dorsal/arxiv
View SchemaCARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation
| Authors | Yutong Song, Jiang Wu, Weijia Zhang, Chengze Shen, Shaofan Yuan, Weitao Lu, Jian Wang, Amir Rahmani, Nikil Dutt, Yu Wang |
|---|---|
| Categories | |
| ArXiv ID | 2601.06352vv1 |
| URL | https://arxiv.org/abs/2601.06352 |
| License | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
Abstract
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.
{
"annotation_id": "f9ee5ce8-3ffd-4694-96fd-e64c0915e326",
"date_created": "2026-02-17T05:53:08.562000Z",
"date_modified": "2026-02-17T05:53:08.562000Z",
"file_hash": "b68eceb95ff9922cf035f512126484e867b0ccfdf251aae2c33837ae2d2250aa",
"private": false,
"record": {
"abstract": "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.",
"arxiv_id": "2601.06352",
"authors": [
"Yutong Song",
"Jiang Wu",
"Weijia Zhang",
"Chengze Shen",
"Shaofan Yuan",
"Weitao Lu",
"Jian Wang",
"Amir Rahmani",
"Nikil Dutt",
"Yu Wang"
],
"categories": [
"cs.AI"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"title": "CARD: Cluster-level Adaptation with Reward-guided Decoding for Personalized Text Generation",
"url": "https://arxiv.org/abs/2601.06352",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "d02ed73b-8f00-4acb-8cda-0c672cace089",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}