Fetching the paper…

Combining Domain and Alignment Vectors to Achieve Better Knowledge-Safety Trade-offs in LLMs · Around