From Instance Training to Instruction Learning: Task Adapters Generation from Instructions

TL;DR

This paper introduces TAGI, a method that generates task-specific adapters from instructions, enabling cross-task generalization without retraining, and outperforms traditional models while reducing computational costs.

Contribution

We propose TAGI, a novel instruction-based adapter generation method that enhances cross-task generalization and reduces training requirements for large language models.

Findings

TAGI matches or outperforms meta-trained models on instruction datasets.

TAGI significantly reduces computational costs compared to traditional methods.

The two-stage training process effectively endows TAGI with cross-task generalization.

Abstract

Large language models (LLMs) have acquired the ability to solve general tasks by utilizing instruction finetuning (IFT). However, IFT still relies heavily on instance training of extensive task data, which greatly limits the adaptability of LLMs to real-world scenarios where labeled task instances are scarce and broader task generalization becomes paramount. Contrary to LLMs, humans acquire skills and complete tasks not merely through repeated practice but also by understanding and following instructional guidelines. This paper is dedicated to simulating human learning to address the shortcomings of instance training, focusing on instruction learning to enhance cross-task generalization. Within this context, we introduce Task Adapters Generation from Instructions (TAGI), which automatically constructs the task-specific model in a parameter generation manner based on the given task instructions without retraining for unseen tasks. Specifically, we utilize knowledge distillation to enhance the consistency between TAGI developed through Learning with Instruction and task-specific models developed through Training with Instance, by aligning the labels, output logits, and adapter parameters between them. TAGI is endowed with cross-task generalization capabilities through a two-stage training process that includes hypernetwork pretraining and finetuning. We evaluate TAGI on the Super-Natural Instructions and P3 datasets. The experimental results demonstrate that TAGI can match or even outperform traditional meta-trained models and other hypernetwork models, while significantly reducing computational requirements.