Fetching the paper…

LeftoverLocals: Listening to LLM Responses Through Leaked GPU Local Memory · Around