EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional Memory

Hongru Cai, Ran Wei, Wenjie Wang, Chengfa Wu, Ning Song et al.|October 7, 2026arXiv

Key Takeaway

Conditional memory architectures can serve as an editable knowledge layer separate from the model's core computation, allowing efficient factual updates without full retraining or catastrophic forgetting.

Summary

EngramEdit enables updating factual knowledge in language models that use conditional memory (n-gram lookup tables) without retraining. It computes target memory states for updated facts across different phrasings, then jointly updates shared embeddings while protecting frequently-used ones to avoid breaking unrelated knowledge.

training efficiency

Key Terms

conditional-memory knowledge-editing n-gram-embeddings catastrophic-forgetting