Adaptive Token Biaser: Knowledge Editing via Biasing Key Entities
This work addresses the need for efficient knowledge updates in LLMs, offering a low-cost solution with broad applicability, though it is incremental as it builds on existing in-context editing approaches.
The paper tackles the problem of outdated knowledge in large language models by introducing ATBias, a decoding technique that enhances in-context editing, achieving up to a 32.3% performance improvement and halving latency compared to state-of-the-art methods.
The parametric knowledge memorized by large language models (LLMs) becomes outdated quickly. In-context editing (ICE) is currently the most effective method for updating the knowledge of LLMs. Recent advancements involve enhancing ICE by modifying the decoding strategy, obviating the need for altering internal model structures or adjusting external prompts. However, this enhancement operates across the entire sequence generation, encompassing a plethora of non-critical tokens. In this work, we introduce $\textbf{A}$daptive $\textbf{T}$oken $\textbf{Bias}$er ($\textbf{ATBias}$), a new decoding technique designed to enhance ICE. It focuses on the tokens that are mostly related to knowledge during decoding, biasing their logits by matching key entities related to new and parametric knowledge. Experimental results show that ATBias significantly enhances ICE performance, achieving up to a 32.3% improvement over state-of-the-art ICE methods while incurring only half the latency. ATBias not only improves the knowledge editing capabilities of ICE but can also be widely applied to LLMs with negligible cost.