Pruning Pretrained Encoders with a Multitask Objective

TL;DR

This paper investigates pruning a single pretrained encoder with a multitask objective to create a compact, versatile model that performs well across multiple downstream tasks, reducing size and computational costs.

Contribution

It demonstrates that multitask pruning can outperform separate single-task models and is effective for low-resource tasks, offering a unified approach to model compression.

Findings

Multitask pruning outperforms ensemble of single-task models on average.

Pruned models are competitive on individual tasks.

Multitask pruning reduces model size for low-resource tasks.

Abstract

The sizes of pretrained language models make them challenging and expensive to use when there are multiple desired downstream tasks. In this work, we adopt recent strategies for model pruning during finetuning to explore the question of whether it is possible to prune a single encoder so that it can be used for multiple tasks. We allocate a fixed parameter budget and compare pruning a single model with a multitask objective against the best ensemble of single-task models. We find that under two pruning strategies (element-wise and rank pruning), the approach with the multitask objective outperforms training models separately when averaged across all tasks, and it is competitive on each individual one. Additional analysis finds that using a multitask objective during pruning can also be an effective method for reducing model sizes for low-resource tasks.