Towards Vision-Language Mechanistic Interpretability: A Causal Tracing Tool for BLIP

TL;DR

This paper adapts a causal tracing tool from language models to vision-language models like BLIP, enabling mechanistic interpretability of image-conditioned text generation and providing an open-source tool for the community.

Contribution

The authors modify a unimodal causal tracing method for use with BLIP, a vision-language model, facilitating causal analysis in multimodal neural mechanisms.

Findings

Causal relevance of later layer representations for all tokens in visual question answering.

Successful adaptation of a unimodal causal tracing tool to a multimodal model.

Open-source release of the BLIP causal tracing tool for community use.

Abstract

Mechanistic interpretability seeks to understand the neural mechanisms that enable specific behaviors in Large Language Models (LLMs) by leveraging causality-based methods. While these approaches have identified neural circuits that copy spans of text, capture factual knowledge, and more, they remain unusable for multimodal models since adapting these tools to the vision-language domain requires considerable architectural changes. In this work, we adapt a unimodal causal tracing tool to BLIP to enable the study of the neural mechanisms underlying image-conditioned text generation. We demonstrate our approach on a visual question answering dataset, highlighting the causal relevance of later layer representations for all tokens. Furthermore, we release our BLIP causal tracing tool as open source to enable further experimentation in vision-language mechanistic interpretability by the community. Our code is available at https://github.com/vedantpalit/Towards-Vision-Language-Mechanistic-Interpretability.