Saved in:
Bibliographic Details
Main Authors: Chen, Ziyi, Yang, Xiaocong, Lin, Jiacheng, Sun, Chenkai, Chang, Kevin Chen-Chuan, Huang, Jie
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2312.11462
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913939300810752
author Chen, Ziyi
Yang, Xiaocong
Lin, Jiacheng
Sun, Chenkai
Chang, Kevin Chen-Chuan
Huang, Jie
author_facet Chen, Ziyi
Yang, Xiaocong
Lin, Jiacheng
Sun, Chenkai
Chang, Kevin Chen-Chuan
Huang, Jie
contents Introduced to enhance the efficiency of large language model (LLM) inference, speculative decoding operates by having a smaller model generate a draft. A larger target model then reviews this draft to align with its output, and any acceptance by the target model results in a reduction of the number of the target model runs, ultimately improving efficiency. However, the drafting process in speculative decoding includes slow autoregressive generation and allocates equal time to generating tokens, irrespective of their importance. These inefficiencies collectively contribute to the suboptimal performance of speculative decoding. To further improve LLM inference, we introduce Cascade Speculative Drafting (CS Drafting), a speculative execution algorithm that incorporates two types of cascades. The Vertical Cascade eliminates autoregressive generation from neural models, while the Horizontal Cascade optimizes time allocation in drafting for improved efficiency. Combining both cascades, CS Drafting achieves greater speedup compared to the baselines in our experiments, while preserving the same output distribution as the target model.
format Preprint
id arxiv_https___arxiv_org_abs_2312_11462
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Cascade Speculative Drafting for Even Faster LLM Inference
Chen, Ziyi
Yang, Xiaocong
Lin, Jiacheng
Sun, Chenkai
Chang, Kevin Chen-Chuan
Huang, Jie
Machine Learning
Computation and Language
Introduced to enhance the efficiency of large language model (LLM) inference, speculative decoding operates by having a smaller model generate a draft. A larger target model then reviews this draft to align with its output, and any acceptance by the target model results in a reduction of the number of the target model runs, ultimately improving efficiency. However, the drafting process in speculative decoding includes slow autoregressive generation and allocates equal time to generating tokens, irrespective of their importance. These inefficiencies collectively contribute to the suboptimal performance of speculative decoding. To further improve LLM inference, we introduce Cascade Speculative Drafting (CS Drafting), a speculative execution algorithm that incorporates two types of cascades. The Vertical Cascade eliminates autoregressive generation from neural models, while the Horizontal Cascade optimizes time allocation in drafting for improved efficiency. Combining both cascades, CS Drafting achieves greater speedup compared to the baselines in our experiments, while preserving the same output distribution as the target model.
title Cascade Speculative Drafting for Even Faster LLM Inference
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2312.11462