Research map / Large language models / Instruction and fine-tuning
PerFit: Exploring Personalization Shifts in Representation Space of LLMs
Jiahong Liu, Wenhao Yu, Quanyu Dai, Zhongyang Li, Jieming Zhu, Menglin Yang, Tat-Seng Chua, Irwin King
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use PerFit representation intervention instead of OPPU per-user LoRA | Llama2-7B personalization · LaMP-2N | accuracy: 0.810 -> 0.818 | Improved |
| use PerFit instead of OPPU | Llama2-7B personalization · LaMP-2M | accuracy: 0.600 -> 0.630 | Improved |
| use PerFit instead of OPPU | Llama2-7B personalization · LaMP-3 | MAE: 0.179 -> 0.179 | No clear effect |
| use PerFit instead of OPPU | Llama2-7B personalization · LaMP-7 | ROUGE-1: 0.539 -> 0.525 | Hurt |
| use PerFit instead of OPPU | Llama2-7B personalization · LaMP-4 | ROUGE-1: 0.191 -> 0.207 | Improved |
| use PerFit trainable parameters instead of OPPU | Llama2-7B personalization · LaMP tasks | parameter reduction: 81.25% to 98.44% | Improved |
| use PerFit instead of existing fine-tuning baselines | Llama2-7B personalization · training time | training time: 17.0% to 35.8% reduction | Improved |
| add Stage-2 personalized training to Stage-1 only (ablation) | PerFit · LaMP-2M | accuracy: 0.466 -> 0.630 | Improved |
| add Stage-2 personalized training to Stage-1 only (ablation) | PerFit · LaMP-2N | accuracy: 0.792 -> 0.818 | Improved |
| add Stage-1 collective training to Stage-2 only (P rank) (ablation) | PerFit · LaMP-2M | accuracy: 0.599 -> 0.630 | Improved |
| raise Stage-2 only rank from P to C+P (ablation) | PerFit · LaMP-2M | accuracy: 0.599 -> 0.620 | Improved |
| intervene at later single layers instead of earlier (ablation) | PerFit · Movie Tagging and News Headline | accuracy, ROUGE | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LoFiT — LoFiT: Localized Fine-tuning on LLM Representations (NeurIPS 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HYDRA: Model Factorization Framework for Black-Box LLM Personalization (NeurIPS 2024 · similarity 0.92)
- Rethinking Personalization in Large Language Models at the Token Level (ICML 2026 · similarity 0.90)
- PersonalLLM: Tailoring LLMs to Individual Preferences (ICLR 2025 · similarity 0.89)
- How Does Personalized Memory Shape LLM Behavior? Benchmarking Rational Preference Utilization in Personalized Assistants (NeurIPS 2026 · similarity 0.89)
- Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-Horizon Agents (NeurIPS 2026 · similarity 0.89)
- MMPB: It’s Time for Multi-Modal Personalization (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
