DISCO: Disentangled Communication Steering for Large Language Models
This work addresses the challenge of fine-grained control over LLM outputs for researchers and practitioners, offering a novel approach to steering vector methods.
The paper tackled the problem of guiding large language model outputs by proposing DISCO, a method that injects steering vectors directly into query and value representation spaces within attention heads, achieving up to 19.1% higher steering efficacy than baselines on models like LLaMA 3.1 8B and Gemma 2 9B.
A variety of recent methods guide large language model outputs via the inference-time addition of steering vectors to residual-stream or attention-head representations. In contrast, we propose to inject steering vectors directly into the query and value representation spaces within attention heads. We provide evidence that a greater portion of these spaces exhibit high linear discriminability of concepts --a key property motivating the use of steering vectors-- than attention head outputs. We analytically characterize the effect of our method, which we term DISentangled COmmunication (DISCO) Steering, on attention head outputs. Our analysis reveals that DISCO disentangles a strong but underutilized baseline, steering attention inputs, which implicitly modifies queries and values in a rigid manner. In contrast, DISCO's direct modulation of these components enables more granular control. We find that DISCO achieves superior performance over a number of steering vector baselines across multiple datasets on LLaMA 3.1 8B and Gemma 2 9B, with steering efficacy scoring up to 19.1% higher than the runner-up. Our results support the conclusion that the query and value spaces are powerful building blocks for steering vector methods.