arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2305.18239cs.CLcs.AI
收藏

从弱教师进行知识蒸馏以扩展预训练语言模型的研究

A Study on Knowledge Distillation from Weak Teacher for Scaling Up Pre-trained Language Models

KAIST(韩国科学技术院);Meta AI

Hayeon Lee, Rui Hou, Jongpil Kim, Davis Liang, Sung Ju Hwang, Alexander Min

AI总结:

本文研究在 NLP 预训练中通过弱教师蒸馏扩展预训练语言模型,系统分析教师质量、损失权重和参数重映射初始化三个关键因素对 DWT 效果的影响。

摘要:

弱教师蒸馏(DWT)是一种将知识从较小、较弱的教师模型迁移到较大的学生模型以提升其性能的方法。以往研究表明,DWT 在视觉领域和自然语言处理(NLP)预训练阶段均可能有效。具体而言,DWT 在实际场景中展现出潜力,例如利用已预训练但较旧或较小的模型来增强新一代或更大的模型,以及在资源预算不足的情况下使用。然而,在 NLP 预训练中,使用 DWT 的最优条件尚未得到充分研究。因此,本研究考察了三个关键因素以优化 DWT,这些因素不同于视觉领域或传统知识蒸馏中所使用的因素。这些因素包括:(i) 教师模型质量对 DWT 有效性的影响,(ii) 调整 DWT 损失权重值的指导原则,以及 (iii) 参数重映射作为 DWT 学生模型初始化技术的影响。

英文摘要:

Distillation from Weak Teacher (DWT) is a method of transferring knowledge from a smaller, weaker teacher model to a larger student model to improve its performance. Previous studies have shown that DWT can be effective in the vision domain and natural language processing (NLP) pre-training stage. Specifically, DWT shows promise in practical scenarios, such as enhancing new generation or larger models using pre-trained yet older or smaller models and lacking a resource budget. However, the optimal conditions for using DWT have yet to be fully investigated in NLP pre-training. Therefore, this study examines three key factors to optimize DWT, distinct from those used in the vision domain or traditional knowledge distillation. These factors are: (i) the impact of teacher model quality on DWT effectiveness, (ii) guidelines for adjusting the weighting value for DWT loss, and (iii) the impact of parameter remapping as a student model initialization technique for DWT.

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。

↑