StyleAdaptedLM: Enhancing Instruction Following Models with Efficient Stylistic Transfer

  • 2025-07-24 10:57:32
  • Pritika Ramu, Apoorv Saxena, Meghanath M Y, Varsha Sankar, Debraj Basu
  • 0

Abstract

Adapting LLMs to specific stylistic characteristics, like brand voice orauthorial tones, is crucial for enterprise communication but challenging toachieve from corpora which lacks instruction-response formatting withoutcompromising instruction adherence. We introduce StyleAdaptedLM, a frameworkthat efficiently transfers stylistic traits to instruction-following modelsusing Low-Rank Adaptation (LoRA). LoRA adapters are first trained on a basemodel with diverse unstructured stylistic corpora, then merged with a separateinstruction-following model. This enables robust stylistic customizationwithout paired data or sacrificing task performance. Experiments acrossmultiple datasets and models demonstrate improved stylistic consistency whilepreserving instruction adherence, with human evaluations confirmingbrand-specific convention uptake. StyleAdaptedLM offers an efficient path forstylistic personalization in LLMs.

 

Quick Read (beta)

loading the full paper ...