Fine-Grained Safety Neurons with Training-Free Continual Projection to Reduce LLM Fine Tuning Risks

Open in new window