Towards Vision-Language Mechanistic Interpretability: A Causal Tracing Tool for BLIP
This work addresses the lack of interpretability tools for vision-language models, which is a problem for researchers in AI interpretability, though it is incremental as it adapts existing methods to a new domain.
The paper tackles the problem of adapting causal tracing tools from unimodal language models to multimodal vision-language models like BLIP, enabling the study of neural mechanisms in image-conditioned text generation, and demonstrates this on a visual question answering dataset, showing causal relevance of later layer representations for all tokens.
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.