DLM-Scope：通过稀疏自编码器实现扩散语言模型的机制可解释性
DLM-Scope: Mechanistic Interpretability of Diffusion Language Models via Sparse Autoencoders
AI总结：
DLM-Scope通过稀疏自编码器首次实现扩散语言模型的机制可解释性，展示了其在特征提取和干预中的有效性。
摘要：
稀疏自编码器（SAEs）已成为在自回归大语言模型（LLMs）中进行机制可解释性研究的标准工具，使研究者能够提取稀疏、可被人类理解的特征并干预模型行为。最近，随着扩散语言模型（DLMs）逐渐成为自回归LLMs的有希望的替代品，开发针对这一新兴模型类别的定制化机制可解释性工具变得至关重要。在本工作中，我们提出了DLM-Scope，这是首个基于SAEs的DLMs可解释性框架，并展示了训练后的Top-K SAEs能够忠实提取可解释的特征。值得注意的是，我们发现插入SAEs对DLMs的影响不同于自回归LLMs：虽然在LLMs中插入SAEs通常会带来损失惩罚，在DLMs中应用到早期层时可以减少交叉熵损失，这一现象在LLMs中要么不存在要么明显更弱。此外，DLMs中的SAE特征能够实现更有效的扩散时间干预，通常优于LLM的引导。此外，我们开创了某些新的基于SAEs的研究方向：我们展示了SAEs可以为DLM解码顺序提供有用信号；并且SAE特征在DLMs的训练后阶段是稳定的。我们的工作为DLMs的机制可解释性奠定了基础，并展示了将SAEs应用于DLM相关任务和算法的巨大潜力。
英文摘要：
Sparse autoencoders (SAEs) have become a standard tool for mechanistic interpretability in autoregressive large language models (LLMs), enabling researchers to extract sparse, human-interpretable features and intervene on model behavior. Recently, as diffusion language models (DLMs) have become an increasingly promising alternative to the autoregressive LLMs, it is essential to develop tailored mechanistic interpretability tools for this emerging class of models. In this work, we present DLM-Scope, the first SAE-based interpretability framework for DLMs, and demonstrate that trained Top-K SAEs can faithfully extract interpretable features. Notably, we find that inserting SAEs affects DLMs differently than autoregressive LLMs: while SAE insertion in LLMs typically incurs a loss penalty, in DLMs it can reduce cross-entropy loss when applied to early layers, a phenomenon absent or markedly weaker in LLMs. Additionally, SAE features in DLMs enable more effective diffusion-time interventions, often outperforming LLM steering. Moreover, we pioneer certain new SAE-based research directions for DLMs: we show that SAEs can provide useful signals for DLM decoding order; and the SAE features are stable during the post-training phase of DLMs. Our work establishes a foundation for mechanistic interpretability in DLMs and shows a great potential of applying SAEs to DLM-related tasks and algorithms.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
