Improving LLM Safety Alignment with Dual-Objective Optimization

Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks. Direct preference optimization (DPO), a widely deployed alignment method, exhibits limitations in both experimental and theoretical contexts as its loss function proves suboptimal for refusal learning. Through gradient-based analysis, we identify these shortcomings and propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge. This approach significantly increases LLM robustness against a wide range of jailbreak attacks, including prefilling, suffix, and multi-turn attacks across both in-distribution and out-of-distribution scenarios. Furthermore, we introduce a method to emphasize critical refusal tokens by incorporating a reward-based token-level weighting mechanism for refusal learning, which further improves the robustness against adversarial exploits. Our research also suggests that robustness to jailbreak attacks is correlated with token distribution shifts in the training process and internal representations of refusal and harmful tokens, offering valuable directions for future research in LLM safety alignment. The code is available at https://github.com/wicai24/DOOR-Alignment

UniFed: A Benchmark for Federated Learning Frameworks

Federated Learning (FL) has become a practical and popular paradigm in machine learning. However, currently, there is no systematic solution that covers diverse use cases. Practitioners often face the challenge of how to select a matching FL framework for their use case. In this work, we present UniFed, the first unified benchmark for standardized evaluation of the existing open-source FL frameworks. With 15 evaluation scenarios, we present both qualitative and quantitative evaluation results of nine existing popular open-sourced FL frameworks, from the perspectives of functionality, usability, and system performance. We also provide suggestions on framework selection based on the benchmark conclusions and point out future improvement directions.