Learning to Generate Task-Specific Adapters from Task Description

TL;DR

This paper introduces Hypter, a hypernetwork-based framework that generates task-specific adapters from descriptions, enhancing the generalization of text-to-text transformers to unseen tasks, with significant improvements demonstrated on benchmark datasets.

Contribution

Hypter is a novel framework that trains a hypernetwork to produce adapters from task descriptions, improving task generalization over traditional fine-tuning methods.

Findings

11.3% improvement on ZEST dataset with BART-Large

Outperforms fine-tuning baselines on ZEST and SQuAD datasets

Enhances generalization to unseen tasks in NLP

Abstract

Pre-trained text-to-text transformers such as BART have achieved impressive performance across a range of NLP tasks. Recent study further shows that they can learn to generalize to novel tasks, by including task descriptions as part of the source sequence and training the model with (source, target) examples. At test time, these fine-tuned models can make inferences on new tasks using the new task descriptions as part of the input. However, this approach has potential limitations, as the model learns to solve individual (source, target) examples (i.e., at the instance level), instead of learning to solve tasks by taking all examples within a task as a whole (i.e., at the task level). To this end, we introduce Hypter, a framework that improves text-to-text transformer's generalization ability to unseen tasks by training a hypernetwork to generate task-specific, light-weight adapters from task descriptions. Experiments on ZEST dataset and a synthetic SQuAD dataset demonstrate that Hypter improves upon fine-tuning baselines. Notably, when using BART-Large as the main network, Hypter brings 11.3% comparative improvement on ZEST dataset.