Multi-Objective Linguistic Control of Large Language Models

TL;DR

This paper introduces MCTune, a method for fine-tuning large language models to control multiple linguistic complexities in their output, improving controllability without sacrificing response quality.

Contribution

It proposes a novel multi-control tuning approach that incorporates multiple linguistic complexity controls during instruction tuning of LLMs.

Findings

Enhanced multi-complexity controllability in LLMs.

Maintained or improved response quality.

Effective on LLaMA2-7B with Alpaca-GPT4 and WizardLM datasets.

Abstract

Large language models (LLMs), despite their breakthroughs on many challenging benchmark tasks, lean to generate verbose responses and lack the controllability of output complexity, which is usually preferred by human users in practice. In this paper, we study how to precisely control multiple linguistic complexities of LLM output by finetuning using off-the-shelf data. To this end, we propose multi-control tuning (MCTune), which includes multiple linguistic complexity values of ground-truth responses as controls in the input for instruction tuning. We finetune LLaMA2-7B on Alpaca-GPT4 and WizardLM datasets. Evaluations on widely used benchmarks demonstrate that our method does not only improve LLMs' multi-complexity controllability substantially but also retains or even enhances the quality of the responses as a side benefit.