Understanding Information Storage and Transfer in Multi-modal Large Language Models
This work addresses the lack of understanding in information mechanisms for MLLMs, which is important for improving model interpretability and reliability in real-world applications, though it is incremental as it extends existing methods from LLMs to MLLMs.
The paper tackles the problem of understanding how Multi-modal Large Language Models (MLLMs) store and transfer information, particularly in factual visual question answering tasks, by extending causal information tracing to multi-modal settings and introducing a test-bed of 9.7K annotated visual questions. Key results show that MLLMs rely on earlier layers for information storage compared to LLMs, with a small subset of visual tokens transferring information, validated by a model-editing algorithm that corrects errors and inserts new information.
Understanding the mechanisms of information storage and transfer in Transformer-based models is important for driving model understanding progress. Recent work has studied these mechanisms for Large Language Models (LLMs), revealing insights on how information is stored in a model's parameters and how information flows to and from these parameters in response to specific prompts. However, these studies have not yet been extended to Multi-modal Large Language Models (MLLMs). Given their expanding capabilities and real-world use, we start by studying one aspect of these models -- how MLLMs process information in a factual visual question answering task. We use a constraint-based formulation which views a visual question as having a set of visual or textual constraints that the model's generated answer must satisfy to be correct (e.g. What movie directed by the director in this photo has won a Golden Globe?). Under this setting, we contribute i) a method that extends causal information tracing from pure language to the multi-modal setting, and ii) VQA-Constraints, a test-bed of 9.7K visual questions annotated with constraints. We use these tools to study two open-source MLLMs, LLaVa and multi-modal Phi-2. Our key findings show that these MLLMs rely on MLP and self-attention blocks in much earlier layers for information storage, compared to LLMs whose mid-layer MLPs are more important. We also show that a consistent small subset of visual tokens output by the vision encoder are responsible for transferring information from the image to these causal blocks. We validate these mechanisms by introducing MultEdit, a model-editing algorithm that can correct errors and insert new long-tailed information into MLLMs by targeting these causal blocks.