arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2206.14366cs.CLcs.AI
收藏

再论基于转换器的语言模型的知识蒸馏

Knowledge Distillation of Transformer-based Language Models Revisited

Alibaba Group(阿里巴巴集团);Zhejiang University(浙江大学)

Chengqiang Lu, Jianwei Zhang, Yunfei Chu, Zhengyu Chen, Jingren Zhou, Fei Wu, Haiqing Chen, Hongxia Yang

摘要:

在过去的几年里,基于transformer的预训练语言模型在工业界和学术界都取得了惊人的成功。然而,大的模型大小和高运行时延迟是将它们应用于实际的严重障碍,特别是在移动电话和物联网(IoT)设备上。为了压缩该模型,近年来围绕知识蒸馏(KD)这一主题,出现了大量的文献。然而,KD如何在基于transformer的模型中工作仍然不清楚。我们梳理了KD的组成部分,并提出了一个统一的KD框架。通过该框架,系统和广泛的实验,花费了超过23,000 GPU小时,从知识类型,匹配策略,宽度-深度权衡,初始化,模型大小等角度进行了全面的分析。我们的实证结果揭示了预训练语言模型的升华,并与以前的最先进的(SOTA)有相对显着的改善。最后,我们为基于transformer的模型中的KD提供了最佳实践指南。

英文摘要:

In the past few years, transformer-based pre-trained language models have achieved astounding success in both industry and academia. However, the large model size and high run-time latency are serious impediments to applying them in practice, especially on mobile phones and Internet of Things (IoT) devices. To compress the model, considerable literature has grown up around the theme of knowledge distillation (KD) recently. Nevertheless, how KD works in transformer-based models is still unclear. We tease apart the components of KD and propose a unified KD framework. Through the framework, systematic and extensive experiments that spent over 23,000 GPU hours render a comprehensive analysis from the perspectives of knowledge types, matching strategies, width-depth trade-off, initialization, model size, etc. Our empirical results shed light on the distillation in the pre-train language model and with relative significant improvement over previous state-of-the-arts(SOTA). Finally, we provide a best-practice guideline for the KD in transformer-based models.

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。

↑