AdaTP: Attention-Debiased Token Pruning for Video Large Language Models

TL;DR

AdaTP introduces a novel attention-debiased token pruning method for Video LLMs, significantly reducing computational costs while maintaining performance by addressing inherent attention biases without additional training.

Contribution

The paper proposes AdaTP, a new token pruning pipeline with debiasing modules that effectively reduce computation in Video LLMs without performance loss.

Findings

Reduces FLOPs to 27.3% of the original model on LLaVA-OneVision-7B.

Maintains performance of vanilla models despite significant pruning.

Achieves state-of-the-art results on multiple video understanding benchmarks.

Abstract

Video Large Language Models (Video LLMs) have achieved remarkable results in video understanding tasks. However, they often suffer from heavy computational overhead due to the large number of visual tokens generated from multiple video frames. Existing visual token compression methods often rely on attention scores from language models as guidance. However, these scores exhibit inherent biases: global bias reflects a tendency to focus on the two ends of the visual token sequence, while local bias leads to an over-concentration on the same spatial positions across different frames. To address the issue of attention bias, we propose $\textbf{A}$ttention-$\textbf{D}$ebi$\textbf{a}$sed $\textbf{T}$oken $\textbf{P}$runing for Video Large Language Models ($\textbf{AdaTP}$), a novel token pruning pipeline for Video LLMs. AdaTP integrates two dedicated debiasing modules into the pipeline, targeting global attention bias and local attention bias, respectively. Without the need for additional training, our method significantly reduces the computational overhead of Video LLMs while retaining the performance of vanilla models. Extensive evaluation shows that AdaTP achieves state-of-the-art performance in various commonly used video understanding benchmarks. In particular, on LLaVA-OneVision-7B, AdaTP maintains performance without degradation while using only up to $27.3\%$ FLOPs compared to the vanilla model. Our code will be released soon.