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.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Visual Question Answering | VQA v2 | Accuracy95.33 | 1429 | |
| Multimodal Evaluation | MME | -- | 727 | |
| Multimodal Understanding | SEED-Bench | -- | 516 | |
| Referring Expression Comprehension | RefCOCO (testA) | Accuracy0.1873 | 346 | |
| Referring Expression Comprehension | RefCOCO (testB) | Accuracy37.72 | 213 | |
| Multi-modal Evaluation | MME | MME Score1.84e+3 | 160 | |
| Multimodal Reasoning | MMBench | MMBench Accuracy (en)82.75 | 61 | |
| Multimodal Understanding | SEED | SEED Score56.66 | 35 | |
| Science Question Answering | SQA | SQA Score68.76 | 22 | |
| Science Question Answering | ScienceQA | SQA Score85.48 | 19 |