VPRune：免训练视觉 token 剪枝框架，高压缩下仍保 LVLM 精度
VPRune: Efficient Training-free Pre-LLM Visual Token Pruning
VPRune 不用训练就能剪掉 LVLM 的视觉 token，FastVLM-1.5B 上高压缩率也掉点少，做端侧部署可以看看。
arXiv 论文提出 VPRune，一个无需训练的 pre-LLM 视觉 token 剪枝框架，用于降低大型视觉语言模型（LVLM）的推理成本。作者归纳出激进剪枝掉点的三个原因：文本引导的选择偏差、被丢弃 token 的信息损失、序列压缩导致的位置失真。对应方案是三个模块：视觉多样性选择、相似度引导的 token 回收、位置保持恢复。在 FastVLM-1.5B 上的实验显示压缩率越高 VPRune 优势越明显，端侧设备测试中端到端推理延迟下降且任务表现保持领先。
VPRune: Efficient Training-free Pre-LLM Visual Token Pruning
Visual token pruning is a promising approach to reducing the inference cost of large vision-language models (LVLMs), yet aggressive token reduction often causes substantial performance degradation. We identify three key factors behind this degradation: text-guided selection bias, information loss from discarded tokens, and positional distortion caused by sequence compaction. Based on these observations, we propose \textbf{VPRune}, a training-free pre-LLM pruning framework consisting of visual-only diversity selection, similarity-guided token recycling, and position-preserving restoration. Experiments on FastVLM-1.5B across multiple vision-language benchmarks demonstrate that VPRune achieves a favorable accuracy--compression trade-off, with particularly pronounced advantages under aggressive compression. Furthermore, evaluations on edge-device show that VPRune effectively reduces end-to-end inference latency while maintaining superior task performance, demonstrating its practicality for resource-constrained LVLM deployment.
