1
TokenMem: Faithful Knowledge Injection for Frozen LLMs
提出TokenMem方法，在冻结LLM上实现高保真知识注入，无需微调即可更新模型知识库。
arXiv:2607.22625v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) enhances large language models (LLMs) with external knowledge, …