arXiv：cs.AI· Myunghoon Ryu, Geunpyo Park, Sungjoon Lee, XinYu Piao, Jong-Kook Kim·· 6 小时前
P2Skill：面向云-端 LLM 推理系统的隐私保护技能蒸馏
P2Skill: Privacy Preserving Skill Distillation for Cloud-Local LLM Inference Systems
AI 导读
P2Skill 是一种基于提示词的技能蒸馏方法，让本地小语言模型自主执行四项技能，并由云端 LLM 从执行失败中迭代优化技能，使本地 SLM 泛化到记忆化的 PII 模式之外。该方法无需隐私专用微调或学习型辅助检测器，最终技能在设定边界内以最低验证泄漏率选出。在真实用户查询上，P2Skill 在三个本地 SLM 中表现最佳者向云端传输的标注标识符少于所有现有隐私保护方法。
正文
Abstract:Cloud-local LLM inference can use the reasoning capability of large cloud models, but cloud-bound requests must exclude personally identifiable information (PII). Existing privacy-preserving methods modify the prompt or the model, which may distort contextual semantics or require additional training. This paper proposes P2Skill, a prompt-based skill distillation method in which a local small language model (SLM) autonomously performs four skills. A cloud LLM iteratively refines the skills from execution failures to let the local SLM generalize beyond memorized PII patterns, and the final skills are selected at the lowest validation leakage within set bounds. P2Skill requires no privacy-specific fine-tuning or learned auxiliary detectors. On real user queries, P2Skill on the best of three local SLMs transmits fewer annotated identifiers than every existing privacy-preserving method that sends text to the cloud and preserves quality more often than the closest of them in leakage, although fully local answering preserves quality more often.
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2608.14094 [cs.CR] |
| (or arXiv:2608.14094v2 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2608.14094 arXiv-issued DOI via DataCite |
Submission history
From: Ryu Myunghoon [view email]
[v1]
Fri, 14 Aug 2026 08:56:11 UTC (4,010 KB)
[v2]
Tue, 13 Oct 2026 08:15:56 UTC (3,538 KB)
来源：arXiv：cs.AI · arxiv.org