计算机科学
现场可编程门阵列
高效能源利用
可扩展性
计算
变压器
延迟(音频)
并行计算
人工神经网络
嵌入式系统
计算机体系结构
计算机硬件
计算机工程
计算科学
人工智能
算法
电气工程
工程类
电信
数据库
电压
作者
Chao Fang,Shouliang Guo,Weiliang Wu,Jun Lin,Zhongfeng Wang,Ming Kai Hsu,Lingzhi Liu
标识
DOI:10.1109/iscas48785.2022.9937659
摘要
Transformers have been an indispensable staple in deep learning. However, it is challenging to realize efficient deployment for Transformer-based model due to their substantial computation and memory demands. To address this issue, we present an efficient sparse Transformer accelerator on FPGA, namely STA, by exploiting N:M fine-grained structured sparsity. Our design features not only a unified computing engine capable of performing both sparse and dense matrix multiplications with high computational efficiency, but also a scalable softmax module eliminating the latency from intermediate off-chip data communication. Experimental results show that our implementation achieves the lowest latency compared to CPU, GPU, and prior FPGA-based accelerators. Moreover, compared with the state-of the-art FPGA-based accelerators, it can achieve up to $12.28\times$ and $51.00\times$ improvement on energy efficiency and MAC efficiency, respectively.
科研通智能强力驱动
Strongly Powered by AbleSci AI