UpSafe$^\circ$C: Upcycling for Controllable Safety in Large Language Models
This addresses safety and controllability issues in LLMs for developers and users, offering a novel approach that is incremental in combining existing techniques like MoE and SFT.
The paper tackles the problem of safety risks in Large Language Models, such as harmful content generation and jailbreak attacks, by proposing UpSafe°C, a framework that achieves robust safety improvements while maintaining competitive performance on general tasks, with a safety temperature mechanism enabling dynamic control at inference time.
Large Language Models (LLMs) have achieved remarkable progress across a wide range of tasks, but remain vulnerable to safety risks such as harmful content generation and jailbreak attacks. Existing safety techniques -- including external guardrails, inference-time guidance, and post-training alignment -- each face limitations in balancing safety, utility, and controllability. In this work, we propose UpSafe$^\circ$C, a unified framework for enhancing LLM safety through safety-aware upcycling. Our approach first identifies safety-critical layers and upcycles them into a sparse Mixture-of-Experts (MoE) structure, where the router acts as a soft guardrail that selectively activates original MLPs and added safety experts. We further introduce a two-stage SFT strategy to strengthen safety discrimination while preserving general capabilities. To enable flexible control at inference time, we introduce a safety temperature mechanism, allowing dynamic adjustment of the trade-off between safety and utility. Experiments across multiple benchmarks, base model, and model scales demonstrate that UpSafe$^\circ$C achieves robust safety improvements against harmful and jailbreak inputs, while maintaining competitive performance on general tasks. Moreover, analysis shows that safety temperature provides fine-grained inference-time control that achieves the Pareto-optimal frontier between utility and safety. Our results highlight a new direction for LLM safety: moving from static alignment toward dynamic, modular, and inference-aware control.