BriefGPT.xyz
May, 2023
ACRoBat：动态深度学习的编译时自动批处理优化
ACRoBat: Optimizing Auto-batching of Dynamic Deep Learning at Compile Time
HTML
PDF
Pratik Fegade, Tianqi Chen, Phillip B. Gibbons, Todd C. Mowry
TL;DR
本文介绍了一种名为ACRoBat的混合静态+动态编译优化和端到端张量代码生成的框架，以实现对动态深度学习计算的高效自动批处理，相比自动批处理的最新框架DyNet，ACRoBat在Nvidia GeForce RTX 3070 GPU上表现提升了8.5倍。
Abstract
dynamic control flow
is an important technique often used to design expressive and efficient
deep learning computations
for applications such as text parsing, machine translation, exiting early out of deep models
→