# Multi-Stage Vision Token Dropping: Towards Efficient Multimodal Large Language Model

arXiv: https://arxiv.org/abs/2411.10803 · HTML: https://mlsanity.com/paper/2411.10803/

## Key facts

- Cited by 68 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 16 of its 49 references that are in the map.
- First posted on arXiv 16 Nov 2024.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)
- Efficient LLM inference and serving / [Visual token pruning in multimodal models](https://mlsanity.com/inference/visual-token-pruning-in-multimodal-models/)

## Builds on (most cited first, 16 of 16)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models](https://mlsanity.com/paper/2304.10592/) (2023)
- [Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond](https://mlsanity.com/paper/2308.12966/) (2023)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [Video-LLaVA: Learning United Visual Representation by Alignment Before Projection](https://mlsanity.com/paper/2311.10122/) (2023)
- [ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools](https://mlsanity.com/paper/2406.12793/) (2024)
- [Video-ChatGPT: Towards Detailed Video Understanding via Large Vision and Language Models](https://mlsanity.com/paper/2306.05424/) (2023)
- [Gemini: A Family of Highly Capable Multimodal Models](https://mlsanity.com/paper/2312.11805/) (2023)
- [VideoChat: Chat-Centric Video Understanding](https://mlsanity.com/paper/2305.06355/) (2023)
- [CogVLM: Visual Expert for Pretrained Language Models](https://mlsanity.com/paper/2311.03079/) (2023)
- [LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models](https://mlsanity.com/paper/2403.15388/) (2024)
- [SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference](https://mlsanity.com/paper/2410.04417/) (2024)
- [DeCo: Decoupling Token Compression from Semantic Abstraction in Multimodal Large Language Models](https://mlsanity.com/paper/2405.20985/) (2024)
- [HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models](https://mlsanity.com/paper/2408.10945/) (2024)
- [Sparse-Tuning: Adapting Vision Transformers with Efficient Fine-tuning and Inference](https://mlsanity.com/paper/2405.14700/) (2024)

## Built on by (most cited first, 30 of 68)

- [Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs](https://mlsanity.com/paper/2506.10967/) (2025)
- [EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models](https://mlsanity.com/paper/2506.10100/) (2025)
- [Don't Just Chase "Highlighted Tokens" in MLLMs: Revisiting Visual Holistic Context Retention](https://mlsanity.com/paper/2510.02912/) (2025)
- [VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models](https://mlsanity.com/paper/2505.22654/) (2025)
- [A Survey of Multimodal Retrieval-Augmented Generation](https://mlsanity.com/paper/2504.08748/) (2025)
- [Global Compression Commander: Plug-and-Play Inference Acceleration for High-Resolution Large Vision-Language Models](https://mlsanity.com/paper/2501.05179/) (2025)
- [FOLDER: Accelerating Multi-modal Large Language Models with Enhanced Performance](https://mlsanity.com/paper/2501.02430/) (2025)
- [Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models](https://mlsanity.com/paper/2505.21200/) (2025)
- [Shifting AI Efficiency From Model-Centric to Data-Centric Compression](https://mlsanity.com/paper/2505.19147/) (2025)
- [LLaVA-Scissor: Token Compression with Semantic Connected Components for Video LLMs](https://mlsanity.com/paper/2506.21862/) (2025)
- [A Survey of Token Compression for Efficient Multimodal Large Language Models](https://mlsanity.com/paper/2507.20198/) (2025)
- [TokenCarve: Information-Preserving Visual Token Compression in Multimodal Large Language Models](https://mlsanity.com/paper/2503.10501/) (2025)
- [Task-Oriented Feature Compression for Multimodal Understanding via Device-Edge Co-Inference](https://mlsanity.com/paper/2503.12926/) (2025)
- [HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit](https://mlsanity.com/paper/2602.23699/) (2026)
- [Prune2Drive: A Plug-and-Play Framework for Accelerating Vision-Language Models in Autonomous Driving](https://mlsanity.com/paper/2508.13305/) (2025)
- [Growing a Multi-head Twig via Distillation and Reinforcement Learning to Accelerate Large Vision-Language Models](https://mlsanity.com/paper/2503.14075/) (2025)
- [N"uwa: Mending the Spatial Integrity Torn by VLM Token Pruning](https://mlsanity.com/paper/2602.02951/) (2026)
- [Can Visual Input Be Compressed? A Visual Token Compression Benchmark for Large Multimodal Models](https://mlsanity.com/paper/2511.02650/) (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)
- [Adaptive Perception for Unified Visual Multi-modal Object Tracking](https://mlsanity.com/paper/2502.06583/) (2025)
- [Why 1 + 1 < 1 in Visual Token Pruning: Beyond Naive Integration via Multi-Objective Balanced Covering](https://mlsanity.com/paper/2505.10118/) (2025)
- [SwimVG: Step-wise Multimodal Fusion and Adaption for Visual Grounding](https://mlsanity.com/paper/2502.16786/) (2025)
- [SalQ-VLM: Fine-Grained Saliency-Guided Quantization for Vision-Language Models](https://mlsanity.com/paper/2508.03351/) (2025)
- [Beyond Intermediate States: Explaining Visual Redundancy through Language](https://mlsanity.com/paper/2503.20540/) (2025)
- [ToDRE: Effective Visual Token Pruning via Token Diversity and Task Relevance](https://mlsanity.com/paper/2505.18757/) (2025)
- [MergeMix: A Unified Augmentation Paradigm for Visual and Multi-Modal Understanding](https://mlsanity.com/paper/2510.23479/) (2025)
- [VLM-Pruner: Buffering for Spatial Sparsity in an Efficient VLM Centrifugal Token Pruning Paradigm](https://mlsanity.com/paper/2512.02700/) (2025)
- [On the Adversarial Robustness of Large Vision-Language Models under Visual Token Compression](https://mlsanity.com/paper/2601.21531/) (2026)
- [IPCV: Information-Preserving Compression for MLLM Visual Encoders](https://mlsanity.com/paper/2512.18747/) (2025)

## Abstract (arXiv)

The vision tokens in multimodal large language models usually exhibit significant spatial and temporal redundancy and take up most of the input tokens, which harms their inference efficiency. To solve this problem, some recent works were introduced to drop the unimportant tokens during inference where the importance of each token is decided only by the information in either the vision encoding stage or the prefilling stage. In this paper, we propose Multi-stage Token Dropping (MustDrop) to measure the importance of each token from the whole lifecycle, including the vision encoding stage, prefilling stage, and decoding stage. Concretely, in the visual encoding stage, MustDrop merges spatially adjacent tokens with high similarity, and establishes a key token set to retain the most vision-critical tokens, preventing them from being discarded in later stages. In the prefilling stage, MustDrop further compresses vision tokens by the guidance of text semantics, with a dual-attention filtering strategy. In the decoding stage, an output-aware cache policy is proposed to further reduce the size of the KV cache. By leveraging tailored strategies in the multi-stage process, MustDrop can more precisely recognize the important and redundant tokens, thus achieving an optimal balance between performance and efficiency. For instance, MustDrop reduces about 88.5% FLOPs on LLaVA with a compression ratio of 92.2% while maintaining comparable accuracy. Our codes are available at urlhttps://github.com/liuting20/MustDrop.