Adapting large language models to individual users remains challenging due to the tension between fine-grained personalization and scalable deployment. We present CARD, a hierarchical framework that achieves effective personalization through progressive refinement. CARD first clusters users according to shared stylistic patterns and learns cluster-specific LoRA adapters, enabling robust generalization and strong low-resource performance. To capture individual differences within each cluster, we propose an implicit preference learning mechanism that contrasts user-authored text with cluster-level generations, allowing the model to infer user-specific style preferences without manual annotation. At inference time, CARD injects personalization exclusively at decoding via lightweight user preference vectors and low-rank logit corrections, while keeping the base model frozen. Experiments on the LaMP and LongLaMP benchmarks show that CARD achieves competitive or superior generation quality compared to state-of-the-art baselines, while significantly improving efficiency and scalability for practical personalized text generation.
@article{arxiv.2601.06352,
title = {CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation},
author = {Yutong Song and Jiang Wu and Weijia Zhang and Chengze Shen and Shaofan Yuan and Weitao Lu and Jian Wang and Yu Wang and Nikil Dutt and Amir M. Rahmani},
journal= {arXiv preprint arXiv:2601.06352},
year = {2026}
}