Toward Preference-aligned Large Language Models via Residual-based Model Steering
Fuente:
arXiv
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914062241103872 |
|---|---|
| author | La Cava, Lucio Tagarelli, Andrea |
| author_facet | La Cava, Lucio Tagarelli, Andrea |
| contents | Preference alignment is a critical step in making Large Language Models (LLMs) useful and aligned with (human) preferences. Existing approaches such as Reinforcement Learning from Human Feedback or Direct Preference Optimization typically require curated data and expensive optimization over billions of parameters, and eventually lead to persistent task-specific models. In this work, we introduce Preference alignment of Large Language Models via Residual Steering (PaLRS), a training-free method that exploits preference signals encoded in the residual streams of LLMs. From as few as one hundred preference pairs, PaLRS extracts lightweight, plug-and-play steering vectors that can be applied at inference time to push models toward preferred behaviors. We evaluate PaLRS on various small-to-medium-scale open-source LLMs, showing that PaLRS-aligned models achieve consistent gains on mathematical reasoning and code generation benchmarks while preserving baseline general-purpose performance. Moreover, when compared to DPO-aligned models, they perform better with huge time savings. Our findings highlight that PaLRS offers an effective, much more efficient and flexible alternative to standard preference optimization pipelines, offering a training-free, plug-and-play mechanism for alignment with minimal data. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_23982 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Toward Preference-aligned Large Language Models via Residual-based Model Steering La Cava, Lucio Tagarelli, Andrea Computation and Language Artificial Intelligence Computers and Society Machine Learning Neural and Evolutionary Computing Preference alignment is a critical step in making Large Language Models (LLMs) useful and aligned with (human) preferences. Existing approaches such as Reinforcement Learning from Human Feedback or Direct Preference Optimization typically require curated data and expensive optimization over billions of parameters, and eventually lead to persistent task-specific models. In this work, we introduce Preference alignment of Large Language Models via Residual Steering (PaLRS), a training-free method that exploits preference signals encoded in the residual streams of LLMs. From as few as one hundred preference pairs, PaLRS extracts lightweight, plug-and-play steering vectors that can be applied at inference time to push models toward preferred behaviors. We evaluate PaLRS on various small-to-medium-scale open-source LLMs, showing that PaLRS-aligned models achieve consistent gains on mathematical reasoning and code generation benchmarks while preserving baseline general-purpose performance. Moreover, when compared to DPO-aligned models, they perform better with huge time savings. Our findings highlight that PaLRS offers an effective, much more efficient and flexible alternative to standard preference optimization pipelines, offering a training-free, plug-and-play mechanism for alignment with minimal data. |
| title | Toward Preference-aligned Large Language Models via Residual-based Model Steering |
| topic | Computation and Language Artificial Intelligence Computers and Society Machine Learning Neural and Evolutionary Computing |
| url | https://arxiv.org/abs/2509.23982 |