Cerberus: Efficient Inference with Adaptive Parallel Decoding and Sequential Knowledge Enhancement

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Liu, Yuxuan, Li, Wenyuan, Cui, Laizhong, Yang, Hailiang
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866912075243061248
author Liu, Yuxuan
Li, Wenyuan
Cui, Laizhong
Yang, Hailiang
author_facet Liu, Yuxuan
Li, Wenyuan
Cui, Laizhong
Yang, Hailiang
contents Large language models (LLMs) often face a bottleneck in inference speed due to their reliance on auto-regressive decoding. Recently, parallel decoding has shown significant promise in enhancing inference efficiency. However, we have identified two key issues with existing parallel decoding frameworks: (1) decoding heads fail to balance prediction accuracy and the parallelism of execution, and (2) parallel decoding is not a universal solution, as it can bring unnecessary overheads at some challenging decoding steps. To address these issues, we propose Cerberus, an adaptive parallel decoding framework introduces the gating mechanism to enable the LLMs to adaptively choose appropriate decoding approaches at each decoding step, along with introducing a new paradigm of decoding heads that introduce the sequential knowledge while maintaining execution parallelism. The experiment results demonstrate that the Cerberus can achieve up to 2.12x speed up compared to auto-regressive decoding, and outperforms one of the leading parallel decoding frameworks, Medusa, with a 10% - 30% increase in acceleration and superior generation quality.
format Preprint
id arxiv_https___arxiv_org_abs_2410_13344
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Cerberus: Efficient Inference with Adaptive Parallel Decoding and Sequential Knowledge Enhancement
Liu, Yuxuan
Li, Wenyuan
Cui, Laizhong
Yang, Hailiang
Computation and Language
Artificial Intelligence
Large language models (LLMs) often face a bottleneck in inference speed due to their reliance on auto-regressive decoding. Recently, parallel decoding has shown significant promise in enhancing inference efficiency. However, we have identified two key issues with existing parallel decoding frameworks: (1) decoding heads fail to balance prediction accuracy and the parallelism of execution, and (2) parallel decoding is not a universal solution, as it can bring unnecessary overheads at some challenging decoding steps. To address these issues, we propose Cerberus, an adaptive parallel decoding framework introduces the gating mechanism to enable the LLMs to adaptively choose appropriate decoding approaches at each decoding step, along with introducing a new paradigm of decoding heads that introduce the sequential knowledge while maintaining execution parallelism. The experiment results demonstrate that the Cerberus can achieve up to 2.12x speed up compared to auto-regressive decoding, and outperforms one of the leading parallel decoding frameworks, Medusa, with a 10% - 30% increase in acceleration and superior generation quality.
title Cerberus: Efficient Inference with Adaptive Parallel Decoding and Sequential Knowledge Enhancement
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2410.13344