LlamaF: An Efficient Llama2 Architecture Accelerator on Embedded FPGAs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Han, Li, Yutong, Ji, Shihao
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910609428185088
author Xu, Han
Li, Yutong
Ji, Shihao
author_facet Xu, Han
Li, Yutong
Ji, Shihao
contents Large language models (LLMs) have demonstrated remarkable abilities in natural language processing. However, their deployment on resource-constrained embedded devices remains difficult due to memory and computational demands. In this paper, we present an FPGA-based accelerator designed to improve LLM inference performance on embedded FPGAs. We employ post-training quantization to reduce model size and optimize for off-chip memory bandwidth. Our design features asynchronous computation and a fully pipelined accelerator for matrix-vector multiplication. Experiments of the TinyLlama 1.1B model on a Xilinx ZCU102 platform show a 14.3-15.8x speedup and a 6.1x power efficiency improvement over running exclusively on ZCU102 processing system (PS).
format Preprint
id arxiv_https___arxiv_org_abs_2409_11424
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LlamaF: An Efficient Llama2 Architecture Accelerator on Embedded FPGAs
Xu, Han
Li, Yutong
Ji, Shihao
Hardware Architecture
Large language models (LLMs) have demonstrated remarkable abilities in natural language processing. However, their deployment on resource-constrained embedded devices remains difficult due to memory and computational demands. In this paper, we present an FPGA-based accelerator designed to improve LLM inference performance on embedded FPGAs. We employ post-training quantization to reduce model size and optimize for off-chip memory bandwidth. Our design features asynchronous computation and a fully pipelined accelerator for matrix-vector multiplication. Experiments of the TinyLlama 1.1B model on a Xilinx ZCU102 platform show a 14.3-15.8x speedup and a 6.1x power efficiency improvement over running exclusively on ZCU102 processing system (PS).
title LlamaF: An Efficient Llama2 Architecture Accelerator on Embedded FPGAs
topic Hardware Architecture
url https://arxiv.org/abs/2409.11424