Beyond Linear Steering: Unified Multi-Attribute Control for Language Models
원문 초록
Controlling multiple behavioral attributes in large language models (LLMs) at inference time is a challenging problem due to interference between attributes and the limitations of linear steering methods, which assume additive behavior in activation space and require per-attribute tuning. We introduce K-Steering, a unified and flexible approach that trains a single non-linear multi-label classifier on hidden activations and computes intervention directions via gradients at inference time. This avoids linearity assumptions, removes the need for storing and tuning separate attribute vectors, and allows dynamic composition of behaviors without retraining. To evaluate our method, we propose two new benchmarks, ToneBank and DebateMix, targeting compositional behavioral control. Empirical results across 3 model families, validated by both activation-based classifiers and LLM-based judges, demonstrate that K-Steering outperforms strong baselines in accurately steering multiple behaviors.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
Attributes Shape the Embedding Space of Face Recognition Models
Ref-Long: Benchmarking the Long-context Referencing Capability of Long-context Language Models
Emergent Semantics Beyond Token Embeddings: Transformer LMs with Frozen Visual Unicode Representations
MambaFusion: Height-Fidelity Dense Global Fusion for Multi-modal 3D Object Detection
The Trilemma of Truth in Large Language Models
Style-Aligned Image Composition for Robust Detection of Abnormal Cells in Cytopathology
토론
이 논문과 연결된 포럼 글이 아직 없습니다.