arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.00296cs.LG
收藏

超越Token探针:基于激活张量与ACT-ViT的幻觉检测

Beyond Token Probes: Hallucination Detection via Activation Tensors with ACT-ViT

Technion(技术离子大学);University of Groningen(格罗宁根大学);Nvidia Research(英伟达研究)

Guy Bar-Shalom, Fabrizio Frasca, Yaniv Galron, Yftah Ziser, Haggai Maron

AI总结:

针对传统探针分类器仅处理孤立层-token对、依赖特定大语言模型的局限,提出受视觉Transformer启发的ACT-ViT模型,将激活张量视作图像处理,支持多LLM训练,幻觉检测性能优于传统方法且具备良好的跨模型、跨数据集迁移能力。

摘要:

检测大语言模型生成文本中的幻觉对其安全部署至关重要。尽管探针分类器展现出应用前景,但它们仅针对孤立的层-token对运行,且具有大语言模型特异性,这限制了其有效性,也阻碍了跨大语言模型的应用。本文提出一种新方法以解决这些缺陷。我们利用激活数据在层×token两个轴上的天然序列结构,主张将完整的激活张量视作图像来处理。我们设计了ACT-ViT,这是一款受Vision Transformer(视觉Transformer)启发的模型,可高效有效地应用于激活张量,且支持同时基于多个大语言模型的数据进行训练。通过涵盖多种大语言模型与数据集的综合实验,我们证明ACT-ViT的性能始终优于传统探针技术,同时部署效率极高。具体而言,我们的架构从多LLM训练中获益显著,在未见过的数据集上实现了优异的零样本性能,且可通过微调有效迁移至新的大语言模型。完整代码可在https://github.com/BarSGuy/ACT-ViT获取。

英文摘要:

Detecting hallucinations in Large Language Model-generated text is crucial for their safe deployment. While probing classifiers show promise, they operate on isolated layer-token pairs and are LLM-specific, limiting their effectiveness and hindering cross-LLM applications. In this paper, we introduce a novel approach to address these shortcomings. We build on the natural sequential structure of activation data in both axes (layers $\times$ tokens) and advocate treating full activation tensors akin to images. We design ACT-ViT, a Vision Transformer-inspired model that can be effectively and efficiently applied to activation tensors and supports training on data from multiple LLMs simultaneously. Through comprehensive experiments encompassing diverse LLMs and datasets, we demonstrate that ACT-ViT consistently outperforms traditional probing techniques while remaining extremely efficient for deployment. In particular, we show that our architecture benefits substantially from multi-LLM training, achieves strong zero-shot performance on unseen datasets, and can be transferred effectively to new LLMs through fine-tuning. Full code is available at https://github.com/BarSGuy/ACT-ViT.

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。

↑