Verifiable Format Control for Large Language Model Generations

TL;DR

This paper introduces a verifiable dataset and training method to improve small LLMs' ability to follow specific formats like JSON, addressing a key limitation in their instruction-following capabilities.

Contribution

It presents a fully verifiable dataset and a training approach that enhances small LLMs' format following abilities without relying on external LLMs for validation.

Findings

Small LLMs struggle with fine-grained format following.

The proposed method improves format following in 7B-level LLMs.

Verifiable dataset enables efficient training without costly API calls.

Abstract

Recent Large Language Models (LLMs) have demonstrated satisfying general instruction following ability. However, small LLMs with about 7B parameters still struggle fine-grained format following (e.g., JSON format), which seriously hinder the advancements of their applications. Most existing methods focus on benchmarking general instruction following while overlook how to improve the specific format following ability for small LLMs. Besides, these methods often rely on evaluations based on advanced LLMs (e.g., GPT-4), which can introduce the intrinsic bias of LLMs and be costly due to the API calls. In this paper, we first curate a fully verifiable format following dataset VFF. In contrast to existing works often adopting external LLMs for instruction-following validations, every sample of VFF can be easily validated with a Python function. Further, we propose to leverage this verifiable feature to synthesize massive data for progressively training small LLMs, in order to improve their format following abilities. Experimental results highlight the prevalent limitations in the format following capabilities of 7B level open-source LLMs and demonstrate the effectiveness of our method in enhancing this essential ability.