Prompt-Based Length Controlled Generation with Multiple Control Types

TL;DR

This paper presents a prompt-based method using reinforcement learning and sample filtering to achieve accurate length-controlled text generation across various control types, improving over existing methods.

Contribution

It introduces a novel prompt-based approach with RL and a standard prompt extractor for flexible, accurate length control in language models, applicable to multiple control types.

Findings

Significantly improves length control accuracy on summarization datasets

Effective across multiple control types and generalizes to unseen prompts

Enhances length control in GPT-style models with high precision

Abstract

Large language models (LLMs) have attracted great attention given their strong performance on a wide range of NLP tasks. In practice, users often expect generated texts to fall within a specific length range, making length controlled generation an important topic, especially for GPT-style models. Existing length control methods mostly focus on a simple control type of "equal to" a target length. Different from them, we propose a prompt-based method to achieve length controlled generation under different control types with high accuracy. In particular, we adopt reinforcement learning (RL) and sample filtering with the reward signal given by rule-based reward models, which enhances the length control ability of models by rewarding outputs that follow certain control instructions. In addition, we introduce a standard prompt extractor to parse arbitrary users' input into standard control instructions. Experiments show that our method significantly improves the accuracy of prompt-based length control on popular summarization datasets like CNNDM and NYT under multiple control types. Moreover, both the standard prompt extractor and RL-tuned model show strong generalization to unseen control prompt templates.