Chain-of-Instructions: Compositional Instruction Tuning on Large Language Models

TL;DR

This paper introduces chain-of-instructions (CoI), a compositional instruction tuning method for large language models that enhances their ability to handle complex, multi-step, and unseen tasks by training them on chained subtasks.

Contribution

The paper proposes a novel CoI-tuning approach that improves LLMs' performance on complex and unseen multi-step instructions by leveraging chained subtasks during training.

Findings

CoI-tuning enhances handling of multi-subtask instructions

Improves generalization to unseen composite tasks

Enables better performance on complex, longer instruction chains

Abstract

Fine-tuning large language models (LLMs) with a collection of large and diverse instructions has improved the model's generalization to different tasks, even for unseen tasks. However, most existing instruction datasets include only single instructions, and they struggle to follow complex instructions composed of multiple subtasks. In this work, we propose a novel concept of compositional instructions called chain-of-instructions (CoI), where the output of one instruction becomes an input for the next like a chain. Unlike the conventional practice of solving single instruction tasks, our proposed method encourages a model to solve each subtask step by step until the final answer is reached. CoI-tuning (i.e., fine-tuning with CoI instructions) improves the model's ability to handle instructions composed of multiple subtasks as well as unseen composite tasks such as multilingual summarization. Overall, our study find that simple CoI tuning of existing instruction data can provide consistent generalization to solve more complex, unseen, and longer chains of instructions.