Transformer Feed-Forward Layers Are Key-Value Memories

Mor Geva, Roei Schuster, Jonathan Berant, Omer Levy

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


Feed-forward layers constitute two-thirds of a transformer model's parameters, yet their role in the network remains under-explored. We show that feed-forward layers in transformer-based language models operate as key-value memories, where each key correlates with textual patterns in the training examples, and each value induces a distribution over the output vocabulary. Our experiments show that the learned patterns are human-interpretable, and that lower layers tend to capture shallow patterns, while upper layers learn more semantic ones. The values complement the keys' input patterns by inducing output distributions that concentrate probability mass on tokens likely to appear immediately after each pattern, particularly in the upper layers. Finally, we demonstrate that the output of a feed-forward layer is a composition of its memories, which is subsequently refined throughout the model's layers via residual connections to produce the final output distribution.
Original languageEnglish
Title of host publicationProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
EditorsMarie-Francine Moens, Xuanjing Huang, Lucia Specia, Scott Wen-tau Yih
PublisherAssociation for Computational Linguistics
Number of pages12
ISBN (Electronic)978-1-955917-09-4
StatePublished - 1 Nov 2021
Event2021 Conference on Empirical Methods in Natural Language Processing, EMNLP 2021 - Hybrid - Online and in Punta Cana, Punta Cana, Dominican Republic
Duration: 7 Nov 202114 Nov 2021


Conference2021 Conference on Empirical Methods in Natural Language Processing, EMNLP 2021
Country/TerritoryDominican Republic
CityPunta Cana


Dive into the research topics of 'Transformer Feed-Forward Layers Are Key-Value Memories'. Together they form a unique fingerprint.

Cite this