# Attend, Transform, or Silence: Operator-Level Visual Skipping for Efficient Multimodal LLM Inference

arXiv: https://arxiv.org/abs/2606.31903 · HTML: https://mlsanity.com/paper/2606.31903/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 18 of its 26 references that are in the map.
- First posted on arXiv 30 Jun 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)

## Builds on (most cited first, 18 of 18)

- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://mlsanity.com/paper/2209.09513/) (2022)
- [Towards VQA Models That Can Read](https://mlsanity.com/paper/1904.08920/) (2019)
- [VizWiz Grand Challenge: Answering Visual Questions from Blind People](https://mlsanity.com/paper/1802.08218/) (2018)
- [A Diagram Is Worth A Dozen Images](https://mlsanity.com/paper/1603.07396/) (2016)
- [LLaVA-OneVision-1.5: Fully Open Framework for Democratized Multimodal Training](https://mlsanity.com/paper/2509.23661/) (2025)
- [Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference](https://mlsanity.com/paper/2405.05803/) (2024)
- [Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs](https://mlsanity.com/paper/2506.10967/) (2025)
- [Token Pruning in Multimodal Large Language Models: Are We Solving the Right Problem?](https://mlsanity.com/paper/2502.11501/) (2025)
- [N"uwa: Mending the Spatial Integrity Torn by VLM Token Pruning](https://mlsanity.com/paper/2602.02951/) (2026)
- [ShortV: Efficient Multimodal Large Language Models by Freezing Visual Tokens in Ineffective Layers](https://mlsanity.com/paper/2504.00502/) (2025)
- [When Token Pruning is Worse than Random: Understanding Visual Token Information in VLLMs](https://mlsanity.com/paper/2512.07580/) (2025)
- [ApET: Approximation-Error Guided Token Compression for Efficient VLMs](https://mlsanity.com/paper/2602.19870/) (2026)
- [Why and When Visual Token Pruning Fails? A Study on Relevant Visual Information Shift in MLLMs Decoding](https://mlsanity.com/paper/2604.12358/) (2026)
- [Large Vision-Language Models Get Lost in Attention](https://mlsanity.com/paper/2605.05668/) (2026)
- [Look Less, Reason More: Block-wise Attention Skipping for Efficient Multimodal LLMs](https://mlsanity.com/paper/2606.08511/) (2026)

## Built on by (most cited first, 1 of 1)

- [Just MLPs: Efficient Visual State Reconstruction for Multimodal Language Models](https://mlsanity.com/paper/2609.34972/) (2026)

## Abstract (arXiv)

Multimodal large language models (MLLMs) increasingly process long visual-token sequences, increasing the overall inference computation. Existing acceleration methods usually remove visual tokens or skip visual-token updates in entire layers, but these coarse strategies may discard fine-grained evidence or suppress useful operators together with redundant ones. In this paper, we study visual-token computation from an answer-observable perspective and find that late visual-token updates can remain large while having little effect on answer-token representations. Motivated by this answer-silent redundancy, we decompose each Transformer layer into attention and FFN operators and show that useful visual computation is often operator-dominant and layer-dependent. We propose an operator-level visual-token skipping framework that preserves the full visual-token sequence while selectively bypassing redundant attention, FFN, or both. Experiments across three MLLM architectures and 10 VQA benchmarks show that our method achieves strong efficiency-accuracy trade-offs, reducing 33.7% TFLOPs on Qwen3-VL while retaining 99.5% of the vanilla model performance.