Preprint
Jul 2026
TF-Engram: A Train-Free Engram with SSD-Backed Memory for Large Language Models
Large Language Models (LLMs) store factual knowledge and domain-specific patterns implicitly in dense Transformer parameters, making knowledge expansion costly through pretraining, fine-tuning, retrieval augmentation, or longer contexts. Engram-style memory offers a compact hidden-state injection pathway, but existing...
Yutang Ma, Kecheng Huang, Xikun Jiang et al.
· 0 citations