Recent multimodal large language models are computationally expensive because Transformers must process a large number of visual tokens. We present ReDiPrune, a training-free token pruning method applied before the vision-language projector, where visual features remain rich and discriminative. Unlike post-projection pruning methods that operate on compressed representations, ReDiPrune selects informative tokens directly from vision encoder outputs, preserving fine-grained spatial and semantic cues. Each token is scored by a lightweight rule that jointly consider text-conditioned relevance and max-min diversity, ensuring the selected tokens are both query-relevant and non-redundant. ReDiPrune is fully plug-and-play, requiring no retraining or architectural modifications, and can be seamlessly inserted between the encoder and projector. Across four video and five image benchmarks, it consistently improves the accuracy-efficiency trade-off. For example, on EgoSchema with LLaVA-NeXT-Video-7B, retaining only 15% of visual tokens yields a +2.0% absolute accuracy gain while reducing computation by more than $6\times$ in TFLOPs. Code is available at https://github.com/UA-CVML/ReDiPrune.

翻译：近年来，多模态大语言模型因Transformer需处理大量视觉令牌而面临高昂计算成本。我们提出ReDiPrune——一种在视觉-语言投影器前应用的无训练令牌剪枝方法，此时视觉特征仍保持丰富且具区分性。与处理压缩表征的后投影剪枝方法不同，ReDiPrune直接从视觉编码器输出中筛选信息性令牌，保留了细粒度的空间与语义线索。每个令牌通过轻量规则评分，该规则联合考量文本条件相关性与最大-最小多样性，确保所选令牌既与查询相关且无冗余。ReDiPrune完全即插即用，无需重新训练或修改架构，可无缝嵌入编码器与投影器之间。在四个视频基准与五个图像基准上，该方法持续提升了精度-效率权衡。例如，在基于LLaVA-NeXT-Video-7B的EgoSchema任务中，仅保留15%视觉令牌即可实现+2.0%的绝对准确率增益，同时将计算量（TFLOPs）降低超过6倍。代码发布于https://github.com/UA-CVML/ReDiPrune。