DocPrune：通过背景、问题和理解感知的词元剪枝实现高效文档问答
DocPrune: Efficient Document Question Answering via Background, Question, and Comprehension-aware Token Pruning
针对文档图像中背景冗余和证据稀疏的特点，提出免训练的词元剪枝方法，可复用于其他视觉语言模型的长文档理解任务，提升推理效率。
原文依据 we propose DOCPRUNE, a training-free document token pruning framework designed for efficient long document understanding.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期视觉语言模型的进展在多种多模态任务中展现出强大性能，包括利用文本、表格和图形中的结构化视觉线索进行文档问答。然而，与自然图像不同，文档图像包含大量背景且仅含稀疏的支持证据，导致大量计算资源浪费，尤其是对长文档而言。我们观察到，现有针对自然图像和视频的词元减少方法未能利用文档特有的结构稀疏性。为此，我们提出DOCPRUNE，一种免训练的文档词元剪枝框架，旨在实现高效的长文档理解。该方法仅保留任务所需的关键词元，同时移除不必要的词元，如背景或与问题无关的词元。此外，它根据模型的理解程度自动选择适当的层来启动词元剪枝。在M3DocRAG基准上的实验表明，DOCPRUNE在编码器和解码器上分别将吞吐量提升3.0倍和3.3倍，同时F1分数提高+1.0，在无需额外训练的情况下实现了更高的准确性和效率。
英文原文摘要 1239 字符
Recent advances in vision-language models have shown strong performance across diverse multimodal tasks, including document question answering that leverages structured visual cues from text, tables, and figures. However, unlike natural images, document images contain large backgrounds and only sparse supporting evidence, leading to the waste of substantial computational resources, especially for long documents. We observe that existing token reduction methods for natural images and videos fall short in utilizing the structural sparsity unique to documents. To address this, we propose DOCPRUNE, a training-free document token pruning framework designed for efficient long document understanding. The proposed method preserves only the essential tokens for the task while removing unnecessary ones, such as background or question-irrelevant tokens. Moreover, it automatically selects the appropriate layers to initiate token pruning based on the model's level of comprehension. Our experiments on the M3DocRAG benchmark show that DOCPRUNE improves throughput by 3.0x and 3.3x in the encoder and decoder, respectively, while boosting the F1 score by +1.0, achieving both higher accuracy and efficiency without any additional training.