Script: Graph-Structured and Query-Conditioned Semantic Token Pruning for Multimodal Large Language Models
This addresses efficiency issues for users of MLLMs handling high-resolution images and videos, representing a strong incremental improvement over existing pruning methods.
The paper tackles the problem of excessive memory and latency in multimodal large language models (MLLMs) due to visual tokens by proposing Script, a plug-and-play pruning method that achieves up to 6.8x prefill speedup and 10x FLOP reduction while retaining 96.88% of original performance on LLaVA-NeXT-7B.
The rapid growth of visual tokens in multimodal large language models (MLLMs) leads to excessive memory consumption and inference latency, especially when handling high-resolution images and videos. Token pruning is a technique used to mitigate this issue by removing redundancy, but existing methods often ignore relevance to the user query or suffer from the limitations of attention mechanisms, reducing their adaptability and effectiveness. To address these challenges, we propose Script, a plug-and-play pruning method that requires no retraining and generalizes across diverse MLLMs. Script comprises two modules: a graph-structured pruning module that removes visually redundant tokens, and a query-conditioned semantic pruning module that preserves query-relevant visual information. Together, they enhance performance on multimodal tasks. Experiments on fourteen benchmarks across image and video understanding tasks show that Script consistently achieves higher model efficiency and predictive accuracy compared to existing pruning methods. On LLaVA-NeXT-7B, it achieves up to 6.8x prefill speedup and 10x FLOP reduction, while retaining 96.88% of the original performance.