Ada-Instruct: Adapting Instruction Generators for Complex Reasoning

TL;DR

Ada-Instruct enhances instruction generation for complex reasoning tasks by fine-tuning open source LLMs with minimal data, enabling the creation of long, intricate instructions that surpass previous methods in complexity and consistency.

Contribution

This paper introduces Ada-Instruct, a novel fine-tuning approach that produces complex, long instructions for reasoning tasks using only ten examples, addressing limitations of prior self-instruct methods.

Findings

Ada-Instruct generates instructions of length ≥ 100 for complex tasks.

It maintains distributional consistency across diverse applications.

It outperforms existing methods in creating intricate instructions.

Abstract

Instructions augmentation is a crucial step for unleashing the full potential of large language models (LLMs) in downstream tasks. Existing Self-Instruct methods primarily simulate new instructions from a few initial instructions with in-context learning. However, our study identifies a critical flaw in this approach: even with GPT4o, Self-Instruct cannot generate complex instructions of length $\ge 100$, which is necessary in complex tasks such as code completion. To address this issue, our key insight is that fine-tuning open source LLMs with only ten examples can produce complex instructions that maintain distributional consistency for complex reasoning tasks. We introduce Ada-Instruct, an adaptive instruction generator developed through fine-tuning. We empirically validated Ada-Instruct's efficacy across different applications. The results highlight Ada-Instruct's capacity to generate long, intricate, and distributionally consistent instructions.