通过智能剪枝文档冗余内容，实现长文档问答的高效准确处理。
DocPrune:Efficient Document Question Answering via Background, Question, and Comprehension-aware Token Pruning
- 基于背景、问题和理解程度动态剪除无关文本块，不依赖训练
- 在M3DocRAG上使编码器吞吐提升3.0倍、解码器提升3.3倍
- 适合需要高速处理长文档的场景，如法律、医疗知识库问答
视觉语言模型在文档问答等多模态任务中表现优异，但文档图像包含大量背景信息而有效证据稀疏，导致计算资源浪费，尤其在长文档场景下。现有针对自然图像和视频的令牌压缩方法难以利用文档特有的结构稀疏性。为此，我们提出DocPrune——一种无需训练、渐进式的文档令牌剪枝框架，仅保留任务必需的令牌（如关键文本），剔除背景或与问题无关的冗余内容，并根据模型理解深度自动选择剪枝起始层。在M3DocRAG数据集上的实验表明，该方法使编码器吞吐提升3.0倍、解码器提升3.3倍，同时F1分数提高1.0，实现高精度与高效率兼顾。
原文摘要 · Abstract (English)
Recent advances in vision-language models have demonstrated remarkable performance across diverse multi-modal 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 inefficient consumption 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 and progressive 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 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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。