Train Large, Then Compress: Rethinking Model Size for Efficient Training and Inference of Transformers

by   Zhuohan Li, et al.

Since hardware resources are limited, the objective of training deep learning models is typically to maximize accuracy subject to the time and memory constraints of training and inference. We study the impact of model size in this setting, focusing on Transformer models for NLP tasks that are limited by compute: self-supervised pretraining and high-resource machine translation. We first show that even though smaller Transformer models execute faster per iteration, wider and deeper models converge in significantly fewer steps. Moreover, this acceleration in convergence typically outpaces the additional computational overhead of using larger models. Therefore, the most compute-efficient training strategy is to counterintuitively train extremely large models but stop after a small number of iterations. This leads to an apparent trade-off between the training efficiency of large Transformer models and the inference efficiency of small Transformer models. However, we show that large models are more robust to compression techniques such as quantization and pruning than small models. Consequently, one can get the best of both worlds: heavily compressed, large models achieve higher accuracy than lightly compressed, small models.


Extremely Low Bit Transformer Quantization for On-Device Neural Machine Translation

Transformer is being widely used in Neural Machine Translation (NMT). De...

When Ensembling Smaller Models is More Efficient than Single Large Models

Ensembling is a simple and popular technique for boosting evaluation per...

Quality and Cost Trade-offs in Passage Re-ranking Task

Deep learning models named transformers achieved state-of-the-art result...

Model Compression

With time, machine learning models have increased in their scope, functi...

It's always personal: Using Early Exits for Efficient On-Device CNN Personalisation

On-device machine learning is becoming a reality thanks to the availabil...

AdapterDrop: On the Efficiency of Adapters in Transformers

Massively pre-trained transformer models are computationally expensive t...

Structural Dropout for Model Width Compression

Existing ML models are known to be highly over-parametrized, and use sig...