AdaptToken: Entropy-based Adaptive Token Selection for MLLM Long Video Understanding

TL;DR

AdaptToken is a novel, training-free framework that uses model uncertainty to adaptively select tokens in long videos, improving efficiency and accuracy for multi-modal large language models.

Contribution

It introduces a principled, entropy-based token selection method that enables global relevance comparison and early stopping in long-video understanding tasks.

Findings

AdaptToken improves accuracy by an average of 6.7 points over baseline models.

AdaptToken-Lite halves inference time with comparable performance.

The method scales effectively to inputs up to 10,000 frames.

Abstract

Long video understanding remains challenging for Multi-modal Large Language Models (MLLMs) due to high memory costs and context-length limits. Prior approaches mitigate this by scoring and selecting frames/tokens within short clips, but they lack a principled mechanism to (i) compare relevance across distant video clips and (ii) stop processing once sufficient evidence has been gathered. We propose AdaptToken, a training-free framework that turns an MLLM's self-uncertainty into a global control signal for long-video token selection. AdaptToken splits a video into groups, extracts cross-modal attention to rank tokens within each group, and uses the model's response entropy to estimate each group's prompt relevance. This entropy signal enables a global token budget allocation across groups and further supports early stopping (AdaptToken-Lite), skipping the remaining groups when the model becomes sufficiently certain. Across four long-video benchmarks (VideoMME, LongVideoBench, LVBench, and MLVU) and multiple base MLLMs (7B-72B), AdaptToken consistently improves accuracy (e.g., +6.7 on average over Qwen2.5-VL 7B) and continues to benefit from extremely long inputs (up to 10K frames), while AdaptToken-Lite reduces inference time by about half with comparable performance. Project page: https://haozheqi.github.io/adapt-token