PLUME Method Enables Training-Free Context Parameterization for LLMs

PLUME introduces a training-free method to internalize contexts into reusable model parameters by constructing a global update representation. It uses a local parameter view and adaptive decoding to manage sequential memory updates without losing unaffected information.