ORBIT: Training-Free Multi-Attribute Behavioral Steering via Orthogonal Subspace Rotation

2026-06-21

AI Summary

This work addresses the challenge of simultaneously controlling multiple behavioral attributes in large language models using training-free activation interventions, which often suffer from norm imbalance and directional cancellation due to naive vector addition. The authors propose a novel training-free orthogonal subspace rotation technique that extends rotation-based intervention to multi-attribute settings. By leveraging singular value decomposition to construct a joint multi-attribute subspace, the method applies norm-preserving orthogonal rotations augmented with adaptive per-token gating and weak projection enhancement, enabling dynamic and balanced co-regulation of multiple attributes. Evaluated on TraitFactory and ToneBank benchmarks across three prominent language models, the approach significantly outperforms existing training-free baselines, achieving precise multi-attribute control while better preserving output coherence.

Abstract

Language models are widely used in assistant settings, where controlling behavioral attributes is often essential. Activation steering modifies hidden-state representations at inference time, providing a lightweight, training-free mechanism that can be toggled at runtime. Existing methods, however, have focused primarily on steering a single attribute at a time. When multiple attributes must be controlled simultaneously, naive summation of per-attribute steering vectors suffers from norm imbalance and directional cancellation, while classifier-based approaches require retraining whenever the attribute set changes. We introduce ORBIT (Orthogonal Rotation-Based Intervention Technique), a training-free extension of rotation-based steering to the multi-attribute setting. Our method constructs a joint subspace from per-attribute steering planes via singular value decomposition and applies a single norm-preserving rotation within that subspace toward a combined target direction. Adaptive per-token gating identifies which attributes need correction at each position, and an optional additive boost strengthens attributes with weak initial projection. We also introduce TraitFactory, a new multi-attribute benchmark that focuses on behavioral tendencies rather than surface-level style. We evaluate ORBIT on TraitFactory and ToneBank across three models (Llama-3.2-3B, Qwen-2.5-7B, Llama-3.1-8B) while steering multiple attributes simultaneously, showing that it achieves stronger and more balanced multi-attribute steering than existing training-free baselines while better preserving output coherence.

Problem

• multi-attribute steering
• training-free control
• behavioral attributes
• language models
• activation steering

Innovation

• activation steering
• multi-attribute control
• orthogonal rotation
• training-free intervention
• behavioral alignment