Constraint Back-translation Improves Complex Instruction Following of Large Language Models

TL;DR

This paper introduces constraint back-translation, a novel data augmentation method that enhances large language models' ability to follow complex instructions by generating high-quality instruction-response pairs with implicit constraints.

Contribution

The paper proposes constraint back-translation to create better training data, significantly improving instruction-following performance of large language models.

Findings

Post-training on CRAB dataset improves instruction-following accuracy.

Constraint back-translation reduces data noise and costs.

Auxiliary training with constraint back-translation enhances model performance.

Abstract

Large language models (LLMs) struggle to follow instructions with complex constraints in format, length, etc. Following the conventional instruction-tuning practice, previous works conduct post-training on complex instruction-response pairs generated by feeding complex instructions to advanced LLMs. However, even advanced LLMs cannot follow complex instructions well, thus limiting the quality of generated data. In this work, we find that existing datasets inherently contain implicit complex constraints and propose a novel data generation technique, constraint back-translation. Specifically, we take the high-quality instruction-response pairs in existing datasets and only adopt advanced LLMs to add complex constraints already met by the responses to the instructions, which naturally reduces costs and data noise. In the experiments, we adopt Llama3-70B-Instruct to back-translate constraints and create a high-quality complex instruction-response dataset, named CRAB. We present that post-training on CRAB improves multiple backbone LLMs' complex instruction-following ability, evaluated on extensive instruction-following benchmarks. We further find that constraint back-translation also serves as a useful auxiliary training objective in post-training. Our code, data, and models will be released to facilitate future research.