Guardado en:
Detalles Bibliográficos
Autores principales: Zhang, Yudi, Zhao, Weilin, Han, Xu, Zhao, Tiejun, Xu, Wang, Cao, Hailong, Zhu, Conghui
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2505.22179
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910973165568000
author Zhang, Yudi
Zhao, Weilin
Han, Xu
Zhao, Tiejun
Xu, Wang
Cao, Hailong
Zhu, Conghui
author_facet Zhang, Yudi
Zhao, Weilin
Han, Xu
Zhao, Tiejun
Xu, Wang
Cao, Hailong
Zhu, Conghui
contents Speculative decoding and quantization effectively accelerate memory-bound inference of large language models. Speculative decoding mitigates the memory bandwidth bottleneck by verifying multiple tokens within a single forward pass, which increases computational effort. Quantization achieves this optimization by compressing weights and activations into lower bit-widths and also reduces computations via low-bit matrix multiplications. To further leverage their strengths, we investigate the integration of these two techniques. Surprisingly, experiments applying the advanced speculative decoding method EAGLE-2 to various quantized models reveal that the memory benefits from 4-bit weight quantization are diminished by the computational load from speculative decoding. Specifically, verifying a tree-style draft incurs significantly more time overhead than a single-token forward pass on 4-bit weight quantized models. This finding led to our new speculative decoding design: a hierarchical framework that employs a small model as an intermediate stage to turn tree-style drafts into sequence drafts, leveraging the memory access benefits of the target quantized model. Experimental results show that our hierarchical approach achieves a 2.78$\times$ speedup across various tasks for the 4-bit weight Llama-3-70B model on an A100 GPU, outperforming EAGLE-2 by 1.31$\times$. Code available at https://github.com/AI9Stars/SpecMQuant.
format Preprint
id arxiv_https___arxiv_org_abs_2505_22179
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Speculative Decoding Meets Quantization: Compatibility Evaluation and Hierarchical Framework Design
Zhang, Yudi
Zhao, Weilin
Han, Xu
Zhao, Tiejun
Xu, Wang
Cao, Hailong
Zhu, Conghui
Computation and Language
Artificial Intelligence
Machine Learning
Speculative decoding and quantization effectively accelerate memory-bound inference of large language models. Speculative decoding mitigates the memory bandwidth bottleneck by verifying multiple tokens within a single forward pass, which increases computational effort. Quantization achieves this optimization by compressing weights and activations into lower bit-widths and also reduces computations via low-bit matrix multiplications. To further leverage their strengths, we investigate the integration of these two techniques. Surprisingly, experiments applying the advanced speculative decoding method EAGLE-2 to various quantized models reveal that the memory benefits from 4-bit weight quantization are diminished by the computational load from speculative decoding. Specifically, verifying a tree-style draft incurs significantly more time overhead than a single-token forward pass on 4-bit weight quantized models. This finding led to our new speculative decoding design: a hierarchical framework that employs a small model as an intermediate stage to turn tree-style drafts into sequence drafts, leveraging the memory access benefits of the target quantized model. Experimental results show that our hierarchical approach achieves a 2.78$\times$ speedup across various tasks for the 4-bit weight Llama-3-70B model on an A100 GPU, outperforming EAGLE-2 by 1.31$\times$. Code available at https://github.com/AI9Stars/SpecMQuant.
title Speculative Decoding Meets Quantization: Compatibility Evaluation and Hierarchical Framework Design
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2505.22179