Empirical Study on Updating Key-Value Memories in Transformer Feed-forward Layers

arXiv:2402.12233v1 Announce Type: new
Abstract: The feed-forward networks (FFNs) in transformers are recognized as a group of key-value neural memories to restore abstract high-level knowledge. In this work, we conduct an empirical ablation study on updating keys (the 1st layer in the FFNs layer) or values (the 2nd layer in the FFNs layer). We compare those two methods in various knowledge editing and fine-tuning tasks of large language models to draw insights to understand FFNs further. Code is available at …

arxiv cs.cl key memories study transformer type value