RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference

• URL: http://arxiv.org/abs/2605.00392v1
• Date: Fri, 01 May 2026 04:30:16 GMT
• Title: RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference
• Abstract summary: We propose a two-stage token pruning method tailored for DeepSeek-OCR.<n>In the first stage, we prioritize high-norm visual tokens that capture salient textual and structural information.<n>In the second stage, the remaining tokens are paired and merged based on optimal transport theory to achieve efficient feature aggregation.
• Score: 17.01369106080539
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: DeepSeek-OCR leverages visual-text compression to reduce long-text processing costs and accelerate inference, yet visual tokens remain prone to redundant textual and structural information. Moreover, current token pruning methods for conventional vision-language models (VLMs) fail to preserve textual fidelity due to improper compression mechanisms. By analyzing the decoding process of DeepSeek-OCR, we find that a distinct two-stage reading trajectory: the model initially prioritizes the majority of high-norm tokens, then subsequently redistributes its attention to the remaining ones. Motivated by this insight, we propose RTPrune, a two-stage token pruning method tailored for DeepSeek-OCR. In the first stage, we prioritize high-norm visual tokens that capture salient textual and structural information. In the second stage, the remaining tokens are paired and merged based on optimal transport theory to achieve efficient feature aggregation. We further introduce a dynamic pruning ratio that adapts to token similarity and textual density for OCR tasks, enabling a better efficiency-accuracy trade-off. Extensive experiments demonstrate state-of-the-art performance, as evidenced by 99.47% accuracy and 1.23$\times$ faster prefill on OmniDocBench, achieved with 84.25% token retention when applied to DeepSeek-OCR-Large.

Related papers

• STAR-Pro: Stage-Wise Token Adaptive Reduction with Progressive Refinement for Efficient Large Vision-Language Models [54.273344678561024]
  Large vision-language models (LVLMs) achieve strong multimodal understanding, but the hundreds to thousands of visual tokens they process impose computational overhead.<n>We conduct two complementary analyses of visual token pruning.<n>We propose STAR-Pro, a training-free two-stage framework for visual token pruning.
  arXiv  Detail & Related papers  (2026-09-05T06:18:28Z)
• FastOCR: Dynamic Visual Fixation via KV Cache Pruning for Efficient Document Parsing [51.905216364362325]
  We propose FastOCR, a training-free framework with two complementary modules.<n>FastOCR retains 98% of the unpruned model's accuracy while attending to only 5% of the visual tokens per decoding step.
  arXiv  Detail & Related papers  (2026-05-17T13:39:47Z)
• Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Models [7.304290212053938]
  This paper empirically revisits the necessity of granular token-level processing.<n>We introduce Affinity Pooling, a training-free, similarity-based token merging mechanism.<n>Our results challenge the necessity of fully distinct token representations, providing new perspectives on LSLM efficiency.
  arXiv  Detail & Related papers  (2026-04-08T09:33:44Z)
• Boosting Document Parsing Efficiency and Performance with Coarse-to-Fine Visual Processing [25.960557727211768]
  Document parsing is a fine-grained task where image resolution significantly impacts performance.<n>We propose PaddleOCR-VL, a novel coarse-to-fine architecture that focuses on semantically relevant regions while suppressing redundant ones.<n>We show that PaddleOCR-VL achieves state-of-the-art performance in both page-level parsing and element-level recognition.
  arXiv  Detail & Related papers  (2026-03-25T14:08:56Z)
• ForensicZip: More Tokens are Better but Not Necessary in Forensic Vision-Language Models [108.5042835056188]
  We introduce ForensicZip, a training-free framework that reformulates token compression from a forgery-driven perspective.<n>ForensicZip achieves $2.97times$ speedup and over 90% FLOPs reduction while maintaining state-of-the-art detection performance.
  arXiv  Detail & Related papers  (2026-03-12T17:30:49Z)
• Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity [54.95089105944234]
  Vision-language models (VLMs) face significant computational inefficiencies caused by excessive generation of visual tokens.<n>We propose PruneSID, a training-free Synergistic Importance-Diversity approach featuring a two-stage pipeline.<n>We demonstrate state-of-the-art performance, achieving 96.3% accuracy on LLaVA-1.5 with only 11.1% token retention, and 92.8% accuracy at extreme compression rates (5.6%) on LLaVA-NeXT.
  arXiv  Detail & Related papers  (2026-03-10T10:31:58Z)
• AgenticOCR: Parsing Only What You Need for Efficient Retrieval-Augmented Generation [35.07704681580893]
  We introduce AgenticOCR, a dynamic parsing paradigm that transforms optical character recognition (OCR) into a query-driven, on-demand extraction system.<n>By autonomously analyzing document layout in a "thinking with images" manner, AgenticOCR identifies and selectively recognizes regions of interest.<n>AgenticOCR has the potential to serve as the "third building block" of the visual document RAG stack.
  arXiv  Detail & Related papers  (2026-02-27T16:09:38Z)
• Focus-Scan-Refine: From Human Visual Perception to Efficient Visual Token Pruning [78.75062483648243]
  Vision-language models (VLMs) often generate massive visual tokens that greatly increase inference latency and memory footprint.<n>We propose Focus-Scan-Refine (FSR), a human-inspired, plug-and-play pruning framework that mimics how humans answer visual questions.<n>FSR consistently improves the accuracy-efficiency trade-off over existing state-of-the-art pruning methods.
  arXiv  Detail & Related papers  (2026-02-05T16:02:48Z)
• Visual Merit or Linguistic Crutch? A Close Look at DeepSeek-OCR [25.00433693229684]
  DeepSeek-OCR claims to decode text tokens exceeding ten times the input visual tokens.<n>We employ sentence-level and word-level semantic corruption to isolate the model's intrinsic OCR capabilities from its language priors.<n>We find that traditional pipeline OCR methods exhibit significantly higher robustness to such semantic perturbations than end-to-end methods.
  arXiv  Detail & Related papers  (2026-01-07T09:01:23Z)
• ToDRE: Visual Token Pruning via Diversity and Task Awareness for Efficient Large Vision-Language Models [59.47738955960352]
  ToDRE is a two-stage and training-free token compression framework.<n>It achieves superior performance by pruning tokens based on token Diversity and token-task RElevance.
  arXiv  Detail & Related papers  (2025-05-24T15:47:49Z)
• Activation-aware Probe-Query: Effective Key-Value Retrieval for Long-Context LLMs Inference [56.71209737306054]
  We propose textbfActQKV, a training-free, textbfActivation-aware approach that dynamically determines probe-textbfQuery and leverages it to retrieve the relevant textbfKV pairs for inference.<n>Experiments on the Long-Bench and $infty$ Benchmarks demonstrate its state-of-the-art performance with competitive inference quality and resource efficiency.
  arXiv  Detail & Related papers  (2025-02-19T08:50:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.