Making Small Language Models Better Multi-task Learners with Mixture-of-Task-Adapters

TL;DR

ALTER leverages a mixture-of-task-adapters to enhance small language models for multi-task NLP, achieving competitive performance while reducing computational costs compared to large models.

Contribution

The paper introduces Mixture-of-Task-Adapters and a two-stage training method to improve multi-task learning in small language models, enabling efficient domain-specific NLP applications.

Findings

MTA architecture captures intra- and inter-task knowledge effectively.

Two-stage training improves multi-task performance with low computational overhead.

Experimental results show competitive performance on various NLP tasks.

Abstract

Recently, Large Language Models (LLMs) have achieved amazing zero-shot learning performance over a variety of Natural Language Processing (NLP) tasks, especially for text generative tasks. Yet, the large size of LLMs often leads to the high computational cost of model training and online deployment. In our work, we present ALTER, a system that effectively builds the multi-tAsk Learners with mixTure-of-task-adaptERs upon small language models (with <1B parameters) to address multiple NLP tasks simultaneously, capturing the commonalities and differences between tasks, in order to support domain-specific applications. Specifically, in ALTER, we propose the Mixture-of-Task-Adapters (MTA) module as an extension to the transformer architecture for the underlying model to capture the intra-task and inter-task knowledge. A two-stage training method is further proposed to optimize the collaboration between adapters at a small computational cost. Experimental results over a mixture of NLP tasks show that our proposed MTA architecture and the two-stage training method achieve good performance. Based on ALTER, we have also produced MTA-equipped language models for various domains.