arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2406.09804cs.AR
收藏

优化多加速器平台上 Transformer 网络的层融合调度

Optimizing Layer-Fused Scheduling of Transformer Networks on Multi-accelerator Platforms

Steven Colleman, Arne Symons, Victor J. B. Jung, Marian Verhelst

AI总结:

本文扩展 DSE 框架 Stream 以支持多类加速器上的 Transformer 调度与层融合探索,发现按注意力头输入大小调整执行调度可显著降低活动特征数据内存需求。

摘要:

Transformer 网络的影响力正迅速增长,但它们伴随着显著的计算复杂性。因此,必须理解如何在现代神经处理器硬件上最优地映射并执行这些网络。迄今为止,关于 Transformer 调度优化的文献一直聚焦于 GPU 和特定 ASIC 上的部署。本工作通过扩展 DSE 框架 Stream,使其支持跨多种硬件架构和不同执行调度的 Transformer,从而实现广泛的硬件/映射探索。经过验证后,我们探索了 Transformer 层/注意力头的最优调度,并研究层融合是否有利于改善延迟、能耗或内存需求。研究表明,通过根据注意力头输入的大小调整执行调度,可以大幅降低活动特征数据的内存需求。

英文摘要:

The impact of transformer networks is booming, yet, they come with significant computational complexity. It is therefore essential to understand how to optimally map and execute these networks on modern neural processor hardware. So far, literature on transformer scheduling optimization has been focusing on deployment on GPU and specific ASICs. This work enables extensive hardware/mapping exploration by extending the DSE framework Stream towards support for transformers across a wide variety of hardware architectures and different execution schedules. After validation, we explore the optimal schedule for transformer layers/attention heads and investigate whether layer fusion is beneficial to improve latency, energy or memory requirements. Our study shows that the memory requirements for active feature data can be drastically reduced, by adapting the execution schedule based on the size of the input of the attention head.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑