FineSteer: A Unified Framework for Fine-Grained Inference-Time Steering in Large Language Models

2026-04-16

AI Summary

Large language models are prone to safety violations and hallucinations during inference, and existing intervention methods struggle to simultaneously achieve effectiveness, utility preservation, and computational efficiency. This work proposes FineSteer, a framework that decouples inference-time intervention into two stages: conditional intervention and fine-grained vector composition. It introduces Subspace-guided Conditional Steering (SCS) to avoid ineffective interventions and a Mixture-of-Steering Experts (MoSE) mechanism to generate query-adaptive intervention vectors. Notably, FineSteer operates without updating model parameters and significantly outperforms current approaches on both safety and factuality benchmarks, while incurring negligible degradation on general-purpose tasks.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Mixture of Experts (MoE)
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large language models (LLMs) often exhibit undesirable behaviors, such as safety violations and hallucinations. Although inference-time steering offers a cost-effective way to adjust model behavior without updating its parameters, existing methods often fail to be simultaneously effective, utility-preserving, and training-efficient due to their rigid, one-size-fits-all designs and limited adaptability. In this work, we present FineSteer, a novel steering framework that decomposes inference-time steering into two complementary stages: conditional steering and fine-grained vector synthesis, allowing fine-grained control over when and how to steer internal representations. In the first stage, we introduce a Subspace-guided Conditional Steering (SCS) mechanism that preserves model utility by avoiding unnecessary steering. In the second stage, we propose a Mixture-of-Steering-Experts (MoSE) mechanism that captures the multimodal nature of desired steering behaviors and generates query-specific steering vectors for improved effectiveness. Through tailored designs in both SCS and MoSE, FineSteer maintains robust performance on general queries while adaptively optimizing steering vectors for targeted inputs in a training-efficient manner. Extensive experiments on safety and truthfulness benchmarks show that FineSteer outperforms state-of-the-art methods in overall performance, achieving stronger steering performance with minimal utility loss. Code is available at https://github.com/YukinoAsuna/FineSteer

Problem

• inference-time steering
• large language models
• safety violations
• hallucinations
• utility preservation

Innovation

• inference-time steering
• fine-grained control
• Subspace-guided Conditional Steering
• Mixture-of-Steering-Experts
• large language models