arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2109.10686cs.CLcs.AIcs.CVcs.LG
收藏

有效扩展:预训练和微调Transformers的见解

Scale Efficiently: Insights from Pre-training and Fine-tuning Transformers

Google Research & DeepMind(谷歌研究院与深度思维)

Yi Tay, Mostafa Dehghani, Jinfeng Rao, William Fedus, Samira Abnar, Hyung Won Chung, Sharan Narang, Dani Yogatama, Ashish Vaswani, Donald Metzler

摘要:

关于Transformer架构的伸缩行为,仍然存在许多悬而未决的问题。这些扩展决策和发现可能是至关重要的,因为训练运行通常伴随着相关的计算成本,这些计算成本具有财务和/或环境影响。本文的目标是从预训练和微调Transformers中展示扩展见解。虽然Kaplan等人对Transformer语言模型的缩放行为进行了全面的研究,但范围仅限于上游(预训练)损失。因此,目前还不清楚这些发现是否转移到下游任务的背景下,pretrain-finetune范式。本文的主要发现如下:(1)我们表明,除了模型大小之外,模型形状对于下游微调也很重要,(2)缩放协议在不同的计算区域中的操作不同,(3)广泛采用的T5-base和T5-large大小是帕累托无效的。为此,我们提出了改进的缩放协议,从而我们重新设计的模型实现了类似的下游微调质量,同时与广泛采用的T5基础模型相比,参数减少了50%,训练速度加快了40%。我们公开发布了100多个不同T5配置的预训练检查点,以方便未来的研究和分析。

英文摘要:

There remain many open questions pertaining to the scaling behaviour of Transformer architectures. These scaling decisions and findings can be critical, as training runs often come with an associated computational cost which have both financial and/or environmental impact. The goal of this paper is to present scaling insights from pretraining and finetuning Transformers. While Kaplan et al. presents a comprehensive study of the scaling behaviour of Transformer language models, the scope is only on the upstream (pretraining) loss. Therefore, it is still unclear if these set of findings transfer to downstream task within the context of the pretrain-finetune paradigm. The key findings of this paper are as follows: (1) we show that aside from only the model size, model shape matters for downstream fine-tuning, (2) scaling protocols operate differently at different compute regions, (3) widely adopted T5-base and T5-large sizes are Pareto-inefficient. To this end, we present improved scaling protocols whereby our redesigned models achieve similar downstream fine-tuning quality while having 50\% fewer parameters and training 40\% faster compared to the widely adopted T5-base model. We publicly release over 100 pretrained checkpoints of different T5 configurations to facilitate future research and analysis.

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。

↑