Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning

In three linesDualSelect, a fine-tuning method for LLMs, jointly selects safety references and compatible task samples to preserve safety alignment during adaptation. Tested on 1B-8B models, it improves Safety Avg. by at least 5.10 points over strongest baselines while maintaining task utility.

Summary generated by Claude — human-verified