כתבה
arXiv cs.CL ·
EngramEdit: עדכון ידע נפרד ב-LLMs דרך ממורא תנאי
EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional Memory
EngramEdit מאפשר עדכון ידע נפרד ב-LLMs דרך ממורא תנאי. התכונה נבדקה באמצעות DeepSeek Engram.
תקציר מקורי באנגליתarXiv:2610.10533v1 Announce Type: new Abstract: Conditional memory architectures such as DeepSeek Engram use input n-grams to look up learned embeddings, expanding the capacity of large language models (LLMs) with limited additional computation. Beyond model scaling, this architecture has demonstrated the potential to decouple factual knowledge storage from general-purpose computation, offering a promising route to updating factual knowledge while keeping the Transformer backbone fixed. Realizing this potential is challenging because different expressions of a fact may activate different n-gram embeddings, while updating shared embeddings can unintentionally change the model's predictions about other facts. We propose EngramEdit for decoupled knowledge updates through conditional memory. E
קרא במקור המקורי
arxiv.org
פתח כתבה מקורית