dorsal/arxiv
View SchemaStyles + Persona-plug = Customized LLMs
| Authors | Yutong Song, Jiang Wu, Shaofan Yuan, Chengze Shen, Jian Wang, Amir Rahmani, Nikil Dutt, Yu Wang |
|---|---|
| Categories | |
| ArXiv ID | 2601.06362vv1 |
| URL | https://arxiv.org/abs/2601.06362 |
| License | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
Abstract
We discover a previously overlooked challenge in personalized text generation: personalization methods are increasingly applied under explicit style instructions, yet their behavior under such constraints remains poorly understood. To balance implicit personalization and explicit style, we formulate personalization as a distributional residual and propose PsPLUG, a lightweight soft-prompt plug-in trained with style-conditioned preference contrasts. Across LaMP benchmark, our framework improves persona alignment, maintains stylistic fidelity, and outperforms retrieval-based and soft-prompt baselines with minimal computation. These results show that residual modeling provides a simple and principled foundation for controllable, style-aware LLM personalization.
{
"annotation_id": "46df29c1-0dc0-4d3c-87fa-f00024cc72a5",
"date_created": "2026-02-17T05:53:07.571000Z",
"date_modified": "2026-02-17T05:53:07.571000Z",
"file_hash": "5d755a40b90e8ab9ac231fd431f601c0118fb6066fb3829930a5df20464632f6",
"private": false,
"record": {
"abstract": "We discover a previously overlooked challenge in personalized text generation: personalization methods are increasingly applied under explicit style instructions, yet their behavior under such constraints remains poorly understood. To balance implicit personalization and explicit style, we formulate personalization as a distributional residual and propose PsPLUG, a lightweight soft-prompt plug-in trained with style-conditioned preference contrasts. Across LaMP benchmark, our framework improves persona alignment, maintains stylistic fidelity, and outperforms retrieval-based and soft-prompt baselines with minimal computation. These results show that residual modeling provides a simple and principled foundation for controllable, style-aware LLM personalization.",
"arxiv_id": "2601.06362",
"authors": [
"Yutong Song",
"Jiang Wu",
"Shaofan Yuan",
"Chengze Shen",
"Jian Wang",
"Amir Rahmani",
"Nikil Dutt",
"Yu Wang"
],
"categories": [
"cs.AI"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"title": "Styles + Persona-plug = Customized LLMs",
"url": "https://arxiv.org/abs/2601.06362",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "b6a556a7-cfee-4fa0-9ecd-3f80f877080b",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}