arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2408.02897cs.LGcs.AI
收藏

一种指标驱动的混合精度训练方法

A Metric Driven Approach to Mixed Precision Training

Google(谷歌公司)

Mitchelle Rasquinha, Gil Tabak

AI总结:

本文提出一种指标驱动的方法,用于在多种8位数据类型中选择低精度数值格式,以降低DNN训练的内存与计算开销,并验证其可扩展语言表示模型训练且适用于其他架构。

摘要:

随着深度学习方法的发展,人们普遍认为增大神经网络规模可以提升模型质量。然而,这是以增加内存和计算需求为代价的,二者同样需要随之增长。为控制硬件成本,研究人员提出了多种效率技术,其中之一是使用低精度数值。近期的加速器引入了几种不同的8位数据类型,以在数值方面帮助适配DNN。在本文中,我们提出了一种指标驱动的方法,以辅助数值格式的选择。我们展示了该方法如何帮助扩展语言表示模型的训练。该技术还可推广到其他模型架构。

英文摘要:

As deep learning methodologies have developed, it has been generally agreed that increasing neural network size improves model quality. However, this is at the expense of memory and compute requirements, which also need to be increased. Various efficiency techniques have been proposed to rein in hardware costs, one being the use of low precision numerics. Recent accelerators have introduced several different 8-bit data types to help accommodate DNNs in terms of numerics. In this paper, we identify a metric driven methodology to aid in the choice of numerics. We demonstrate how such a methodology can help scale training of a language representation model. The technique can be generalized to other model architectures.

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。

↑