RCP: Representation Consistency Pruner for Mitigating Distribution Shift in Large Vision-Language Models

• URL: http://arxiv.org/abs/2604.04972v1
• Date: Sat, 04 Apr 2026 13:31:45 GMT
• Title: RCP: Representation Consistency Pruner for Mitigating Distribution Shift in Large Vision-Language Models
• Abstract summary: Large Vision-Language Models (LVLMs) suffer from prohibitive inference costs due to the massive number of visual tokens processed by the language decoder.<n>Existing pruning methods often lead to significant performance degradation because the irreversible removal of visual tokens causes a distribution shift in the hidden states that deviates from the pre-trained full-token regime.<n>We propose Consistency Representation Pruner, which integrates cumulative visual token pruning with a delayed repair mechanism.
• Score: 24.3914653184824
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large Vision-Language Models (LVLMs) suffer from prohibitive inference costs due to the massive number of visual tokens processed by the language decoder. Existing pruning methods often lead to significant performance degradation because the irreversible removal of visual tokens causes a distribution shift in the hidden states that deviates from the pre-trained full-token regime. To address this, we propose Representation Consistency Pruner, which we refer to as RCP, as a novel framework that integrates cumulative visual token pruning with a delayed repair mechanism. Specifically, we introduce a cross-attention pruner that leverages the intrinsic attention of the LLM as a baseline to predict cumulative masks, ensuring consistent and monotonic token reduction across layers. To compensate for the resulting information loss, we design a delayed repair adapter denoted as DRA, which caches the essence of pruned tokens and applies FiLM-based modulation specifically to the answer generation tokens. We employ a repair loss to match the first and second-order statistics of the pruned representations with a full-token teacher. RCP is highly efficient because it trains only lightweight plug-in modules while allowing for physical token discarding at inference. Extensive experiments on LVLM benchmarks demonstrate that RCP removes up to 88.9\% of visual tokens and reduces FLOPs by up to 85.7\% with only a marginal average accuracy drop, and outperforms prior methods that avoid fine-tuning the original model on several widely used benchmarks.

Related papers

• Calibrate Before Reason: Robust Visual Token Reduction against Semantic Drift in VLMs [26.936352183867854]
  CaRe consists of two mutually complementary modules: 1) Perturbation-Robust Anchoring, which identifies anchors with stable model-side influence under multi-directional perturbations; 2) Confidence-Gated Token, which extracts reliable calibration signals from unselected tokens and injects them into anchors.<n>While pruning 94.4% of visual tokens, our method retains 96.4% of the original full-token performance, delivering up to 2.30 times faster end-to-end inference speed relative to unpruned vanilla models.
  arXiv  Detail & Related papers  (2026-07-30T05:36:29Z)
• Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learning [40.28153252530189]
  Large vision-language models can reason over multimodal inputs by generating textual chains of thought.<n>A key capability exhibited in CoT reasoning is self-reflection: revisiting earlier decisions and correcting previous errors.<n>Existing LVLMs often fail to properly attend to visual inputs during reflection, limiting their ability to translate feedback into grounded corrections.<n>We propose a novel reinforcement learning training framework VRRL, with two components explicitly designed to elicit visually grounded self-reflection.
  arXiv  Detail & Related papers  (2026-07-02T17:53:15Z)
• ERA: Entropy-Guided Visual Token Pruning with Rectified Attention for Efficient MLLMs [77.57757084153083]
  We propose ERA, an Entropy-guided visual token pruning framework with Rectified Attention for efficient MLLMs.<n> ERA comprises three crucial components: Dual-view Entropy Pruning (DEP), Bias-aware Token Recycling (BTR), and Logit-preserving Attention Rectification (LAR)
  arXiv  Detail & Related papers  (2026-06-30T17:20:29Z)
• Evading Visual Aphasia: Contrastive Adaptive Semantic Token Pruning for Vision-Language Models [52.78477729846771]
  We introduce COAST (COntrastive Adaptive Semantic Token Pruning), a training-free pruning framework that casts compression as adaptive semantic routing.<n> COAST consistently outperforms strong pruning baselines across token budgets and generalizes across multiple LVLM families.
  arXiv  Detail & Related papers  (2026-05-10T09:07:04Z)
• From Early Encoding to Late Suppression: Interpreting LLMs on Character Counting Tasks [49.57538588967748]
  Large language models (LLMs) exhibit failures on elementary symbolic tasks such as character counting in a word, despite excelling on complex benchmarks.<n>We uncover a consistent phenomenon across modern architectures, including LLaMA, Qwen, and Gemma.<n>We show that symbolic reasoning failures in LLMs are not due to missing representations or insufficient scale, but arise from structured interference within the model's computation graph.
  arXiv  Detail & Related papers  (2026-04-01T11:40:12Z)
• FrugalPrompt: Reducing Contextual Overhead in Large Language Models via Token Attribution [3.4666771782038652]
  Large language models (LLMs) owe much of their stellar performance to expansive input contexts, yet such verbosity inflates monetary costs, carbon footprint, and inference-time latency.<n>We introduce FrugalPrompt, a novel prompt compression framework for LLMs, which retains only the most semantically significant tokens.<n>We evaluate the approach across four NLP tasks: Sentiment Analysis, Commonsense QA, Summarization, and Mathematical Reasoning.
  arXiv  Detail & Related papers  (2025-10-18T10:22:13Z)
• Grounding-Aware Token Pruning: Recovering from Drastic Performance Drops in Visual Grounding Caused by Pruning [10.271284217509482]
  Multimodal Large Language Models (MLLMs) have demonstrated strong performance in visual grounding.<n>We observe that pruning significantly weakens the model's grounding ability, leading to incorrect predictions and drastic performance degradation.<n>We propose Grounding-Aware Token Pruning (GAP), a simple yet effective adjustment to position IDs that recovers REC accuracy back to 51.42%.
  arXiv  Detail & Related papers  (2025-06-27T03:11:22Z)
• SINDER: Repairing the Singular Defects of DINOv2 [61.98878352956125]
  Vision Transformer models trained on large-scale datasets often exhibit artifacts in the patch token they extract. We propose a novel fine-tuning smooth regularization that rectifies structural deficiencies using only a small dataset.
  arXiv  Detail & Related papers  (2024-07-23T20:34:23Z)
• Autoregressive Speech Synthesis without Vector Quantization [135.4776759536272]
  We present MELLE, a novel continuous-valued token based language modeling approach for text-to-speech synthesis (TTS)<n>MELLE autoregressively generates continuous mel-spectrogram frames directly from text condition.<n>MELLE mitigates robustness issues by avoiding the inherent flaws of sampling vector-quantized codes.
  arXiv  Detail & Related papers  (2024-07-11T14:36:53Z)
• TPC-ViT: Token Propagation Controller for Efficient Vision Transformer [6.341420717393898]
  Vision transformers (ViTs) have achieved promising results on a variety of Computer Vision tasks. Previous approaches that employ gradual token reduction to address this challenge assume that token redundancy in one layer implies redundancy in all the following layers. We propose a novel token propagation controller (TPC) that incorporates two different token-distributions.
  arXiv  Detail & Related papers  (2024-01-03T00:10:33Z)
• Iterative Token Evaluation and Refinement for Real-World Super-Resolution [77.74289677520508]
  Real-world image super-resolution (RWSR) is a long-standing problem as low-quality (LQ) images often have complex and unidentified degradations. We propose an Iterative Token Evaluation and Refinement framework for RWSR. We show that ITER is easier to train than Generative Adversarial Networks (GANs) and more efficient than continuous diffusion models.
  arXiv  Detail & Related papers  (2023-12-09T17:07:32Z)
• Learning to Mask and Permute Visual Tokens for Vision Transformer Pre-Training [55.12082817901671]
  We propose a new self-supervised pre-training approach, named Masked and Permuted Vision Transformer (MaPeT)<n>MaPeT employs autoregressive and permuted predictions to capture intra-patch dependencies.<n>Our results demonstrate that MaPeT achieves competitive performance on ImageNet, compared to baselines and competitors under the same model setting.
  arXiv  Detail & Related papers  (2023-06-12T18:12:19Z)

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.