Pith. sign in

REVIEW 7 cited by

Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2405.05803 v3 pith:DSLNLQLD submitted 2024-05-09 cs.CV cs.AI

classification cs.CVcs.AI
keywords tokensmllmslayersvisualattentioninferenceinformationlayer
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Multimodal large language models (MLLMs) demand considerable computations for inference due to the extensive parameters and the additional input tokens needed for visual information representation. Herein, we introduce Visual Tokens Withdrawal (VTW), a plug-and-play module to boost MLLMs for rapid inference. Our approach is inspired by two intriguing phenomena we have observed: (1) the attention sink phenomenon that is prevalent in LLMs also persists in MLLMs, suggesting that initial tokens and nearest tokens receive the majority of attention, while middle vision tokens garner minimal attention in deep layers; (2) the presence of information migration, which implies that visual information is transferred to subsequent text tokens within the first few layers of MLLMs. As per our findings, we conclude that vision tokens are unnecessary in the deep layers of MLLMs. Thus, we strategically withdraw them at a certain layer, enabling only text tokens to engage in subsequent layers. To pinpoint the ideal layer for VTW, we initially analyze a limited set of tiny datasets and choose the first layer that meets the Kullback-Leibler divergence criterion. Our VTW approach can cut computational overhead by over 40\% across diverse multimodal tasks while maintaining performance.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. SlimVLM: Sensitivity-aware Dynamic Structured Pruning with Adaptive Visual Token Selection for Efficient Vision-Language Models

    cs.CV 2026-08 conditional novelty 6.0 of 10

    Visual-token attention filtering improves structured pruning of vision-language models, keeping 94% of average benchmark accuracy after removing 20% of parameters.

  2. IKOD: Mitigating Visual Attention Degradation in Large Vision-Language Models

    cs.CV 2025-08 unverdicted novelty 6.0 of 10

    IKOD reduces hallucination in vision-language models by merging KV states to derive image-focused shorter-sequence logits and combining them with normal decoding, without training.

  3. Mitigating Information Loss under High Pruning Rates for Efficient Large Vision Language Models

    cs.CV 2025-08 conditional novelty 6.0 of 10

    ACCM recovers information lost in high-rate visual token pruning by generating a question-guided caption from discarded tokens and selecting the best candidate, improving pruned LVLM accuracy with fewer FLOPs.

  4. LLaVA-Scissor: Token Compression with Semantic Connected Components for Video LLMs

    cs.CV 2025-06 conditional novelty 6.0 of 10

    A training-free token compression method using semantic connected components in space and time keeps video understanding accuracy high even when retaining only 5-10% of visual tokens.

  5. Short-LVLM: Compressing and Accelerating Large Vision-Language Models by Pruning Redundant Layers

    cs.CV 2025-07 conditional novelty 5.0 of 10

    A training-free layer pruning framework for large vision-language models, combining token importance scoring with subspace-compensated weight projection, preserves most accuracy while speeding inference.

  6. Fast3D: Accelerating 3D Multi-modal Large Language Models for Efficient 3D Scene Understanding

    cs.CV 2025-07 conditional novelty 5.0 of 10

    Fast3D prunes up to 90% of object-centric visual tokens in 3D MLLMs while preserving about 96.8% of original benchmark performance, using a trained attention predictor and adaptive layer-wise pruning.

  7. GreedyPrune: Retenting Critical Visual Token Set for Large Vision Language Models

    cs.CV 2025-06 conditional novelty 5.0 of 10

    A training-free token pruning method that combines cosine-similarity saliency with greedy redundancy removal to preserve accuracy at high compression ratios.

Pith tools