SDSAT: Accelerating LLM Inference through Speculative Decoding with Semantic Adaptive Tokens

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Chengbo, Zhu, Yong
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929299487981568
author Liu, Chengbo
Zhu, Yong
author_facet Liu, Chengbo
Zhu, Yong
contents We propose an acceleration scheme for large language models (LLMs) through Speculative Decoding with Semantic Adaptive Tokens (SDSAT). The primary objective of this design is to enhance the LLM model's ability to generate draft tokens more accurately without compromising the model's accuracy. The core strategies involve: 1) Fine-tune the model by incorporating semantic adaptive tokens that possess flexible decoding capabilities without changing its structure, allowing them to generate high-quality draft tokens. 2) By employing a training method that does not affect the standard tokens, the model can acquire parallel decoding abilities atop its original framework with minimal training overhead. 3) We have designed the "two-step-draft-then-verify" generation strategies using both greedy search and nucleus sampling. Experiments conducted on the CodeLlama-13B and 7B models have yielded speed increases of over 3.5X and 3.0X, respectively. Please refer to https://github.com/hasuoshenyun/SDSAT.
format Preprint
id arxiv_https___arxiv_org_abs_2403_18647
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SDSAT: Accelerating LLM Inference through Speculative Decoding with Semantic Adaptive Tokens
Liu, Chengbo
Zhu, Yong
Computation and Language
We propose an acceleration scheme for large language models (LLMs) through Speculative Decoding with Semantic Adaptive Tokens (SDSAT). The primary objective of this design is to enhance the LLM model's ability to generate draft tokens more accurately without compromising the model's accuracy. The core strategies involve: 1) Fine-tune the model by incorporating semantic adaptive tokens that possess flexible decoding capabilities without changing its structure, allowing them to generate high-quality draft tokens. 2) By employing a training method that does not affect the standard tokens, the model can acquire parallel decoding abilities atop its original framework with minimal training overhead. 3) We have designed the "two-step-draft-then-verify" generation strategies using both greedy search and nucleus sampling. Experiments conducted on the CodeLlama-13B and 7B models have yielded speed increases of over 3.5X and 3.0X, respectively. Please refer to https://github.com/hasuoshenyun/SDSAT.
title SDSAT: Accelerating LLM Inference through Speculative Decoding with Semantic Adaptive Tokens
topic Computation and Language
url https://arxiv.org/abs/2403.18647