research-article

A length adaptive algorithm-hardware co-design of transformer on FPGA through sparse attention and dynamic pipelining

Authors:

Caiwen DingAuthors Info & Claims

DAC '22: Proceedings of the 59th ACM/IEEE Design Automation Conference

Pages 1135 - 1140

https://doi.org/10.1145/3489517.3530585

Published: 23 August 2022 Publication History

Abstract

Transformers are considered one of the most important deep learning models since 2018, in part because it establishes state-of-the-art (SOTA) records and could potentially replace existing Deep Neural Networks (DNNs). Despite the remarkable triumphs, the prolonged turnaround time of Transformer models is a widely recognized roadblock. The variety of sequence lengths imposes additional computing overhead where inputs need to be zero-padded to the maximum sentence length in the batch to accommodate the parallel computing platforms. This paper targets the field-programmable gate array (FPGA) and proposes a coherent sequence length adaptive algorithm-hardware co-design for Transformer acceleration. Particularly, we develop a hardware-friendly sparse attention operator and a length-aware hardware resource scheduling algorithm. The proposed sparse attention operator brings the complexity of attention-based models down to linear complexity and alleviates the off-chip memory traffic. The proposed length-aware resource hardware scheduling algorithm dynamically allocates the hardware resources to fill up the pipeline slots and eliminates bubbles for NLP tasks. Experiments show that our design has very small accuracy loss and has 80.2 × and 2.6 × speedup compared to CPU and GPU implementation, and 4 × higher energy efficiency than state-of-the-art GPU accelerator optimized via CUBLAS GEMM.

References

[1]

Nikolas Adaloglou. Why Multi-head Self Attention Works: Math, Intuitions and 10+1 Hidden Insights. https://theaisummer.com/self-attention/, 2021. [Online; accessed August 30, 2021].

[2]

Jean-Baptiste Cordonnier et al. On the Relationship Between Self-Attention and Convolutional Layers. In ICLR, 2019.

[3]

Ashish Vaswani et al. Attention is All You Need. In Advances in neural information processing systems, pages 5998--6008, 2017.

[4]

Kaiming He et al. Deep Residual Learning for Image Recognition. In CVPR, 2016.

[5]

Alexey Dosovitskiy et al. An Image is Worth 16× 16 Words: Transformers for Image Recognition at Scale. In ICLR, 2021.

[6]

Ze Liu et al. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. In CVPR, 2021.

[7]

Kevin Lin et al. End-to-End Human Pose and Mesh Reconstruction with Transformers. In CVPR, 2021.

[8]

Pranav Rajpurkar et al. Know What You Don't Know: Unanswerable Questions for SQuAD. In ACL, pages 784--789, 2018.

[9]

Manzil Zaheer et al. Big Bird: Transformers for Longer Sequences. In H. Larochelle, M. Ranzato, R. Hadsell, M. F. Balcan, and H. Lin, editors, NeurIPS, volume 33, pages 17283--17297. Curran Associates, Inc., 2020.

[10]

Zihao Ye et al. BP-Transformer: Modelling Long-range Context via Binary Partitioning. arXiv preprint arXiv:1911.04070, 2019.

[11]

Nikita Kitaev et al. Reformer: The Efficient Transformer. In ICLR, 2019.

[12]

Tae Jun Ham et al. A^ 3: Accelerating Attention Mechanisms in Neural Networks with Approximation. In 2020 HPCA, pages 328--341. IEEE, 2020.

[13]

Hanrui Wang et al. SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning. In 2021 HPCA, pages 97--110. IEEE, 2021.

[14]

Tae Jun Ham et al. ELSA: Hardware-Software Co-design for Efficient, Lightweight Self-Attention Mechanism in Neural Networks. In ISCA. IEEE, 2021.

[15]

NVIDIA. TensorRT. Retrived fromhttps://developer.nvidia.com/tensorrt. Online; accessed: October 6, 2021.

[16]

Merity Stephen et al. Pointer Sentinel Mixture Models. ICLR, 2016.

[17]

Thomas Wolf et al. Transformers: State-of-the-art natural language processing. In EMNLP, pages 38--45, 2020.

[18]

Shiyang Chen et al. Et: re-thinking self-attention for transformer models on gpus. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, pages 1--18, 2021.

[19]

Shaoyi Huang et al. Hmc-tran: A tensor-core inspired hierarchical model compression for transformer-based dnns on gpu. In Proceedings of the 2021 on Great Lakes Symposium on VLSI, pages 169--174, 2021.

[20]

Jiarui Fang et al. Turbotransformers: an efficient gpu serving system for transformer models. In Proceedings of the 26th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, pages 389--402, 2021.

[21]

Zeke Wang et al. Benchmarking High Bandwidth Memory on FPGAs. arXiv preprint arXiv:2005.04324, 2020.

[22]

Panjie Qi et al. Accommodating transformer onto fpga: Coupling the balanced model compression and fpga-implementation optimization. In GLSVLSI, 2021.

[23]

Hongwu Peng et al. Binary complex neural network acceleration on fpga. In 2021 IEEE 32nd ASAP, pages 85--92. IEEE, 2021.

[24]

Hongwu Peng et al. Optimizing FPGA-based Accelerator Design for Large-Scale Molecular Similarity Search. In ICCAD, 2021.

[25]

Jacob Devlin et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In 2019 ACL, pages 4171--4186, 2019.

[26]

Victor Sanh et al. DistilBERT, a Distilled Version of BERT: Smaller, Faster, Cheaper and Lighter. arXiv preprint arXiv:1910.01108, 2019.

[27]

Yinhan Liu et al. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 2019.

[28]

Pranav Rajpurkar et al. SQuAD: 100,000+ Questions for Machine Comprehension of Text. In EMNLP, pages 2383--2392, Austin, Texas, November 2016. Association for Computational Linguistics.

[29]

Ido Dagan et al. Recognizing textual entailment: Rational, evaluation and approaches-erratum. Natural Language Engineering, 16(1):105--105, 2010.

[30]

Bill Dolan et al. Automatically constructing a corpus of sentential paraphrases. In Third International Workshop on Paraphrasing. AFNLP, 2005.

[31]

Wei Zhang et al. TernaryBERT: Distillation-aware Ultra-low Bit BERT. In 2020 EMNLP, pages 509--521, 2020.

[32]

Panjie Qi et al. Accelerating framework of transformer by hardware design and model compression co-optimization. In 2021 ICCAD, pages 1--9. IEEE, 2021.

Cited By

Avila-Garzon CBacca-Acosta J(2024)Thirty Years of Research and Methodologies in Value Co-Creation and Co-DesignSustainability10.3390/su1606236016:6(2360)Online publication date: 13-Mar-2024
https://doi.org/10.3390/su16062360
Qin YLou WWang CGong LZhou X(2024)Enhancing Long Sequence Input Processing in FPGA-Based Transformer Accelerators through Attention FusionProceedings of the Great Lakes Symposium on VLSI 202410.1145/3649476.3658810(599-603)Online publication date: 12-Jun-2024
https://dl.acm.org/doi/10.1145/3649476.3658810
Guo CXue FLeng JQiu YGuan YCui WChen QGuo M(2024)Accelerating Sparse DNNs Based on Tiled GEMMIEEE Transactions on Computers10.1109/TC.2024.336594273:5(1275-1289)Online publication date: 14-Feb-2024
https://dl.acm.org/doi/10.1109/TC.2024.3365942
Show More Cited By

Recommendations

Algorithm-hardware Co-design of Attention Mechanism on FPGA Devices
Special Issue ESWEEK 2021, CASES 2021, CODES+ISSS 2021 and EMSOFT 2021
Multi-head self-attention (attention mechanism) has been employed in a variety of fields such as machine translation, language modeling, and image processing due to its superiority in feature extraction and sequential data analysis. This is benefited from ...
Hardware-software co-design of AES on FPGA
ICACCI '12: Proceedings of the International Conference on Advances in Computing, Communications and Informatics

This paper presents a compact hardware-software co-design of Advanced Encryption Standard (AES) on the field programmable gate arrays (FPGA) designed for low-cost embedded systems. The design uses MicroBlaze, a soft-core processor from Xilinx. The ...
Design and evaluation of a hardware/software FPGA-based system for fast image processing

We evaluate the performance of a hardware/software architecture designed to perform a wide range of fast image processing tasks. The system architecture is based on hardware featuring a Field Programmable Gate Array (FPGA) co-processor and a host ...

Comments

Information & Contributors

Information

Published In

cover image ACM Conferences

DAC '22: Proceedings of the 59th ACM/IEEE Design Automation Conference

July 2022

1462 pages

ISBN:9781450391429

DOI:10.1145/3489517

General Chair:
Rob Oshana
NXP

Copyright © 2022 ACM.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected]

Sponsors

SIGDA: ACM Special Interest Group on Design Automation
IEEE CEDA

Publisher

Association for Computing Machinery

New York, NY, United States

Publication History

Published: 23 August 2022

Permissions

Request permissions for this article.

Request Permissions

Check for updates

Author Tags

Qualifiers

Research-article

Funding Sources

NSF CRII Award
U.S. DOE Office of Science, Office of Advanced Scientific Computing Research
NSF CAREER Award

Conference

DAC '22

Sponsor:

SIGDA

DAC '22: 59th ACM/IEEE Design Automation Conference

July 10 - 14, 2022

California, San Francisco

Acceptance Rates

Overall Acceptance Rate 1,770 of 5,499 submissions, 32%

Upcoming Conference

DAC '25

Sponsor:
sigda

62nd ACM/IEEE Design Automation Conference

June 22 - 26, 2025

San Francisco , CA , USA

Contributors

Other Metrics

View Article Metrics

Bibliometrics & Citations

Bibliometrics

Article Metrics

21
Total Citations
View Citations
1,225
Total Downloads

Downloads (Last 12 months)596
Downloads (Last 6 weeks)41

Reflects downloads up to 15 Sep 2024

Other Metrics

View Author Metrics

Citations

Cited By

Avila-Garzon CBacca-Acosta J(2024)Thirty Years of Research and Methodologies in Value Co-Creation and Co-DesignSustainability10.3390/su1606236016:6(2360)Online publication date: 13-Mar-2024
https://doi.org/10.3390/su16062360
Qin YLou WWang CGong LZhou X(2024)Enhancing Long Sequence Input Processing in FPGA-Based Transformer Accelerators through Attention FusionProceedings of the Great Lakes Symposium on VLSI 202410.1145/3649476.3658810(599-603)Online publication date: 12-Jun-2024
https://dl.acm.org/doi/10.1145/3649476.3658810
Guo CXue FLeng JQiu YGuan YCui WChen QGuo M(2024)Accelerating Sparse DNNs Based on Tiled GEMMIEEE Transactions on Computers10.1109/TC.2024.336594273:5(1275-1289)Online publication date: 14-Feb-2024
https://dl.acm.org/doi/10.1109/TC.2024.3365942
Song HXu LWang YWu XWang MWang Z(2024)HSViT: A Hardware and Software Collaborative Design for Vision Transformer via Multi-level Compression2024 IEEE International Symposium on Circuits and Systems (ISCAS)10.1109/ISCAS58744.2024.10557837(1-5)Online publication date: 19-May-2024
https://doi.org/10.1109/ISCAS58744.2024.10557837
Li JXu CFeng BZhao H(2023)Credit Risk Prediction Model for Listed Companies Based on CNN-LSTM and Attention MechanismElectronics10.3390/electronics1207164312:7(1643)Online publication date: 30-Mar-2023
https://doi.org/10.3390/electronics12071643
Peng HYu JNie Y(2023)Efficient Neural Network for Text Recognition in Natural Scenes Based on End-to-End Multi-Scale Attention MechanismElectronics10.3390/electronics1206139512:6(1395)Online publication date: 15-Mar-2023
https://doi.org/10.3390/electronics12061395
Sun RPang YLi W(2023)Efficient Lung Cancer Image Classification and Segmentation Algorithm Based on an Improved Swin TransformerElectronics10.3390/electronics1204102412:4(1024)Online publication date: 18-Feb-2023
https://doi.org/10.3390/electronics12041024
Xu CLi JFeng BLu B(2023)A Financial Time-Series Prediction Model Based on Multiplex Attention and Linear Transformer StructureApplied Sciences10.3390/app1308517513:8(5175)Online publication date: 21-Apr-2023
https://doi.org/10.3390/app13085175
Pan ZSharma AHu JLiu ZLi ALiu HHuang MGeng TWilliams BChen YNeville J(2023)Ising-trafficProceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence and Thirty-Fifth Conference on Innovative Applications of Artificial Intelligence and Thirteenth Symposium on Educational Advances in Artificial Intelligence10.1609/aaai.v37i8.26121(9354-9363)Online publication date: 7-Feb-2023
https://dl.acm.org/doi/10.1609/aaai.v37i8.26121
Tuli SJha N(2023)TransCODE: Co-Design of Transformers and Accelerators for Efficient Training and InferenceIEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems10.1109/TCAD.2023.328344342:12(4817-4830)Online publication date: 6-Jun-2023
https://dl.acm.org/doi/10.1109/TCAD.2023.3283443
Show More Cited By

View Options

Get Access

Login options

Check if you have access through your login credentials or your institution to get full access on this article.

Full Access

Get this Publication

View options

PDF

View or Download as a PDF file.

eReader

View online with eReader.

Media

Figures

Other

Tables

View Table of Contents