在FPGA上实现低延迟Transformer推理，支持物理实时分析。
Low Latency Transformer Inference on FPGAs for Physics Applications with hls4ml
- 用hls4ml在FPGA上部署多头注意力等核心模块
- 三类模型在VU13P芯片上延迟均低于2微秒
- 兼容任意TensorFlow构建的Transformer，适合高能物理应用
本研究利用hls4ml在现场可编程门阵列（FPGA）上高效实现Transformer架构。我们提出了多头注意力、Softmax和归一化层的实现策略，并评估了三种不同模型。这些模型在VU13P FPGA芯片上的部署延迟均低于2微秒，展示了其在实时应用中的潜力。hls4ml对任意基于TensorFlow构建的Transformer模型的兼容性，进一步提升了该工作的可扩展性与适用范围。
原文摘要 · Abstract (English)
This study presents an efficient implementation of transformer architectures in Field-Programmable Gate Arrays(FPGAs) using hls4ml. We demonstrate the strategy for implementing the multi-head attention, softmax, and normalization layer and evaluate three distinct models. Their deployment on VU13P FPGA chip achieved latency less than 2us, demonstrating the potential for real-time applications. HLS4ML compatibility with any TensorFlow-built transformer model further enhances the scalability and applicability of this work. Index Terms: FPGAs, machine learning, transformers, high energy physics, LIGO
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。