arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.04819cs.CVcs.CL
收藏

语言模型内部的视觉表示

Visual Representations inside the Language Model

University of Washington(华盛顿大学);University of California Los Angeles(加州大学洛杉矶分校);Allen Institute for AI(人工智能研究院)

Benlin Liu, Amita Kamath, Madeleine Grunde-McLaughlin, Winson Han, Ranjay Krishna

AI总结:

本研究通过分析LLaVA-OneVision等MLM的视觉键值标记,发现值标记蕴含丰富感知信息,但键标记存在伪影,且33.3%的感知信息未输出,提出文本前缀可提升感知能力,为多模态模型可解释性提供新视角。

摘要:

尽管可解释性工作已经分析了VIT编码器和Transformer激活,我们仍不理解多模态语言模型(MLMs)为何在感知密集型任务上表现不佳。我们通过考察流行MLMs(LLaVA-OneVision、Qwen2.5-VL和Llama-3-LLaVA-NeXT)如何处理其视觉键值(key-value)标记,提供了一个未被充分研究的视角。我们首先研究视觉信息在语言模型中的流动,发现图像值(value)标记编码了足够的信息,可以在零样本情况下执行若干感知密集型任务:分割、语义对应、时间对应和指代表达检测。我们发现,虽然语言模型确实增强了从输入视觉编码投影中接收的视觉信息——我们揭示这与整体MLM感知能力相关——但在若干任务上,它包含的视觉信息少于未经MLM微调的等效视觉编码器(SigLIP)。此外,我们发现,在语言模型后期层中,与输入无关的图像键(key)标记对应的视觉信息包含伪影,这些伪影降低了整体MLM的感知能力。接下来,我们讨论控制语言模型中的视觉信息,表明在图像输入中添加文本前缀可改善视觉表示的感知能力。最后,我们揭示,如果语言模型能够更好地控制其视觉信息,它们的感知能力将显著提升;例如,在BLINK基准的Art Style问题中,33.3%的情况下,语言模型中存在的感知信息未呈现在输出中!我们的发现揭示了键值标记在多模态系统中的作用,为MLMs的更深层机制可解释性铺平了道路,并为其视觉编码器和语言模型组件的训练提出了新方向。

英文摘要:

Despite interpretability work analyzing VIT encoders and transformer activations, we don't yet understand why Multimodal Language Models (MLMs) struggle on perception-heavy tasks. We offer an under-studied perspective by examining how popular MLMs (LLaVA-OneVision, Qwen2.5-VL, and Llama-3-LLaVA-NeXT) process their visual key-value tokens. We first study the flow of visual information through the language model, finding that image value tokens encode sufficient information to perform several perception-heavy tasks zero-shot: segmentation, semantic correspondence, temporal correspondence, and referring expression detection. We find that while the language model does augment the visual information received from the projection of input visual encodings-which we reveal correlates with overall MLM perception capability-it contains less visual information on several tasks than the equivalent visual encoder (SigLIP) that has not undergone MLM finetuning. Further, we find that the visual information corresponding to input-agnostic image key tokens in later layers of language models contains artifacts which reduce perception capability of the overall MLM. Next, we discuss controlling visual information in the language model, showing that adding a text prefix to the image input improves perception capabilities of visual representations. Finally, we reveal that if language models were able to better control their visual information, their perception would significantly improve; e.g., in 33.3% of Art Style questions in the BLINK benchmark, perception information present in the language model is not surfaced to the output! Our findings reveal insights into the role of key-value tokens in multimodal systems, paving the way for deeper mechanistic interpretability of MLMs and suggesting new directions for training their visual encoder and language model components.

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。

↑