Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

ETC: Extreme Token Compression via Task-aware Visual Information Distillation in VLMs

About

In Vision-Language Models (VLMs), high-resolution images produce a large number of visual tokens, resulting in high computational costs and KV-cache overhead during inference. To address this problem, we propose an Extreme Token Compression (ETC) framework that minimizes task loss when reducing the number of input tokens based on the principle of variational information distillation. Specifically, from an information-theoretic perspective, we show that minimizing task loss requires the compact representation to preserve the instruction-aware sufficient statistic of the task-relevant visual information for prediction. In practice, ETC leverages text-to-image cross-attention to weight the original visual features to approximate the latent instruction-aware predictive statistic. Moreover, ETC introduces a variational information distillation, enabling the compact representation to preserve the essential information to recover this predictive statistic. Experiments on LLaVA-1.5-7B and Qwen3-VL-2B show that ETC remains effective even under single-token compression, substantially reducing KV-cache overhead while retaining strong task performance.

Yiling Gao, Hongchen Wei, Zhenzhong Chen• 2026

Related benchmarks

TaskDatasetResultRank
Visual Question AnsweringVQA v2
Accuracy95.33
1429
Multimodal EvaluationMME--
727
Multimodal UnderstandingSEED-Bench--
516
Referring Expression ComprehensionRefCOCO (testA)
Accuracy0.1873
346
Referring Expression ComprehensionRefCOCO (testB)
Accuracy37.72
213
Multi-modal EvaluationMME
MME Score1.84e+3
160
Multimodal ReasoningMMBench
MMBench Accuracy (en)82.75
61
Multimodal UnderstandingSEED
SEED Score56.66
35
Science Question AnsweringSQA
SQA Score68.76
22
Science Question AnsweringScienceQA
SQA Score85.48
19
Showing 10 of 12 rows

Other info

Follow for update