Cache the Encoder Within:Compact, Reusable Memory across LLM Queries

EncBank introduces a method to treat a pretrained LLM's lower layers as a reusable document encoder, compactly storing outputs to reduce redundant encoding…