Title: Enabling Instant Personalization of Large Language Models Through Hypernetworks

Original: arXiv:2510.16282v2 Announce Type: replace Abstract: Personalized large language models (LLMs) tailor content to individual preferences using user profiles or histories. However, existing parameter-efficient fine-tuning (PEFT) methods, such as the ``One-PEFT-Per-User'' (OPPU) paradigm, require training a separate adapter for each user, making them computationally expensive and impractical for real-time updates. We introduce Profile-to-PEFT, a scalable framework that employs a hypernetwork, trained end-to-end, to map a user's encoded profile directly to a full set of adapter parameters (e.g., LoRA), eliminating per-user training at deployment. This design enables instant adaptation, generalization to unseen users, and privacy-preserving local deployment. Experimental results demonstrate that our method outperforms both prompt-based personalization and OPPU while using substantially fewer computational resources at deployment. The framework exhibits strong generalization to out-of-distribution users and maintains robustness across varying user activity levels and different embedding backbones. The proposed Profile-to-PEFT framework enables efficient, scalable, and adaptive LLM personalization suitable for large-scale applications.

Rewrite: Personalized large language models (LLMs) customize output by leveraging individual user histories and profiles. Nevertheless, current parameter-efficient fine-tuning (PEFT) approaches, including the "One-PEFT-Per-User" (OPPU) strategy, necessitate the training of distinct adapters for every single user. This requirement renders the process computationally burdensome and unsuitable for immediate, real-time adjustments. To address these limitations, we present Profile-to-PEFT, a highly scalable system that utilizes an end-to-end trained hypernetwork. This architecture directly translates an encoded user profile into a complete set of adapter parameters, such as those used in LoRA, thereby removing the need for per-user training during deployment. This approach facilitates immediate adaptation, allows for effective generalization to new, unseen users, and supports privacy-focused local implementation. Our experiments show that this method surpasses both OPPU and prompt-based personalization techniques, all while consuming significantly fewer computational resources during the deployment phase. Furthermore, the framework demonstrates robust performance across diverse user activity levels and various embedding backbones, showing strong generalization capabilities even with out-of-distribution users. Ultimately, the Profile-to-PEFT framework provides a highly efficient, scalable, and adaptive solution for LLM personalization, making it well-suited for large-scale deployment.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC