LatentLens: Revealing Highly Interpretable Visual Tokens in LLMs

TL;DR

LatentLens is a novel interpretability method that reveals the semantic content of visual tokens in large vision-language models by mapping their representations to natural language descriptions, demonstrating high interpretability across models and layers.

Contribution

This work introduces LatentLens, a new approach for interpreting visual tokens in VLMs by leveraging nearest neighbor search in a large text corpus, revealing their semantic meanings.

Findings

Most visual tokens are interpretable across models and layers.

LatentLens provides more meaningful and fine-grained descriptions than previous methods.

Common interpretability methods underestimate visual token interpretability.

Abstract

Transforming a large language model (LLM) into a Vision-Language Model (VLM) can be achieved by mapping the visual tokens from a vision encoder into the embedding space of an LLM. Intriguingly, this mapping can be as simple as a shallow MLP transformation. To understand why LLMs can so readily process visual tokens, we need interpretability methods that reveal what is encoded in the visual token representations at every layer of LLM processing. In this work, we introduce LatentLens, a novel approach for mapping latent representations to descriptions in natural language. LatentLens works by encoding a large text corpus and storing contextualized token representations for each token in that corpus. Visual token representations are then compared to their contextualized textual representations, with the top-k nearest neighbor representations providing descriptions of the visual token. We evaluate this method on 10 different VLMs, showing that commonly used methods, such as LogitLens, substantially underestimate the interpretability of visual tokens. With LatentLens instead, the majority of visual tokens are interpretable across all studied models and all layers. Qualitatively, we show that the descriptions produced by LatentLens are semantically meaningful and provide more fine-grained interpretations for humans compared to individual tokens. More broadly, our findings contribute new evidence on the alignment between vision and language representations, opening up new directions for analyzing latent representations.