Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wang, Peipei, Guan, Wu, Liang, Liping, Wang, Zhijun, Luo, Hanqing, Zhang, Zhibin
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2507.14139
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866913948788326400
author Wang, Peipei
Guan, Wu
Liang, Liping
Wang, Zhijun
Luo, Hanqing
Zhang, Zhibin
author_facet Wang, Peipei
Guan, Wu
Liang, Liping
Wang, Zhijun
Luo, Hanqing
Zhang, Zhibin
contents This paper introduces SpeedLLM, a neural network accelerator designed on the Xilinx Alevo U280 platform and optimized for the Tinyllama framework to enhance edge computing performance. Key innovations include data stream parallelism, a memory reuse strategy, and Llama2 operator fusion, which collectively reduce latency and energy consumption. SpeedLLM's data pipeline architecture optimizes the read-compute-write cycle, while the memory strategy minimizes FPGA resource demands. The operator fusion boosts computational density and throughput. Results show SpeedLLM outperforms traditional Tinyllama implementations, achieving up to 4.8* faster performance and 1.18* lower energy consumption, offering improvements in edge devices.
format Preprint
id arxiv_https___arxiv_org_abs_2507_14139
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SpeedLLM: An FPGA Co-design of Large Language Model Inference Accelerator
Wang, Peipei
Guan, Wu
Liang, Liping
Wang, Zhijun
Luo, Hanqing
Zhang, Zhibin
Hardware Architecture
This paper introduces SpeedLLM, a neural network accelerator designed on the Xilinx Alevo U280 platform and optimized for the Tinyllama framework to enhance edge computing performance. Key innovations include data stream parallelism, a memory reuse strategy, and Llama2 operator fusion, which collectively reduce latency and energy consumption. SpeedLLM's data pipeline architecture optimizes the read-compute-write cycle, while the memory strategy minimizes FPGA resource demands. The operator fusion boosts computational density and throughput. Results show SpeedLLM outperforms traditional Tinyllama implementations, achieving up to 4.8* faster performance and 1.18* lower energy consumption, offering improvements in edge devices.
title SpeedLLM: An FPGA Co-design of Large Language Model Inference Accelerator
topic Hardware Architecture
url https://arxiv.org/abs/2507.14139