, Li, Juntao, Xia, Qingrong, Duan, Xinyu, Wang, Zhefeng, Huai, Baoxing, Zhang, Min
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908534194569216
author Wang, Jikai
Tian, Zhenxu
Li, Juntao
Xia, Qingrong
Duan, Xinyu
Wang, Zhefeng
Huai, Baoxing
Zhang, Min
author_facet Wang, Jikai
Tian, Zhenxu
Li, Juntao
Xia, Qingrong
Duan, Xinyu
Wang, Zhefeng
Huai, Baoxing
Zhang, Min
contents Recent works have revealed the great potential of speculative decoding in accelerating the autoregressive generation process of large language models. The success of these methods relies on the alignment between draft candidates and the sampled outputs of the target model. Existing methods mainly achieve draft-target alignment with training-based methods, e.g., EAGLE, Medusa, involving considerable training costs. In this paper, we present a training-free alignment-augmented speculative decoding algorithm. We propose alignment sampling, which leverages output distribution obtained in the prefilling phase to provide more aligned draft candidates. To further benefit from high-quality but non-aligned draft candidates, we also introduce a simple yet effective flexible verification strategy. Through an adaptive probability threshold, our approach can improve generation accuracy while further improving inference efficiency. Experiments on 8 datasets (including question answering, summarization and code completion tasks) show that our approach increases the average generation score by 3.3 points for the LLaMA3 model. Our method achieves a mean acceptance length up to 2.39 and speed up generation by 2.23.
format Preprint
id arxiv_https___arxiv_org_abs_2505_13204
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification
Wang, Jikai
Tian, Zhenxu
Li, Juntao
Xia, Qingrong
Duan, Xinyu
Wang, Zhefeng
Huai, Baoxing
Zhang, Min
Computation and Language
Recent works have revealed the great potential of speculative decoding in accelerating the autoregressive generation process of large language models. The success of these methods relies on the alignment between draft candidates and the sampled outputs of the target model. Existing methods mainly achieve draft-target alignment with training-based methods, e.g., EAGLE, Medusa, involving considerable training costs. In this paper, we present a training-free alignment-augmented speculative decoding algorithm. We propose alignment sampling, which leverages output distribution obtained in the prefilling phase to provide more aligned draft candidates. To further benefit from high-quality but non-aligned draft candidates, we also introduce a simple yet effective flexible verification strategy. Through an adaptive probability threshold, our approach can improve generation accuracy while further improving inference efficiency. Experiments on 8 datasets (including question answering, summarization and code completion tasks) show that our approach increases the average generation score by 3.3 points for the LLaMA3 model. Our method achieves a mean acceptance length up to 2.39 and speed up generation by 2.23.
title Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification
topic Computation and Language
url https://arxiv.org/abs/2505.13204