SAG: Style-Aligned Article Generation via Model Collaboration

TL;DR

This paper introduces a collaborative training framework combining large and small language models to generate styled articles, achieving state-of-the-art results and improved style consistency while reducing hallucinations.

Contribution

It proposes a novel method that leverages LLMs and SLMs together, including a self-improvement technique and a new benchmark for style-aligned article generation.

Findings

Achieves 0.78 higher ROUGE-L score than GPT-4.

Improves BLEU-4 score by 0.55 over baseline.

Maintains low hallucination rate.

Abstract

Large language models (LLMs) have increased the demand for personalized and stylish content generation. However, closed-source models like GPT-4 present limitations in optimization opportunities, while the substantial training costs and inflexibility of open-source alternatives, such as Qwen-72B, pose considerable challenges. Conversely, small language models (SLMs) struggle with understanding complex instructions and transferring learned capabilities to new contexts, often exhibiting more pronounced limitations. In this paper, we present a novel collaborative training framework that leverages the strengths of both LLMs and SLMs for style article generation, surpassing the performance of either model alone. We freeze the LLMs to harness their robust instruction-following capabilities and subsequently apply supervised fine-tuning on the SLM using style-specific data. Additionally, we introduce a self-improvement method to enhance style consistency. Our new benchmark, NoteBench, thoroughly evaluates style-aligned generation. Extensive experiments show that our approach achieves state-of-the-art performance, with improvements of 0.78 in ROUGE-L and 0.55 in BLEU-4 scores compared to GPT-4, while maintaining a low hallucination rate regarding factual and faithfulness.